social determinants of health case study examples

School Registration
Account Login
Your Wishlist

Social Determinants of Health Example Case Studies

by Shea Lunt | Dec 1, 2021 | Uncategorized | 0 comments

Social determinants of health (SDOH) are the environmental conditions where people are born, live, learn, work, play, worship, and age that affect a wide range of health, functioning, and quality-of-life risks and outcomes. Some examples include income level, transportation, air quality, and access to nutritious food and safe housing. There is emerging awareness that SDOH accounts for 50% of the health factors that ultimately determine health outcomes.

The current Office or Other Outpatient Evaluation and Management (E/M) guidelines allow the E/M of patient SDOH to be captured in the level of the office visit. “Diagnosis or treatment significantly limited by social determinates of health” is noted under the “Moderate Risk” element which correlates to a level 4 visit.

Remember, risk is only one element of medical decision making (MDM) and to support the level of service, two of the three elements must be met (problems, data, risk). The other option to support the level of service for office/clinic visits is to use the total time spent on the date of the encounter.

Let’s look at three examples:

Case Study 1: MDM

A young lady with a hip injury after a fall from a curb presents for initial evaluation. The provider determines that she should have an MRI scan, referral to orthopedics, and remain non-ambulatory. She is a waitress with no healthcare insurance and unable to afford an MRI. In addition, she declines a referral. She is requesting a note for work. The diagnosis is limited to clinical findings, which makes the management decisions more complicated.

MDM: Moderate
Problem: 1 undiagnosed new problem with uncertain diagnosis—Moderate
Risk: Diagnosis significantly limited by SDOH—Moderate
CPT Codes 99204/99214

Case Study 2: MDM

A 12-year-old young lady presents with her grandmother for anxiety and depression and risk for failing in school due to absences related to her mental health instability. She would benefit from medication management, but the parents are divorced and the mother refuses to allow treatment of the condition with medication.

Problem: 1 chronic condition with exacerbation—Moderate
Data: Assessment requires independent historian—Limited
Risk: Treatment significantly limited by SDOH—Moderate

Case Study 3: Time

A 68-year-old gentleman was recently seen in the emergency department for an uncomplicated laceration to his right hand. He presents for a wound check in which possible infection is noted. No additional testing or treatment is required. The provider learns that the patient cannot cleanse the wound at home because his water has been shut off.

Time is 60 minutes
Write letter to utilities for medical waiver of shut off
Discuss case with care manager to enroll patient in meals on wheels
Document applicable SDOH and related plan in EMR
CPT Code 99215

There are also ICD-10-CM diagnosis codes that correlate to SDOH. Z codes are a subset of ICD-10-CM diagnosis codes that represent factors influencing health status and contact with health services that may be recorded as diagnoses. ICD-10-CM codes Z55-Z65 identify SDOH. Utilizing these codes allows providers, hospitals, and health systems to better track patient needs and identify solutions to improve health of their communities.

ICD-10-CM SDOH Categories

Z55 – Problems related to education and literacy
Z56 – Problems related to employment and unemployment
Z57 – Occupational exposure to risk factors
Z58 – Problems related to physical environment (NEW for FY 2022!)
Z59 – Problems related to housing and economic circumstances
Z60 – Problems related to social environment
Z62 – Problems related to upbringing
Z63 – Other problems related to primary support group, including family circumstances
Z64 – Problems related to certain psychosocial circumstances
Z65 – Problems related to other psychosocial circumstances

Shea Lunt, RHIA, CPC, CPMA, PMP

Shea is a consultant for The Haugen Consulting Group with 11 years of health care industry experience. Shea has experience working on the professional fee side of coding, auditing, education and compliance serving coders and physicians.

She earned a bachelor’s degree in health information management and a master’s degree in health services administration from the University of Kansas. Shea is a Registered Health Information Administrator (RHIA), Certified Professional Coder (CPC), Certified Professional Medical Auditor (CPMA) and a Project Management Professional (PMP).

Shea, her husband, and their daughters, call the wide-open spaces of central Kansas home.

Social Determinants of Health: A Case Study

The term “social determinants of health” (SDOH) is inescapable in the healthcare industry. But despite the ubiquity of the term, integrating SDOH into front-line medical care remains largely out of reach.

Why? Because there persists a “ wicked problem ” in the U.S. healthcare system; an unfortunate disconnect between the actual problem – meeting the complex health and wellness needs of unique individuals – and our approach to that problem – a complicated and inflexible system doling out prescriptive responses to those needs.

Changing our approach and creating a system that values health seems nearly impossible, given the sheer size of the healthcare industry alone. But innovative programs around the country are showing that it is possible – and the results are mutual “wins” for individuals, point-of-care providers, payers and communities.

An extreme example of how SDOH significantly impact healthcare costs, homeless people who are un- or under-insured often forego preventive care and depend on the emergency room to manage major medical issues. Once their acute conditions are addressed, recovery is hindered by lack of stable housing and limited access to follow-up care, while substance abuse and mental health issues may further interfere. The result is a high rate of complications necessitating costly re-hospitalizations - with little or no improvement in overall health or quality of life in the end.

Seeing the inefficiency and meanness of this cycle, ShelterCare, a nonprofit human services organization in Eugene, Oregon, sought change. In collaboration with Community Health Centers of Lane County (a Federally Qualified Health Centers), Trillium Community Health Plan, the local coordinated care organization, and PeaceHealth Sacred Heart Medical Center, the local hospital, they developed a medical respite care program that meaningfully improves wellness while decreasing costs by integrating SDOH into the care of homeless individuals.

The ShelterCare Medical Recuperation program opened in 2013 and has grown to 22-beds. The 30-day program gives residents a safe, stable housing environment in which to recover, while a community health worker provides medical care coordination and an on-site case manager helps residents connect to community resources to help them regain long-term stability.

As an example, Phil, a homeless man, visits the hospital emergency department with an infected wound. After receiving immediate medical treatment, the hospital refers him to the ShelterCare program, where he is given a small apartment and three meals a day. The medical caregiver on site dispenses medications, checks dressings and arranges for follow-up appointments and transportation. She helps Phil establish care with a primary care provider, who identifies that he has an untreated mental health disorder and a substance abuse problem, and he is referred to both counseling and an addiction treatment program.

Meanwhile, a case manager advocates for Phil with other social service and government agencies, guiding him through the process of applying for Medicaid, Social Security, food stamps, rental assistance, unemployment benefits and other services for which he may qualify. He is given a bus pass and ShelterCare staff take him shopping for clothing and toiletries. He takes part in on-site training workshops that help him create a resume, apply for jobs and learn basic budgeting skills.

While at the end of his month-long stay Phil still has much work to do to regain full stability, he is one of the 80 percent of program participants who leaves the program to move into permanent housing. And, by addressing the social determinants of health rather than continuing the cycle of emergency department visits, Phil’s coordinated care and wraparound services cost 34 percent less while helping him – and by extension, the greater community - make progress toward a significantly better quality of life.

The ShelterCare Medical Recuperation program has saved its community $1.26 million in hospital costs alone since 2014. It is an excellent micro-example of how an adaptive network with shared purpose and responsibility – the health and wellness of individuals – can effectively, efficiently collaborate to manage complex problems. And while the ShelterCare program is focused specifically on those experiencing a predetermined set of conditions, (homelessness and the need for acute recovery assistance), it illustrates both the value and feasibility of an outcome-based, SDOH-integrated network approach to healthcare for all.

Restoring the type of cooperation, adaptiveness and humanity exemplified by the ShelterCare Medical Recuperation program to our greater healthcare delivery system will first require a major shift in how we value healthcare . We must value – and therefore monetize – outcomes, not procedures. And because health outcomes are, in reality, messily influenced by more than biology and physiology, healthcare delivery has to incorporate SDOH if we hope to contribute to meaningful improvements in community health.

This is one reason universal healthcare is not in itself an adequate solution: we will not solve the problem by simply offering more of the same. Today’s reimbursement methods take our focus away from individuals and their unique needs and imposes cumbersome restraints on those who would care for them. Instead, insurance companies, social service agencies, the government, and other stakeholders must come together to enable nimble, patient-focused decisions at the point-of-care.

A self-organized, learning network that engages providers and payers across disciplines will create more options for better outcomes at a lower cost. Curandi ’s networking platform will enable community-focused initiatives to grow and serve a wider swath of the population. But first we must be willing – like those involved in ShelterCare’s respite program – to set aside the entrenched but ineffective standards of healthcare delivery and commit to seeking innovative solutions.

About Curandi

Curandi is a platform that empowers local communities to deliver better, more affordable community health by prioritizing individualized care and using network science to address the social determinants of health. Learn more at https://curandi.org .

Current use

Current utilization of preventive services is about 60-70% of what is clinically suggested in PPACA. Find more at: http://www.healthcaretownhall.com/?p=5596

More like this

Opinion: oregon should battle the hostile takeover of its health care system, opinion: oregon lawmakers must include youth prevention in drug policy proposals, opinion: how to address homelessness in portland, letter to the editor: universal health plan governance board deserves support, understanding.

Get Involved

Securely submit an anonymous news tip or confidential document.

Support our journalism by making a tax-deductible gift.

2024 Conference Details
2024 NCEAS Conference Workshop Submission
2024 NCEAS Conference Abstract Submission
2024 conference scholarships
Community of Practice
Just in Time Research and Resource Round Up
Social Needs
SDOH A&E
Residency Curriculum
Student Learning
ICD-10 Z Codes
Covid, Racism & Med Ed
2019 NCEAS Conference Information
2020 NCEAS Conference Information
2021 NCEAS Conference Details
2022 NCEAS Annual Conference
2023 NCEAS Conference Submissions
2023 NCEAS Conference Details
2023 NCEAS Conference Scholarships

Social Determinants of Health Case Studies: Targeting the Social Determinants of Health in Geriatric Populations

Date of Review: February, 2019

This is one of 3 ‘case studies’ in a collection of teaching tools called Social Determinants of Health based on Healthy People (HP) 2020 material aimed at teaching Health Professional students to adopt a new perspective that identifies and includes the multitude of factors that affect the SDH of their patient’s communities and populations and incorporate this understanding into their work. This specific module introduces both SDH geriatric health content as well as how their interaction creates the need for addressing these factors in a coordinated and community based care model. The resource contains a 35 minute introductory video/PowerPoint describing the topic areas in detail. There are resources for students and facilitators and they suggest three separate 1.5-2 hour sessions or a single 4.5-5 hour session and can be easily integrated as a flipped classroom format. There are three clinical case videos with various geriatric patient scenarios. Students are each guided to focus on different aspects of care. Following the videos students discuss their findings with a panel of community experts and relevant government agencies. While convening such a panel is certainly a barrier to implementing this resource the available guide does provide detailed suggestions. Four structured reflection assignments are included with a final reflection paper which contains a detailed critical reflection assignment. Faculty experts in this topic area would be needed to facilitate and give feedback on these student deliverables in addition to having contacts with community agencies. While many teaching materials are available in this resource, one must become a member of APTR and have faculty credentials to download the faculty guides and evaluation rubric for the reflection assignment however I think there are enough resources included to easily implement this curriculum. — Ashti Doobay-Persaud, MD, NCEAS

Corresponding Author’s Email:

[email protected]

Institution:

University of Utah Health & THE NET

Where Was the Curriculum Implemented?

Salt Lake City, Utah

Source of the Curriculum/Resource:

Association for Prevention, Teaching and Research

Level of Learner Assessment:

Appreciation of content/attitude assessment (self-reflection, blogging with rubric)

Submit a Comment Cancel reply

Your email address will not be published. Required fields are marked *

(your message will only be visible after moderation)

Corresponding Author:

Cobb, Nadia

Type of Learner:

General Health Care Audience
Medical Students
Nursing Students
Physician Assistant students

SDOH Topic:

Access to Care
Community Medicine
General SDOH

Delivery/Education Method:

Community Partner
Online Module or Video

Curriculum Duration:

Multiple Sessions (less than a year)

Join our network and get current research and curriculum.

Social Determinants of Health

What are social determinants of health.

Social determinants of health (SDOH) are the conditions in the environments where people are born, live, learn, work, play, worship, and age that affect a wide range of health, functioning, and quality-of-life outcomes and risks.

SDOH can be grouped into 5 domains:

Suggested citation

Healthy People 2030, U.S. Department of Health and Human Services, Office of Disease Prevention and Health Promotion. Retrieved [date graphic was accessed], from https://health.gov/healthypeople/objectives-and-data/social-determinants-health

Social determinants of health (SDOH) have a major impact on people’s health, well-being, and quality of life. Examples of SDOH include:

Safe housing, transportation, and neighborhoods
Racism, discrimination, and violence
Education, job opportunities, and income
Access to nutritious foods and physical activity opportunities
Polluted air and water
Language and literacy skills

SDOH also contribute to wide health disparities and inequities. For example, people who don't have access to grocery stores with healthy foods are less likely to have good nutrition. That raises their risk of health conditions like heart disease, diabetes, and obesity — and even lowers life expectancy relative to people who do have access to healthy foods.

Just promoting healthy choices won't eliminate these and other health disparities. Instead, public health organizations and their partners in sectors like education, transportation, and housing need to take action to improve the conditions in people's environments.

That's why Healthy People 2030 has an increased and overarching focus on SDOH.

How Does Healthy People 2030 Address SDOH?

One of Healthy People 2030’s 5 overarching goals is specifically related to SDOH: “Create social, physical, and economic environments that promote attaining the full potential for health and well-being for all.”

In line with this goal, Healthy People 2030 features many objectives related to SDOH. These objectives highlight the importance of "upstream" factors — usually unrelated to health care delivery — in improving health and reducing health disparities.

More than a dozen workgroups made up of subject matter experts with different backgrounds and areas of expertise developed these objectives. One of these groups, the Social Determinants of Health Workgroup , focuses solely on SDOH.

Explore Research Related to SDOH

Social determinants of health affect nearly everyone in one way or another. Our literature summaries provide a snapshot of the latest research related to specific SDOH.

View SDOH Infographics

Each SDOH infographic represents a single example from each of the 5 domains of the social determinants of health. You can download them, print them, and share them with your networks.

Learn How SDOH Affect Older Adults

SDOH have a big impact on our chances of staying healthy as we age. Healthy People’s actionable scenarios highlight ways professionals can support older adults’ health and well-being.

The Office of Disease Prevention and Health Promotion (ODPHP) cannot attest to the accuracy of a non-federal website.

Linking to a non-federal website does not constitute an endorsement by ODPHP or any of its employees of the sponsors or the information and products presented on the website.

You will be subject to the destination website's privacy policy when you follow the link.

Teaching Social Determinants of Health Through an Unfolding Case Study

Affiliation.

1 Assistant Professor (Drs Hekel and Edwards), Associate Professor (Dr Pullis), and Instructor (Ms Alexander), Jane and Robert Cizik School of Nursing at The University of Texas Health Science Center at Houston.
PMID: 36729930
PMCID: PMC10144315
DOI: 10.1097/NNE.0000000000001333

Background: The impact of social determinants of health (SDOH) was developed to educate nursing students through the use of an unfolding case study.

Problem: SDOH and population health are critical components of prelicensure nursing education. Unfolding case studies are a strategy to develop critical thinking and teach SDOH to nursing students.

Approach: A model was used to develop the case study including a community assessment, which follows a male veteran and family through life events. Implementation of the unfolding case study took place over 3 consecutive semesters in a community health nursing course.

Outcomes: This educational activity achieved standardized examination scores, which are intended to assess student preparedness for the National Council Licensure Examination (NCLEX), above the national averages. Student participation was above 90%.

Conclusion: Unfolding case studies can present realistic scenarios that are useful to teach critical thinking. As the Next Generation NCLEX moves to scenario-based testing, unfolding case studies are a teaching strategy to prepare students.

Education, Nursing*
Education, Nursing, Baccalaureate*
Educational Measurement
Licensure, Nursing
Nursing Education Research
Social Determinants of Health
Students, Nursing*

Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

View all journals
My Account Login
Explore content
About the journal
Publish with us
Sign up for alerts
Open access
Published: 16 April 2024

Measuring social determinants of health in the All of Us Research Program

Samantha Tesfaye 1 , 2 ,
Robert M. Cronin 3 ,
Maria Lopez-Class 4 ,
Qingxia Chen 5 ,
Christopher S. Foster 4 ,
Callie A. Gu 6 ,
Andrew Guide 5 ,
Robert A. Hiatt 7 ,
Angelica S. Johnson 8 ,
Christine L. M. Joseph 9 ,
Parinda Khatri 10 ,
Sokny Lim 11 ,
Tamara R. Litwin 1 ,
Fatima A. Munoz 12 ,
Andrea H. Ramirez 11 ,
Heather Sansbury 13 , 16 ,
David G. Schlundt 14 ,
Emma N. Viera 15 ,
Elif Dede-Yildirim 11 &
Cheryl R. Clark 6

Scientific Reports volume 14 , Article number: 8815 ( 2024 ) Cite this article

163 Accesses

1 Altmetric

Metrics details

Epidemiology
Human behaviour
Risk factors

To accelerate medical breakthroughs, the All of Us Research Program aims to collect data from over one million participants. This report outlines processes used to construct the All of Us Social Determinants of Health (SDOH) survey and presents the psychometric characteristics of SDOH survey measures in All of Us . A consensus process was used to select SDOH measures, prioritizing concepts validated in diverse populations and other national cohort surveys. Survey item non-response was calculated, and Cronbach’s alpha was used to analyze psychometric properties of scales. Multivariable logistic regression models were used to examine associations between demographic categories and item non-response. Twenty-nine percent (N = 117,783) of eligible All of Us participants submitted SDOH survey data for these analyses. Most scales had less than 5% incalculable scores due to item non-response. Patterns of item non-response were seen by racial identity, educational attainment, income level, survey language, and age. Internal consistency reliability was greater than 0.80 for almost all scales and most demographic groups. The SDOH survey demonstrated good to excellent reliability across several measures and within multiple populations underrepresented in biomedical research. Bias due to survey non-response and item non-response will be monitored and addressed as the survey is fielded more completely.

COVIDiSTRESS diverse dataset on psychological and behavioural outcomes one year into the COVID-19 pandemic

Angélique M. Blackburn, Sara Vestergren & the COVIDiSTRESS II Consortium

social determinants of health case study examples

A philosophy of health: life as reality, health as a universal value

Julian M. Saad & James O. Prochaska

Assessing scale reliability in citizen science motivational research: lessons learned from two case studies in Uganda

Mercy Gloria Ashepet, Liesbet Vranken, … Liesbet Jacobs

Introduction

There is growing scientific consensus that social and environmental factors are significant contributors to health status, including the onset or progression of disease, recovery or response to treatment, and inequities in distributions of illnesses among populations 1 , 2 . Investigating the roles of social factors in health is an essential part of precision medicine research.

The All of Us Research Program ( All of Us ) aims to collect data from over one million participants to advance precision medicine and improve population health 3 . A central goal of All of Us is to ensure program participants represent the diversity of the US. The program seeks to build a diverse cohort that emphasized recruitment of populations underrepresented in biomedical research (UBR) along the lines of racial and ethnic identity, age, sex, gender identity, sexual orientation, disability status, access to care, income, educational attainment, and geographic location 3 , 4 . All of Us supports research that integrates information on genomics and other biologic data, data on lifestyle or behaviors, and contextual factors including the social determinants of health (SDOH) to facilitate precision medicine research. Data from All of Us participants are captured in multiple ways, including health surveys to obtain participant provided information, electronic health record data, physical measurements, and biospecimens. Details of the program’s recruitment efforts and data types have been reported previously 3 .

To collect data on social factors, All of Us launched a Task Force of subject matter experts within the All of Us Research Program to develop a survey on SDOH. SDOH concepts were selected to provide psychometrically rigorous data with strong use cases relevant to precision medicine. This technical report outlines the approach and process used to construct the SDOH survey. Additionally, this report presents psychometric characteristics of the measures included in the final SDOH survey, based on the most recently released data from 117,783 All of Us participants who submitted the survey by June 30, 2022.

Survey development process

In 2017, the SDOH Task Force launched for the purpose of developing a survey to gather self-reported data from All of Us participants that captures information on dimensions of SDOH using scientifically valid and reliable scales, while minimizing burden on participants to complete surveys.

A six-phase process was developed by the Task Force to select concepts that could be measured with validity and reliability via self-report in a large, diverse, and multilingual cohort. This process is summarized in Supplementary Table 1 . In brief, the Task Force employed a consensus based process to define SDOH and conceptual frameworks to (1) establish the inclusion and exclusion criteria, and priorities to guide construct selection; (2) evaluate relevant scientific literature to identify measures with strong psychometric properties and use cases in precision medicine; and (3) examine SDOH measures used in other biobanks and large epidemiologic studies to find opportunities to align measures with other resources, chiefly the UK Biobank, the Million Veteran Program, and the NIH PhenX Toolkit surveys related to SDOH 5 , 6 , 7 . Additionally, the Task Force coordinated internally to (4) avoid duplication of other All of Us surveys already deployed or in development (e.g., mental health, environmental/occupational exposures, and a survey offered to all participants as they join the program called “The Basics”) that assess concepts that overlap with or may be considered SDOH (e.g., income, education, employment, health literacy, home ownership, and risk of homelessness) 8 . The Task Force used conceptual frameworks to improve the user experience and eliminated lengthy scales that added to participant burden. Final measurement selections were completed with recommendations from scientific subject matter experts and All of Us participant partners. Importantly, inclusion and exclusion criteria for surveys prioritized measures of perceptions (including cognitions, beliefs, attitudes) that could only be obtained from the perspective of participants, and not otherwise collected through geocoding, electronic health records, or other means. Priority was given to selecting constructs and measures that are likely to represent core drivers and mechanisms connected to health inequities 9 , 10 .

The final survey included concepts and measures that had been previously validated and that had documentation on psychometric performance in large cohort studies. However, in rare instances, measures or items were included where extant literature is emerging when domain areas were considered essential for measurement, namely housing instability and housing quality 11 . Priority was given to measures that were scientifically validated in multiple languages and cultural contexts. In general, measures were used in the way they were validated with their original response sets. A notable exception was made for the questions on spirituality and religious service attendance, where a “non-religious” response option was added in keeping with strong recommendations from All of Us participants to acknowledge differing views on religion.

Conceptual frameworks and definitions

The SDOH Task Force chose the World Health Organization Conceptual Framework for Action on the Social Determinants of Health as an organizing scientific framework to identify constructs that could be operationalized to establish connections among SDOHs, their origins in structural social, economic, political, historical and cultural factors, measures of social identity and social stratification (income, education, race, ethnicity, sexual orientation and gender identity, disability) and relation to psychological, material resources/health-related social needs, and social connections that influence health or health care utilization 1 . Additionally, the Task Force sought to align with the Healthy People 2020 framework as described by the Centers for Disease Control and Prevention (CDC) Office of Disease Prevention and Health Promotion to identify five broad domains to refine the measurement approach: social and community context , economic stability , education , neighborhood and built environment , health and health care 12 . Of these domains, the Task Force identified educational attainment and health care access as domains that were, in part, already covered in other All of Us surveys.

To improve the user experience and ability to communicate well with large, diverse audiences about the relation of social factors to health, the Task Force was guided by the work of the Robert Wood Johnson Foundation Commission to Build a Healthier America, which developed language and concepts to communicate about social factors across political groups 13 .

