Social Determinants of Health (SDOH): Definition, Examples and Data

What Are Social Determinants of Health?

Social determinants of health (SDOH) are the non medical conditions in which people are born, grow, live, work and age that shape their health outcomes. The World Health Organization defines them as the non medical factors that influence health outcomes, together with the wider forces and systems that shape daily life: economic policies, development agendas, social norms, social policies and political systems.

In the United States the working definition comes from Healthy People 2030, the decade long objectives program run by the Office of Disease Prevention and Health Promotion (ODPHP) inside HHS. It describes SDOH 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. CMS uses the phrase social drivers of health for the same idea, and the American Hospital Association prefers societal factors that influence health. All three share the SDOH acronym.

The word that matters is conditions. A determinant is a property of a place, an economy or a policy, not a diagnosis a person carries. A patient does not have a social determinant. A patient lives inside several of them, and what a clinic records is the individual level trace they leave: a positive food insecurity screen, an eviction notice, no car. That distinction drives how the data has to be modeled.

SDOH vs social needs vs social risk factors vs health equity

Alderwick and Gottlieb separated these terms in a 2019 Milbank Quarterly paper, and the National Academies adopted the vocabulary the same year in Integrating Social Care into the Delivery of Health Care. Determinants sit at the community level. Social risk factors are the adverse conditions an individual is exposed to. Social needs are the subset of those risks the person actually wants help with. Health equity is the outcome all of this work is meant to move.

For a data team the four terms map to four different data shapes. Determinants are area level variables joined by census tract, such as the Area Deprivation Index from the University of Wisconsin Neighborhood Atlas or the CDC and ATSDR Social Vulnerability Index. Social risks are screened observations coded to LOINC. Social needs are goals paired with consent. Health equity is a stratified quality report. Confusing them produces models that assign a neighborhood average to a person, which is the single most common analytical error in this field.

The Five Domains of Social Determinants of Health

Healthy People 2030 organizes SDOH into five domains. When a US health system, payer or state Medicaid agency says SDOH without further qualification, this is almost always the framework behind it. Each domain has published objectives and a literature summary on the ODPHP site, and each maps to a recognizable set of screening questions and ICD-10-CM codes.

One source of confusion is worth clearing up. The five domains above are not the same five things CMS asks hospitals to screen for. The CMS hospital measures and most Medicaid HRSN programs use five need categories: food insecurity, housing instability, transportation needs, utility difficulties and interpersonal safety. Those are individual level needs that fall mostly inside the economic stability and neighborhood domains. A data dictionary that treats the two lists as one will break the first time a reviewer asks for a domain level report.

Why Social Determinants of Health Matter

The estimate most people quote comes from the County Health Rankings and Roadmaps model, published by the University of Wisconsin Population Health Institute. That model weights the modifiable factors that shape health at 40 percent for social and economic factors, 30 percent for health behaviors, 20 percent for clinical care and 10 percent for the physical environment. It is a weighting scheme built from the literature, not a measured causal share, but it is the source behind every slide that says clinical care drives only a fifth of outcomes. An older Health Affairs analysis by McGinnis and colleagues in 2002 put medical care at roughly 10 to 15 percent of preventable mortality, which points the same direction.

Federal policy already treats social risk as a performance variable. The 21st Century Cures Act required CMS to stratify the Hospital Readmissions Reduction Program into peer groups by the share of dual eligible patients, starting with fiscal year 2019, because safety net hospitals were being penalized for readmissions their patients' circumstances drove. The IMPACT Act of 2014 directed ASPE to study how social risk affects Medicare value based programs, which it did in reports to Congress in 2016 and 2020. For Medicare Advantage, CMS is replacing the Star Ratings reward factor with a Health Equity Index beginning with the 2027 Star Ratings.

The operational reasons are more concrete than the policy ones. Each of the mechanisms below is well documented and shows up in ordinary utilization data once someone looks for it.

