Effects of Patient Complexity and Social Determinants of Health

Researchers Dive into Healthcare Utilization

08/28/17  Portland, Ore.

By Anna Templeton and the OCHIN/OHSU study team: Erika Cottrell, Katie Dambrun, Ashley Kroll, Thuy Lee, Jee Oakley, Jean O'Malley

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. A large body of research suggests that SDH may contribute as much or even more to health outcomes than clinical factors. However, care quality metrics typically emphasize disease specific, clinical outcomes, such as the number or severity of chronic conditions (also referred to as clinical complexity), and do not include measurement or adjustment for the SDH that impact health (also referred to as social complexity). In this study, we assessed whether accounting for both social and clinical complexity better explained differences in quality of care and health service utilization rates than accounting for patient clinical complexity alone.

Over a twelve month period, ADVANCE (Accelerating Data Value across a National Community Health Center Network) and OneFlorida, two of the Patient-Centered Outcomes Research Institute funded Clinical Data Research Networks (CDRN), worked closely with patient, provider, and health system stakeholders to develop and test measures assessing the combined effect of social and clinical complexity on quality of care. ADVANCE and OneFlorida comprise socially vulnerable patient populations largely reliant on Medicaid and safety net services like community health centers. We assessed clinical complexity using the Charlson Comorbidity Index, a weighted measure that predicts the risk of mortality and resource utilization for patients with a range of comorbid conditions. To assess social complexity, we utilized publicly available data on the community-level SDH of the neighborhoods where patients live.

In the first phase of the study, stakeholders prioritized community-level SDH characteristics and clinical quality metric outcomes to include in our analysis. A total of 71 stakeholders participated, representing patient, clinician, and health systems groups in the ADVANCE and OneFlorida networks. Stakeholders reviewed a list of 16 SDH categories and 7 clinical quality outcomes then ranked them according to perceived importance (see Table below). The top three SDH categories were socioeconomic status, neighborhood inequality, and transportation. The top three clinical outcomes were chronic disease management, risk factor screening, and preventive care. 

Stakeholder identified variables

Clinical Quality Metrics
Community-Level SDH
  • Care of chronic diseases (e.g. diabetes control)
  • Risk factor screening (e.g. alcohol/ drug misuse)
  • Preventive care visits (e.g. well child visits)
  • Cancer screening
  • Emergency Department visits
  • Neighborhood socioeconomic status(e.g. education, income)
  • Neighborhood resources (e.g. healthy food access)
  • Transportation
  • Substance Abuse
  • Neighborhood economic conditions (e.g. level of inequality)
  • Race / ethnicity
  • Physical environment

In Phase 2, we assessed the relationship between clinical complexity (using the Charlson Comorbidity Index), social complexity (using community-level SDH variables), and quality of care. Our sample included Medicaid patients in Florida and Oregon and patients in the electronic health record systems for subsets of OneFlorida and ADVANCE. All established patients, defined as attending a primary care appointment in 2015 with at least one visit at any prior point, were included. Patient addresses were geocoded and linked to US census and American Community Survey data to describe census tract or zip code level SDH factors in the communities where patients lived.

As expected, patients in our sample live in more socially vulnerable communities than the average US population. For example, the mean Social Deprivation Index (SDI) –a composite measure of the socioeconomic conditions in a given neighborhood –for our population was 84, which means 84% of study patients live in neighborhoods with poor socioeconomic indicators. Almost half of the patients in our sample live in the most vulnerable quartile of neighborhoods (e.g. those with the highest SDI scores) in the US, whereas less than 8 percent live the least vulnerable quartile of neighborhoods (e.g. those with the lowest SDI scores). Our analyses revealed that accounting for patients' social and clinical complexity explains more of the variation in clinic-level quality of care –such as Hemoglobin A1c control or Emergency Department utilization –than does accounting for clinical complexity alone. For example, when looking at the percent of patients with poor diabetes control, those with higher levels of clinical complexity (e.g. a Charlson score higher than 2) had worse outcomes. Interestingly, if we were to assume an SDI level of 50 (the US mean) instead of 84 (the mean of our population), the percent of diabetics in poor control dropped by 2%. Although this effect may seem modest, even this slight decrease from adjusting for social complexity of patient panels could result in improved performance metrics for providers caring for patients within safety-net clinics

In the third phase of this study we again engaged stakeholders to ask if our preliminary results were consistent with their own observations and experiences. Most thought the preliminary findings were "in line with what we see anecdotally, so it's great to see the data showing this." Discussing implications of the findings, stakeholders strongly endorsed provider performance metrics and funding models for service delivery adjusted for the social complexity of patient panels. Stakeholders recommended policy and practice applications including innovations to existing care models, higher allocation of resources in vulnerable communities, and ways to integrate SDH with routine patient assessments in primary care practice. Several stakeholders were surprised the social-clinical adjustment did not have a larger effect in our sample and recommended future studies that include comparison with more affluent, privately insured populations.

For the fourth and final phase, we are developing a toolkit for other data networks to use in assessing their own patient populations, surveying other CDRN's to assess their interest and capacity to replicate the analyses conducted in this study, and pursuing additional mechanisms for broad dissemination of our findings. The study toolkit will include key lessons learned, strategies for stakeholder engagement, and resources for conducting similar research (e.g. data extraction and analysis code). Detailed study findings will be presented at local and national conferences, and are in preparation for submission to peer reviewed journals.

When we discussed findings with stakeholders, one patient advisor remarked, "It seems obvious that [social deprivation] was associated with poor diabetes control –sometimes the point of research is to confirm obvious things." In addition to demonstrating social and clinical complexity in safety net populations and the effects of combining social and clinical risk factors in provider performance measures, this study shows the value of engagement in tailoring the design and interpretation of large scale analyses to stakeholder priorities. Although future research is needed to further refine our models, our initial findings suggest measures that adjust quality measures for higher levels of social risk can help providers serving vulnerable populations better demonstrate the quality of care they deliver and justify more intensive resource use or allocation.

For any questions or more information regarding this project, please contact Katie Dambrun, dambrunk@ochin.org.