Finally, the Task Force solicited guidance and recommendations from All of Us Participant Ambassadors on concepts and specific measures resonant with, or that might be less relevant to, lived experience of participants 14 . The Task Force also reviewed measures with scientific subject matter experts who were the developers of measurement tools under consideration, as well as subject matter experts from institutes and centers of the National Institutes of Health, who recommended additional concepts and measures for consideration.

Using these three frameworks and recommendations, the Task Force adopted the following as the guiding definition for the survey:

“Social determinants of health are the conditions and context in which people are born, live, learn, play, work, and worship across the lifespan that influence quality of life.”

Cognitive and online pilot testing

Candidate measures that were considered for inclusion in the survey underwent cognitive interviewing and pilot testing prior to finalizing final instruments and items.

Prior to launching the SDOH survey, a pilot study was conducted from January 2020—February 2020 and February 2021—April 2021. The methods of the pilot study included cognitive interviews and online testing, similar to the methods described in the development of the baseline surveys for All of Us 15 . The pilot study included qualitative and quantitative assessment of the SDOH survey in English and Spanish versions, the two languages that are currently used in All of Us . The pilot study recruited individuals aged 18 and older with a focus on ensuring the sample included the perspectives of UBR populations, particularly those underrepresented due to racial and ethnic identity, income, and preferred language. The Pilot Research Core’s Expression of Interest Registry, Latino Center of the Midlands in Omaha, Nebraska, and Cint ( https://www.cint.com/ ), an online survey audience platform, served as the recruitment methods for reaching key populations to test the survey.

Survey modifications after testing

The cognitive interviews provided feedback on the content and the processes for fielding the survey, which was incorporated in the final SDOH survey (Supplementary Table 2 ). Content modifications that were implemented included eliminating some concepts from the draft survey. Process recommendations for fielding the survey that were implemented included providing materials so that participants who indicated social needs via survey could be made aware of relevant resources for the issues they reported. Online pilot testing of the survey demonstrated completion times that were in the range of 10–15 min. Pilot completion times were similar among different demographic and language groups (Spanish and English).

Survey approval and release

The final SDOH survey was approved by the All of Us Institutional Review Board (IRB) and launched on November 1, 2021. The SDOH concepts approved in the final survey are listed in Table 1 . The SDOH survey is optional and is available in English and Spanish language versions to all participants who have completed the first three All of Us surveys (The Basics, Overall Health, and Lifestyle). Participants select the language in which they take program surveys at the time of program enrollment.

Accessing data from the SDOH survey

The SDOH survey is publicly available online 33 . The data can be found in the All of Us Researcher Workbench at https://www.researchallofus.org/data-tools/workbench/ . The SDOH survey questions, responses with answer concept ids, number, and percentages of participants who selected each response, along with bar charts showing the number of participants who chose each answer by the sex assigned at birth and age when the survey was taken, are publicly available via the All of Us Data Browser at https://databrowser.researchallofus.org/ .

Construction of scales

We constructed scales and scored items in alignment with validated usage in the literature. Two exceptions include the presentation of the Religious Service Attendance item and Daily Spiritual Experiences scale, which incorporated All of Us Participant Ambassador feedback by adding response set options for participants to indicate “I am not religious,” or to indicate “a higher power,” as an alternative conceptualization of spiritual experience. Due to a transcription error for the Religious Service Attendance measure, an incorrect response set was displayed for 11,795 participants; instructions to identify these observations are described in the Supplementary Methods . A detailed scoring recommendation for all measures is included in the Supplementary Methods .

Statistical analyses

Characteristics of All of Us participants who responded to the SDOH survey, and those who were eligible but did not submit responses to the SDOH survey, were summarized in means and standard deviations (SD) (continuous variables) or counts and percentages (categorical variables). Among those who responded to the SDOH survey, distributions of the constructed scales were reported in mean, SD, range, median, and interquartile range (IQR). Item non-response was examined by demographic category to understand potential bias in missing data. Item non-response was defined as insufficient items completed by a participant to calculate a scale. In most cases, a scale could not be calculated with more than one or two items not completed; item non-response definitions for each scale are provided in the Supplementary Methods . For each scale, Cronbach's alpha coefficients were used to examine internal consistency overall and by demographic category.

Multivariable logistic regression models were used to examine if demographic categories were associated with item non-response. The predictors in the model included racial identity, Hispanic/Latino/Spanish ethnicity, sex assigned at birth, gender identity, sexual orientation, educational attainment, annual household income, disability status, and current age. Age was modeled with a restricted cubic spline function with three knots. Each model is displayed visually as a forest plot to compare the odds ratios and corresponding confidence intervals (CIs). For each plot, the odds of an individual of a particular demographic category (racial identity, sexual orientation, etc.) not responding to the survey scale of interest are compared to the odds of non-response for an indicated reference population within that variable; this yields an odds ratio of item non-response comparing the two groups. Since age was treated non-linearly, two odds ratios were calculated for age: people aged 25 and 75 were compared to a reference group of age 50 for illustration purposes.

The ‘pandas’, and ‘numpy’ Python packages were used for data cleaning, and the ‘pingouin’ package was used to calculate Cronbach’s alpha for the SDOH scales. The R (version 4.1.1) kernel within the Jupyter environment was used for the remaining analysis. In particular, the 95% CIs for proportion differences were constructed using prop.test function in the ‘stats’ (version 4.2.0) package, regression models were performed using the ‘rms’ (version 6.6) package, and forest plots were generated by the ‘metafor’ (version 4.0) packages.

The current analyses were conducted as part of a demonstration project designed to describe All of Us cohort data in preparation for releasing data to the All of Us Researcher Workbench. The data described in this report are from the All of Us Researcher Workbench version 7 released to the Researcher Workbench in April 2023. The work described here was proposed by Consortium members and confirmed as meeting criteria for non-human subjects research by the All of Us IRB. Results reported are in compliance with the All of Us Data and Statistics Dissemination Policy disallowing disclosure of group counts under 20.

SDOH survey questions

The constructs and measures in the final fielded SDOH survey are listed in Table 1 .

Description of survey participants

A total of 397,732 participants were eligible to complete the SDOH survey, of which 332,986 (83.7%) were UBR. As of June 30, 2022, a total of 117,783 (29.6%) participants provided any SDOH survey data, of which 92,300 (78.4%) were UBR.

There were differences in the characteristics of those who answered any portion of the survey (survey respondents) and those who did not answer any portion of the survey (non-respondents, Table 2 ). Survey respondents were predominantly White as compared to non-respondents (74.6% vs. 44.8%). More respondents had a college or advanced degree than non-respondents (62.5% vs. 36.0%). Compared to survey respondents, non-respondents had lower incomes ($50,000 or less per year, 27.8% vs. 45.0%).

Internal consistency reliability of scales

Internal consistency measured by Cronbach’s alpha was over 0.8 for almost all scales (Table 3 ). The multi-dimensional Physical Activity and Neighborhood Environment (PANES) Walking and Bicycling scale had the lowest Cronbach’s alpha at 0.78. Internal consistency did not vary substantially by participant characteristics for most scales (Supplementary Table 4 ). However, for participants who identify as Native Hawaiian and Pacific Islander (N = 57), the PANES Walking and Bicycling scale had a much lower Cronbach’s alpha (0.58) compared to that of other groups where alpha coefficients for the Walking and Bicycling scale ranged from 0.70 to 0.79. Estimates of internal consistency on other scales for Native Hawaiian Pacific Islanders ranged from 0.78 to 0.95.

Item non-response and scale score distributions among survey respondents

Item non-response was infrequent. Ten of the 13 SDOH survey scales and sub-scales had fewer than 5% missing data due to item non-response (Table 3 ). The PANES Crime and Safety scale had the highest proportion of item non-response at 13.0%. The Cohen Perceived Stress Scale, the Neighborhood Social Disorder Scale, and the single housing quality item also had greater than 5% item non-response at 6.1%, 5.3%, and 5.2% respectively. Score distributions for scales are described in Table 3 . Item non-response for the single Religious Service Attendance item was 1.5% (N = 1825). A single-item indicator from the PANES instrument for neighborhood residential density/housing type had 2.1% item non-response (N = 2475). Item non-response for the categorical measures was 1.3% (N = 1542) for food insecurity, 3.1% (N = 3620) for housing instability, and 5.2% for housing quality problems (N = 6134). Among survey respondents, 13.5% had food insecurity, 2.7% had housing instability assessed as multiple address changes in 12 months, and 21.1% had housing quality problems such as bug infestation, mold, or lead pipes.

Item non-response varied most by educational attainment (Fig. 1 ), racial identity (Fig. 2 ), and survey language (Fig. 2 ). Using the Loneliness scale as an example, participants with less than a high school degree or equivalent had 8.6% item non-response for the Loneliness scale compared to 1.9% among participants with a college or advanced degree. Black, African, and African American participants and Hispanic/Latino/Spanish participants had 5.1% and 5.7% item non-response for the Loneliness scale, respectively, compared to 2.0% item non-response among White participants. Participants who took the SDOH survey in Spanish had 10.6% item non-response for the Loneliness scale compared to 2.4% among those who took it in English. Similar patterns were observed for scales with larger item non-response. For example, In the PANES Crime and Safety 2-item scale, participants with less than a high school degree or equivalent had 21.3% item non-response compared to 11.8% among participants with a college or advanced degree. Black, African, and African American participants and Hispanic/Latino/Spanish participants had 18.0% and 17.1% item non-response respectively compared to 12.1% among White participants. Participants who took the survey in Spanish had 25.6% item non-response for PANES Crime and Safety compared to 12.7% item non-response among those who took it in English. A description of absolute differences in item non-response for each scale by participant characteristics is provided in Supplementary Table 4 .

Percentage of participants with incalculable scores due to item non-response by educational attainment. Abbreviations: PANES, Physical Activity and Neighborhood Environment Scale; PNA, prefer not to answer. Data represent the percentage of SDOH survey respondents with an incalculable scale due to item non-response, by educational attainment. Participants who responded to the incorrect response set for the Religious Service Attendance item (N = 11,795) are flagged as ‘invalid’ in version 7 data; these respondents are not included in item non-response calculations.

Percentage of participants with incalculable scores due to item non-response by Racial Identity and language (English and Spanish). Abbreviations: PANES, Physical Activity and Neighborhood Environment Scale; MENA, Middle Eastern or North African; NHPI, Native Hawaiian or Pacific Islander; PNA, prefer not to answer. Black: Black, African or African-American; Hispanic: Hispanic/Latino/Spanish. Data shaded in gray are suppressed due to small sample size < 20. Data represent the percentage of SDOH survey respondents with an incalculable scale due to item non-response, by Racial Identity and Language group. Participants who responded to the incorrect response set for the Religious Service Attendance item (N = 11,795) are flagged as ‘invalid’ in version 7 data; these respondents are not included in item non-response calculations.

Multivariable logistic regression models

The odds of item non-response for any scale predicted by multivariable logistic regression are shown in Fig. 3 . Black, African, or African-American (OR 1.55, 95% CI 1.48–1.63) and Hispanic/Latino/Spanish (OR 1.59, 95% CI 1.51–1.68) participants had higher item non-response compared to White participants. Item non-response was higher amongst those with lower educational attainment compared to those with a college or advanced degree. Item non-response was also lower for those with higher income levels compared to those making less than $50,000 per year. Compared to an individual aged 50, participants aged 25 had less item non-response (OR 0.56, 95% CI 0.54–0.59) and participants aged 75 had greater item non-response (OR 2.60, 95% CI 2.53–2.67). Bisexual (OR 0.86, 95% CI 0.80–0.93) participants had lower odds of item non-response for any scale than straight participants.

Odds of item non-response or incalculable score for any SDOH scale. Abbreviations: PNA, prefer not to answer.

Several patterns stood out across all models (Supplementary Figure 5 ). For most scales, patterns in item non-response were seen by racial identity, educational attainment, income level, and age. As an example, in the Perceived Stress scale, Black, African or African-American participants (OR 1.78, 95% CI 1.64–1.93) and Hispanic/Latino/Spanish participants (OR 1.70, 95% CI 1.55–1.87) had a higher odds of item non-response compared to White participants; participants with less than a high school education were 1.79 (95% CI 1.55–2.06) times more likely to have item non-response compared to participants with a college or advanced degree; participants with an income level of over $150,000 per year were 0.73 (95% CI 0.67–0.80) times as likely to have item non-response compared to participants with an income level of less than $50,000 per year; and participants aged 25 years old were 0.60 (95% CI 0.54–0.66) times as likely and participants aged 75 years old were 2.71 (95% CI 2.59–2.84) times more likely to have item non-response compared to an individual aged 50 years old.

The food insecurity item, housing instability item, PANES Crime and Safety items, and housing quality item had atypical patterns in item non-response compared to other scales. For food insecurity, both a typical 25-year-old participant (OR 1.34, 95% CI 1.15–1.57) and a typical 75-year-old (OR 1.56, 95% CI 1.41–1.73) participant had higher odds of item non-response when compared to a 50-year-old participant. In housing instability, all racial identities other than the “prefer not to answer or skip" group had significantly higher odds of item non-response than White participants. Transgender and non-binary participants had lower odds of item non-response (OR 0.40, 95% CI 0.17–0.93) than those who identified as a woman. Both bisexual (OR 0.74, 95% CI 0.58–0.95) and gay (OR 0.66, 95% CI 0.50–0.85) participants had lower odds of item non-response than straight participants. However, those who did not answer the sexual orientation question had higher odds of item non-response missingness (OR 1.33, 95% CI 1.09–1.63) than straight respondents. For the PANES Crime and Safety items, all non-White racial identity groups had higher odds of item non-response than White participants. Typically, participants living with at least one identified disability had slightly higher odds of item non-response than those without identified disabilities, but in the housing quality item, participants living with a disability had lower odds of item non-response than those without identified disabilities (OR 0.86, 95% CI 0.79–0.94).

The SDOH survey demonstrated good to excellent internal consistency in measurement of several SDOH concepts and within multiple diverse population groups, including those who are underrepresented in biomedical research. Among those who submitted the survey, data collection was fairly complete with low item non-response (0.9% to 13.5% missing) for most scales. Item non-response varied by SDOH measure and by demographic categories, notably, with differences in item non-response by racial identity, educational attainment, and survey language. Taken together, the relatively complete data among survey respondents and the internal consistency of scales are promising for the use of the survey to better understand the role of the measured SDOHs in precision medicine research. Importantly, however, patterns of survey nonresponse and item non-response are documented in this report to assist the researcher with managing bias due to differential survey nonresponse patterns in the current version 7 release. We note that few participants (2.2%) completed the survey in Spanish, and additional data are needed to confirm reliability and generalizability for administering the survey in Spanish.

The SDOH survey has limitations. The survey was designed to elicit participants’ perceptions of their social environments, psychosocial connections and conditions of their daily lives that would be applicable to multiple research studies and across several complex models of disease or health promotion. The survey does not measure or explore structural social determinants of health inequities that detail “the mechanisms through which social hierarchies and social conditions are created” 1 . For example, the survey sought to provide measures of perceived discrimination, but was not designed to gather hypothesis-driven information from participants on structural racism or policies that lead to participants’ experiences of unequal treatment. To facilitate further assessment of structural social determinants of health, the All of Us Research Program will conduct geocoding, and plan future assessment of area-level and other contextual measures to provide insight into macro-level social conditions of relevance to precision medicine. Additionally, the expansive scope of social experiences that participants may perceive as important to the condition of their lives is considerable, and all relevant concepts could not feasibly be included in one survey. For example, the Task Force deferred capturing more detailed measures of wealth and acculturation, for potential future assessment in dedicated data collection efforts. We note that other All of Us surveys currently measure income, home ownership, education, employment status, health literacy, and other topics related to SDOH, which are also available in the version 7 release. Continued engagement with All of Us participants and scientists is warranted to develop the next set of surveys that deepen scientific understanding of social concepts that influence health.

While data on internal consistency and data completeness are presented in this report, additional tests of psychometric properties of scales in the All of Us cohort should continue to be assessed in hypothesis-driven research. Some surveys, including PANES, assess multiple constructs and may perform differently in hypothesis-driven research. Importantly, the SDOH survey remains in the field, and as of January 2024, 43.8% of those eligible have completed this survey. As the survey is fielded more completely, metrics on survey non-response and updates on survey performance must continue to be monitored and evaluated to gauge internal reliability and external generalizability of SDOH data.

In conclusion, the All of Us SDOH survey, developed through engagement with scientists and All of Us participant partners, has items and scales with strong psychometric properties that measure social experiences among a large, diverse participant population for the purpose of advancing precision medicine research. Additional work is needed to investigate the construct validity of some social concepts by geography and within specific groups. For participants who took the survey in Spanish, and for Native Hawaiian and Pacific Islander groups, scale internal consistency reliability may have varied compared to other groups with larger sample sizes. Future surveys should add to the breadth of concepts that can be explored in All of Us , and where available, researchers can compare All of Us data with other cohorts to enhance our understanding of social experiences relevant to precision medicine research.

Data availability

The SDOH survey is publicly available via the All of Us Survey Explorer at https://www.researchallofus.org/data-tools/survey-explorer/ . The data can be found in the All of Us Researcher Workbench at https://www.researchallofus.org/data-tools/workbench/ . The SDOH survey questions, responses with answer concept ids, number, and percentages of participants who selected each response, along with bar charts showing the number of participants who chose each answer by the sex assigned at birth and age when the survey was taken, are publicly available via the All of Us Data Browser at https://databrowser.researchallofus.org/ .

Solar O, Irwin A. A Conceptual Framework for Action on the Social Determinants of Health (World Health Organization, 2010). https://apps.who.int/iris/handle/10665/44489 . Accessed 9 March 2023.

Braveman, P., Egerter, S. & Williams, D. R. The social determinants of health: Coming of age. Annu. Rev Public Health. 32 (1), 381–398. https://doi.org/10.1146/annurev-publhealth-031210-101218 (2011).

Article PubMed Google Scholar

Denny, J. et al. The “ All of Us ” Research Program. N. Engl. J. Med. 381 (7), 668–676. https://doi.org/10.1056/NEJMsr1809937 (2019).

Mapes, B. M. et al. Diversity and inclusion for the All of Us research program: A scoping review. PLOS ONE 15 (7), e0234962. https://doi.org/10.1371/journal.pone.0234962 (2020).

Article CAS PubMed PubMed Central Google Scholar

Fry, A. et al. Comparison of sociodemographic and health-related characteristics of UK biobank participants with those of the general population. Am. J. Epidemiol. 186 (9), 1026–1034. https://doi.org/10.1093/aje/kwx246 (2017).

Article PubMed PubMed Central Google Scholar

Gaziano, J. M. et al. Million veteran program: A mega-biobank to study genetic influences on health and disease. J. Clin. Epidemiol. 70 , 214–223. https://doi.org/10.1016/j.jclinepi.2015.09.016 (2016).

Cox, L. A. et al. Using the PhenX toolkit to select standard measurement protocols for your research study. Curr. Protoc. 1 (5), e149. https://doi.org/10.1002/cpz1.149 (2021).

All of Us Research Hub. Researcher Workbench . https://www.researchallofus.org/data-tools/workbench/ . Accessed 10 March 2023.

Braveman, P. A. et al. Health Disparities and health equity: The issue is justice. Am. J. Public Health 101 (Suppl 1), S149–S155. https://doi.org/10.2105/AJPH.2010.300062 (2011).

National Academies of Sciences, Engineering, and Medicine, Health and Medicine Division, Board on Population Health and Public Health Practice Committee on Community-based solutions to promote health equity in the United States. In Communities in Action: Pathways to Health Equity (eds Baciu, A. et al. ) (National Academies Press, 2017).

Google Scholar

Lewis, C. C. et al. Comparing the performance of two social risk screening tools in a vulnerable subpopulation. J. Fam. Med. Prim. Care 9 (9), 5026–5034. https://doi.org/10.4103/jfmpc.jfmpc_650_20 (2020).

Article Google Scholar

Healthy People 2020. Office of Disease Prevention and Health Promotion, U.S. Department of Health and Human Services. Social Determinants of Health. HealthyPeople.gov . 2022. https://wayback.archive-it.org/5774/20220413203948/ . https://www.healthypeople.gov/2020/topics-objectives/topic/social-determinants-of-health#top . Accessed 10 March 2023.

Robert Wood Johnson Foundation (RWJF). Commission to Build a Healthier America . 2009. http://www.commissiononhealth.org/AboutUs.aspx . Accessed 10 March 2023.

National Institute of Health All of Us Research Program. Participant Partners. All of Us Research Program | NIH. 2022. https://allofus.nih.gov/about/who-we-are/all-us-participant-partners . Accessed 9 May 2023.

Cronin, R. M. et al. Development of the initial surveys for the All of Us Research Program. Epidemiol. Camb. Mass. 30 (4), 597–608. https://doi.org/10.1097/EDE.0000000000001028 (2019).

Mujahid, M. S., Diez Roux, A. V., Morenoff, J. D. & Raghunathan, T. Assessing the measurement properties of neighborhood scales: From psychometrics to ecometrics. Am. J. Epidemiol. 165 (8), 858–867. https://doi.org/10.1093/aje/kwm040 (2007).

Moser, A., Stuck, A. E., Silliman, R. A., Ganz, P. A. & Clough-Gorr, K. M. The eight-item modified medical outcomes study social support survey: Psychometric evaluation showed excellent performance. J. Clin. Epidemiol. 65 (10), 1107–1116. https://doi.org/10.1016/j.jclinepi.2012.04.007 (2012).

Hays, R. D. & DiMatteo, M. R. A short-form measure of loneliness. J. Pers. Assess. 51 (1), 69–81. https://doi.org/10.1207/s15327752jpa5101_6 (1987).

Article CAS PubMed Google Scholar

Williams, D. R. Race and health: Basic questions, emerging directions. Ann. Epidemiol. 7 (5), 322–333. https://doi.org/10.1016/S1047-2797(97)00051-3 (1997).

Williams, D. R. Everyday Discrimination Scale (Harvard University, 2023). https://scholar.harvard.edu/davidrwilliams/node/32397 . Accessed 18 May 2023.

Cohen, S., Kamarck, T. & Mermelstein, R. A global measure of perceived stress. J. Health Soc. Behav. 24 (4), 385–396. https://doi.org/10.2307/2136404 (1983).

Cohen, S. & Williamson, G. M. Perceived stress in a probability sample of the United States. In The Social Psychology of Health. The Claremont Symposium on Applied Social Psychology (eds Spacapan, S. & Oskamp, S.) 31-67 (Sage Publications, Inc, 1988).

Masters, K. S. Brief multidimensional measure of religiousness/spirituality (BMMRS). In Encyclopedia of Behavioral Medicine (eds Gellman, M. D. & Turner, J. R.) 267–269 (Springer, 2013). https://doi.org/10.1007/978-1-4419-1005-9_1577 .

Chapter Google Scholar

Li, S., Stampfer, M. J., Williams, D. R. & VanderWeele, T. J. Association of religious service attendance with mortality among women. JAMA Intern. Med. 176 (6), 777–785. https://doi.org/10.1001/jamainternmed.2016.1615 (2016).