• Transportation insecurity produces missed appointments, late refills and post discharge visits that never happen, which is why no show models improve when a transportation item is added.
• Food insecurity forces a choice between medication and groceries, and it undermines glycemic control in a way no dosing change can fix.
• Housing instability makes discharge plans fail. Wound care, oxygen, refrigerated insulin and a follow up address all assume a stable home.
• Utility shutoffs disable CPAP machines, nebulizers, home oxygen concentrators and insulin refrigeration at the same time.
• Social isolation carries a mortality risk that meta analyses by Holt-Lunstad place on the order of established clinical risk factors, and it predicts who will not answer a follow up call.

This is why population health management programs now treat social risk as a first class input next to claims, labs and encounter history. A risk score built only on clinical data explains part of who returns to the emergency department. The social variables explain a good deal of the rest.

SDOH Screening Tools: PRAPARE, AHC HRSN and Hunger Vital Sign

Screening is how a community level determinant becomes an individual level, codable observation. Three instruments dominate US practice, and Epic, Oracle Health, MEDITECH and athenahealth all ship at least one of them as a flowsheet or questionnaire. They differ in length, in who owns them and in which domains they cover.

PRAPARE

PRAPARE, the Protocol for Responding to and Assessing Patients' Assets, Risks and Experiences, was developed by the National Association of Community Health Centers with the Association of Asian Pacific Community Health Organizations and the Oregon Primary Care Association. It has roughly 20 questions across a core set and an optional set: personal characteristics such as race, ethnicity, farmworker and veteran status and language; family and home; money and resources; social and emotional health; and optional items on incarceration, refugee status and safety. It is free to use, mapped to LOINC (panel 93025-5), SNOMED CT and ICD-10-CM, and it is the default instrument in federally qualified health centers.

AHC HRSN Screening Tool

The CMS Innovation Center built the Accountable Health Communities Health Related Social Needs Screening Tool for the AHC Model, which ran from 2017 to 2023. Its 10 core questions cover housing instability and quality, food insecurity (using the two Hunger Vital Sign items), transportation, utilities and interpersonal safety (using the four HITS items). A supplemental set adds financial strain, employment, family and community support, education, physical activity, substance use, mental health and disabilities. Because the five core categories became the five domains of the CMS hospital SDOH measures, this tool is the one most hospitals adopted.

Hunger Vital Sign

The Hunger Vital Sign is a two question food insecurity screen developed by Children's HealthWatch and validated by Hager and colleagues in 2010 against the 18 item USDA Household Food Security Survey. The two statements ask whether, in the past 12 months, the household worried food would run out before there was money to buy more, and whether the food bought did not last and there was no money to get more. An answer of often true or sometimes true to either item is a positive screen. It is embedded inside both PRAPARE and the AHC tool, and it carries its own LOINC panel, 88121-9.

Since the CY 2024 Physician Fee Schedule there is also a payment hook. HCPCS code G0136 covers administration of a standardized, evidence based SDOH risk assessment lasting 5 to 15 minutes, no more than once every six months, billable with an evaluation and management visit, the annual wellness visit or a behavioral health visit. Many programs now move the questionnaire to pre visit digital intake so the answers arrive coded to LOINC before the clinician opens the chart. That is how Bonami wires SDOH items into its AI patient intake agent: the patient answers on their own device, the responses are stored as structured observations, and a positive screen raises a task instead of vanishing into a PDF.

Z Codes, Quality Measures and Reporting Requirements

ICD-10-CM Z codes Z55 to Z65

Chapter 21 of ICD-10-CM, Factors influencing health status and contact with health services, contains the block Z55 to Z65, titled Persons with potential health hazards related to socioeconomic and psychosocial circumstances. These are the SDOH Z codes. They are reported as secondary diagnoses on professional and facility claims, and the ICD-10-CM Official Guidelines (Section I.B.14) allow coders to assign them from documentation by any clinician involved in the patient's care, including social workers, community health workers, case managers and nurses. Since fiscal year 2022 the guidelines also permit patient self reported information, as long as a clinician signs off and incorporates it into the record.

