This study explored the statistical association of key social determinants of health with specific health outcomes to validate impact and then created weighted categorical matrices.
Objectives: This study explored the contributions of social determinants of health (SDOH) to measures of population health—specifically cost, hospitalization rates, rate of emergency department utilization, and health status—in Texas.
Study Design: The study associated common SDOH metrics from public data sources (county specific) with health plan enrollment data (including demographics, counties, and zip codes) and medical and pharmaceutical annual claims data.
Methods: Following correlation analyses to reduce variables, the contribution of each SDOH individually and by category to the health outcomes was evaluated. Separate matrices for age populations (under age 19, general population [all ages], and ≥ 65 years) were created with assigned weights of influence for categories and the factors within each category.
Results: The contributions of the categories varied by population, confirming that different SDOH influence populations to varying degrees. This was reflected in each model. The largest contributor to cost for the general population and for the group 65 years and older was factors grouped as health outcomes (such as perceived health), at 43.5% contribution and 37.7% contribution, respectively. Yet for the population younger than 19 years, the largest contributor to cost was socioeconomic factors (such as unemployment rate), at 40.2%. The other performance measures also varied by population and the mix and weight of determinants.
Conclusions: This study and the developed population-based matrices can provide a valuable framework for reporting the impact of SDOH on health care quality. The variation suggests the need for further research on how age groups react to the social environment.
Am J Manag Care. 2021;27(3):e89-e96. https://doi.org/10.37765/ajmc.2021.88603
The study proposed a conceptual framework useful for the integration of social and behavioral data into population health strategies at the patient, health system, and community levels. This study:
A growing recognition is occurring of the role of “place,” which is defined as various environments and settings (eg, school, church, workplace, neighborhood).1 Place has been found to affect health outcomes and provider quality-of-care scores across the United States.2-5 However, most research on health care quality to date has focused on select quality measures and reported on a health plan or provider level using claims or medical record data collected at the individual patient level. This methodology often overlooks the role of place and social determinants of health (SDOH) characteristics.2,6 Little is known about whether, or how, population-level care quality scores differ by county-level demographic characteristics and, in particular, community SDOH factors.5 The integration of health care quality data and SDOH data is particularly important to support the transition to value-based care and to inform health plans and providers on how to target interventions, programs, and funding into communities to improve health outcomes.7
Multiple metrics, indices, and rankings exist for assessing community health, including county-level, community, and neighborhood indicators; well-being indices; deprivation indices; and health indicators and rankings.8-12 A recognized ranking in the United States is the County Health Rankings (CHRs), created by the University of Wisconsin Population Health Institute.13,14 The CHRs provide a set of rankings that are simple to interpret and use for public health policy decision makers to spur population health action. The conceptual framework for the index includes modifiable health factors—health behaviors, clinical care (access), health outcomes, social and economic factors, and physical environment. Previous studies of the CHRs found that some factors are more closely associated with health outcomes than others.15 Henning-Smith et al extended the CHRs by assessing differences between rural and urban counties modeled in Medicare. They found inconsistent associations across 3 quality measures used (preventable hospitalization2 and diabetic and mammography screening) and the direction of association varied, indicating that risk adjustment for sociodemographic data alone is not sufficient to address the role of place.2
The CDC created the social vulnerability index, which focuses on community preparedness, resilience, and planning for natural hazards.16 The index contains data in 4 themes: (1) socioeconomic status (SES), (2) household composition, (3) minority status and language, and (4) housing and transportation. The index was created for natural hazards; however, utility has been suggested for access-to-care studies, particularly given the inclusion of social factors (transportation, housing).16
To date, there are no methods for integrated assessment of social and behavioral determinants of health and for statistical assessment of how these factors contribute to key health metrics. In addition, there are no methods available that focus specifically on health care quality data for age-specific populations or that are designed to support health plans in health care quality reporting. This study seeks to fill that gap and to build on existing CHRs and indices. Using county-level data on SDOH, health care performance and outcomes measures, and population characteristics, SDOH indices for health care quality were created and validated. The results from this study will help to illuminate the role of place-based inﬂuences on population health and quality-of-care measures. These results may also help health plans and health care providers to better identify and improve the health of their populations by accounting for the nontreatment factors affecting risk and variation. Additionally, the findings are valuable to policy makers to inform care delivery and health care policy.
The state of Texas and its 254 counties served as the study area for index creation. “Place” for this analysis was the county in which the insured member resided.
Creation of the Aggregated Index
The study started by using the CHRs,13,14 which provided data compiled from a variety of national and state data sources. The CHRs group more than 30 measures by general category: health outcomes, health behaviors, clinical care, social and economic factors, and physical environment. Each category contains subcategories, which are defined as 1 or more measures from data sources. Each measure stands alone and is distinguished by a specific unit of measure, such as percentage, rate per 1000, and air particulate matter density.