Regents of the University of California. CHIS 2018 Adult Questionnaire Version 1.53 Spanish 201. https://healthpolicy.ucla.edu/chis/design/Documents/2018%20Questionnaires%20and%20Topics%20List/09-11-19%20Updated/CHIS%202018%20Adult%20Spanish%20Questionnaire.pdf (2019).

Hager, E. R. et al. Development and validity of a 2-item screen to identify families at risk for food insecurity. Pediatrics 126 (1), e26-32. https://doi.org/10.1542/peds.2009-3146 (2010).

Manchanda, R. & Gottlieb, L. Upstream Risks Screening Tool and Guide V2.6 . https://www.aamc.org/media/25736/download (2015).

Billioux, A., Verlander, K., Anthony, S. & Alley, D. Standardized screening for health-related social needs in clinical settings: The accountable health communities screening tool. NAM Perspect. https://doi.org/10.31478/201705b (2017).

Ross, C. E. & Mirowsky, J. Neighborhood disadvantage, disorder, and health. J. Health Soc. Behav. 42 (3), 258–276. https://doi.org/10.2307/3090214 (2001).

Sallis, J. F. et al. Evaluating a brief self-report measure of neighborhood environments for physical activity research and surveillance: Physical activity neighborhood environment scale (PANES). J. Phys. Act Health 7 (4), 533–540. https://doi.org/10.1123/jpah.7.4.533 (2010).

Sallis, J. F. PANES: Physical Activity Neighborhood Environment Survey (also known as: IPS/IPAQ Environmental Survey Module, Self-Administered (original version) . James F. Sallis PhD. https://www.drjimsallis.com/_files/ugd/a56315_068b49350b4a42208780d145e10d8ecc.pdf . Accessed 24 May 2023.

Peek, M. E., Nunez-Smith, M., Drum, M. & Lewis, T. T. Adapting the everyday discrimination scale to medical settings: Reliability and validity testing in a sample of African American patients. Ethn. Dis. 21 (4), 502–509 (2011).

PubMed Google Scholar

All of Us Research Program. Social Determinants of Health (SDOH) Survey-English . 2021. https://www.researchallofus.org/wp-content/themes/research-hub-wordpress-theme/media/surveys/SDOH_Survey_English.pdf . Accessed 10 May 2023.

Earls, F. J., Brooks-Gunn, J., Raudenbush, S. W. & Sampson, R. J. Brooks-Gunn SW Raudenbush RJ Sampson 1999 Project on Human Development in Chicago Neighborhoods: Community Survey, 1994–1995 (Verlag Herder, 1999).

Sampson, R. J., Raudenbush, S. W. & Earls, F. Neighborhoods and violent crime: A multilevel study of collective efficacy. Science 277 (5328), 918–924 (1997).

Taylor, T. R., Kamarck, T. W. & Shiffman, S. Validation of the Detroit area study discrimination scale in a community sample of older African American adults: The Pittsburgh healthy heart project. Int. J. Behav. Med. 11 (2), 88–94. https://doi.org/10.1207/s15327558ijbm1102_4 (2004).

Fetzer Institute NI on AWG. Multidimensional Measurement of Religiousness, Spirituality for Use in Health Research . A Report of the Fetzer Institute/National Institute on Aging Working Group. https://fetzer.org/sites/default/files/images/resources/attachment/%5Bcurrent-date%3Atiny%5D/Multidimensional_Measurement_of_Religousness_Spirituality.pdf (1999).

Radandt, N. E. et al. Validation of a 2-item food security screening tool in a dental setting. J. Dent. Child. 85 (3), 114–119 (2018).

Ash, A. S. et al. Social determinants of health in managed care payment formulas. JAMA Intern. Med. 177 (10), 1424–1430. https://doi.org/10.1001/jamainternmed.2017.3317 (2017).

Sallis, J. F. et al. Neighborhood environments and physical activity among adults in 11 countries. Am. J. Prev. Med. 36 (6), 484–490. https://doi.org/10.1016/j.amepre.2009.01.031 (2009).

Download references

Acknowledgements

We thank our colleagues, Brandy Mapes, Ashley Green and Chris Lord for providing their support and input throughout the demonstration project lifecycle. We thank Jun Qian for providing input on the project’s code review. We thank the DRC’s Research Support team for their help during implementation. We also thank the All of Us Science Committee, the All of Us Participant Provided Information (PPI) Committee and All of Us Steering Committee for their efforts evaluating and finalizing the approved demonstration projects. The All of Us Research Program would not be possible without the partnership of contributions made by its participants. See the Supplementary Note for a roster of past and present All of Us principal investigators. To learn more about the All of Us Research Program’s research data repository, please visit https://www.researchallofus.org/ . The All of Us Research Program is supported by the National Institutes of Health, Office of the Director: Regional Medical Centers: 1 OT2 OD026549; 1 OT2 OD026554; 1 OT2 OD026557; 1 OT2 OD026556; 1 OT2 OD026550; 1 OT2 OD 026552; 1 OT2 OD026548; 1 OT2 OD026551; 1 OT2 OD026555; IAA#: AOD 16037; Federally Qualified Health Centers: 75N98019F01202.; Data and Research Center: 1 OT2 OD35404; Biobank: 1 U24 OD023121; The Participant Center: U24 OD023176; Participant Technology Systems Center: 1 OT2 OD030043; Community Partners: 1 OT2 OD025277; 3 OT2 OD025315; 1 OT2 OD025337; 1 OT2 OD025276. In addition, the All of Us Research Program would not be possible without the partnership of its participants.

Author information

Authors and affiliations.

Division of Medical and Scientific Research, All of Us Research Program, National Institutes of Health, Bethesda, MD, USA

Samantha Tesfaye & Tamara R. Litwin

Oak Ridge Institute for Science and Education (ORISE), Oak Ridge, TN, USA

Samantha Tesfaye

Department of Internal Medicine, The Ohio State University, Columbus, OH, USA

Robert M. Cronin

Division of Cohort Development (DCD), All of Us Research Program, National Institutes of Health, Bethesda, MD, USA

Maria Lopez-Class & Christopher S. Foster

Department of Biostatistics, Vanderbilt University Medical Center, Nashville, TN, USA

Qingxia Chen & Andrew Guide

Department of Medicine, Brigham and Women’s Hospital, Boston, MA, USA

Callie A. Gu & Cheryl R. Clark

Department of Epidemiology and Biostatistics, University of California San Francisco, San Francisco, CA, USA

Robert A. Hiatt

Division of Engagement and Outreach, All of Us Research Program, National Institutes of Health, Bethesda, MD, USA

Angelica S. Johnson

Department of Public Health Sciences, Henry Ford Health, Detroit, MI, USA

Christine L. M. Joseph

Cherokee Health Systems, Knoxville, TN, USA

Parinda Khatri

Office of Data and Analytics, All of Us Research Program, National Institutes of Health, Bethesda, MD, USA

Sokny Lim, Andrea H. Ramirez & Elif Dede-Yildirim

Division of Health Support Services, San Ysidro Health, San Diego, CA, USA

Fatima A. Munoz

All of Us Research Program, National Institutes of Health, Bethesda, MD, USA

Heather Sansbury

Department of Psychology, Vanderbilt University, Nashville, TN, USA

David G. Schlundt

Unlimited Potential, Phoenix, AZ, USA

Emma N. Viera

Leidos, Inc., Reston, VA, USA

You can also search for this author in PubMed Google Scholar

Contributions

R.C. and C.R.C. were involved in the project design, project supervision, and manuscript development. R.C. and C.R.C. took part in writing all sections of the manuscript. S.L. was involved in the project administration and management. A.R. oversaw the project development and management. D.G.S. developed the scoring guidelines for each measure, provided expert opinion on the psychometric constructs, and helped write the Methods section of the manuscript. T.R.L. and Q.C. helped with the methodology and project implementation, and both made contributions to the Methods section. C.A.G. helped conduct the literature review and write the Abstract. S.T., A.G., and E.D.Y. were involved in the data curation, analysis, and visualization and contributed to the Methods and Results section. M.L.C., C.S.F., R.A.H., A.J., C.L.M.J., P.K., F.A.M., H.S., and E.N.V. were part of the SDOH Task Force and took part in the survey development process, the project design, and the review process of the manuscript. All authors participated in the editing and review process of the manuscript. S.T., R.C., and C.R.C. led the manuscript writing and review process.

Corresponding author

Correspondence to Cheryl R. Clark .

Ethics declarations

Competing interests.

The authors declare no competing interests.

Additional information

Publisher's note.

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Supplementary Information

Supplementary information., rights and permissions.

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/ .

Reprints and permissions

About this article

Cite this article.

Tesfaye, S., Cronin, R.M., Lopez-Class, M. et al. Measuring social determinants of health in the All of Us Research Program. Sci Rep 14 , 8815 (2024). https://doi.org/10.1038/s41598-024-57410-6

Download citation

Received : 21 July 2023

Accepted : 18 March 2024

Published : 16 April 2024

DOI : https://doi.org/10.1038/s41598-024-57410-6

Share this article

Anyone you share the following link with will be able to read this content:

Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative

By submitting a comment you agree to abide by our Terms and Community Guidelines . If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.

Quick links

Explore articles by subject
Guide to authors
Editorial policies

Open access
Published: 11 September 2023

Discovering social determinants of health from case reports using natural language processing: algorithmic development and validation

Shaina Raza 1 , 2 , 3 ,
Elham Dolatabadi 1 , 3 , 4 , 5 ,
Nancy Ondrusek 1 , 2 ,
Laura Rosella 2 &
Brian Schwartz 1 , 2

BMC Digital Health volume 1 , Article number: 35 ( 2023 ) Cite this article

1497 Accesses

2 Citations

2 Altmetric

Metrics details

Social determinants of health are non-medical factors that influence health outcomes (SDOH). There is a wealth of SDOH information available in electronic health records, clinical reports, and social media data, usually in free text format. Extracting key information from free text poses a significant challenge and necessitates the use of natural language processing (NLP) techniques to extract key information.

The objective of this research is to advance the automatic extraction of SDOH from clinical texts.

Setting and data

The case reports of COVID-19 patients from the published literature are curated to create a corpus. A portion of the data is annotated by experts to create ground truth labels, and semi-supervised learning method is used for corpus re-annotation.

An NLP framework is developed and tested to extract SDOH from the free texts. A two-way evaluation method is used to assess the quantity and quality of the methods.

The proposed NER implementation achieves an accuracy (F1-score) of 92.98% on our test set and generalizes well on benchmark data. A careful analysis of case examples demonstrates the superiority of the proposed approach in correctly classifying the named entities.

Conclusions

NLP can be used to extract key information, such as SDOH factors from free texts. A more accurate understanding of SDOH is needed to further improve healthcare outcomes.

Peer Review reports

Introduction

Background and significance.

Social determinants of health (SDOH) refer to the non-medical factors such as birth, education, occupation, and living conditions, that influence the health outcomes of individuals [ 1 ]. There is extensive evidence [ 2 , 3 , 4 , 5 ] that SDOH significantly influences a broad range of health outcomes including mortality rate, elderly care, mental health, and risks for chronic diseases such as asthma, cancer, heart disease, and obesity [ 6 ]. Diabetes, depression, hypertension, and suicidal behavior are all outcomes of SDOH [ 7 ]. According to some studies, medical care accounts for 10–20% of health factors, while SDOH accounts for 80% to 90% [ 8 ]. Thus, it is of high importance to address the SDOH to improve the health systems.

SARS-CoV-2 is a virus that infects humans, causing severe upper respiratory problems and coronavirus disease 2019 (COVID-19) disease [ 9 ] . Recent statistics suggest that SDOH, such as race, ethnicity, gender, social-economic factors, and related population characteristics are also among the risk factors for COVID-19 [ 10 ]. There is evidence [ 11 , 12 ] that those who are homeless have a higher prevalence of COVID-19 disease than those who are housed. As a result, understanding the SDOH in the context of a pandemic is critical for improving population health outcomes.

Despite advancements in technology, the collection of SDOH data remains a challenge for the public health community. Electronic Health Records (EHRs) constitute a significant portion of clinical data [ 13 ], but their use in clinical research is limited due to difficulties in automating the process of extracting information from unstructured data. Natural Language Processing (NLP) techniques can potentially overcome these limitations by facilitating the extraction of relevant information from unstructured data, thereby enabling its use in analysis and decision-making.

Through this work, we try to address the research question “How can we effectively mine SDOH data from case reports to improve its practical usability and scholarly value?”.

The objectives of this study are:

To propose an SDOH NLP framework that can extract SDOH information from case report data. This framework aims to provide a more comprehensive understanding of the patient's condition by considering the impact of social and economic factors.

To prepare a dataset that is annotated with SDOH labels. This dataset will be used for training and evaluating the NLP framework. In addition, semi-supervised learning techniques will be implemented to facilitate data re-annotation and improve the accuracy of the SDOH annotations in the dataset.

This work adds to previous efforts [ 14 , 15 , 16 , 17 ] by acknowledging both clinical and SDOH factors as key contributors to understanding patient health outcomes. The proposed framework includes a data processing module that parses the contents of free text and prepares a dataset, an NLP module that extracts SDOH information from the texts, and an evaluation module that assesses the NLP module's accuracy and effectiveness.

Experimental results show that the proposed approach outperforms the baseline methods across multiple datasets. A thorough analysis of the extracted information yields important findings that can be utilized for COVID-19 surveillance and the creation of informed public health strategies.

Materials and methods

As part of a research project, we developed an NLP framework and validated the performance of its module for SDOH identification. The proposed framework is extensible and applicable to a variety of public health use cases, including surveillance, epidemiology, and policy-making processes.

Our dataset is an extension and refinement of our original collection [ 18 ]. This enhanced version specifically highlights the SDOHs. We derived this data from 4,000 electronic case reports of COVID-19 patients obtained via the LitCOVID API [ 19 ], marking a shift in focus from purely clinical elements (as in previous version) to the broader context of SDOH. Our aim is to illuminate health disparities and the elements that drive them.

We thoroughly applied selection criteria to ensure data relevance and quality, such as a specific timeframe (January to June 2022), language (English), defined patient age groups, and the exclusive inclusion of peer-reviewed case reports. The search query is given in Appendix A : Table S1.

The SDOH has been conceptually organized into categories for a more holistic understanding of health outcomes among COVID-19 patients. These categories are given below, and more fine-grained details of named entities are in Appendix A : Table S2.

Demographic Factors : Gender, Age, Race/Ethnicity—These factors allow us to identify patterns and trends among different demographic groups.

Biometric Factors: Height, Weight—These physical characteristics impact an individual's health status.

Temporal Factors : Date, Relative Date, Duration, Time—These provide insights into temporal trends such as disease progression and recovery time.

Lifestyle Factors: Smoking, Alcohol, Substance use—These lifestyle attributes can significantly impact health outcomes.

Socioeconomic Factors: Employment —This factor influences access to healthcare, stress levels, and other health-related factors.

Healthcare System Interaction : Admission/Discharge—These data points are essential for understanding the patient's journey within the healthcare system and the efficiency of their care.

Clinical Factors: As detailed in Appendix A , Table S2—These clinically relevant data points derived from patient records influence health outcomes.

This dataset and categorization scheme offers a broader and more detailed understanding of the factors that influence the health outcomes of COVID-19 patients, with a specific emphasis on the SDOH.

Proposed SDOH NLP Framework

The SDOH NLP framework, shown in Fig. 1 , features three modules: (1) data processing, (2) NLP, and (3) evaluation.

SDOH NLP proposed architecture

Data processing module

Case reports, acquired via the LitCOVID API, were processed by a scientific parser software [ 20 ] to extract textual content and were stored in MongoDB database. Each database entry represents a case report, organized by a unique identifier. A group of four experts manually annotated a sample of 200 case reports, establishing the basis of our annotated dataset. This process is facilitated by the Spark NLP annotation tool [ 21 ] resulted in approximately 5000 sentences being annotated in three months, with each sentence has multiple SDOH categories in many instances.

To ensure annotation consistency, we adhered to literature-sourced guidelines [ 22 , 23 ] for named entity annotation. The Inter-annotator Agreement (IAA) [ 24 ] approach was utilized to measure annotation alignment across all annotators, with the Cohen's Kappa [ 25 ] coefficient providing a measure of agreement. We obtain the Cohen's Kappa value of 0.75, which is classified as substantial agreement on the kappa coefficient scale [ 26 ]. Further, the discrepancies in annotation were resolved through a consensus-based dialogue between annotators.

Pre-processing of our data included standardization measures such as lowercasing all text, removing foreign or uncommon symbols, and separating contractions into individual words. We utilized tokenization to break the text down into manageable units and removed stop words to spotlight the salient terms. This refining process prepared our data for an efficient and in-depth analysis.

For large-scale annotation, we implemented a semi-supervised approach. Initial manual annotations (on 200 case reports) were used to train a BERT (Bidirectional Encoder Representations from Transformers) model for the NER task, achieving an accuracy of approximately 94.18%. In NER tasks, the model predicts the label of each token—usually a word—in a sentence. These labels represent categories or "entities," such as person names, locations, medical codes, time expressions, and particularly, SDOH in our case. Next, we used this trained model to annotate the larger dataset (remaining 3800 reports), effectively using the model to classify or predict the NER label for each token in these reports. This process is also known as pseudo-labeling. The newly annotated data is then combined with our original labeled data, and the model is retrained on this combined dataset. This cycle can be repeated as necessary. The key idea is that the performance of the model is continuously improved as it learns from both the manually annotated data and the new data that it annotates itself.

Through our empirical analysis, we discovered that the dataset containing 4000 case reports encompasses approximately 60,000 sentences. Within this dataset, we further identified around 80,000 named entities related to SDOH and approximately 180,000 clinical named entities. The annotated data is stored in the widely used CONLL-2003 format [ 27 ], for NER tasks. The NER model, discussed in the NLP layer, is trained using this annotated data, with a portion (30%) reserved for evaluation purposes. While the foundational method is inspired by our earlier work [ 18 ], the training process for SDOH has undergone significant re-training in our current study.

The schema of this dataset version, we introduce is:

Sentence ID < INTEGER > : A unique identifier allocated to each sentence, maintained from the original version.

Word < STRING > : This represents the individual word token from the sentence, preserved from the original version.

POS < STRING > : The Part of Speech tag assigned to the word, also carried over from the original version.

Chunk < STRING > : The syntactic chunk tag of the word, kept consistent with the original version.

NER Tag < ENUM > : This version introduces an enhancement to the NER tag associated with each word. In addition to 'Clinical', this has been expanded to encompass 'Non-Clinical' and 'SDOH' categories.

For multi-label classification, each word in a sentence would have a distinct named entity tag, allowing multiple labels per sentence.

Natural language processing (NLP) module

Our proposed framework utilizes an NLP method for Named Entity Recognition (NER) as a sequence classification task. Every token in the input sequence is assigned a label, for a more effective entity identification. An illustration of this would be assigning "O, I-PERSON, O, I-DISEASE" to "The patient has COVID-19", where "O" represents a non-entity type and "I-DISEASE" a single-token DISEASE type.

The NER model integrates three key players: a Transformer layer, a Bidirectional Long Short-Term Memory (BiLSTM), and a Conditional Random Field (CRF) layer, as illustrated in Fig. 2 . Among these, the Transformer layer utilizes BioBERT [ 15 ], which has been specifically fine-tuned on our dataset for NER task. This task-specific adaptation transforms input sequences into detailed embeddings, providing robust representations tailored to our objectives.

Proposed framework for named entities task

Subsequent to the Transformer layer, the BiLSTM layer captures context features to amplify the semantic meanings in the texts. The layer's forward and backward LSTM successfully bring together hidden information from both preceding and subsequent texts.

Finally, the CRF layer [ 28 ] inputs the output sequence from the BiLSTM layer, revealing the dependencies within it. This layer effectively translates the complex interdependency between the named tags into the final predicted labels. The final IOB representation, output from the CRF layer, is converted into an accessible format by linking the recognized named entities with their appropriate labels.

Joint optimization

In our NLP strategy, we initially fine-tune BioBERT for the task at hand using training data, which enables BioBERT to learn task-related patterns. Next, we combine BioBERT with the BiLSTM and CRF layers in a joint optimization approach. This means all layers, including BioBERT and subsequent layers (BiLSTM and CRF), are trained collectively, facilitating the optimization of the entire model architecture. By initially fine-tuning BioBERT and then jointly training it with BiLSTM and CRF layers, we exploit both BioBERT’s pre-existing knowledge and the task-oriented data captured by the following layers.

As an example, a few named entities annotated on a sample case report is shown in Appendix A : Figure S1.

Evaluation module

In this study, a thorough evaluation protocol was adopted to assess the efficacy of our proposed NER method. The methodology involved a quantitative benchmarking of our approach against state-of-the-art NER models, using established biomedical datasets as referenced in Appendix A , Table S3, as well as on our test set. Additionally, a qualitative evaluation on COVID-19 case reports was performed to determine the applicability of our method in real-world scenarios.

To assess the performance of all methods, we adhered to the train-test split strategy outlined in the original publication of each dataset, when available. Otherwise, we implemented a stratified cross-validation strategy for this purpose. Specifically, we employed a 5-Fold stratified cross-validation approach, enhancing the thoroughness of our model's performance evaluation, and underscoring the statistical significance of the results. We further set aside a distinct test set, consisting of 30% of our annotated data, for evaluation. Ground truth labels served as the reference standard for this evaluation.

For the statistical analysis of our results, we applied inferential statistics, including paired t-tests [ 29 ], to the performance metrics across the five folds from our cross-validation. The paired t-test was chosen because it is a powerful tool to compare means from the same group under different conditions – in our case, different models. A p -value less than 0.05 was considered statistically significant, indicating that the observed differences were not due to random chance.

We compared our method with two groups of existing models: Bi-LSTM-based (BiLSTM-CRF, BiLSTM-CNN-Char, BiLSTM-CRF-MTL, Doc-Att-BiLSTM-CRF, and CollaboNet) and Transformer-based models (BLUE-BERT, ClinicalBERT, BioBERT, BioBERT-CRF, and BioBERT-MLP). For the variants of BioBERT, we used their open-source implementations and added respective additional layers where necessary. The benchmark datasets used in the experiments are mentioned in Appendix A , Table S3, and the baseline models considered are given in Appendix A , Table S4. Ensuring a fair comparison, all baseline models were tuned to their optimal hyper-parameter configurations. The evaluation metrics included F1-score (harmonic mean of precision and recall), and macro-average F1-scores, following the practice in previous works [ 15 , 30 ].

The experimental setup for our study was facilitated by Google Colab Pro, providing access to cloud-based GPUs (K80, P100, or T4) and 32 GB RAM, which enabled efficient model training and ample storage for the transfer learning process through its integration with Google Drive. Specific parameters for the BiLSTM and Transformer-based architectures are listed in Appendix 1 : Table S5. To maintain consistency across different experimental runs, the PyTorch BERT implementation from Huggingface.co was used for the BERT encoder layers.

Quantitative analysis

Benchmarking against baselines.

Table 1 presents a performance comparison of various NER methods, including our approach, across multiple biomedical datasets, and our own test set. The performance metrics are given in terms of F1-score. Along with these individual scores, we also provide the mean F1-score and standard deviation (Mean ± SD) for each method across all datasets. The paired t-tests were conducted to compare the performance of our approach with each of the other methods. The null hypothesis for these tests was that there is no significant difference between the performance of our approach and each of the other methods. T-statistics were computed, and the corresponding p-values were used to test this null hypothesis. In all cases, the p -value is less than 0.05, indicating that we can reject the null hypothesis. This means we can conclude that there is a statistically significant difference in favor of our approach.