Use has been low. The CMS Office of Minority Health found that only about 1.6 percent of Medicare fee for service beneficiaries had a Z code on a 2019 claim, with Z59.0 homelessness, Z63.4 death of a family member, Z60.2 living alone, Z59.5 extreme poverty and Z63.0 spousal relationship problems the five most common. The reasons are structural: no payment attaches to the code, screening results live in a flowsheet the coder never sees, and coding policies were written for physician documentation. The FY 2022 and FY 2023 additions (food insecurity, housing instability, transportation insecurity, financial insecurity, material hardship) gave programs codes that match their screening items for the first time, and the CY 2024 Community Health Integration (G0019, G0022) and Principal Illness Navigation (G0023, G0024) services put a fee behind the follow up work.

CMS Hospital IQR: SDOH-1 and SDOH-2

The FY 2023 IPPS final rule adopted two measures into the Hospital Inpatient Quality Reporting Program. SDOH-1, Screening for Social Drivers of Health, is the share of admitted patients aged 18 and older screened for all five HRSN domains: food insecurity, housing instability, transportation needs, utility difficulties and interpersonal safety. SDOH-2, Screen Positive Rate, is the share of those screened who screened positive, reported per domain. Reporting was voluntary for calendar year 2023 and mandatory for 2024. CMS then removed both measures from the Hospital IQR Program in the FY 2026 IPPS final rule, beginning with the CY 2024 reporting period, so the workflows most hospitals built in 2023 now serve accreditation, state programs and internal analytics rather than a federal submission.

NCQA HEDIS Social Need Screening and Intervention

NCQA introduced the Social Need Screening and Intervention measure (SNS-E) in HEDIS Measurement Year 2023, reported through Electronic Clinical Data Systems. It counts members screened for food, housing and transportation insecurity during the measurement period, and, among those who screen positive, the share who received a corresponding intervention within 30 days. The value sets use LOINC and SNOMED CT codes aligned with the Gravity Project, which means a plan can only earn credit for screens and interventions that reach it as structured data.

The Joint Commission requirements

On January 1, 2023 The Joint Commission put new requirements to reduce health care disparities into effect for hospitals, ambulatory care and behavioral health organizations, first as leadership standard LD.04.03.08 and, from July 1, 2023, as National Patient Safety Goal NPSG.16.01.01. The elements of performance require an organization to designate a leader for disparities work, assess patients' health related social needs and share community resource information, identify disparities by stratifying quality and safety data by sociodemographic characteristics, write an action plan, act when goals are not met and report progress to stakeholders at least annually. The Joint Commission does not prescribe an instrument or a patient population, which is why most accredited hospitals kept the AHC tool they had already deployed for SDOH-1.

SDOH Data Standards: Gravity Project, FHIR and USCDI v3

SDOH data standards come in three layers. The Gravity Project decides what the concepts are and which codes represent them. The HL7 FHIR SDOH Clinical Care implementation guide decides how they travel between systems. USCDI decides which of them a certified EHR must be able to store and exchange.

Gravity Project

The Gravity Project started in 2018 at the Social Interventions Research and Evaluation Network (SIREN) at UCSF with Robert Wood Johnson Foundation funding, and became an HL7 FHIR Accelerator in 2019. Its job is consensus terminology. For each of more than 20 SDOH domains, including food insecurity, housing instability, homelessness, inadequate housing, transportation insecurity, financial insecurity, material hardship, employment, education, veteran status, stress, social connection, intimate partner violence, elder abuse, health literacy, digital access and utility insecurity, it publishes value sets that map screening questions and answers to LOINC, health concerns to SNOMED CT and ICD-10-CM, goals to LOINC and SNOMED CT, and interventions to SNOMED CT, CPT and HCPCS. When a code is missing, Gravity submits the request to the code committee, which is how most of the FY 2022 and FY 2023 Z codes came to exist.