To compare the overall SDOH across counties, the disparate measures were converted into determinant-specific scores for each. Ranking was converted from a metric value into an indicator between the values of 1 and 5, with 1 being a favorable or reasonable level of social or behavioral conditions and 5 being undesirable. This allowed the summation of a total score for each component, using the scores of each contributing factor.
CHRs were supplemented by additional factors supported by previous studies, which contributed to the robustness of the models (eg, access to primary care and mental health professionals).17-19 Furthermore, social and behavioral determinants were hypothesized to vary by age and population group, which resulted in different matrices for children and adolescents younger than 19 years, persons 65 years and older, and the general population (all ages). The categories used for measurement were health outcomes (child and infant mortality, life expectancy, and residents’ perception of their health), health behaviors (eg, smoking, alcohol use, sexually transmitted diseases, teen pregnancy), access (access to primary care physicians and behavioral health professionals), social and economic environment (eg, education, income, community safety, violence), and physical environment (eg, water quality, air quality, housing, food environment). Specific metrics within each category are identified in the tables and eAppendix (available at ajmc.com).
Statistical analyses were conducted to identify measures that were highly correlated for subcomponent reduction. Once relevant measures were identified, they were analyzed to assess the contribution of each measure to select health performance indicators: total cost per member per year, rate of acute inpatient hospitalization, rate of emergency department (ED) utilization, and 2 measures of health status: 3M Clinical Risk Groups (CRGs) and corresponding patient severity scores. CRGs are derived from standard claims data and pharmaceutical data to assign a level associated with severity of disease to each enrollee in a population.14
Health care outcomes were derived from health care claims and enrollment data from 2016. Data sources included Texas Medicaid and Children’s Health Insurance Program managed care claims, Optum Clinformatics DataMart claims data (commercial health plans and Medicare Advantage plans), and Medicare claims data. Member ages were derived from the enrollment files for each data source. Cost was computed as the allowed amount, inclusive of patient pay portions and coordination of benefit amounts. Acute inpatient admissions and ED visits were identified by type of bill codes, revenue codes, and place of service codes. CRG severity scores were derived by the 3M software using the full claims history per member. Results were computed for more than 10 million insured members within the 3 population groups in each of the 254 counties in Texas and reported as a per-member per-year amount for cost, a rate per year for acute inpatient admissions and ED visits, and a CRG and severity for the risk value.
Validation of the Aggregated Index
Pearson correlation coefficients were calculated to identify possible collinearity between SDOH categories. Regression models of various types were developed to assess weights for the impact of SDOH categories on the 5 outcomes. For medical cost, acute inpatient admissions, and ED visits, linear regression models were used. For 3M CRGs and CRG severity level, proportional odds models were employed to assess the odds of each CRG level and corresponding severity. Statistical analyses were performed using SAS 9.4 (SAS Institute). Models were developed separately for each age group and included sex, age, and insurance types (Medicare, Medicaid, Medicare Advantage, and commercial) as covariates along with SDOH variables. Race and income could not be included in the models because that information was not available in all data sources. Akaike information criterion values of full models and reduced models (intercept only) were compared to determine the value of SDOH variables to the model.
The final resulting percentages of impact for each category are displayed in the conceptual frameworks shown in Table 1, Table 2, and Table 3. The tables identify the percentage of overall contribution by category, as well as the independent SDOH variables within each category. For each outcome and age group model, the percentage of variance explained by each SDOH was calculated from the model results (Table 4 and Table 5). Percentages were derived by dividing the coefficient of interest for a social determinant over the sum of the absolute values of all social determinant coefficients. These percentages were then averaged across the outcomes for the age group, giving an overall value to be used as the weighting system (Table 4). For the 3M CRG severity outcome, additional steps were taken to account for the highest possible severity score differing between CRGs. For example, CRGs of 1 and 2 have no severity score; a CRG of 3 has a maximum severity score of 2; CRGs of 4, 5, 8, and 9 have a maximum severity score of 4; and for CRGs 6 and 7, the severity score can go up to 6. For this, the weight for all variables was calculated for each individual CRG (1 through 9) and then averaged for each age group before being included in the average calculation from the 4 other outcome results (eAppendix Table).
Overall, the individual subcategories of social determinants were mostly shown to be independent based on Pearson correlation coefficients, with a few exceptions. The coefficients and their positive or negative effect are presented in the eAppendix. Food environment and residing in a food desert showed an anticipated correlation for all population groups. Access metrics (access to primary care physicians, access to dentists, and access to mental health professionals) were highly correlated in all populations as well. Dental access was removed from the model because it was highly correlated with other access variables and dental care was not a designated health outcome.