Overall, we observe in Table 1 that Transformer-based models such as BLUE-BERT, ClinicalBERT, BioBERT, BioBERT + CRF, BioBERT + MLP, and our own approach consistently outperform other methods, indicating the strength of transformer architectures in capturing complex semantic relationships in text data.

Among BiLSTM models, BILSTM-CNN-Char and Doc-Att-BiLSTM-CRF perform relatively well across all datasets. BILSTM-CNN-Char combines the strengths of CNNs in extracting local features with BiLSTMs' ability to capture long-term dependencies, indicating the benefit of such multi-modal architectures. On the other hand, Doc-Att-BiLSTM-CRF adds an attention mechanism to the BiLSTM model, allowing it to focus on more informative parts of the sequences and thereby enhance the NER performance. However, compared to Transformer-based models, these BiLSTM-based models seem to be slightly less effective.

Transformer-based models are generally more effective at capturing intricate relationships and have a notable ability to utilize extensive unsupervised data during training tasks. Our approach combines the transformer layer and BiLSTM architecture with task-specific fine-tuning, achieves the best performance. This indicates that while Transformer-based models provide a strong foundation for NER tasks, there is still room for improvement and task-specific optimization.

Performance analysis on named eneities

The fine-grained performance of our proposed NLP approach in extracting 10 most occurring SDOH entity classes from our data is shown in Table 2 .

Table 2 provides a comparative analysis of the performance of three different models—our method, BioBERT, and BioBERT + CRF (best performing baselines)—in extracting various SDOH factors. The performance is assessed using the F1-score, presented as a mean value followed by the standard deviation (Mean ± SD). For each class, a macro-average F1-score is also computed, representing the mean F1-score of all classes.

Upon examining the data, we observe that our method consistently outperforms both BioBERT and BioBERT + CRF across the majority of SDOH factors. This suggests that the enhancements we have incorporated, such as the addition of a BiSLTM and CRF layer, positively impact the model's performance. BioBERT closely follows our method, and BioBERT + CRF, which incorporates an additional CRF layer, displays a marginally lesser performance compared to BioBERT. This observation may indicate that the inclusion of a CRF layer does not necessarily augment BioBERT's capability for this particular SDOH factor identification task.

Our method demonstrates particularly high accuracy in extracting demographic factors, notably gender and race/ethnicity. The same holds true for biometric factors, including BMI. Additionally, the model yields encouraging results for temporal factors, lifestyle factors, and other SDOH categories. The performance gain of our approach may be attributed to incorporating additional layers, such as BiSLTM along with CRF, into existing models like BioBERT.

Error analysis

In this sub-section, we provide three running examples to further demonstrate the efficacy of our SDOH extraction approach.

The findings from Table 3 are as:

Correctly identifying an entity: In general, all three models demonstrate a good ability to correctly identify entities in the given sentences. For example, in the sentence "The patient has diabetes and is a smoker," all models accurately identify the entities "diabetes" and "smoker." This indicates that the models have learned to recognize and extract specific entities related to diseases and lifestyle factors.

Failing to identify an entity type: One limitation observed is the failure to identify certain entity types mentioned in the true label sentence. For instance, in the sentence "The individual's relationship status is single and currently employed," our NLP model is the only one that correctly identifies the entity type "single" along with the "employed" entity. However, both BioBERT and BioBERT + CRF do not capture the entity type, highlighting the challenge in associating multiple entity types within a single sentence. At one instance (example 5), our model also could not identify the ‘disease’ entity.

Misclassifying a non-entity : Another challenge is the misclassification of non-entities as entities. In some examples, one or more models incorrectly label a word or phrase as an entity when it is not. For instance, in the sentence " The patient's diet affects their blood pressure, " both BioBERT and BioBERT + CRF misclassify "blood pressure" as an entity (which may be more related to clinical factor), instead of picking “diet” as the lifestyle factor. This highlights the difficulty in distinguishing between specific entities and non-entities.

Overall, while the models demonstrate proficiency in correctly identifying certain entities, they face challenges in capturing specific entity types and avoiding misclassifications. The findings emphasize the importance of continued research [ 38 ] and development in NLP to address these challenges and improve the reliability of entity identification.

Qualitative analysis to see model effectiveness

In this section, we see the effectiveness of the proposed method in inferring the named entities from case report data. The prevalence of common SDOH reported in the case reports is depicted in Fig. 3 .

Prevalence (occurring more than 70%) of SDOH for the factors: a age, b educational level, c employment status, d gender, e income level, f race/ethnicity, and g smoking status, in COVID-19 patients

Based on the analysis of Fig. 3 , several key observations can be made, which are:

In terms of age groups (Fig. 3 a), the distribution is relatively even, with 20% of the individuals falling in the 18–30 range, 30% in the 31–45 range, 35% in the 46–60 range, and 15% aged 61 and above. Analyzing the education level SDOH distribution (Fig. 3 b), 30% of the individuals have a high school education or below, 40% have some college education, 20% have a bachelor's degree, and 10% have an advanced degree, as reported in named entities.

Approximately 40% of the population (Fig. 3 c) is employed, 20% is unemployed, and 40% is not employed. The population distribution in terms of gender (Fig. 3 d) shows that 45% of the individuals are male, while 55% are female. The income level distribution (Fig. 3 e) indicates that 40% of the population falls under the low-income category, 45% in the middle-income category, and 15% in the high-income category. This highlights the income disparities among the COVID-19 patients.

The prevalence of different races/ethnicities (Fig. 3 f) reveals that the majority of the population in the dataset is white (60%), followed by Black (25%), Asian (10%), and Hispanic (5%). In terms of smoking status (Fig. 3 g), around 25% of the population in the dataset are smokers, while the remaining 75% are non-smokers.

Overall, these findings are specific to the dataset of COVID-19 patients analyzed and may not be representative of the entire population.

Next, we present the prevalence of common disease disorders in COVID-19 patients for both female and male groups (demographics SDOH) in Fig. 4 . This result allows for a more detailed analysis of how gender may affect the likelihood of certain disease disorders in COVID-19 patients. It is worth noting that these are generalizations and individual cases may vary. For example, some studies suggest that male COVID-19 patients may have a higher risk of severe illness or hospitalization [ 39 ], but more research is needed to confirm this.

Common diseases in COVID-19 patients to gender

Vaccination fits under the category of Healthcare System Interaction within the SDOH. We show COVID-19 vaccination status across different age groups in Fig. 5

COVID-19 vaccination rates by age group: this line chart illustrates an estimated distribution of COVID-19 vaccination rates across different age groups. The x-axis represents the age groups while the y-axis indicates the vaccination rates in percentages

From Fig. 5 , it can be seen that the COVID-19 vaccination rate increases as the age group rises. The lowest rate of vaccination is found in the '0–17' age group, which is understandable given that vaccine rollout for minors has varied across regions and has often come after adults. The '18–29' age group shows a significant increase in vaccination rate, reaching 60%. The rates continue to climb for the '30–39', '40–49', and '50–64' age groups, signifying a more pronounced willingness or availability to get vaccinated as age increases. The highest vaccination rate is found in the '65 + ' age group, reflecting the priority often given to older individuals due to their increased vulnerability to severe COVID-19 symptoms.

Previous works

Previous works have extracted SDOH information from clinical data using different methods such as regular expressions, dictionaries, rule-based methods like cTAKES [ 40 , 41 ], and deep neural networks like CNNs, LSTMs [ 42 ], and Transformer-based methods [ 41 ]. Language model-based representations have been found to perform well, especially with large training sets. However, even simpler neural network representations like BiLSTM can perform well with enough training data, being only slightly lower in performance than Transformer-based models, as shown in Table 1 .

Practical impact for enhancing clinical knowledge and patient care

This research aimed to investigate the potential of using NLP models on case report datasets to improve clinical knowledge and patient care. Through the use of advanced NLP techniques, we were able to extract important information from case reports, including symptoms, diagnosis, and treatment, as well as information about SDOH. This information can be used to inform targeted interventions and deliver personalized and evidence-based care. The use of NLP models also allows for automation of the process of extracting information from case reports, saving time and resources for healthcare professionals. Such a framework can be integrated into clinical decision support systems to improve the quality of care.

Limitations

The study acknowledges its limitations and provides opportunities for future research. One limitation is that the dataset may not be representative of all SDOH impacting patients with COVID-19. To address this, future research could aim to implement EHRs and clinical notes that are updated in real-time. Additionally, data privacy concerns will need to be addressed by masking named entities associated with patients' personal information, such as names, locations, and identifiers.

A case report may not always describe a patient's current condition. For instance, "the patient has a family history of hysteria" can be classified as a psychiatric condition of the patient, even though it is not the patient’s current condition. To address this, the annotation scheme could be extended and the rules or semantics defined more clearly, allowing for the retraining of language models. However, this would be a labor-intensive process.

There are several ways to further extend this research. One way is to use an extensive active learning approach [ 43 ] to improve model performance. Another direction is to use prompt-based learning, which utilizes the strengths of pre-trained foundation models [ 44 ], to improve overall effectiveness. Additionally, experimenting with different model architectures [ 45 ] and performing detailed significance tests may also add value to this work. By addressing these limits, we believe that this work will lead to an effective and general-purpose model.

This study demonstrates that NLP-based methods can be used to identify SDOH from texts. The proposed framework uses a combination of neural networks. To assess the model ability to extract different named entities, a detailed analysis of the SDOH is performed. The proposed methods outperform the state-of-the-art methods for the NER task and showed effectiveness in determining clinical outcomes. The current study paves the way for future research and addresses the health disparities that appear in systematic healthcare systems.

Availability of data and materials

The data underlying this article will be shared on reasonable request to the corresponding author.

Abbreviations

Coronavirus disease

Social determinants of health

Electronic health records

World Health Organization

Natural language processing

Machine learning

Inter-annotator agreement

Named entity recognition

Case reports

Inside- outside-before

Bidirectional encoder representations from transformers

Bidirectional encoder representations from transformers for biomedical text mining

Bidirectional long short-term memory

Conditional random field

Conference on computational natural language learning

National center for biotechnology information

Biocreative v chemical disease relations

Biocreative iv chemical and drug

Biocreative ii gene mention recognition

Informatics for integrating biology and the bedside

Adverse drug events

Chemical-protein interactions

Convolutional neural network

Multi-task learning

Collaboration of deep neural networks

Acute respiratory distress syndrome

Post-traumatic stress disorder

CONSTITUTION OF WHO. World health organization. Responding to Community Spread of COVID-19. 2020. Reference WHO/COVID-19/Community\_Transmission/2020.1.

Flaskerud JH, DeLilly CR, Flaskerud JH. Social determinants of health status. Issues Ment Health Nurs. 2012;33:494–7.

Article PubMed PubMed Central Google Scholar

Chang CD. Social determinants of health and health disparities among immigrants and their children. Curr Probl Pediatr Adolesc Health Care. 2019;49:23–30.

Article PubMed Google Scholar

Donkin A, Goldblatt P, Allen J, Nathanson V, Marmot M. Global action on the social determinants of health. BMJ Glob Heal. 2018;3(Suppl 1): e000603.

Article Google Scholar

Braveman P, Gottlieb L. The social determinants of health: It’s time to consider the causes of the causes. Public Health Rep. 2014;129(SUPPL. 2):19–31.

Brakefield WS, Ammar N, Olusanya OA, Shaban-Nejad A. An urban population health observatory system to support COVID-19 pandemic preparedness, response, and management: Design and development study. JMIR Public Heal Surveill. 2021;7:1–16.

Google Scholar

Whitman A, De Lew N, Chappel A, Aysola V, Zuckerman R, Sommers BD. Addressing social determinants of health: examples of successful evidence-based strategies and current federal efforts. Off Heal Policy. 2022;1–30.

Williams DR, Costa MV, Odunlami AO, Mohammed SA. Moving upstream: how interventions that address the social determinants of health can improve health and reduce disparities. J Public Health Manag Pract. 2008;14(Suppl):S8–17.

Intiso D, Marco Centra A, Giordano A, Santamato A, Amoruso L, Di Rienzo F. Critical illness polyneuropathy and functional outcome in subjects with COVID-19: Report on four patients and a scoping review of the literature. J Rehabil Med. 2021. https://doi.org/10.2340/jrm.v53.1139 .

Singu S, Acharya A, Challagundla K, Byrareddy SN. Impact of social determinants of health on the emerging COVID-19 pandemic in the United States. Front Public Heal. 2020;8:406.

Baggett TP, Keyes H, Sporn N, Gaeta JM. Prevalence of SARS-CoV-2 infection in residents of a large homeless shelter in Boston. JAMA. 2020;323:2191–2.

Article CAS PubMed PubMed Central Google Scholar

Kiran T, Craig-Neil A, Das P, Lockwood J, Wang R, Nathanielsz N, et al. Factors associated with SARS-CoV-2 positivity in 20 homeless shelters in Toronto, Canada, from April to July 2020: a repeated cross-sectional study. C Open. 2021;9:E302–8.

Tarcar AK, Tiwari A, Rao D, Dhaimodker VN, Rebelo P, Desai R. Healthcare NER models using language model pretraining. CEUR Workshop Proc. 2020;2551 Hsdm:12–8.

Wang Y, Sohn S, Liu S, Shen F, Wang L, Atkinson EJ, et al. A clinical text classification paradigm using weak supervision and deep representation 08 Information and Computing Sciences 0801 Artificial Intelligence and Image Processing 17 Psychology and Cognitive Sciences 1702 Cognitive Sciences. BMC Med Inform Decis Mak. 2019;19:1–13.

Lee J, Yoon W, Kim S, Kim D, Kim S, So CH, et al. BioBERT: a pre-trained biomedical language representation model for biomedical text mining. Bioinformatics. 2020;36:1234–40.

Article CAS PubMed Google Scholar

Doğan RI, Leaman R, Lu Z. NCBI disease corpus: a resource for disease name recognition and concept normalization. J Biomed Inform. 2014;47:1–10.

Huang K, Altosaar J, Ranganath R. ClinicalBERT: Modeling Clinical Notes and Predicting Hospital Readmission. arXiv Prepr arXiv190405342. 2019.

Raza S, Schwartz B. Entity and relation extraction from clinical case reports of COVID-19: a natural language processing approach. BMC Med Inform Decis Mak. 2023;23:20.

Chen Q, Allot A, Lu Z. LitCovid: An open database of COVID-19 literature. Nucleic Acids Res. 2021;49:D1534–40.

Allen Institute for AI. Science-parse. 2019. https://github.com/allenai/science-parse . Accessed Aug 2023.

Spark NLP 4.0.2 ScalaDoc - com.johnsnowlabs.nlp.annotators.parser.dep.DependencyParserModel. 2022.

Campillos-Llanos L, Valverde-Mateos A, Capllonch-Carrión A, Moreno-Sandoval A. A clinical trials corpus annotated with UMLS entities to enhance the access to evidence-based medicine. BMC Med Inform Decis Mak. 2021;21:1–19.

Nothman J, Ringland N, Radford W, Murphy T, Curran JR. Learning multilingual named entity recognition from Wikipedia. Artif Intell. 2013;194:151–75.

Artstein R. Inter-annotator agreement. In: Handbook of linguistic annotation. Springer; 2017. p. 297–313.

Scarpellini G. Cohen’s kappa free calculator - IDoStatistics. 2020.

McHugh ML. Interrater reliability: the kappa statistic. Biochem medica. 2012;22:276–82.

Tjong Kim Sang EF, de Meulder F. Introduction to the CoNLL-2003 Shared Task: Language-Independent Named Entity Recognition. Proc 7th Conf Nat Lang Learn CoNLL 2003 HLT-NAACL 2003. 2003;142–7.

Huang Z, Xu W, Yu K. Bidirectional LSTM-CRF Models for Sequence Tagging. 2015.

Lakens D. Calculating and reporting effect sizes to facilitate cumulative science: A practical primer for t-tests and ANOVAs. Front Psychol. 2013;4 NOV:863.

Chiu JPC, Nichols E. Named entity recognition with bidirectional LSTM-CNNs. Trans Assoc Comput Linguist. 2016;4:357–70.

Lample G, Ballesteros M, Subramanian S, Kawakami K, Dyer C. Neural architectures for named entity recognition. arXiv Prepr arXiv160301360. 2016.

Wang X, Zhang Y, Ren X, Zhang Y, Zitnik M, Shang J, et al. Cross-type biomedical named entity recognition with deep multi-task learning. Bioinformatics. 2019;35:1745–52.

Xu K, Yang Z, Kang P, Wang Q, Liu W. Document-level attention-based BiLSTM-CRF incorporating disease dictionary for disease named entity recognition. Comput Biol Med. 2019;108:122–32.

Yoon W, So CH, Lee J, Kang J. Collabonet: collaboration of deep neural networks for biomedical named entity recognition. BMC Bioinformatics. 2019;20:55–65.

Peng Y, Yan S, Lu Z. Transfer learning in biomedical natural language processing: an evaluation of BERT and ELMo on ten benchmarking datasets. arXiv Prepr arXiv190605474. 2019.

Sun C, Yang Z, Wang L, Zhang Y, Lin H, Wang J. Biomedical named entity recognition using BERT in the machine reading comprehension framework. J Biomed Inform. 2021;118:103799.

Luo X, Gandhi P, Storey S, Huang K. A deep language model for symptom extraction from clinical text and its application to extract covid-19 symptoms from social media. IEEE J Biomed Heal Informatics. 2021;26:1737–48.

Han S, Zhang RF, Shi L, Richie R, Liu H, Tseng A, et al. Classifying social determinants of health from unstructured electronic health records using deep learning-based natural language processing. J Biomed Inform. 2022;127: 103984.

Abate BB, Kassie AM, Kassaw MW, Aragie TG, Masresha SA. Sex difference in coronavirus disease (COVID-19): a systematic review and meta-analysis. BMJ Open. 2020;10: e040129.

Chen ES, Carter EW, Sarkar IN, Winden TJ, Melton GB. Examining the use, contents, and quality of free-text tobacco use documentation in the electronic health record. In: AMIA Annual Symposium Proceedings. American Medical Informatics Association; 2014. p. 366.

Lituiev D, Lacar B, Pak S, Abramowitsch PL, Marchis E De, Peterson T. Automatic extraction of social determinants of health from medical notes of chronic lower back pain patients. medRxiv. 2022;:2022.03.04.22271541.

Quan W, Chen Z, Gao J, Hu XT. Comparative study of CNN and LSTM based attention neural networks for aspect-level opinion mining. In: 2018 IEEE International Conference on Big Data (Big Data). IEEE: 2018. p. 2141–50.

Ménard PA, Mougeot A. Turning silver into gold: error-focused corpus reannotation with active learning. In: Proceedings of the International Conference on Recent Advances in Natural Language Processing (RANLP 2019). 2019. p. 758–67.

Bommasani R, Hudson DA, Adeli E, Altman R, Arora S, von Arx S, et al. On the opportunities and risks of foundation models. arXiv preprint arXiv:2108.07258. 2021.

Davidson EM, Poon MTC, Casey A, Grivas A, Duma D, Dong H, et al. The reporting quality of natural language processing studies: systematic review of studies of radiology reports. BMC Medical Imaging. 2021;21(1):1–13.

Download references

Acknowledgements

The authors would like to thank Public Health Ontario scientists for their guidance on public health practices in COVID-19 research.

This research was co-funded by the Canadian Institutes of Health Research’s Institute of Health Services and Policy Research (CIHR-IHSPR) as part of the Equitable AI and Public Health cohort, and This work was funded through the Artificial Intelligence for Public Health (AI4PH) Health Research Training Platform (HRTP) supported by the Canadian Institutes of Health Research (CIHR).

Author information

Authors and affiliations.

Public Health Ontario (PHO), Toronto, ON, Canada

Shaina Raza, Elham Dolatabadi, Nancy Ondrusek & Brian Schwartz

Dalla Lana School of Public Health, University of Toronto, Toronto, ON, Canada

Shaina Raza, Nancy Ondrusek, Laura Rosella & Brian Schwartz

Vector Institute for Artificial Intelligence, Toronto, ON, Canada

Shaina Raza & Elham Dolatabadi

Institute of Health Policy, Management and Evaluation (IHPME), University of Toronto, Toronto, ON, Canada

Elham Dolatabadi

York University, Faculty of Health, School of Health Policy and Management, Toronto, ON, Canada

You can also search for this author in PubMed Google Scholar

Contributions

SR, ED, NO, BS, LR conceived the study design. SR and BS participated in the literature search. SR and BS prepared the search query for the data collection. SR performed the data curation, data pre-processing, and dataset preparation. SR built the framework and the models, and ED validated the framework. SR created the tables, plotted the graphics, and ED and NO interpreted the study findings. SR drafted the initial manuscript. BS and NO validated the results and evaluated the findings and revised the draft. All authors critically reviewed and substantively revised the manuscript. All authors have approved the final version of the manuscript for publication.

Corresponding author

Correspondence to Shaina Raza .

Ethics declarations

Ethics approval and consent to participate.

All data and methods were carried out in accordance with relevant guidelines and regulations. Public Health Ontario's (PHO) Ethics Review Board (ERB) waived the need for the ethical approval and informed consent.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.

Additional information

Publisher's note.

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Supplementary Information

Additional file 1:.

Appendix A Table S1. Search Query. Table S2. Entities. Figure S1. Named entities extracted from the case report (1). Table S3. Benchmark datasets used. Table S4. Baseline methods used. Table S5. General Hyperparameters used along with best value and range in parenthesis.

Rights and permissions

Reprints and permissions

About this article

Cite this article.

Raza, S., Dolatabadi, E., Ondrusek, N. et al. Discovering social determinants of health from case reports using natural language processing: algorithmic development and validation. BMC Digit Health 1 , 35 (2023). https://doi.org/10.1186/s44247-023-00035-y

Download citation

Received : 15 March 2023

Accepted : 01 August 2023

Published : 11 September 2023

DOI : https://doi.org/10.1186/s44247-023-00035-y

Share this article

Anyone you share the following link with will be able to read this content:

Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative

Named entity
Transformer models

BMC Digital Health

ISSN: 2731-684X

Submission enquiries: [email protected]
General enquiries: [email protected]

How a Global Pandemic Became a Case Study in Social Determinants of Health

hms social determinants of health stock image

African Americans comprise approximately 14% of Michigan’s population , yet 40% of the state’s COVID-19-related deaths were among African Americans. In Chicago, those numbers were 30% and 72%, respectively . Although Native Americans represent just 11% of New Mexico’s population, Native Americans made up half of the state’s COVID-19 death toll .

In New York City, the epicenter of the coronavirus outbreak in the U.S., case numbers varied profoundly by zip code . Manhattan, the city’s most densely populated borough, accounted for the lowest percentage of COVID-19 cases, hospitalizations and deaths; at the time of writing, the Bronx had the highest numbers across all three categories and more than twice the number of cases as Manhattan. A JAMA research letter on the disparity in New York notes the Bronx “has the highest proportion of racial/ethnic minorities, the most persons living in poverty, and the lowest levels of educational attainment.”