HL7 FHIR SDOH Clinical Care implementation guide

The SDOH Clinical Care IG, first published as STU 1 in 2021 and built on US Core, models the whole loop from question to outcome. Screening answers are Observations. A recognized problem is a Condition carrying both a SNOMED CT concept and a Z code. What the patient wants to change is a Goal. The referral is a ServiceRequest, the tracking of that referral through the community organization is a Task, the completed service is a Procedure, and permission to share with a non HIPAA entity is a Consent. HL7 also maintains 360X, a Direct messaging based closed loop referral specification, for partners who cannot yet speak FHIR.

USCDI v3 SDOH data elements

ONC published USCDI version 3 in July 2022 with four SDOH elements: SDOH Assessment, SDOH Goals, SDOH Interventions and SDOH Problems/Health Concerns. The HTI-1 final rule of January 2024 made USCDI v3 the baseline for certified health IT with a compliance date of January 1, 2026. Every certified EHR in the United States must therefore store these elements and expose them through its US Core FHIR API, which removes the usual excuse that the vendor cannot surface social data.

Getting these resources in and out of Epic, Oracle Health or a state HIE through the certified API is a FHIR integration job with a few specific traps: Observations for screening answers must reference the QuestionnaireResponse they came from, the Condition needs both code systems or the Z code never reaches the claim, and Task status must be writable by an outside party or the loop never closes.

Community Referral Platforms and Closed Loop Referrals

A closed loop referral is one where the referring organization learns whether the community based organization received the referral, accepted it, delivered the service and resolved the need. Anything short of that is an open loop: a printed resource list, a phone number, a hope. Closed loop status is what separates a screening program from a social care program, and it is the data HEDIS SNS-E and Medicaid HRSN contracts actually score.

Three names come up in every US procurement. findhelp, formerly Aunt Bertha, maintains a national directory of free and reduced cost programs searchable by zip code and integrates with the major EHRs so a referral can be sent from the chart. Unite Us builds contracted networks in which community organizations agree to accept electronic referrals and report outcomes back; North Carolina's NCCARE360 network, which supports the Healthy Opportunities Pilots, runs on it. 211 is the national information and referral service reached by dialing 2-1-1, operated locally by United Way affiliates and other agencies under the Inform USA (formerly AIRS) standards, and its call data is coded to the 211 LA County Taxonomy of Human Services. Health systems typically use one of the first two for the referral workflow and 211 as a directory and after hours source.

• A shared directory with a taxonomy. Two organizations cannot exchange a referral if one codes it as housing and the other as shelter. The 211 taxonomy and Gravity value sets are the two candidates.
• Consent that fits the receiver. Most community organizations are not HIPAA covered entities, substance use records fall under 42 CFR Part 2, and several states add their own rules. The FHIR Consent resource exists to carry the answer with the referral.
• A status vocabulary. Received, accepted, in progress, completed, unable to contact, declined by client, and closed without service are the minimum states; without them the closure rate is meaningless.
• A bidirectional interface. FHIR Task or 360X into the EHR so the care team sees status where they work, not in a separate portal.
• Receiving capacity. A referral sent to a food pantry with a six week wait is not a closed loop, it is a delayed decline. Network adequacy has to be measured before screening volume is scaled.

The patient side of the loop is ordinary engagement work: confirm the referral was received, remind about the appointment, ask whether the need was met. Those touches run through the same channels as any patient engagement software, and the answer to the last question is the only outcome measure that matters to the person being referred.

Related Guides

These guides cover the neighboring concepts this article refers to, each written in the same answer first format.

• Healthcare Interoperability Explained: HIE, TEFCA, USCDI, Cures Act and SMART on FHIR: the four levels of interoperability, TEFCA, USCDI and SMART on FHIR.
• HIPAA Privacy Rule vs Security Rule: What Each Covers, Who Complies: what the Privacy Rule and Security Rule each govern and who must comply.
• What Is Revenue Cycle Management? The 2026 Guide to Healthcare RCM: the full revenue cycle, its 12 steps and the KPIs that measure it.