The associations of the various categories in each population matrix with the total medical cost per member per year are shown in Table 4. The largest contributor to cost for the general population and for the group older than 65 years was health outcomes, at 43.5% and 37.7% contribution, respectively. Yet for the population younger than 19 years, the largest contributor to cost was social economic factors, at 40.2%, where high rates of unemployment, single-parent households, and injury rates were associated with higher annual health care costs.
The contributions of the categories varied by population, thus confirming that different social determinants affect populations to varying degrees and may reflect the variation in subcategories for each model. The impact of different and even the same determinants on age-delineated population groups stresses the need to further evaluate how age groups react to or are affected by the environment in which they live.
Models also illustrate that health outcome factors were the highest contributor to the acute inpatient admission rate per year for the general population, at 48.2%. However, for the population 65 years and older, health behavior factors were the highest contributor, at 43.5%, with health outcomes accounting for another 31.9%; together they influenced the rate by 75.4%. The population younger than 19 years experienced greater influence on the inpatient rate per county from the combined factors of physical environment and social/economic environment at 60%. Access was a contributor across all populations.
A markedly different pattern was revealed for the contribution of the various categories to the rates of ED visits per year. For the general population, there was a strong relationship with health behaviors contributing 56.8% to the variations by county. The population 65 years and older also had the strongest relationship with health behaviors (47.2%) but additionally had a strong relationship with health outcomes contributing 21.7% to the variations across counties. The group younger than 19 years had a strong relationship with health outcomes contributing 34.6% to the variations across counties in the rate of ED utilization. Interestingly, access was a significant but low contributor for the general population and the group younger than 19 years.
A higher CRG risk score is an indicator of poorer health status and greater severity of illness. Table 4 identifies the association of the SDOH categories with the variations in the mean CRG score per county by population group. Health outcomes remained the top contributor for the general population and the group 65 years and older, with contribution rates of 45.4% and 67.3%, respectively. For the group younger than 19 years, access played a large role at 33.5%, as did physical environment at 45.1%.
Using the resulting percentages created by the model parameters, conceptual framework matrices were developed for each age group as seen in Tables 1 through 3. Percentages taken from the model output were rounded to the nearest 5% for ease of use.
This study expanded upon a widely used US SDOH index (Wisconsin’s CHRs) through the addition of variables and the use of regression models to determine the impact of SDOH and their aggregated categories for 3 commonly reported health care quality outcomes. In their 2015 analysis of the CHRs at the national level, Hood et al found that health behaviors were the biggest predictor of outcomes in Texas (for mortality and morbidity).20 In this analysis of Texas counties, health behaviors were found to be the largest predictor for some outcomes and some population segments; however, the overall impact varied. Variation was found in both the suite of factors and weights for each population segment and for the impact of these factors on health care quality outcomes. In particular, in younger populations, physical environment and socioeconomic factors showed the strongest effect on the rate of acute inpatient admissions in the study. The inpatient admission rates of the population 65 years and older were most affected by health behaviors and health outcomes. In the conceptual framework, health behaviors are rates of behavioral variables at the county level, such as the proportion of smokers or the rate of obesity (see eAppendix). For the general population, health outcomes and socioeconomic factors had the largest impact on rate of acute inpatient admissions. The association between access to care and acute inpatient admission rates was consistent across all 3 age groups.
ED utilization rates were most influenced by health behaviors for the general population and the population 65 years and older. Interestingly, although access factors were significant for their relationship to ED use, their contribution was low for the group younger than 19 years and the general population at 2.2% and highest for the group 65 years and older at 11%. For the general population, socioeconomic factors contributed 18%. For the population 65 years and older, socioeconomic factors contributed 11%. For those younger than 19 years, health outcomes and socioeconomic categories were the highest impactors at 35% and 28%, respectively. ED utilization and preventable ED utilization have been consistently linked to populations of lower SES and underserved populations across the United States and are costly in terms of both health care costs and quality of care.21,22 These results are consistent and find an association between SES and ED utilization. Importantly, socioeconomic factors are upstream and influence health behaviors, leading to a compound effect on outcomes.
It has been widely noted that place contributes to the variation in key health and quality outcomes such as cost, resource use, and risk scores indicating health severity and disease status. However, a recent literature review of risk prediction and segmentation models found that only 20% incorporated geography.23 It further found that use of demographic data was typically limited to age and gender, and inclusion of community social determinants factors was uncommon.23 In their recent study, Henning et al2 found that failing to account for geographic location and associated community-level factors in quality adjustment models may lead to biased quality scores for Medicare populations.