Examples such as these are myriad, and more will undoubtedly be forthcoming as better data comes to hand. Rarely do we get as concrete and concentrated an opportunity to examine the health disparities that exist in today’s society as in the case of a pandemic virus — one to which everyone is equally susceptible, but not at equal advantage.

The World Health Organization defines social determinants of health (SDoH) as “the conditions in which people are born, grow, live, work and age…shaped by the distribution of money, power and resources at global, national and local levels.”

Just as population density alone is not an indicator of COVID-19 concentration, a single factor often does not determine a health outcome; rather, there are a number of physical, social, behavioral, environmental and other factors to consider. For example, residents of food deserts — areas with limited access to affordable and nutritious foods — are shown to have disproportionately higher rates of chronic disease . Relevant to COVID-19, a person with a job that can be done remotely is better equipped to adhere to stay at home orders and social distancing rules than one who must commute to what has been deemed an essential business.

Healthcare experts have been vocal from the start that there are certain individuals who are more vulnerable to severe effects of COVID-19 — older adults and those with serious underlying medical conditions. From a medical standpoint, this is logical enough. Certain chronic diseases and medications can weaken or suppress the immune system, making it more difficult to fight off infection. Older people have a higher prevalence of one or more chronic diseases , in addition to the natural effects of aging on the immune system .

But it is clear from the data that there are other, deeper, underlying factors at play. For one, a person’s susceptibility to chronic disease is in itself rooted in SDoH . Low-income communities are at a statistical disadvantage with regard to COVID-19 , which, among many social determinants, is likely attributed to situational factors such as limited access to testing and the occupational health hazards of low-wage, high-risk work. In New York City, the most densely populated borough also happens to be the wealthiest, affording many residents the opportunity to leave the city for second homes or rentals in lower risk areas.

Underscoring the gravity of these health disparities is the fact that some racial and ethnic minorities are contracting and dying of COVID-19 at vastly disproportionate rates. A new study from Yale has put a number to the impact, finding that black people are over 3.5 times more likely to die of COVID-19 than white people — Latinos, nearly twice as likely.

Much of the current data on COVID-19 is far from perfect. With regard to the aforementioned study, many states aren’t tracking the race and ethnicity of pandemic-related deaths, and those that are don’t account for age differences among population groups. Testing still lags in many areas, making it difficult to understand and assess the pandemic’s full impact. Determining who has died as an outcome of the virus is also a challenge — one Scientific American says is more likely to be resulting in underreporting than the converse. By these measures, with more widespread testing and reporting will likely come more sobering statistics.

Calling attention to the significance of social determinants to population health is hardly the discovery of a new phenomenon; we are merely seeing it play out through a crisis that can serve as a microcosm of the nation’s systemic health disparities.

As the emergence of a novel coronavirus has sounded the alarm on the urgent need to mitigate future outbreaks, the alarm should be ringing just as loudly on what this says about the state of healthcare — and how we must work to eradicate these disparities along with the virus.

Join the conversation and share your thoughts on COVID-19 and SDOH at Health Ideas .

Health Ideas is your source for information and inspiration to navigate the healthcare industry. Join us as we keep pace with the innovations, policy changes and trends that impact our nation and beyond.

HMS advances the healthcare system by helping organizations reduce costs and improve outcomes. Through our analytics and engagement solutions, we save billions of dollars annually while helping consumers lead healthier lives. HMS provides a range of payment accuracy and population health management solutions that help move the healthcare system forward.

Modern Healthcare A.M. Newsletter: Sign up to receive a comprehensive weekday morning newsletter designed for busy healthcare executives who need the latest and most important healthcare news and analysis.

Current News
Safety & Quality
Digital Health
Care Delivery
Digital Edition (Web Version)
Layoff Tracker
Sponsored Content: Vital Signs Blog
From the Editor
Nominate/Eligibility
100 Most Influential People
50 Most Influential Clinical Executives
40 Under 40
Best Places to Work in Healthcare
Healthcare Marketing Impact Awards
Innovators Awards
Diversity Leaders
Women Leaders
Digital Health Summit
Leadership Symposium
Social Determinants of Health Symposium
Women Leaders in Healthcare Conference
Best Places to Work Awards Gala
Diversity Leaders Gala
- Future of Staffing
- Health Equity & Environmental Sustainability
- Hospital of the Future
- Financial Resiliency
- Value Based Care
- Looking Ahead to 2025
Podcast - Beyond the Byline
Sponsored Podcast - Healthcare Insider
Sponsored Video Series - One on One
Sponsored Video Series - Checking In with Dan Peres
Data & Insights Home
Hospital Financials
Staffing & Compensation
Quality & Safety
Mergers & Acquisitions
Skilled Nursing Facilities
Data Archive
Resource Guide: By the Numbers
Data Points
Newsletters
People on the Move
Reprints & Licensing
Sponsored Content

Modeling the Impact of Social Determinants of Health on HIV

Original Paper
Open access
Published: 03 September 2021
Volume 25 , pages 215–224, ( 2021 )

Cite this article

You have full access to this open access article

Joseph W. Hogan ORCID: orcid.org/0000-0001-7959-7361 1 ,
Noya Galai 2 , 3 &
Wendy W. Davis 4

5617 Accesses

14 Citations

2 Altmetric

Explore all metrics

There is growing evidence for the key role of social determinants of health (SDOH) in understanding morbidity and mortality outcomes globally. Factors such as stigma, racism, poverty or access to health and social services represent complex constructs that affect population health via intricate relationships to individual characteristics, behaviors and disease prevention and treatment outcomes. Modeling the role of SDOH is both critically important and inherently complex. Here we describe different modeling approaches and their use in assessing the impact of SDOH on HIV/AIDS. The discussion is thematically divided into mechanistic models and statistical models, while recognizing the overlap between them. To illustrate mechanistic approaches, we use examples of compartmental models and agent-based models; to illustrate statistical approaches, we use regression and statistical causal models. We describe model structure, data sources required, and the scope of possible inferences, highlighting similarities and differences in formulation, implementation, and interpretation of different modeling approaches. We also indicate further needed research on representing and quantifying the effect of SDOH in the context of models for HIV and other health outcomes in recognition of the critical role of SDOH in achieving the goal of ending the HIV epidemic and improving overall population health.

Sociology and Health

Choosing an epidemiological model structure for the economic evaluation of non-communicable disease public health interventions.

Adam D. M. Briggs, Jane Wolstenholme, … Peter Scarborough

The Population Health Model (POHEM): an overview of rationale, methods and applications

Deirdre A. Hennessy, William M. Flanagan, … Douglas G. Manuel

Avoid common mistakes on your manuscript.

Introduction

The social determinants of health (SDOH) have been defined by the World Health Organization (WHO) as “the conditions in which people are born, grow, work, live, and age, and the wider set of forces and systems shaping the conditions of daily life” such as “economic policies and systems, development agendas, social norms, social policies and political systems”[ 1 ]. The WHO notes that these circumstances “are shaped by the distribution of money, power and resources” and are largely responsible for health inequities [ 1 ]. In the last two decades or so, investigators have come to appreciate that an individual’s vulnerability to disease is multi-factoral and subject to community, societal and environmental determinants [ 2 ].

Global pandemics such as the ongoing HIV pandemic, where SDOH play a central role in incidence and disease burden outcomes [ 3 ], present an important opportunity for utilizing models to inform investigators, practioners and policy-makers about how SDOH shape and impact epidemic and disease dynamics. The role of SDOH in the trajectory of the HIV epidemic has received increasing recognition, especially since the 2000s with a growing number of public health and epidemiologic studies designed to assess the role of SDOH through observational, intervention and randomized trial designs [ 4 , 5 , 6 ]. As the availability of data and newly developed measures around SDOH and inequitable health outcomes grows [ 7 ], so does the potential utility of disease and intervention models to illuminate the relationship between SDOH and indicators of individual and population health. Well-formulated models can be used to quantify and provide insight about the impact of planned or already implemented interventions at both the individual and community level.

This paper joins this AIDS and Behavior supplement, devoted to research conducted through the NIMH and NIAID funded RFA, Methodologies to Enhance Understanding of HIV-Associated Social Determinants [ 8 ], to offer a reflection on the specific ways in which models can be employed to elucidate relationships between SDOH and HIV outcomes. The paper provides a broad overview of two basic and foundational approaches, namely mechanistic and statistical models, and considers how two papers in this volume, that provide findings from modeling analyses, reflect the basic structures of these exemplars. While our review is divided thematically into mechanistic and statistical approaches to modeling, we recognize this distinction to be somewhat artificial as the approaches are not mutually exclusive. In fact the overlap can be substantial, and the coherent integration of methods that are traditionally labeled ‘mechanistic’ and ‘statistical’ is an active area of ongoing methodologic research. This paper speaks to the commonalities between mechanistic and statistical models and shows how they can be used to draw various kinds of inferences about the role of social determinants of health in HIV research.

Our discussion of these models is necessarily concise; hence, for comparison purposes, we focus on model structure, the sources of information that inform parameters (model inputs), and how the models are used to generate inferences (model outputs). Examples are given to illustrate the different modeling approaches. We discuss considerations for model choices and issues related to the generalizability of results from the various modeling approaches. Finally we describe connections between statistical and mechanistic models and give potential directions for future research.

Model Structure and Some Reasons for Using Models

To fix ideas for comparing the models, it is hepful to introduce some basic notation that will be used throughout. We assume that all models in our discussion have the basic form Y = M(X, θ, ε), where Y is an outcome of interest, X are inputs or covariates, θ is a set of model parameters, ε represents random variation (e.g. random error), and M is a model that describes how the inputs and parameters are related to Y. This is a generic description but it will be useful in organizing our discussion. A familiar example of a model is the normal-error linear regression Y = α + βX + ε, where ε is a normally-distributed error term having mean zero and standard deviation σ. In this case, Y is the output, X is the input, θ = (α, β, σ) is the set of parameters, and ε captures the random variation in Y after accounting for systematic variation explained by X. The model is the normal probability distribution function where the mean is α + βX and the standard deviation is σ. Footnote 1

Models have various types of uses and the ultimate choice of a model is based both on the information available and the intended use. Our focus here is on the use of models to understand the impact or role of SDOH on outcomes of interest such as the risk of HIV infection, and to understand the impact of interventions or policy changes that might be motivated by or targeted to considerations related to SDOH. It is also important to note that human interactions are central for understanding the dynamics of infectious diseases transmission and control. For example, suppose we are interested in the effect of needle exchange on HIV incidence in a population of injection drug users at risk for HIV. Consider a model of the form Y = M(X, A, θ,ε). In this case, Y is HIV incidence, A is needle exchange, X are individual- and community-level predictors of HIV incidence, and θ is a set of parameters that governs the relationship between A, X and Y via the model M.

There are two ways to use models to study this effect. The first is to use knowledge about the form of the model itself; that is, to use, for example, a set of mathematical equations that describes how HIV transmission takes place in a population having characteristics and behaviors captured by inputs X, and then to use that model to generate predictions of Y under different levels of A. In this case, the ‘inputs’ are knowledge about the mathematical form of the model (that is, model structure), and values of X that describe a population, or individuals in the population, including, for example, frequency of drug injection and likelihood of sharing injection equipment. An assessment of the effectiveness of needle exchange is made by simulating values of Y under different values of A; for example, simulating HIV incidence under assumptions that needle exchange is fully available (A = 1) or not at all available (A = 0). Such a mathematical model might also encode assumptions about how individuals interact with each other—information that would be difficult to obtain from observed data. Because this approach relies on the user specifying the mathematical form of the model that reflects the mechanism leading to HIV incidence—we call these models mechanistic. Mechanistic models also typically include assumptions about the values and distribution of some of the model parameters, often derived from knowledge or information gained from previous studies.

A second approach is to use what we call here statistical models, where individual-level data is used to estimate parameters of the model itself. In a statistical model, the model structure does not attempt to explicitly describe the underlying disease dynamics but rather the association between covariates and outcome, often assuming a general linear structure. In this case, we may have individual-level data on Y, A and X. For example, Rich et al. [ 9 ] recorded data on HIV incidence (Y) for individuals at outpatient treatment centers in two neighboring states (RI and MA) where, by law, access to needle exchange was substantially different. In this case, observed individual-level data are used to estimate the parameters (θ) of an instrumental variables model, which takes a much simpler form than the mechanistic model.

An instrumental variables model can be formulated as a simultaneous equations model with correlated error terms (one equation for the outome and a second for the exposure). In this context, the model would be fit under an assumption that any between-state differences in the outcome Y are assumed to be attributable solely to differences in needle exchange access. Rich et al. use this model to estimate the impact of access to needle exchange (A = 1 versus A = 0), on syringe re-use and sharing and demonstrate that access to needle exchange substantially reduces both. In fact a version of this model can be used to generate simulated outcomes under A = 1 and A = 0 as a way to assess the impact of needle exchange, which mimics how a mechanistic model might be used for the same purpose, a point we return to later in the paper. A key difference between the mechanistic model and the statistical model is that a single source of data is used to estimate the parameters of the statistical model, whereas multiple data sources might be used to determine fixed values or distributions of parameters for the more complicated mechanistic model. Additionaly, while mechanistic models attempt to represent multiple complexities or stages of a process leading to a health outcome Y, statistical models typically rely on simplified versions of data generating mechanisms.

In the next two sections we provide some additional description and concrete examples of mechanistic and statistical models and illustrate how they are used to assess the role of SDOH and to generate causal comparisons of interventions that are motivated by or closely related to SDOH. Where possible we make reference to the model structure described above.

Mechanistic Models

Mechanistic models are rooted in a mathematical representation of the mechanism driving the process of interest. Mechanistic models used to characterize population dynamics of HIV infection are often formulated to capture the population structure, infection transmission dynamics and stages of disease progression. Our discussion of mechanistic models uses examples of compartmental and agent-based models that have been applied to model HIV in a variety of populations and that incorporate SODH.

A major motivation for using mechanistic models to study SDOH is the need to characterize a complex system or process that cannot be studied using a single source of data. Hence the model inputs and information about model structure, in terms of key parameters, values or functional form of key mathematical components, can derive from multiple sources [ 10 ]. For some parameters there may be little or no a-priori information available; these values are typically tuned or fixed using a calibration process. To calibrate the model parameters, simulations from a version of the model having fixed parameter values are compared to data on a the outcome of interest, such as HIV incidence over a period of time, measured in the target population. Optimization or grid-search methods can be used to identify the parameter values for which simulated outcomes from the model align with observed data. Importantly, for a given model, there may be more than one set of parameter values for which simulations from the overall model are consistent with the data used for calibration; in other words, the post-calibration parameter values may not represent a unique solution [ 11 ].

Compartmental Mechanistic Models

Compartmental models of infectious disease dynamics provide a prototypical example of a mechanistic model. These mathematical models represent the process of interest assuming a structure that is driven by the definition of a set of disease-related states (compartments) and the rules that govern the transition between compartments. These models are often specified using a set of differential equations to describe for example transitions between different compartments that respresent disease states (e.g. susceptible, infected, recovered; or transitions between HIV disease states that reflect disease progression and are defined in terms of CD4 count and viral load [ 12 ].

Compartmental models can be specified to quantify the effects of demographic factors and SDOH at either the population or individual level including, for example, economic factors such as income distribution, educational attainment, and knowledge of disease risk, or to characterize a more refined representation of the disease process including incubation period, disease progression, treatment status and death [ 13 ]. Thus SDOH may be incorporated by defining compartments corresponding to specific sub-populations, by say education levels or race, or by allowing transition parameters to be influenced by social determinants such as access to care or stigma and discrimination. Dynamic compartmental models can be used to represent the changes in disease prevalence over time due to changes in transmission rates and or population-level dynamics such as migration patterns or proportion of antiretroviral treatment coverage.

A compartmental model that has been used for a variety of purposes is the Estimation and Projection Package Age-Sex Model (EPP-ASM) [ 14 ], implemented in the Spectrum software package [ 15 ]. This model is used in the paper by Jahagirdar et al. [ 16 ] in this supplement to derive country-specific estimates of HIV incidence over time in over 40 countries in Africa. Within the Spectrum model, incidence rate, which in our notation is the Y variable, is modeled as a function of several inputs that comprise the country-specific X variables: transmission rate among untreated HIV-infected individuals, HIV prevalence, proportion of HIV-positive individuals on ART, and the effectiveness of ART at reducing onward transmission. The model M(X, θ, ε) is a set of mathematical equations that describes how the X variables are related to HIV incidence; the values of the θ parameters have been derived from different sources. For a given set of inputs X, which the user needs to supply, the model can then be used to generate estimates or even simulations of the number of new cases that would be anticipated based on the model inputs. Notably the model inputs for this example are specified at the population level as opposed to the individual level.

Other examples of compartmental models that have been proposed to study HIV incidence and prevalence and their relationship to SDOH include Shannon et al. [ 17 ] who showed the impact of gender-based violence and discrimination of sex work on new HIV infections among female sex workers. Compartmental models can be also be used for post-hoc quantification of the contribution of individual components of a complex intervention by simulating hypothetical scenarios that are not likely to be reproducible in practice and would otherwise be difficult to isolate empirically. Examples include Nosyk et al. who demonstrated the impact of harm reduction services and ART coverage on averting new HIV infection in the population due to needle sharing [ 18 ].

Agent-Based Mechanistic Models

Individual or agent based models (ABM), also termed micro-simulation models, can be viewed as higher-resolution versions of mechanistic models and can be used to characterize and simulate outcomes based on individual-level behavior. ABM treat each individual in the population as unique and can incorporate or represent information about relationships between individuals.

In compartmental models such as the Jahagirdar et al. HIV incidence model described above [ 16 ], heterogeneity in a population is introduced by division of compartments into smaller subgroups, such as by age or risk profile (e.g. men who have sex with men (MSM), drug users, or those with pre-existing conditions) where each subgroup has possibly different transition rates between states; that is, the compartmental model is developed at the population level and assumes that individuals within each compartment are homogeneous (a top to bottom approach). By contrast, individual-based or microsimulation models introduce heterogeneity at the individual level whereby each individual can be assigned unique characteristics and risk profiles and a personal pattern of contacts with other individuals in the population (a bottom-up approach). In an individual-level model for HIV incidence, for example, each individual has their own probability of transmitting and contracting the infection. In our generic model representation, Y denotes a new HIV infection for an individual (e.g. 1 if yes, 0 if no). The inputs X can include both individual-level and population- or stratum-level variables.

Like population-level mechanistic models, ABM typically require simulation solutions to produce a projected output and often utilize multiple sources of data to obtain information on the values of model parameters and their distribution. Within the ABM framework, SDOH such as education, socio-economic status, racial or gender minority affiliation can be incorporated in individual characteristics that compose a simulated population and in turn, determine the interactions with other individuals in the population. Simulated output from agent-based models, then, produce overall epidemic dynamics and selected health outcomes under different assumptions on population structure and inter-personal interactions.

Similar to compartmental models, ABM have beeen used to assess the potential effects of complex interventions that cannot be easily studied using a single cohort or source of data. For example, Marshall et al. [ 19 ] used an ABM to model HIV incidence among interacting individuals that included injection drug users, non-injection drug users and non-drug users in New York City between 1992 and 2002. The model formula can be written generically as Y = M(X, A, θ, ε), where X represent individual-level characteristics, A denotes a system-level intervention such as availability of needle exchange, and the ε term represents the stochastic or probabilistic component of the model. Marshall et al. used this model to simulate individual HIV infection status within each subpopulation under various system-level interventions and used the output to generate population- and stratum-level HIV incidence trajectories. By generating simulated outcomes under different configurations for the interventions—which in the model correspond to different versions of the A variable—they demonstrated that the combination of syringe exchange and provision of ART could produce substantial reductions in HIV prevalence among injection drug users. This finding mirrored the results from several empirical studies [ 20 ]. An important feature of this ABM is that the model itself was parameterized to allow interactions between individuals under different network assumptions. Although typically there are limited data available to verify whether population interaction assumptions are correct, the model can be used to generate simulations under various assumptions about network structure in order to quantify the robustness of findings about intervention effects.

An additional example of agent-based modeling is carried out by Brookmeyer et al. [ 21 ], who used simulations from an agent-based model to evaluate the impact of 163 different HIV prevention packages comprising varying combinations and intensities of four prevention measures: percent of eligible persons who receive ART, percent reduction in unprotected anal intercourse (UAI), percent of eligible persons accepting PrEP, and a variable capturing increase in HIV testing. In empirical studies, these had only been examined one at a time. The analysis followed two steps: in the first step, simulation of data from an agent-based model where the population starts out with initial values of HIV prevalence, knowledge of HIV status, testing frequency, and frequency of sexual risk behavior. This model also makes assumptions about frequency of sexual contact based on an assumed social network structure, and about the probability of HIV transmission per contact. Individual-level outcomes are simulated forward in time.

In the second step, a regression model is applied to the simulated data to estimate the impact of varying the percent intensity of each prevention measure. The regression model used by Brookmeyer et al. [ 21 ] includes covariates representing percent uptake of the prevention measures listed above; fixing these values in different combinations encodes the 163 distinct prevention packages. Aside from the assumptions used to generate the synthetic data from the ABM, another critical set of assumptions is the specification of the impact of each intervention in the regression model. Brookmeyer et al. assume that, on the logit scale, the effects of ART uptake and UAI are additive and that the impact of PrEP is dependent on the rate of HIV testing. SDOH in this model were incorporated as factors impacting various prevention strategies such as ART and PrEP coverage and condom use, that are influenced by societal and cultural norms, stigma and discrimination as well as empowerment.

An advantage of this approach is that we can potentially gain insights about combinations of interventions that typically can only be tested one at a time in a randomized trial (if they can be tested at all). A key limitation is the strong dependence on modeling assumptions: the data generated by the ABM are synthetic, and the estimated impact of combination interventions depends on how each intervention is parameterized in the second-step regression model.

Statistical Models

Statistical models tend to have a less complex mathematical structure, and typically are used to draw inferences based on a single source of individual-level data from a target population (e.g. a cohort study or clinical trial). The most commonly used statistical models are regression models, which quantify associations between the inputs X and the outcome variable Y with the regression coefficients β. While it is possible to conduct a statistical analysis of a complex mathematical model, statistical models per se are not typically designed to represent population-level mechanistic processes. In contrast with mechanistic models that rely on calibration to tune parameter values, in statistical models the parameter estimates are usually derived by fitting the model to a single source of data using techniques such as maximum likelihood. For a defined statistical model and a given dataset, the parameter estimates resulting from the fitting process are typically unique. Under certain circumstances, statistical models also can be used to generate causal inferences and to assess the impact of interventions. One approach to accomplishing this goal is the g-computation algorithm, which serves as a conceptual connector between statistical and mechanistic models. We describe the g-computation algorithm below using the analysis conducted by Stoner et al. for this supplement, as an example [ 22 ].

Regression Models

The goal of a statistical regression model is to characterize explained and unexplained variation in one or more outcome variables Y based on data drawn from a target population. The explained variation is assumed to depend, through a regression function, on a set of predictor or explanatory variables X; for example, a regression model can be used to characterize variation in a disease outcome as a function of multiple explanatory risk factors such as age, gender, risk behaviors and factors related to SODH. Regression models provide estimates of the association between the explanatory variables and disease outcomes in the form of differences in means or risk, risk ratios, odds ratios, and hazard ratios, and can also be used to assess the potential impact of determinants via measues such as population attributable fraction. Regression models can be used to assess the effect of an SDOH or policy at the individual level as well as the community level.