Medicaid 1115 Waivers and Medicare Advantage Supplemental Benefits

Medicaid Section 1115 demonstrations

Section 1115 of the Social Security Act lets HHS approve state demonstrations that waive ordinary Medicaid rules, and since 2022 it has been the main vehicle for paying for social care with federal Medicaid dollars. CMS approvals for Oregon, Massachusetts, Arizona and Arkansas in late 2022 set the pattern, and a November 2023 CMCS informational bulletin laid out a formal health related social needs framework: housing supports including up to six months of rent or temporary housing, utility assistance, home modifications, nutrition supports including up to six months of medically tailored meals, and case management, with HRSN spending capped at 3 percent of a state's total annual Medicaid expenditure.

The named programs are the ones your platform will be asked to integrate with. North Carolina's Healthy Opportunities Pilots launched in 2022 with a fee schedule of 29 services across housing, food, transportation and interpersonal safety. California's CalAIM offers 14 Community Supports through Medi-Cal managed care plans alongside Enhanced Care Management. New York's Social Care Networks, approved in January 2024, contract regional lead entities to screen, navigate and pay community organizations. Oregon covers climate related devices such as air conditioners and air filters for eligible members. In March 2025 CMS rescinded the 2023 framework guidance and indicated it would not approve new or extended HRSN spending authority, so approved demonstrations continue under their existing terms while new state proposals face a narrower path.

Whatever the policy direction, the data requirements these programs created are now baked into state contracts: screening results in a standard format, eligibility determination against program criteria, service authorization, invoicing from organizations that have never billed a health plan, and outcome reporting back to the state. Several states require managed care organizations to report screening and referral rates by domain, which is the same structure as the HEDIS measure.

Medicare Advantage supplemental benefits

Medicare Advantage got its SDOH authority from the CHRONIC Care Act, passed inside the Bipartisan Budget Act of 2018. From contract year 2019 CMS widened the definition of primarily health related supplemental benefits to include adult day care, in home support and caregiver support. From contract year 2020, Special Supplemental Benefits for the Chronically Ill (SSBCI) allowed plans to offer benefits that are not primarily health related to chronically ill enrollees: food and produce, meals beyond a short post discharge window, non medical transportation, pest control, indoor air quality equipment, general supports for living such as housing assistance, and structural home modifications. The CY 2025 MA final rule tightened the standard by requiring plans to hold evidence that an SSBCI benefit has a reasonable expectation of improving health, and it requires mid year notices to enrollees about unused supplemental benefits beginning in 2026.

For plans the engineering problem is eligibility and utilization. An SSBCI benefit is tied to a chronic condition and a social need, both of which have to be documented, and the benefit is worthless if the member never uses the card. That is why grocery and transportation benefit data increasingly flows through member engagement platforms, where it can trigger outreach when a benefit sits idle and feed the Health Equity Index calculation that now shapes Star Ratings.

How SDOH Data Feeds Risk Stratification and Population Health Analytics

Three kinds of SDOH data reach an analytics platform, and they should be kept in separate tables with separate provenance. Individual screened risks arrive as LOINC coded observations and Z codes on claims. Area level context arrives by geocoding the patient address to a census tract or block group and joining the Area Deprivation Index, the Social Vulnerability Index or state specific indices. Program data arrives as referral status histories and supplemental benefit utilization. Each has a different update cadence, a different consent posture and a different level of trust.

