BACKGROUND

In an age of value-based payment, primary care providers are increasingly scrutinized on performance metrics that purport to measure quality of care, including patterns of health care use and health care outcomes of their patient populations. Social determinants of health (SDH), including the economic, social, and environmental characteristics of communities where people live, affect a wide range of health outcomes and risks. Research suggests that SDH contribute as much or even more to health outcomes than health care does. Although health care quality metrics frequently account for patient clinical complexity (ie, the number or severity of chronic conditions), they typically do not account for patient social complexity (ie, the presence of adverse SDH, or “social risk factors,” known to affect health). In this study, we leveraged data from the National Patient-Centered Clinical Research Network (PCORnet), a distributed network of clinical research networks (CRNs) that brings together data from millions of patients across different health systems, to assess whether accounting for both social complexity and clinical complexity better explains differences in quality-of-care metrics than patient clinical complexity alone.

OBJECTIVES

The specific aims of this study were to do the following: 1. Engage stakeholders to identify how clinic-level measures of comorbidity and social complexity are useful to health systems leaders, clinicians, and patients in delivering quality clinical care, improving treatment adherence and health, and managing population health, resources, and decision-making. 2. Use diverse sources of data to assess the association between social and clinical complexity and quality-of-care metrics. 3. Evaluate the relative contributions of patient- and community-level SDH in explaining variation in quality-of-care metrics. We engaged key stakeholders at several points throughout the life cycle of the study to evaluate the applications of our findings for clinical decision-making, system improvement, and/or health care payment and policy, and to identify future directions and considerations for this line of research.

METHODS

We used patient electronic health record (EHR) data from 3 PCORnet CRNs linked to Medicaid claims data and geocoded community-level data from the American Community Survey (ACS). The first round of analyses was conducted in the Accelerating Data Value Across a National Community Health Center Network (ADVANCE) and OneFlorida CRNs, both of which include a large network of safety-net community health centers (CHCs) that provide care to some of the nation's most vulnerable populations, most of whom have low income, are uninsured or publicly insured, and have high levels of social complexity. Based on input from stakeholders and advisors who reviewed the first-round results, we partnered with the Patient Outcomes to Advance Learning (PORTAL) CRN to conduct a second round of analysis in a broader sample of patients. Relative to the ADVANCE and OneFlorida CRNs, PORTAL clinics provide care primarily to privately insured patients with lower levels of social complexity. Primary outcome measures, prioritized based on input from stakeholders, included diabetes control (ie, the percentage of patients with diabetes with hemoglobin A [HbA] > 9%) and emergency department (ED) use. Clinical complexity was measured at the patient level using the Charlson Comorbidity Index (CCI) and a count of mental and behavioral health conditions. Social complexity was measured at the community level using the Social Deprivation Index (SDI), a composite measure of 7 demographic variables from the 5-year ACS, calculated as national percentile ranking (ranging from 0 to 100) based on all Zip Code Tabulation Areas and census tracts in the country. To address aim 2, we used logistic regression models to estimate the impact of adjusting for both clinical complexity and social complexity on the odds of poor diabetes control or ED use. In aim 3, we included additional information on household income (measured as a percentage of the federal poverty level) and social-risk screening data to further understand the relative impact of patient- and community-level measures of social complexity.

RESULTS

Even after accounting for patient-level clinical complexity, community-level social complexity was significantly associated with increased odds of ED use, measured as both a dichotomous (any use) variable and a count variable, as well as with poor HbA control (defined as HbA >9%). This relationship was significant for patients receiving care in both of the safety-net CHC networks (ADVANCE and OneFlorida), as well as within the PORTAL network, which has a much higher proportion of commercially insured patients. For patients in the OneFlorida and ADVANCE cohorts, there were 3% and 5% increased odds of poor diabetes control, respectively, for each 10-point increase in SDI percentile score. Moreover, we found that ADVANCE and PORTAL patients living in the most deprived quartile of census tracts (defined as those with SDI percentile score >75) had significantly higher odds of poor diabetes control (38% and 22% higher odds, respectively) and ED use (28% and 26%, respectively), relative to those in the least deprived quartile of census tracts (SDI percentile score ≤25). Preliminary analyses that included both patient- and community-level SDH measures suggested that after controlling for patient-level SDH, the effect of community-level SDH was no longer significant.

CONCLUSIONS

Even after accounting for patient clinical complexity, social complexity was an independent contributor to diabetes health outcomes and ED use. Although community-level SDH measures were associated with variation in outcomes, results of preliminary analysis using patient-level measures suggest that, when available, patient-level measures may be preferable. Providers caring for patients with more social risk factors may benefit from having their performance metrics adjusted for their patients' social risk factors. As the United States moves toward value-based purchasing models of reimbursement, more research and refinement of a combined measure of social risk and clinical complexity may contribute to more effective and equitable value-based purchasing models that incentivize the care of socially complex patients.

LIMITATIONS

Patients were geocoded to the lowest level of geography possible using the available address information, but the availability and accuracy of patient address data in the EHR may have led to some inaccuracies in patient location. Although patients in the ADVANCE and PORTAL cohorts were geocoded at the census tract level, patients within the OneFlorida network were geocoded at the Zip Code Tabulation Area level; future analyses should explore smaller units of geography. For analyses comparing patient- and community-level measures, patient-level social risk data were only available for a small and biased sample of patients. Additional research should be conducted to further explore the concordance between patient- and community-level measures of social risk as more patient-level social-risk screening data become available. Finally, among ADVANCE patients, ED-use data were only available for Medicaid patients in Oregon; we were unable to obtain access to Washington Medicaid data within the study time frame.