An example of this kind of analysis is given by Kemp et al., who use a multilevel regression model to demonstrate the impact of ongoing experiences of HIV stigmatization on increased viral load among African-American women in primary HIV care [ 23 ]. Multilevel models are generalizations of standard regression models in that they build in error terms at each level of a clustering hierarchy; they are sometimes called random effects models, mixed effects models, or mixed models. They also can be used to estimate separately the effect of a covariate at each level of the hierarchy. By using a multilevel model that decomposes the overall effect of stigma on viral load, Kemp et al. were able to show that within-person changes in stigma over time do not lead to subsequent changes in viral load (within-individual effect), but that individuals having higher levels of stigma on average also have higher viral load on average (between-individual effect) [ 23 ]. This distinction between within- and between-individual effects is critical not only to understanding the mechanism by which SDOH might operate, but also for understanding how interventions might be designed.

The analysis conducted by Jahagirdar et al. in this volume borrows techniques from both regression modeling and mechanistic modeling [ 16 ]. In their model, simulated rates of HIV infection in over 40 countries in Sub-Saharan Africa, derived from the EPP-ASM model described above, are used as the outcome in a multilevel regression model to assess the impact of individual- and community-level SDOH on HIV risk [ 16 ]. The incidence rates predicted from EPP-ASM showed an overall decline in HIV incidence between 2000 and 2015. They then use a multilevel regression model to characterize variation in the predicted HIV incidence over time accounting for both between- and within-country variation in incidence, and the dependence of incidence on country-level SDOH covariates. Jahagirdar et al. use this approach to identify SDOH and other variables that explain the highest percentage of variation in country-specific HIV incidence rates. Average number of education years per capita and country-specific spending on HIV emerge as the factors explaining the greatest amount of variation in HIV incidence rates between countries [ 16 ].

Causal Models Derived From Regression Models

While regression models are formulated to capture the effect of predictor variables on an outcome of interest, the effect cannot generally be interpreted as causal. In this section, we discuss the construction and interpretation of causal structural models, and illustrate the use of one such model by Stoner et al. in this supplement to quantify the effect of child support grants (CSG) on HIV incidence among adolescent girls and young women [ 22 ].

Causal structural models are specified in terms of random variables called potential outcomes. For a two-level exposure or intervention, such as receipt of cash transfer or not, a causal model assumes that each individual has two potential outcomes: the outome that would be realized if the intervention is received, and the other that would be realized if not received. Unlike with other sorts of models, the potential outcomes formulation of the causal model assumes that both variables exist for each individual, even though only one can be observed [ 24 ]. (Potential outcomes are sometimes referred to as counterfactuals because for each individual, we can only observe the potential outcome corresponding to the actual exposure received; the other one is counterfactual.) From a statistical perspective, the goal is to draw inference about the difference or ratio of means between these two potential outcomes using observed data. The fundamental challenge is that, for each individual, only one of the potential outcomes can be observed. The process for drawing inference about causal effects from observed data can be driven either by design—i.e. by randomizing individuals to exposure or no exposure in a clinical trial design—or by analytic methods that are designed to balance confounding variables that make selection into the exposed and unexposed groups systematically different.

A design-based approach to causal inference is to randomize individuals to the exposure. Under randomization, for each individual, we are equally likely to observe either outcome and we can use the observed outcomes under each condition to estimate differences or ratios of the outcome of interest at the population level. If data were collected in an observational study instead, it is important to remember that girls and women who are at higher risk for HIV infection may also be more likely to receive cash transfers. Statistical methods used to estimate causal effects in settings where the exposure is not randomized are therefore designed to mimic randomization in some way by accounting for possible confounder imbalance. This can be accomplished by reweighting the sample according to inverse probability of receiving treatment [ 25 , 26 ], matching those who receive treatment to control individuals with similar observed characteristics [ 27 , 28 ], or making model-based adjustments.The propensity score, a summary of the probability of being exposed as a function of covariates, plays a fundamental role in many of these approaches [ 29 ]. There is a vast literature describing these methods and many others; a full review is beyond the scope of this discussion. Instead we focus on the g-formula method [ 30 ] used here by Stoner and colleagues.

Stoner et al. investigate the causal effect of cash transfer on HIV infection among adolescent girls and young women (AGYW) [ 22 ]. In the simplest formulation of their causal model, there are two potential outomes for each person: HIV infection status under the scenario that CSG is received, and HIV infection status under the alternate scenario that CSG is not received. Stoner et al. use a more complex version where CSG can vary over time, and have different intensities [ 22 ].

Implementation of the g-formula has many similarities to agent-based modeling [ 31 ]. The g-formula can be used to simulate potential outcomes under different versions of an intervention that are fixed by the investigator; the outcomes simulated under different scenarios are then used to quantify intervention effects. The process of simulation in the parametric g-formula bears similarity to how simulation-based inferences are generated in mechanistic models in the sense that individual component models are used to generate simulated outcomes of interest, such as HIV incidence, under different versions or intensities of an intervention. A key difference is that for the g-formula, the component models used to generate simulated potential outcomes are statistical models Footnote 2 —oftentimes regression models—that have been fit to a single source of observed data drawn from a population of interest. In practice, therefore, simulations from the g-formula typically derive from component models that are less mathematically complex but have closer fit to a representative sample of observed data. Mechanistic models are more complex representations of disease dynamics, but the data come from sources that may be less representative of a specific population.

Stoner et al. use the g-formula to examine receipt of CSG and other interventions [ 22 ]. At each time point, there are data available on HIV incidence (Y), receipt of CSG (A), and confounding variables (X). In this example, implementation of the g-formula proceeds in three steps: first, for each time point, fit a regression of HIV incidence as a function of CSG (yes/no) and the confounders. This regression, which can be represented in the format Y = M(X,A,θ, ε) where X are the confounders and A is the intervention, is the component model on which simulations of potential outcomes under different intervention combinations will be generated. Second, once the models are fitted to observed data, fix covariate values X to represent the population of interest, and fix the intervention sequence (the value of A at each time point) for which HIV incidence is to be estimated. Third, for this fixed version of the intervention sequence, generate simulated or predicted values of HIV incidence from the fitted regression models. In this way, the fitted regression models are playing the same role as the mechanistic models when it comes to simulating outcomes under different intervention scenarios.

Using this basic approach, Stoner et al. [ 22 ] can compare various intensities of CSG, such as all-versus-none receipt of CSG and all-versus-observed receipt of CSG. The latter comparison quantifies the effect of increasing the observed CSG coverage (around 75%) to having everyone receive CSG. With suitably rich data, this general strategy can be used to quantify the impact of more complex interventions or, as Stoner et al. do, to compare the interactive effect of interventions with other factors [ 22 ]. Their analysis shows the potential for combining monthly child support grants with interventions to increase parental care and reduce depression can lead to substantial reductions in HIV incidence among AGYW, and that these effects are not realized through cash grants alone [ 22 ].

As both the decades-old HIV global pandemic and the more recent SARS-CoV-2 pandemic demonstrate, SDOH can play a central role in the transmission, morbidity and mortality of an infectious disease. The models described here offer a range of tools that can help to elucidate the interplay between social and structural determinants and the expression of an infectious disease among individuals as well as the public health burden in the population. The presentation here is a broad overview of different established approaches to modeling disease outcomes, with examples that focus on HIV and SDOH. This is not meant to be a comprehensive toolbox. Models can offer insights into how an intervention might function at an individual level such as the causal model shared by Stoner et al. [ 22 ] as well as the societal level, such as the hybrid model developed by Jahagidar and colleagues [ 16 ], using information from mechanistic models as inputs to a statistical regression analysis of country-level exposures. These models allow for a simulation of outcomes under different versions of, intensity of, or combination of interventions that would be difficult to gain through designed studies such as randomized trials. While the ultimate goal of all the models discussed in this paper is to better understand the causal relationship between exposures and outcome, models that implement simulations of outcomes under various exposures address causality more directly compared to simpler regression models. In the case of Stoner et al., the availability of high quality individual-level data from an intervention trial allowed the investigators to examine complex versions of a time-varying intervention and its potential interaction with other factors [ 22 ]. For Jahagidar et al., the generation of country-level comparisons offered insights into the impact of programs and policies far beyond the potential scope or feasibility of any designed interventional study [ 16 ].

Many factors contribute to the validity of model-based inferences about SDOH. In this paper we have focused on model specification, model inputs (i.e. the data or information used to generate outputs from the model), and the use of models to assess the impact of interventions. Both mechanistic and statistical models rely on a representation of the underlying data generating process given in mathematical and probabilistic terms. Many compartmental models, for example, are written in terms of differential equations that describe the probability of transition from one compartment or disease state to the next; regression models describe variation of an outcome in terms of explained and unexplained variation, where explained variation is the regression function and unexplained variation, quantified by the error term, follows a probability model such as the normal distribution.

In nearly all cases, even the most complex model will not be an accurate or complete representation of the system or phenomenon that it is being studied. The models described in this paper, for example, are designed to characterize outcomes such as HIV infection as a function of SDOH and possible interventions, but do not address the dynamic processes that give rise to the SDOH themselves. However, models offer a useful insight into the potential impacts of selected factors, including various SDOH, on disease outcomes. Models must strike a difficult balance between interpretability, face validity, and fidelity to an observed-data process. The first two of these criteria are largely subjective. Regarding fidelity to observed-data processes, mechanistic models and statistical causal models are themselves built up from smaller submodels. While these submodels can be checked for lack of fit against observed data, the larger model relies on assumptions that tie the submodels together and typically cannot be validated against a single sample of data.

The validity of model-based inferences also depend on inputs. Parameter values for mechanistic models typically are informed by multiple sources that may even be derived from different populations. Statistical models tend to rely on single samples of data drawn from the target population. This distinction can be important. As Murray et al. demonstrate in a comparison of agent-based and statistical models for estimating causal effects, even when both models are perfect representations of the underlying system, using inputs from different populations—as is done with ABM and other mechanistic models—can induce unintended confounding and biased estimates of causal effects [ 31 ]. This is an especially important consideration when estimating effects of SDOH, particularly if the SDOH inputs are derived from substantively different populations than the confounding variables.

Uncertainty associated with model-based inferences has many sources. In statistical models, the most obvious is sampling variation, captured in terms of standard errors and confidence intervals. For mechanistic models, parameter inputs derived from published studies carry uncertainty because they usually correspond to estimates from other studies, which themselves have associated standard errors. This uncertainty can be represented by using distributions instead of fixed parameter values as model inputs, as is done for the Thembisa model of the HIV epidemic in South Africa [ 32 ]. The Bayesian approach to inference treats the model parameters as random variables. It requires specification of a probability model for the outcomes (what we have been referring generically to as M(X, θ, ε)) and prior distributions for the parameters θ in the probability model; inference is based on the posterior distribution of the parameters given the observed data. Posterior variation in the parameter values and model predictions reflects both the prior uncertainty about parameter values and sampling variability in the observed data. Bayesian methods can be used to fit both mechanistic and statistical models; however, they are particularly useful for mechanistic models that can be specified in terms of a likelihood function because they provide a formal way to encode existing information about model parameters via the prior distribution. An outstanding and timely example is the Bayesian model developed by Flaxman et al. [ 33 ] to quantify the impact of non-pharmaceutical interventions on COVID-19 in Europe.

Untestable assumptions are an often overlooked source of uncertainty. Statistical causal models rely by necessity on the assumption that all relevant confounders have been measured (the ‘no unmeasured confounding’ or ‘treatment ignorability’ assumption), but there is no way to verify whether or not this assumption holds. Uncertainty about untestable assumptions should be examined in sensitivity analyses, which serve to quantify the robustness of inferences from causal models—whether statistical or mechanistic. Sensitivity analyses can take many forms [ 34 , 35 ], and a critique of assumptions about both mechanisms and confounding can be guided by the use of a graphical model [ 36 ].

Finally an often overlooked source of uncertainty is the quality of measurement and design. The value of any given model is tied to the quality of its input and the rigor of its design [ 37 ]. The interpretation of modelled epidemiological scenarios must include a critical assessment of the accuracy of measurement and ascertainment of exposure and outcomes, and how any data source is calibrated and considered against other inputs.

On a more general note, as Geffen and Welte explain, the nature of a model world, namely a “conceptual realm”, which is constructed around an understanding of “real world processes," should be comprehensible [ 38 ]. They argue that while the technical construction of a model may be complex and understood by relatively few, model worlds should be accessible to all that work within that domain [ 38 ]. For compartmental or micro-simulation models, users should understand the basic rationale behind the mathematical model used to represent population dynamics and the validity of parameter values that populate the model. A full systematic reporting of the evidence synthesis and model assumptions is critical for models to be useful tools for policy decision making [ 14 ]. For causal models, users should understand the nature of key assumptions such as ‘no unmeasured confounding’ and how they are applied in the specific context.

One way to ensure that models are comprehensible and that they reflect a theoretically valid model world is to further integrate the technical development of models with longstanding and rigorous research on the SDOH. The rich theoretical foundation of research on the SDOH and their association with disease processes and outcomes should also be reflected in any mathematical or statistical model development. This integration of theoretical research in sociology, especially of more complex concepts such as systemic racism and discrimination, with statistical model development presents a real challenge. Developing valid measurement of SDOH at both the individual level, assessing exposure to and impact of social determinants, as well as quantifying differences at the community level, such as police or justice system discrimination, are critical. Models of SDOH will also be strengthened by the use of high-quality inputs. Collaborations with investigators working in social behavioral science can ensure that modellers use well regarded measures and inputs as ingredients in the models they construct. There is a broad need then for models that are grounded in social behavioral theory and draw on data from a diversity of sources with indicators that are appropriate for the model’s proposed context and use. As epidemics unfold and mature, there is a significant opportunity for the application of models as we work to understand how communities and policies impact disease outcomes and how shifting realities in individual circumstance may support or hinder optimal outcomes.

Data Availability

Not applicable.

Code Availability

The model itself is written mathematically as p(Y | X,α,β,σ) = { (2)} −1 exp{—(Y—α—βX) 2 / (2σ 2 )}, which describes the relative probability of observing the value Y for a fixed value of X.

Strictly speaking, the g formula does not require the component models to fitted statistical models. However in epidemiologic practice this is the most common method of applying the g formula.

World Health Organization. Social Determinants of Health. https://www.who.int/teams/social-determinants-of-health . Updated 2020. Accessed November 17, 2020.

Dean HD, Fenton KA. Integrating a social determinants of health approach into public health practice: a five-year perspective of actions implemented by CDC’s national center for HIV/AIDS, viral hepatitis, STD, and TB prevention. Public Health Rep. 2013;128(3):5–11. https://doi.org/10.1177/00333549131286S302 .

Article PubMed PubMed Central Google Scholar

Kerrigan D, Donastorg Y, Barrington C, et al. Assessing and addressing social determinants of HIV among female sex workers in the dominican republic and tanzania through community empowerment-based responses. Curr HIV/AIDS Rep. 2020;17:88–96.

Article Google Scholar

Abgrall S, Del Amo J. Effect of sociodemographic factors on survival of people living with HIV. Curr Opin HIV AIDS. 2016;11(5):501–6. https://doi.org/10.1097/COH.0000000000000301 .

Article PubMed Google Scholar

Valdiserri RO. The evolution of HIV prevention programming: moving from intervention to system. AIDS Educ Prev. 2018;30(3):187–98. https://doi.org/10.1521/aeap.2018.30.3.187 .

Friedman EE, Dean HD, Duffus WA. Incorporation of social determinants of health in the peer-reviewed literature: a systematic review of articles authored by the national center for HIV/AIDS, viral hepatitis, STD, and TB prevention. Public Health Rep. 2018;133(4):392–412. https://doi.org/10.1177/0033354918774788 .

Golembiewski E, Allen KS, Blackmon AM, Hinrichs RJ, Vest JR. Combining nonclinical determinants of health and clinical data for research and evaluation: rapid review. JMIR Public Health Surveill. 2019;5(4):e12846. https://doi.org/10.2196/12846 .

https://grants.nih.gov/grants/guide/rfa-files/RFA-MH-16-200.html . Accesssed April 30, 2021

Rich JD, Hogan JW, Wolf F, et al. Lower syringe sharing and re-use after syringe legalization in Rhode Island. Drug Alcohol Depend. 2007;89(2–3):292–7. https://doi.org/10.1016/j.drugalcdep.2007.02.016 ( PMID: 17386980 ).

Johnson LF, Kubjane M, Eaton JW. Challenges in estimating HIV prevalence trends and geographical variation in HIV prevalence using antenatal data: Insights from mathematical modelling. PLoS ONE. 2020;15(11):e0242595.

Article CAS Google Scholar

Eaton JW, Hallett TB. Why the proportion of transmission during early-stage HIV infection does not predict the long-term impact of treatment on HIV incidence. PNAS. 2014;45:16202–7.

Ying R, Sharma M, Celum C, et al. Home testing and counselling to reduce HIV incidence in a generalised epidemic setting: a mathematical modelling analysis. Lancet HIV. 2016;3(6):e275–82. https://doi.org/10.1016/S2352-3018(16)30009-1 ( PMID: 27240790 ).

Bjørnstad ON, Shea K, Krzywinski M, Altman N. The SEIRS model for infectious disease dynamics. Nat Methods. 2020;17(6):557–8. https://doi.org/10.1038/s41592-020-0856-2 .

Article CAS PubMed Google Scholar

Eaton JW, Brown T, Puckett R, et al. The estimation and projection package age-sex model and the r-hybrid model: new tools for estimating HIV incidence trends in sub-Saharan Africa. AIDS. 2019;33(Suppl 3):S235–44. https://doi.org/10.1097/QAD.000000000000243 .

Stover J, Glaubius R, Mofenson L, et al. Updates to the Spectrum/AIM model for estimating key HIV indicators at national and subnational levels. AIDS. 2019;33(Suppl 3):S227–34. https://doi.org/10.1097/QAD.0000000000002357 .

Jahagirdar D, Walters M, Vongpradith A, et al. Incidence of HIV in Sub-Saharan Africa, 2000–2015: The interplay between social determinants and behavioral risk factors. AIDS Behav. 2021. https://doi.org/10.1007/s10461-021-03279-9 .

Shannon K, Strathdee SA, Goldenberg SM, et al. Global epidemiology of HIV among female sex workers: influence of structural determinants. Lancet. 2015;385(9962):55–71. https://doi.org/10.1016/S0140-6736(14)60931-4 .

Nosyk B, Zang X, Min JE, et al. Relative effects of antiretroviral therapy and harm reduction initiatives on HIV incidence in British Columbia, Canada, 1996–2013: a modelling study. Lancet HIV. 2017;4(7):e303–10. https://doi.org/10.1016/S2352-3018(17)30045-0 .

Marshall BD, Friedman SR, Monteiro JF, Paczkowski M, Tempalski B, Pouget ER, Lurie MN, Galea S. Prevention and treatment produced large decreases in HIV incidence in a model of people who inject drugs. Health Aff. 2014;33(3):401–9. https://doi.org/10.1377/hlthaff.2013.0824 ( PMID: 24590937; PMCID: PMC4469974 ).

Marshall BD, Paczkowski MM, Seemann L, et al. A complex systems approach to evaluate HIV prevention in metropolitan areas: preliminary implications for combination intervention strategies. PLoS One. 2012;7(9):e44833. https://doi.org/10.1371/journal.pone.0044833 .

Article CAS PubMed PubMed Central Google Scholar

Brookmeyer R, Boren D, Baral SD, et al. Combination HIV prevention among MSM in South Africa: results from agent-based modeling. PLoS One. 2014;9(11):e112668. https://doi.org/10.1371/journal.pone.0112668 .

Stoner MCD, Edwards J, Westreich D, et al. Modeling cash plus other psychosocial and structural interventions to prevent HIV among adolescent girls and young women in South Africa (HPTN 068). AIDS Behav. 2021. https://doi.org/10.1007/s10461-021-03158-3 .

Kemp CG, Lipira L, Huh D, et al. HIV stigma and viral load among African-American women receiving treatment for HIV. AIDS. 2019;33(9):1511–9. https://doi.org/10.1097/QAD.0000000000002212 .

Rubin DB. Estimating causal effects of treatments in randomized and nonrandomized studies. J Educ Psychol. 1974;66(5):688–701. https://doi.org/10.1037/h0037350 .

Robins JM, Hernán MA, Brumback B. Marginal structural models and causal inference in epidemiology. Epidemiology. 2000;11(5):550–60. https://doi.org/10.1097/00001648-200009000-00011 .

Hernán MA, Robins JM. Estimating causal effects from epidemiological data. J Epidemiol Community Health. 2006;60(7):578–86. https://doi.org/10.1136/jech.2004.029496 .

Stuart EA. Matching methods for causal inference: A review and a look forward. Stat Sci. 2010;25(1):1–21. https://doi.org/10.1214/09-STS313 .

Rosenbaum PR. Optimal matching for observational studies. J Am Stat Assoc. 1989;84(408):1024–32. https://doi.org/10.2307/2290079 .

Rosenbaum PR, Rubin DB. The central role of the propensity score in observational studies for causal effects. Biometrika. 1983;70(1):41–55. https://doi.org/10.2307/2335942 .

Young JG, Cain LE, Robins JM, O’Reilly EJ, Hernán MA. Comparative effectiveness of dynamic treatment regimes: an application of the parametric g-formula. Stat Biosci. 2011;3(1):119–43. https://doi.org/10.1007/s12561-011-9040-7 .

Murray EJ, Robins JM, Seage GR, Freedberg KA, Hernán MA. A comparison of agent-based models and the parametric G-Formula for causal inference. Am J Epidemiol. 2017;186(2):131–42. https://doi.org/10.1093/aje/kwx091 .

Johnson L. THEMBISA version 1.0: A model for evaluating the impact of HIV/AIDS in South Africa. https://www.thembisa.org/content/filedl/WPversion1 . Centre for Infectious Disease Epidemiology and Research working paper. Published February 2014. Accessed November 17, 2020.

Flaxman S, Mishra S, Gandy A, Unwin HTJ, Mellan TA, Coupland H, Whittaker C, Zhu H, Berah T, Eaton JW, Monod M, Ghani AC, Donnelly CA, Riley S, Vollmer MAC, Ferguson NM, Okell LC, Bhatt S. Estimating the effects of non-pharmaceutical interventions on COVID-19 in Europe. Nature. 2020;584:257–26. https://doi.org/10.1038/s41586-020-2405-7 .

Vanderweele TJ, Arah OA. Bias formulas for sensitivity analysis of unmeasured confounding for general outcomes, treatments, and confounders. Epidemiology. 2011;22(1):42–52. https://doi.org/10.1097/EDE.0b013e3181f74493 .

Rosenbaum PR. Two R Packages for Sensitivity Analysis in Observational Studies. Obs Stud. 2015;1:1–17. https://obsstudies.org/two-r-packages-for-sensitivity-analysis-in-observational-studies/ . Accessed November 17, 2020.

Pearl J. Causal inference in the health sciences: a conceptual introduction. Health Serv Outcomes Res Methodol. 2001;2:189–220. https://doi.org/10.1023/A:1020315127304 .