• Risk stratification. Adding transportation insecurity, housing instability and social isolation as features to readmission, emergency department and no show models is the most common first project. The features are sparse until screening coverage is high, which is why screening rate is an analytics metric and not only a compliance one.
• Care management prioritization. A patient with three positive screens and a recent discharge outranks a patient with the same clinical profile and none, and the care manager needs to see why.
• Quality stratification. The Joint Commission requires it, HEDIS stratifies selected measures by race and ethnicity, and payers increasingly ask for outcomes by ADI quintile.
• Payment and rate setting. CMS already uses dual eligibility and low income subsidy status in several models, HRRP uses dual share peer groups, and MassHealth adjusts managed care rates with a neighborhood stress score built from census data.
• Network and program planning. Positive screen counts by domain and tract tell a health system where to contract for food, housing and transportation capacity before it refers patients into a queue.

The modeling itself is a predictive analytics problem with two extra constraints. First, area level variables describe the tract, not the person, so they belong in the model as context and never as a substitute for a screen. Second, models trained on utilization or cost learn that people who use less care are healthier, when often they simply have less access. Obermeyer and colleagues showed in Science in 2019 how a widely deployed algorithm using cost as a proxy for need under referred Black patients to care management for exactly that reason. Fairness testing by race, ethnicity, language and ADI is part of the model, not a review step after it.

What Bonami builds here is the pipeline and the models, not the claims about outcomes: address geocoding and index enrichment, FHIR and claims ingestion for Observations and Z codes, a feature store that keeps the three data kinds distinct, and stratified reporting on top of a data analytics platform. The point where it becomes useful is when a score changes an action. A discharge model that knows a patient screened positive for transportation insecurity can trigger a ride benefit and a live call instead of a generic reminder, which is the kind of branch a post discharge follow up agent is built to run, and it is the reason the population health layer has to sit on top of the screening layer rather than beside it.

A minimal SDOH data model

• Screening event: patient, instrument, LOINC panel, date, setting, who administered it and how (self, staff, interpreter).
• Item responses: one row per question with the LOINC answer code, never a summary score alone.
• Derived risks: SNOMED CT concept plus Z code, with the rule that produced each one, so a positive screen can be traced to its questions.
• Needs and goals: what the patient asked for, in priority order, with the consent decision attached.
• Referrals: ServiceRequest, receiving organization, full Task status history with timestamps, and the reported outcome.
• Area context: tract level indices keyed to the address and its effective dates, stored apart from the person level tables.

Implementation Pitfalls

Most SDOH programs fail for the same handful of reasons, and none of them are about the screening instrument. They are about what happens after the screen and where the data goes.

• Screening without referral capacity. Asking a patient whether they have enough food and then offering nothing is worse than not asking. The National Academies report framed this as an ethical obligation: assistance capacity has to be in place before screening is scaled. Measure community partner capacity by domain before choosing screening volume.
• Data silos. The screen lives in an EHR flowsheet, the referral in a vendor portal, the outcome in a spreadsheet at the community organization, and the Z code is never assigned because the coder never sees any of it. Map every step to Gravity value sets and route screening results to coding on the same day they are collected.
• Consent treated as a checkbox. Sharing a positive interpersonal safety screen with an outside organization is not the same as sharing a food insecurity screen. Consent has to be per domain and per receiver, revocable, and carried with the referral. Partners outside HIPAA and records under 42 CFR Part 2 need their own handling.
• Bias in proxies and models. Applying a tract level index to an individual, using cost as a stand in for need, or training on populations that have historically been under screened all produce models that direct resources away from the people the program exists to reach.
• Duplicated screening. The health plan, the hospital and the primary care clinic each ask the same ten questions, and the patient stops answering honestly by the third time. Exchange the screen through FHIR instead of repeating it.
• Compliance metrics mistaken for outcomes. SDOH-1 counted screens, not resolved needs. Report screening rate, positive rate, referral acceptance, closure and patient reported resolution as five separate numbers and resist collapsing them.

A program that avoids these six looks unremarkable from the outside. Patients are asked once, on their own device or by someone they trust. Positive screens become coded observations, Z codes and tasks the same day. Referrals go to partners who have agreed to take them and report back. Status comes back into the chart. Analysts can see, by domain and by tract, where needs are being met and where they are not. None of that requires new science. It requires the standards described above, applied without shortcuts.