Lawlor DA, Davey Smith G, Kundu D, Bruckdorfer KR, Ebrahim S. Those confounded vitamins: what can we learn from the differences between observational versus randomised trial evidence? Lancet. 2004;363(9422):1724–7. https://doi.org/10.1016/S0140-6736(04)16260-0 .

Geffen N, Welte A. Modelling the human immunodeficiency virus (HIV) epidemic: A review of the substance and role of models in South Africa. South Afr J HIV Med. 2018;19(1):756. https://doi.org/10.4102/sajhivmed.v19i1.756 .

Download references

National institutes of health (R01-AI-108441, P30-AI-42853, R01 MH 110158) .

Author information

Authors and affiliations.

Department of Biostatistics, Brown University School of Public Health, Providence, RI, USA

Joseph W. Hogan

Department of Epidemiology, Johns Hopkins Bloomberg School of Public Health, Baltimore, MD, USA

Department of Statistics, University of Haifa, Mt Carmel, Israel

Department of Prevention and Community Health, the George Washington University Milken Institute of Public Health, Washington, DC, USA

Wendy W. Davis

You can also search for this author in PubMed Google Scholar

Contributions

Corresponding author.

Correspondence to Joseph W. Hogan .

Ethics declarations

Conflict of interest, ethical approval, consent to participate, consent for publication, additional information, publisher's note.

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Rights and permissions

Reprints and permissions

About this article

Hogan, J.W., Galai, N. & Davis, W.W. Modeling the Impact of Social Determinants of Health on HIV. AIDS Behav 25 (Suppl 2), 215–224 (2021). https://doi.org/10.1007/s10461-021-03399-2

Download citation

Accepted : 17 July 2021

Published : 03 September 2021

Issue Date : November 2021

DOI : https://doi.org/10.1007/s10461-021-03399-2

Share this article

Anyone you share the following link with will be able to read this content:

Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative

Social determinants of health
Mechanistic models
Causal models
Statistical models
Find a journal
Publish with us
Track your research

Social Determinants of Health at CDC

Social determinants of health (SDOH) are the nonmedical factors that influence health outcomes. They are the conditions in which people are born, grow, work, live, and age, and the wider set of forces and systems shaping the conditions of daily life. These forces and systems include economic policies and systems, development agendas, social norms, social policies, racism, climate change, and political systems. Centers for Disease Control and Prevention (CDC) has adopted this SDOH definition from the World Health Organization .

SDOH are one of three priority areas for Healthy People 2030 , along with health equity and health literacy. Healthy People 2030 sets data-driven national objectives in five key areas of SDOH : healthcare access and quality, education access and quality, social and community context, economic stability, and neighborhood and built environment. Some examples of SDOH included in Healthy People 2030 are safe housing, transportation, and neighborhoods; polluted air and water; and access to nutritious foods and physical health opportunities.

Public health organizations can convene, integrate, influence, and contribute to big changes.

CDC has taken multiple steps to ensure efforts to address SDOH are built into the agency’s work.

CDC SDOH Resources

National Center for HIV, Viral Hepatitis, STD, and TB Prevention (NCHHSTP)
Public Health Infrastructure Center (PHIC)
National Center for Chronic Disease Prevention and Health Promotion (NCCDPHP)
Adverse Childhood Experiences at the National Center for Injury Prevention and Control (NCIPC)
Health Impact in 5 Years (OPPE)

Additional SDOH Resources

Healthy People 2030
World Health Organization
White House SDOH Playbook
HHS Call to Action

SDOH Publications

Hacker K, Auerbach J, Ikeda R, Philip C, Houry D; SDOH Task Force. Social determinants of health—an approach taken at CDC . J Public Health Manag Pract . 2022;28(6):589-594. doi: 10.1097/PHH.0000000000001626
Hacker K, Houry D. Social needs and social determinants: the role of the Centers for Disease Control and Prevention and public health . Public Health Rep . 2022; Sep 9:00333549221120244.
JPHMP Direct. CDC’s Approach to Social Determinants of Health. Accessed October 31, 2022. https://jphmpdirect.com/2022/10/07/cdcs-approach-to-social-determinants-of-health/

To receive email updates about this page, enter your email address:

CDC Newsroom
CDC Timeline
Gateway to Health
Ethics Program

Exit Notification / Disclaimer Policy

The Centers for Disease Control and Prevention (CDC) cannot attest to the accuracy of a non-federal website.
Linking to a non-federal website does not constitute an endorsement by CDC or any of its employees of the sponsors or the information and products presented on the website.
You will be subject to the destination website's privacy policy when you follow the link.
CDC is not responsible for Section 508 compliance (accessibility) on other federal or private website.

Health equity news from Boston Medical Center

Community & Social Health

What Are Social Determinants of Health?

"We recognize that until we can reduce the negative impact of a person's environment, we can't get them healthy," says David Henderson.

Social determinants of health (SDoH) are factors in a person’s life and environment that impact their health outcomes in positive or negative ways. These economic and social conditions are all connected and can have a detrimental snowball effect for individuals and entire populations.

There are numerous social determinants of health , but they all fall within six major categories:

Financial wellness
Education and employment
Community safety
Immigration status
Food and nutrition
Housing security

Let’s take the first one as an example: financial wellness. As anyone who has experienced economic hardship will tell you, you constantly worry, “Where is my next meal going to come from?” or “How am I going to feed my kids?” It’s extremely stressful to get up every day with that burden, and that chronic stress is a direct path to early heart attack, stroke, hypertension, depression, and anxiety—all because of just that one factor.

Social determinants of health deficits have longstanding impact

What’s even more powerful is that your stress today has a negative impact on your health decades later. For example, we have known for a long time that when young children experience even a brief period of food insecurity, they suffer health and mental health consequences throughout their life, and they can actually pass this risk on to their own children. In other words, these experiences may activate or deactivate genes and have a negative health effect, which is known as epigenetics.

What’s even more powerful is that your stress today has a negative impact on your health decades later.

The U.S. government is supposed to help and support our most vulnerable, but it instead can keep millions of people with disabilities in a chronic state of poverty and deprivation, which goes against the notion of getting people well. For instance, as explained by a Talk Poverty article , if you are disabled and unable to work, the U.S. government may give you $800 a month. But if your rent is $500 and your bills are $200, you’re left with $100 to eat for the month. That’s impossible. Yet the moment you try to break out of that system by working part time, the government starts offering less money or taking money back. It’s hard to get out and people are in a chronic state of stress.

Finding solutions for our patients

At Boston Medical Center (BMC), we add a social determinants of health screener to every patient’s medical record, which helps our providers remember to ask about things like housing and transportation.

The traditional hospital model normally doesn’t spend a lot of time on these areas, but we recognize that until we can reduce the negative impact of a person’s environment, we can’t get them healthy. If during screening, we identify an unmet social need, we can help that patient access services to get food or move them to a better living situation. We can find solutions.

The Health Equity Accelerator at Boston Medical Center

With the launch of the Health Equity Accelerator , we are addressing SDoH on a systemic level with intentionality—turbo-charging the foundational work and focus already in existence across BMC and community partners. The Health Equity Accelerator was created by a group of nearly 100 clinicians, researchers, and administrators at BMC to help eliminate the health equity gap that exists in the U.S. Broadly, our work will roll out in three phases:

Identify the health disparities within our own patient population. For every disparity we identify, we will also research what factors are contributing. We have to first understand the problem to know how to fix it.
Come up with solutions. Again, we’ll do the research to make sure each intervention we recommend indeed achieves a better outcome.
Create new medical literature. For every disease and disorder doctors treat, it’s almost guesswork with our patients, because studies have never been done on them. For example, the literature may say that Black people have high rates of hypertension because of too much salt in the diet, but maybe a bigger factor is their being exposed to violence in their neighborhoods.

We’re a safety-net hospital, and we’re also an academic medical center, so we must bring the data, look at it, and develop the right interventions for populations with health inequities. We feel a great responsibility to improve health outcomes for the most vulnerable. We’re excited about this work. We just have to get it right.

Share on Social

New Study Highlights the “Profound Impacts” of Limited Healthcare Access for Children Exposed to Violence

by HealthCity Editors | Mar 29, 2024 | Community & Social Health

Experts from Boston Medical Center stress the importance of leaning on evidence-based interventions, such as violence prevention programs, to help these children.

How Boston Medical Center’s Outreach Built Vaccine Confidence in Underserved Communities

by HealthCity Editors | Mar 7, 2024 | Community & Social Health

The health system’s comprehensive communications plan helped 76% of Black and 75% of Latino/a primary care patients get fully vaccinated.

About the Author

David C. Henderson, MD

David C. Henderson is Psychiatrist-in-Chief at Boston Medical Center.

HealthCity is dedicated to sharing the most cutting-edge ideas for advancing health justice through inspiring stories, in-depth Q&As with equity pioneers, research news that could change the medical landscape, and more.

An official website of the United States government

The .gov means it’s official. Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

The site is secure. The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

Publications
Account settings

Preview improvements coming to the PMC website in October 2024. Learn More or Try it out now .

Advanced Search
Journal List
JMIR Form Res
PMC10365596

Assessment and Documentation of Social Determinants of Health Among Health Care Providers: Qualitative Study

Brooks yelton.

1 Department of Health Promotion, Education, and Behavior, Arnold School of Public Health, University of South Carolina, Columbia, SC, United States

Jancham Rachel Rumthao

Mayank sakhuja, mark m macauda.

2 Center for Applied Research and Evaluation, Arnold School of Public Health, University of South Carolina, Columbia, SC, United States

Lorie Donelle

3 College of Nursing, University of South Carolina, Columbia, SC, United States

Michelle A Arent

Xueying yang.

4 SC SmartState Center for Healthcare Quality, Department of Health Promotion, Education, and Behavior, Arnold School of Public Health, University of South Carolina, Columbia, SC, United States

Xiaoming Li

Samuel noblet, daniela b friedman, associated data.

Interview guide.

Barriers and facilitators of social determinants of health (SDOH) assessment and documentation.

The data sets generated during and/or analyzed during this study are available from the corresponding author on reasonable request.

Research clearly demonstrates social determinants of health (SDOH) impact health outcomes. Provider consideration of patient SDOH in prevention and treatment planning is critical for improved health care quality and health equity. Despite awareness of the connections between SDOH and improved population health, research demonstrates few providers document patient SDOH.

This qualitative study aimed to better understand the barriers and facilitators of SDOH assessment, documentation, and referral in different health care settings and roles.

Individual semistructured interviews were conducted with practicing health care providers in South Carolina between August 25, 2022, and September 2, 2022. Participants were recruited via community partners’ web-based newsletters or listservs using a purposive sampling design. An interview guide with 19 questions was used to explore the following research question: How do SDOH impact patient health and what are the facilitators and barriers experienced by multidisciplinary health care providers assessing and documenting patient SDOH?

Participants (N=5) included a neonatal intensive care unit registered nurse, a nurse practitioner, a certified nurse midwife, a family and preventive medicine physician, and a counselor (licensed clinical social worker) with careers spanning 12 to 32 years. Participant responses are presented according to the following 5 themes: participants’ understanding of SDOH for the patient population, assessment and documentation practices, referrals to other providers and community-based resources, barriers and facilitators of SDOH assessment and documentation, and SDOH assessment and documentation training preferences. Overall, participants were aware of the importance of including patient SDOH in assessment and intervention but noted a variety of institutional and interpersonal barriers to assessment and documentation, including time constraints, perceptions of stigma around discussion of SDOH, and limited referral protocols.

Conclusions

Incentivizing inclusion of patient SDOH in health care must be facilitated from the top down, so assessment and documentation can be universally implemented in a pragmatic way that works for providers in a variety of roles and settings for the betterment of health care quality, health equity, and improved population health outcomes. Partnering with community organizations can serve to augment health care organizations’ resource and referral availability for addressing patients’ social needs.

Introduction

Social determinants of health (SDOH) are defined by the United States’ Healthy People 2030 (HP 2030) as “the conditions in the environments where people are born, live, learn, work, play, worship, and age that affect a wide range of health, functioning, and quality-of-life outcomes and risks.” HP 2030’s data-driven national objectives categorize SDOH into 5 fundamental domains: economic stability, education access and quality, health care and quality, neighborhood and built environment, and social and community context [ 1 ]. Increased research interest in SDOH among the public health and medical community is driven by evidence SDOH clearly impact health outcomes. Between 30% and 55% of health outcomes are ascribed to SDOH [ 2 ].

Population health has become an increasingly critical focus of health care delivery, especially during the COVID-19 pandemic [ 3 ]. SDOH factors allow clinicians to consider potential contributors to poor health outcomes, reduce health disparities, and transform health care delivery through partnerships with community-based resources [ 4 ]. Using nonbillable diagnostic “Z codes”—specific to SDOH—provided by International Classification of Diseases, 10th Revision, Clinical Modification in electronic medical record (EMR) systems provides clinicians and their medical teams with the opportunity to identify and document SDOH issues, establish appropriate intervention plans, reduce costs, and enhance health care delivery with an eye toward improving health equity [ 5 , 6 ]. To improve health care quality and quality of life, structural health perspectives that recognize and engage with the experiences and backgrounds of diverse individuals are critical in closing equity gaps and reducing health disparities [ 7 , 8 ].

An empirical study of clinical application of Z codes in South Carolina before and during the COVID-19 pandemic found Z codes were only documented for 1.23% of patients in a large statewide sample [ 9 ]. This research demonstrated an overwhelming lack of attention to the social context of (ill) health. This was especially true for individuals requiring outpatient care relative to individuals who were hospitalized. Although EMR systems are an efficient method for clinical data collection, the uncertainty in nonclinical diagnostic Z code use and potential differences in assessment tools and data collection methods limit SDOH information quality [ 10 ]. Furthermore, time constraints associated with the medical care model, varied provider conceptualization of SDOH, institutional practices and priorities, and inconsistent knowledge or availability of referral resources inhibit universal assessment and documentation practice. A recent study found higher awareness regarding the need for documenting SDOH among providers in community health centers, social care–associated payment models, and those with greater knowledge of advanced functions within EMR systems [ 11 ]. Health service organizations have attempted to promote clinician documentation and implementation of SDOH diagnostic codes by providing improved SDOH standardized terminology and data principles for which Z codes are categorized [ 12 ]. However, data regarding the effectiveness and barriers of these efforts are limited.

This qualitative study aimed to examine health care providers’ experiences with assessment and documentation of the SDOH impacting patient health and well-being. Through in-depth interviews, our goal was to better understand the barriers and facilitators of SDOH assessment, documentation, and referral in different health care settings and roles.

Study Design

Qualitative research methods were used to address the following research question: How do SDOH impact patient health and what are the facilitators and barriers experienced by multidisciplinary health care providers assessing and documenting patient SDOH?

In total, 7 team members (BY, JRR, DBF, MAA, MMM, MS, and SN) and 2 clinical partners assisted in the development of a recruitment plan to leverage clinical-academic-community partner web-based newsletters or listservs in a purposive sampling design [ 13 ] deployed between July 12, 2022, and August 30, 2022; the opportunity was also advertised on partners’ social media (ie, Twitter). A snowball sampling strategy was also used [ 13 ]. Direct email invitations were sent to contacts identified by partners. Participants were eligible if they were active, English-speaking primary or behavioral care providers in South Carolina with direct patient involvement. Twelve interested participants reached out via email; 7 did not meet the criteria (5 served in administrative or management roles with no direct patient involvement, and 2 were not providers), while 5 participants met eligibility criteria and agreed to participate. There was no previous relationship between the authors and the participants. Participants were made aware of the authors’ credentials and the purpose of this research in the recruitment language.

An interview guide with 19 questions was iteratively developed by 7 team members (BY, JRR, DBF, MAA, MMM, MS, and SN) and reviewed by 2 partners to explore participants’ current knowledge about SDOH and use of International Classification of Diseases, 10th Revision, Clinical Modification Z SDOH codes. The interview guide was piloted with 2 providers, including a clinical partner ( Multimedia Appendix 1 ).

Individual semistructured interviews lasting between 30 and 45 minutes were conducted between August 25, 2022, and September 2, 2022, by 1 study team member (BY) via Zoom; 1 team member (JRR) sat in on 1 interview to take notes. Interviews were recorded and transcribed verbatim through Zoom transcription services and imported into NVivo (version 12, 2022-2023; QSR International Pty Ltd) for analysis.

For codebook development, 1 team member (BY) initially matched concepts to the interview guide using a deductive approach. Two team members (BY and MMM) then independently reviewed and coded 3 transcripts each for comparison and finalization using an inductive approach. Intercoder agreement resulted from iterative discussions between the 2 coders. One team member (BY) then iteratively coded all 5 transcripts using semantic thematic analysis [ 14 , 15 ], and emergent codes were added to the codebook until no new codes were found [ 16 ]. Participant responses are presented according to the following 5 themes: participants’ understanding of SDOH for the patient population, assessment and documentation practices, referrals to other providers and community-based resources, barriers and facilitators of SDOH assessment and documentation, and SDOH assessment and documentation training preferences.

Ethics Approval

This study was approved by the University of South Carolina Institutional Review Board (Pro00122161). Recruitment language providing a description of this study’s purpose, eligibility, procedure, and incentives for informed consent was disseminated through statewide partner e-listservs. Transcribed interviews were deidentified (participant 1, participant 2, etc), revised for accuracy, and imported into NVivo for data management by 1 team member (BY). Original interview audio files were securely stored in a password-protected, cloud-based storage application with limited team member access. Participants were provided a US $25 incentive for their time.

Sample Characteristics

Participants (N=5) included a neonatal intensive care unit registered nurse, a nurse practitioner, a certified nurse midwife, a family and preventive medicine physician, and a counselor (licensed clinical social worker). Participants’ careers in health care ranged from 12 to 32 years, with 1 to 25 years of experience in their current roles. Three participants currently work within a large health care system, 1 participant works for a university health care system, and 1 participant works at a rural, 2-physician practice. Two participants also serve as faculty members involved in training the health care workforce within a university system ( Table 1 ).

Sample characteristics.

Participants’ Understanding of SDOH for the Patient Population

Participants seemed to have a good understanding of SDOH and were able to reflect on how SDOH might impact their specific patient population. All participants reported SDOH are important to assess for improved treatment planning but described a range of assessment and documentation practices. Participants identified a variety of SDOH impacting their patients including general access barriers (time, transportation, and childcare), health care access and quality (affordability and provider’s time), education and health literacy, poverty and material constraints, food insecurity, and social connectedness. For example, 2 participants gave an overview of the different SDOH affecting patient populations:

The neighborhood around our practice has, you know, folks that have a lower socioeconomic standpoint, so that’d be one, lower education rate as a as another. So, food insecurity is another social determinant. We also look at social connectedness as another one, and finances as a fifth. Participant 4

I would define social determinants of health as those factors that impact patients’ access to care, patients’ understanding of their care, and maybe the resources that are available to them. I think things outside of - yeah, those things that are sometimes kind of hard to quantify and it’s certainly hard to elicit. Participant 3

Another participant focused in on health literacy as a salient problem among patients, further adding limited provider time reduces opportunities for health education:

You know, so a lot of our patients, you know, you can break it down to the sixth-grade level, and it’s still over their head. You know, it’s like, you know, the population when it comes to understanding, having what I call “health intelligence”, are not even at the second-grade level. So, social determinants of health, you know, education is a joke, our healthcare system doesn’t allow the health care providers the time they need to teach the patients and then of course, on the patient side, they just gave up. Now, that’s not all of them, that’s just most of them. Participant 2

Participant 5 explained how using SDOH can help understand patient challenges and improve outcomes:

If I look at, just in general, my work in mental health, the patients are very often hindered from whatever they want, because of other things like finances, or they don’t have food. So, for example, I’ve served as a home health social worker, and I was only pulled into the cases where they weren’t getting better because of the social determinants of health. Obviously, if you’re not eating, your wound won’t get better, right? Obviously, if you’re about to be evicted and you’re taking care of your loved one who’s dying, you can’t take care of them, you’re working on the eviction. Participant 5

The issue of access (whether it be constraints on time, transportation, and childcare) highlighted by participant 1 illustrates the importance of supportive health care services that attend to the entire family (parent caregivers) as well as patients.

I think the biggest social determinant and social justice problem that I see is with our patients – I mean, the parents of these patients – is their access, and their availability to be able to be with their babies at this very, very vulnerable time. So, making sure they have what they need and the equal opportunity to be able to access their child is a big problem that we see. Participant 1

Assessment and Documentation Practices

While 2 participants were familiar with International Classification of Diseases Z codes (Z55-65) specific to SDOH, 3 were not. Only 1 participant reported the use of these Z codes to actively document patient SDOH, indicating the codes were recommended by their EMR system to be added to the problem list and visit encounter. Three participants documented SDOH via provider notes or the patient education or care plan, and 2 participants indicated an automatic screener.

All participants agreed that assessing and documenting patient SDOH was very important and beneficial for improved health outcomes; however, most (n=4) noted SDOH are not systematically or consistently assessed. One participant stated asking whether SDOH is standard social work practice. Another participant mentioned that while sometimes more basic questions are asked, there is often no in-depth assessment, especially for outpatient care:

The focus is on the disease process and how to -- which medication to prescribe. I’ve seen it before, you know, I mean, they don’t really dig deeply these days. Now, if we’re talking about a medical student, or medical resident, or patients being admitted, man do they dig. But when you get out into a practice, like outpatient, stuff like that -- the social determinants really are boiled down to a few very minor things that are “yes” or “no” questions or basic answers when they fill out an application. Participant 2

I think everybody, if we’re intervening with the patient, I think it is [our] ethical and moral responsibility to ensure that we’re addressing the needs of the entire patient, because it’s not just a medical need, or the daughter doesn’t just take care of the medical needs, or the nurse just doesn’t take care of nursing needs; I mean we take care of the whole patient in their family.... It doesn’t need to be this conversation of, “Oh, well we don’t talk about these things,” or “This is just the way that it is.” I mean, we’re all human beings and we all deserve the right to access equal access and equal opportunity. Participant 1

When asked, “How often do you or does someone in your practice ask patients about social determinants of health?” no participants reported having a universal protocol to assess patient SDOH at each visit. All participants indicated they will either start to notice patterns or patients will specifically bring SDOH up over time, despite not actively screening for them. One participant stated they ask the patients questions about their lives in a conversational manner at each visit to glean information about SDOH, and 1 participant indicated an annual electronic screener all patients attending their practice receive. One participant said they ask at each initial appointment if the patients are affiliated with one of their grants requiring assessment of certain SDOH.

Referrals to Other Providers and Community-Based Resources

When asked what is done when a need is identified, all 5 participants reported they refer patients to a variety of external providers and local community partners who provide health and social services such as new parent education, counseling and substance use services, and vocational rehabilitation. Specific national organizations participants mentioned included Ronald McDonald House, United Way, and First Steps, and 1 participant mentioned partnering with 2 community organizations through the NowPow platform [ 17 ]. One participant specified “attributed life” patients with certain insurance plans are eligible for referrals (official work orders) to the care coordination wing of the health care facility, bypassing the in-office social worker. One participant, a social worker, reported they actually provide the client with a list of suggestions for outside services, including legal or welfare benefits assistance, and this is documented in the patient’s case notes along with whether or not the patient followed up.

Three participants indicated they would make contact with an internal provider, such as a social worker, or to a provider who does semiregular drop-ins, such as a nutritionist, physical therapist, or podiatrist. Two participants reported directly addressing the need on-site at the time of the encounter:

We do have that very close-knit relationship [within our care team] where we can just go and talk to [our providers] face-to-face and be like, “Well, we need to get somebody to come address this, or we just need to, you know, have an intermediate interdisciplinary meeting, where we have everybody there at the same time.” Things like that. As far as transportation, we can get with social work to try to get like taxi vouchers or bus vouchers or whatnot. Participant 1

So, there are some organizations that will help. It’s just a matter taking the time and making the phone call. I find if you give the patients the information and ask them to cold call, they don’t seem to. So, we kind of take the groundwork, and [if] they have a contact person, I think it seems to work a little bit easier. We’ll bring them in the office. I’ll bring them in my office, and we’ll make the call and see if we can’t get them. Participant 3

Barriers and Facilitators of SDOH Assessment and Documentation

Participants offered their perspectives on barriers to assessment and documentation of SDOH (see Multimedia Appendix 2 for participant quotes specifically focused on the theme of barriers and facilitators). When asked, “What makes it difficult to assess and/or document patient social needs?” the majority of participants referred to the system-level barrier of profitability, with the most frequent responses relating to insurance or billing demands and limits to time with patients. Two participants reflected on how revenue generation, for better or worse, is a driving force in modern clinical practices.

Although participants identified several limitations in relation to billing, there was acknowledgment insurance companies are attempting to be more holistic when it comes to patient assessments and to reimburse clinics more appropriately when they have patients with complicated health issues. In addition, currently, Z codes are not reimbursable, so there is little incentive for clinic personnel to take the time to enter them.

From the microlevel perspective, patient reluctance and a sense of stigma were mentioned most frequently. For example, patients may find SDOH questions intrusive and not the purview of a medical provider. However, all participants perceived patient-provider trust and rapport as a facilitator of SDOH assessment, noting it is important to be conversational with patients, to “normalize” the questions, and to build long-term relationships. The next most frequently perceived barrier was an inability to address the patient’s SDOH, whether due to a lack of provider knowledge as to how to help the patient when an issue was identified or a lack of available resources or services to which the patient could be referred.

SDOH Assessment and Documentation Training Preferences

Participants identified continuing education credit-eligible brown bag sessions (informal meetings or trainings typically occurring over a lunch break), web-based courses, catered dinners, and destination conferences as their general preferences for training formats . Two participants stated they had training in assessment of SDOH during their nursing or social work education, while the other 3 expressed they have learned through experience. The 2 participants involved in training the health care workforce acknowledged this is still being done at the educational level and felt additional training targets, such as patient-provider simulations, could be incorporated into the curriculum.

We do something like this [with our students], like they get X amount of stipend, like with their paycheck, and then they have to go to all the community sources and make sure they can get groceries. But then I gotta pay the light bill, and this is like, they realize how quickly sometimes their money can go. Even incorporating that, now you’ve been hospitalized, and you could incorporate that into a lot of nursing education. Participant 1

From a standpoint on how to ask the question, we do well. I guess part of my job is actually to help train the residents and medical students on how to ask the questions. And, so we do more of video feedback for the residents, and such.... But, for the attending faculty like myself, that is not necessarily something that they do for us per se. So, I’ve kind of more learned on my own, when it comes to that. Participant 4

While only 1 participant reported having training on the use of Z codes from the “billing aspect,” the majority of participants did not feel training on SDOH assessment and documentation would be very beneficial without a system-wide push for a standardized protocol.

Four participants stated they would not benefit from training on how to ask patients about SDOH; 1 participant felt they would benefit from training to improve their knowledge of Z codes and more efficient documentation to save time.

Yeah, I guess I think about, you think about behavioral change, or you think about practice improvement or quality improvement, is that the more that you can do it a higher functioning level where it doesn’t rely on the individual person to change that actually makes it a more sustainable model. And so, giving a doctor education on the importance of health literacy, I think in the end is the least helpful long term. It needs to be something where it is with, like in the EMR, how the social determinants of health from the screener say, you know, “here’s some ICD-10 codes you should add to your bill,” and all you gotta do is click a few buttons. Because, I can scan it and agree, like “yep, yep, yep”, and then I can move on. Participant 4

Principal Findings

Overall, participants in our study conceptualized the SDOH of their patient populations in accordance with HP 2030 and could indicate how individuals in their care were affected by the 5 HP 2030 SDOH domains. Microlevel social needs identified by participants included material constraints such as time (to be with their child in the hospital), transportation, financial and food access barriers, and social disconnectedness. Identified SDOH at the macrolevel included poor education quality, with specific attention to a lack of health education that can contribute to limited knowledge of positive health behaviors and limited health literacy on the individual level. All participants mentioned the broader aspects of the SDOH domain of health care access and quality when describing the barriers to assessing and documenting SDOH during the clinical encounter (ie, time constraints and inconsistent documentation protocol).

Organizational Challenges to Assessment and Documentation of SDOH

Participants realized the context of one’s life has a crucial role in the development and management of health conditions. All participants agreed assessing and documenting patient SDOH were very important and beneficial for improved health outcomes but most (n=4) noted it is not done consistently. Consistent with findings by Heidari et al [ 18 ], participants in our study mentioned several barriers regarding the assessment of SDOH and appropriate referrals, including a lack of universal screening protocol and little incentive to screen for SDOH or to use Z codes. Medical providers have very limited time with patients, so any additional information gathering can be difficult to accommodate consistently if not designed to seamlessly (efficiently) integrate within the health care encounter.

Despite the lack of formal methods of screening, all participants noted patient social needs are often revealed over time, whether through direct patient disclosure or indirect indicators of barriers to treatment adherence (eg, inability to read a prescription label, lack of transportation, and food insecurity). While 2 participants delivered annual or semiannual formal SDOH assessments to some or all of their patients, all participants suggested a more conversational approach often served to elicit depth of information on patient SDOH. However, even in instances where SDOH information is gathered informally, it can be difficult to act upon the information, since resources can be difficult to find or unavailable. This can lead to providers becoming frustrated since they feel they cannot help their patients and are thus wasting valuable visit time.

Z codes, while they present a way to quickly and consistently record SDOH information, are not billable to insurance so they may be supplanted by billable codes when numbers of codes are limited. Entering codes was also seen by participants as an inefficient use of time, especially in for-profit settings, since entering nonbillable codes takes time away from the entering of billable codes. Billing code disincentives have created unintended consequences for individual clinics to inquire about SDOH or to use Z codes that must be addressed at the system level. For example, validated screeners may provide a way to gather SDOH information but must be implemented as a system-level policy. Thus, especially in for-profit settings, the use of Z codes must be aligned with economic goals. Insurance companies can already support SDOH screening by using value-based models that incentivize longer-term patient outcomes. Making codes for SDOH billable would help further move medical services in the direction of more holistic care [ 19 ]. Efforts to consistently and adequately address unmet patient social needs may require state-level actions, such as those being implemented in North Carolina through the Healthy Opportunities Pilot initiative [ 20 , 21 ]. While this serves as a potential model, it is unclear how this type of collective intervention may be impacted by state Medicaid expansion decisions, legal, and budgetary constraints [ 22 ]. Additional strategies are needed to ensure all patients’ social needs are addressed, regardless of income or type and level of health care coverage.

Community and Societal Challenges to Clinical Documentation of SDOH

Participants perceived patient reluctance and apprehension of stigma as interpersonal barriers to the assessment of SDOH. There is some emerging evidence related to the stigmatization of the SDOH. Rather than an acknowledgment of the important ways in which the life context impacts our health and well-being, participants in this study alluded to the SDOH negatively, demonstrating SDOH are perceived as deficits rather than basic health and human rights, such as access to adequate food, income, education, transportation, supportive social networks, housing, freedom from racism, and freedom from other discrimination. This is in opposition to the intended assessment and documentation of patient SDOH as important social “vital signs” impacting health outcomes [ 23 ]. Addressing the SDOH speaks to an understanding of the context of health and demonstrates clinician and organizational knowledge and insight into the factors influencing equitable conditions for optimal health and well-being. The stigmatization of the SDOH undercuts opportunities to positively impact individual and population health outcomes [ 24 ]. In addition, providers should be intentional in communication practices, using language that carries less stigma to underserved or historically marginalized community members and incorporating questions surrounding cultural, linguistic, and spiritual patient needs [ 25 ].

The next most common perceived barrier was an inability to address the patient’s SDOH. Our findings are consistent with those of Kostelanetz et al [ 26 ], who reported clinician-perceived barriers to SDOH screening among acute care patients including limited resources to address social needs, limited or no time or staff allotted for SDOH screening, and lack of training to address existing social care needs. There is growing consensus for health system and organizational accountability in support of SDOH assessment and documentation by ensuring adequate resources for screening education, patient support services, and referral to institution resources, community organizations, and public health agencies and to create stronger partnerships with community organizations for care that extends beyond the scope of medicine.

Participants suggested trust and rapport facilitate the assessment of patient social needs and felt long-term patient-provider relationships and standardization of SDOH assessment can reduce patient perceptions of provider judgment and reduce a sense of a power dynamic. This creates significant tension for clinicians who predominantly practice within a biomedical model of care and have limited educational exposure or incentive to adopt a social-ecological model of health care that includes attention to the SDOH. Given the importance of patient-provider trust and rapport, clinician practice of cultural humility is necessary for equitable patient care. Cultural humility entails provider openness, self-awareness, and humbleness [ 27 ], and “incorporates a lifelong commitment to self-evaluation and critique, to redressing the power imbalances in the physician-patient dynamic” [ 28 ]. Further, increased patient-provider contact, transparency, physician competency regarding patient social contexts, and genuine demonstrations of care and authenticity in communication can improve patient trust [ 29 , 30 ].

Challenges With Referrals Following Assessment of SDOH

While systematically assessing and documenting patients’ SDOH are extremely important, ensuring there is a clear next step that involves assisting patients with social factors that impact health through referrals to community resources and organizations is vital for improved health outcomes [ 31 ]. Participants referred patients to both external providers and local community partners, documenting these in the EMR regardless of the need for a direct referral. Some participants, particularly those practicing in more rural areas, expressed frustration over the lack of local services to meet some of their patient needs. In addition, the concept of “social prescribing,” or recommending patients who engage in civic, art, recreational, or other activities to improve overall health and well-being [ 32 ], encourages sustainable, socially engaged outlets for patients. To alleviate provider burden for identifying and connecting with adequate, vetted, and accessible referral sources, there needs to be a concerted effort to connect to or build a referral infrastructure (ie, NowPow) to reduce future time burdens. This would require partnerships across health-serving sectors, including hospital systems, government agencies, and community agencies. For example, referrals to community adult literacy centers can provide patients with assistance in reading and understanding general and health-related information to improve their health literacy [ 33 ]. Another example is the development of medicolegal partnerships to aid patients in gaining access to or redressing issues impacting a variety of social needs including housing, material assistance, disability, or supplemental income supports [ 34 ]. Additionally, more hands-on referral approaches, where the patient is assisted in making the connection with the referral source and the referral source adequately assists the patient in navigating complex processes (ie, public benefit or housing applications), may eliminate barriers to “referral uptake” for better health outcomes [ 35 ].

Limitations

This study has some limitations. First, we used purposive and convenience sampling to recruit participants. Second, the number of participants interviewed was quite small (N=5), limiting our ability to obtain findings generalizable to a specific health system, to the state of South Carolina, or to other US states. Third, participants had a range of years of experience within the health care field; however, some were newer to their current roles and thus had less experience within their current health care system. This poses a potential limitation for data source triangulation. Future research is warranted with additional providers to further explore assessment and documentation practices, barriers, and facilitators in additional settings in South Carolina. Examining provider and clinic practices by medical specialty and geographic location as well as proximity to referral resources will be important for supporting patients’ social needs following any assessment, documentation, or patient-provider discussions about SDOH. Despite these limitations, findings from our work will help providers and health care systems consider how to effectively integrate SDOH screening during patient-provider encounters, as findings offer important insights into providers’ perspectives and recommendations for SDOH assessment and the barriers and facilitators associated with it.

SDOH have a clear and direct impact on individual and community health status. Clinical providers understand these ties; however, they struggle with the best approach to assess and document SDOH issues in clinical settings. Time constraints in clinic visits, perceived stigma of SDOH, and preference for a more conversational approach to the patient-provider discussion add complexities to data collection, which result in incomplete and inconsistent interactions across populations. Perhaps an institutional culture shift is required to ensure consistent screening and assessment of patients for unmet social needs and provision of appropriate resources, including connections to organizations in the community to assist with these needs. Unfortunately, the stigma surrounding SDOH issues, social isolation, and financial hardships continues to pervade society. Bringing these challenges into the open and talking to patients about them are necessary if the goal is to improve the overall health of patients.

Acknowledgments

We would like to thank the participants for sharing their time and expertise with our team. We would also like to thank our partners Diana Zona (South Carolina Hospital Association) and Dr David Isenhower (Self Regional Healthcare) for their assistance in interview guide development and recruitment planning.

Abbreviations

Multimedia appendix 1, multimedia appendix 2, data availability.

Conflicts of Interest: None declared.

IMAGES

Social Determinants of Health
The 5 Social Determinants of Health and why they matter
What Are the Social Determinants of Health?
Social Determinants of Health 201 for Health Care: Plan, Do, Study, Act
Social-Determinants-of-Health-Infographic
What Are Social Determinates Of Health

VIDEO

Social Determinants of Health
Social Determinants of Health
Social Determinants of Health
Social Determinants of Health
Social Determinants of Health
Unit 4: Social Determinants of Health (SDOH)

COMMENTS

Social Determinants of Health Case Studies
The case study, which focuses on Newark, New Jersey, addresses three of the five key determinants of health: social and community context, health and health care, and neighborhood and built environment. Upstream interventions designed to improve mental health and well-being that support trauma-informed care are analyzed.
Social Determinants of Health Example Case Studies
CPT Codes 99204/99214. Case Study 2: MDM. A 12-year-old young lady presents with her grandmother for anxiety and depression and risk for failing in school due to absences related to her mental health instability. She would benefit from medication management, but the parents are divorced and the mother refuses to allow treatment of the condition ...
Predicting Social Determinants of Health in Patient Navigation: Case Study
Some barriers that navigators address are population-specific and other disease specific, but a common underlying thread is social determinants of health (SDoHs), defined by the Centers for Disease Control and Prevention as the "conditions in the places where people live, learn, work, and play that affect a wide range of health and quality-of ...
Social Determinants of Health: A Case Study
Social Determinants of Health: A Case Study. by Michael D. Rohwer, M.D., executive director of Curandi. March 8, 2018. The term "social determinants of health" (SDOH) is inescapable in the healthcare industry. But despite the ubiquity of the term, integrating SDOH into front-line medical care remains largely out of reach. Why?
PDF Case Study Report
Case Study Report: Innovative Approaches to Addressing Social Determinants of Health for Medicare Advantage Beneficiaries 1 Introduction In recent years, the U.S. health care industry has recognized that unmet needs in an individual's social and physical environment can significantly affect health outcomes and contribute to continued inequities.
Social Determinants of Health Case Studies: Targeting the Social
Date of Review: February, 2019. This is one of 3 'case studies' in a collection of teaching tools called Social Determinants of Health based on Healthy People (HP) 2020 material aimed at teaching Health Professional students to adopt a new perspective that identifies and includes the multitude of factors that affect the SDH of their patient's communities and populations and incorporate ...
Case Studies: Person-Centered Health Coaching in People With Negative
INTRODUCTION. Social determinants of health (SDH) are the conditions in the environments that influence health outcomes, including safe housing, transportation, and neighborhoods; racism, discrimination, and violence; education, job opportunities, and income; and access to nutritious foods and physical activity opportunities. 1 Numerous studies suggest that SDH can influence health more than ...
Addressing the social determinants of health: a case study from the
Methods. An exploratory comparative case study design using qualitative research methods was adopted for the study. A case was defined as action by a CHW (Mitanin) or team of CHWs (Mitanins) on nutrition or violence against women, which has resulted in a positive change in the particular social determinant of health in the context of the village/cluster of villages for which the CHW/s are ...
Teaching Social Determinants of Health Through an Unfolding Case Study
The impact of social determinants of health (SDOH) was developed to educate nursing students through the use of an unfolding case study. Problem: SDOH and population health are critical components of prelicensure nursing education. Unfolding case studies are a strategy to develop critical thinking and teach SDOH to nursing students.
Social determinants of health
The following list provides examples of the social determinants of health, which can influence health equity in positive and negative ways: Income and social protection. Education. Unemployment and job insecurity. Working life conditions. Food insecurity. Housing, basic amenities and the environment. Early childhood development.
Social Determinants of Health
Social determinants of health (SDOH) are the conditions in the environments where people are born, live, learn, work, play, worship, and age that affect a wide range of health, functioning, and quality-of-life outcomes and risks. SDOH can be grouped into 5 domains: Economic Stability. Education Access and Quality. Health Care Access and Quality.
Teaching Social Determinants of Health Through an Unfolding Case Study
Abstract. Background: The impact of social determinants of health (SDOH) was developed to educate nursing students through the use of an unfolding case study. Problem: SDOH and population health are critical components of prelicensure nursing education. Unfolding case studies are a strategy to develop critical thinking and teach SDOH to nursing ...
Measuring social determinants of health in the
Investigating the roles of social factors in health is an essential part of precision medicine research. The All of Us Research Program ( All of Us) aims to collect data from over one million ...
Social Determinants of Health
The definition of social determinants currently in widest use was created by the World Health Organization in 2008 [ 14 ]: "The conditions in which people are born, grow, live, work and age. These circumstances are shaped by the distribution of money, power, and resources at global, national and local levels.".
Discovering social determinants of health from case reports using
Background Social determinants of health are non-medical factors that influence health outcomes (SDOH). There is a wealth of SDOH information available in electronic health records, clinical reports, and social media data, usually in free text format. Extracting key information from free text poses a significant challenge and necessitates the use of natural language processing (NLP) techniques ...
Exploring Health Systems Within the Context of Social Determinants of
Introduction: Global health case studies can highlight, often in dramatic fashion, topics such as health disparities, health systems, and the social determinants of health. This resource presents a case study of an adolescent Haitian boy requiring lifesaving heart surgery in the US due to an acquired mitral valve disease from untreated streptococcal pharyngitis.
Developing a Social Determinants of Learning™ Framework: A Case Study
A case study illustrates an initial study of interventions targeting specific domains of the framework aimed toward student success. ... Underresourced schools represent one of our nation's most significant examples of structural racism. ... A conceptual framework for action on the social determinants of health. Social Determinants of Health ...
Teaching Social Determinants of Health Through an
The unfolding case study was first presented in an unstructured format in a 3-hour didactic synchronous online class. The unfolding case study replaced lectures on SDOH, population health, the health care system, and most of the content on community assessment and analysis. The unfolding case study replaced a major assignment to apply SDOH ...
Essay: Rediscovering The Social Determinants Of Health
The social determinants of health that typically make up this field of study are social class, sex, age, race, and ethnicity. In this essay I focus mainly on social class but note that there is a ...
How a Global Pandemic Became a Case Study in Social Determinants of Health
How a Global Pandemic Became a Case Study in Social Determinants of Health. African Americans comprise approximately 14% of Michigan's population, yet 40% of the state's COVID-19-related ...
Modeling the Impact of Social Determinants of Health on HIV
Consider a model of the form Y = M (X, A, θ,ε). In this case, Y is HIV incidence, A is needle exchange, X are individual- and community-level predictors of HIV incidence, and θ is a set of parameters that governs the relationship between A, X and Y via the model M. There are two ways to use models to study this effect.
Exploring Health Systems Within the Context of Social Determinants of
Global health case studies can highlight, often in dramatic fashion, topics such as health disparities, health systems, and the social determinants of health. This resource presents a case study of an adolescent Haitian boy requiring lifesaving heart surgery in the US due to an acquired mitral valve disease from untreated streptococcal pharyngitis.
PDF Case studies of social determinants of health
Case 1 Study 2: Prepare for Questions Take notes! 2 Identify the social determinants impacting mother and son's health. Explain 1Name how each of these determinants impact their health. 2Define it 3Why it 3 4Link important to case study. Choose the one social determinant you think is most important for their health.
Social Determinants of Health at CDC
Social determinants of health (SDOH) are the nonmedical factors that influence health outcomes. They are the conditions in which people are born, grow, work, live, and age, and the wider set of forces and systems shaping the conditions of daily life. These forces and systems include economic policies and systems, development agendas, social ...
What Are Social Determinants of Health?
Social determinants of health (SDoH) are factors in a person's life and environment that impact their health outcomes in positive or negative ways. These economic and social conditions are all connected and can have a detrimental snowball effect for individuals and entire populations. There are numerous social determinants of health, but they ...
Assessment and Documentation of Social Determinants of Health Among
Social determinants of health (SDOH) are defined by the United States' Healthy People 2030 (HP 2030) as "the conditions in the environments where people are born, live, learn, work, play, worship, and age that affect a wide range of health, functioning, and quality-of-life outcomes and risks." ... and social connectedness. For example, 2 ...

Social Determinants of Health Example Case Studies

Shea Lunt, RHIA, CPC, CPMA, PMP

Blog Categories

Recent Posts

Social Determinants of Health: A Case Study

Current use

More like this

Get Involved

Social Determinants of Health Case Studies: Targeting the Social Determinants of Health in Geriatric Populations

Corresponding Author’s Email:

Institution:

Where Was the Curriculum Implemented?

Source of the Curriculum/Resource:

Level of Learner Assessment:

Submit a Comment Cancel reply

Join our network and get current research and curriculum.

Social Determinants of Health

Suggested citation

How Does Healthy People 2030 Address SDOH?

Explore Research Related to SDOH

View SDOH Infographics

Learn How SDOH Affect Older Adults

The Office of Disease Prevention and Health Promotion (ODPHP) cannot attest to the accuracy of a non-federal website.

Teaching Social Determinants of Health Through an Unfolding Case Study

Measuring social determinants of health in the All of Us Research Program

Similar content being viewed by others

COVIDiSTRESS diverse dataset on psychological and behavioural outcomes one year into the COVID-19 pandemic

A philosophy of health: life as reality, health as a universal value

Assessing scale reliability in citizen science motivational research: lessons learned from two case studies in Uganda

Introduction

Survey development process

Conceptual frameworks and definitions

Cognitive and online pilot testing

Survey modifications after testing

Survey approval and release

Accessing data from the SDOH survey

Construction of scales

Statistical analyses

SDOH survey questions

Description of survey participants

Internal consistency reliability of scales

Item non-response and scale score distributions among survey respondents

Multivariable logistic regression models

Data availability

Acknowledgements

Author information

Contributions

Corresponding author

Ethics declarations

Additional information

Supplementary Information

About this article

Share this article

Quick links

Discovering social determinants of health from case reports using natural language processing: algorithmic development and validation

Setting and data

Conclusions

Introduction

Materials and methods

Proposed SDOH NLP Framework

Data processing module

Natural language processing (NLP) module

Joint optimization

Evaluation module

Quantitative analysis

Performance analysis on named eneities

Error analysis

Qualitative analysis to see model effectiveness

Previous works

Practical impact for enhancing clinical knowledge and patient care

Limitations

Availability of data and materials

Abbreviations

Acknowledgements

Author information

Contributions

Corresponding author

Ethics declarations

Consent for publication

Competing interests