This study and the developed population-based matrices can provide valuable information for reporting efforts on quality in health care, specifically for health plans reporting on provider results. As the executive order on “Improving Price and Quality Transparency in American Healthcare to Put Patients First” (issued June 24, 2019)24 is implemented to develop a Health Quality Roadmap that aims to establish, adopt, and improve reporting on quality measures across publicly funded health systems, the need to incorporate SDOH should be addressed. The matrices offered in this study serve as an excellent beginning to enhance the transparency and quality reporting efforts within the Roadmap to be developed.24 Although the findings were based on Texas metrics, the results may be generalizable to other states because the metrics methods are consistent. Furthermore, with the evidence from this study, interventions from health plans that address such social determinants can be developed and targeted to the relevant populations. A recent study by the Center for Population Health Information Technology at Johns Hopkins Bloomberg School of Public Health supports the prospect that interventions that target social and behavioral risk factors can improve health outcomes.25 The study proposed a conceptual framework useful for the integration of social and behavioral data into population health strategies at the patient, health system, and community levels.
Limitations to this study are primarily associated with the assignment of population statistics to a geographic area of county, which can vary greatly in size and population diversity. For example, much of the metropolitan area of Houston, Texas, is in Harris County, which includes pockets of very low SES and areas of very high SES. It would have been preferable to use a more refined geographic level, but not all data came in zip codes, so counties were the only consistently available geographic indicator. Additionally, counties that had no reported index value were assigned the mean for the state, which may not accurately reflect conditions. Further limitations included the inability to consider race and income of the study populations.
This study and the developed population-based matrices can provide a valuable framework for reporting the impact of SDOH on health care quality. The variation suggests the need for further research on how age groups react to the social environment.
Author Affiliations: Department of Management, Policy and Community Health Practice, UTHealth School of Public Health (TMK, CS, LH), Houston, TX.
Source of Funding: None.
Author Disclosures: The authors report no relationship or financial interest with any entity that would pose a conflict of interest with the subject matter of this article.
Authorship Information: Concept and design (TMK, CS, LH); acquisition of data (TMK, CS, LH); analysis and interpretation of data (TMK, CS, LH); drafting of the manuscript (TMK, CS, LH); critical revision of the manuscript for important intellectual content (TMK, CS, LH); and statistical analysis (CS).
Address Correspondence to: Trudy Millard Krause, DrPH, Department of Management, Policy and Community Health Practice, UTHealth School of Public Health, RAS 1017, 1200 Pressler St, Houston, TX 77030. Email: Trudy.M.Krause@uth.tmc.edu.
1. The Institute of Medicine. The Future of the Public's Health in the 21st Century. The National Academies Press; 2003. doi:10.17226/10548
2. Henning-Smith C, Prasad S, Casey M, Kozhimannil K, Moscovice I. Rural-urban differences in Medicare quality scores persist after adjusting for sociodemographic and environmental characteristics. J Rural Health. 2019;35(1):58-67. doi:10.1111/jrh.12261
3. Graham G, Ostrowski M, Sabina A. Defeating the zip code health paradigm: data, technology, and collaboration are key. Health Affairs. August 6, 2015. Accessed December 12, 2019. https://www.healthaffairs.org/do/10.1377/hblog20150806.049730/full/
4. What works? strategies to improve rural health. County Health Rankings & Roadmaps. July 2016. Accessed August 8, 2016. https://www.countyhealthrankings.org/reports/what-works-strategies-improve-rural-health
5. Report to Congress: social risk factors and performance under Medicare’s value-based purchasing programs. Ofﬁce of the Assistant Secretary for Planning and Evaluation. December 21, 2016. Accessed March 13, 2017. https://aspe.hhs.gov/pdf-report/report-congress-social-risk-factors-and-performance-under-medicares-value-based-purchasing-programs
6. Knighton AJ, Savitz L, Belnap T, Stephenson B, VanDerslice J. Introduction of an area deprivation index measuring patient socioeconomic status in an integrated health system: implications for population health. EGEMS (Wash DC). 2016;4(3):1238. doi:10.13063/2327-9214.1238
7. Phillips RL, Liaw W, Crampton P, et al. How other countries use deprivation indices—and why the United States desperately needs one. Health Aff (Millwood). 2016;35(11):1991-1998. doi:10.1377/hlthaff.2016.0709
8. Environmental scan of existing domains and indicators to inform development of a new measurement framework for assessing the health and vitality of communities. National Committee on Vital and Health Statistics. June 2016. Accessed October 6, 2016. https://www.ncvhs.hhs.gov/wp-content/uploads/2016/06/NCVHS-Indicators-Envirn-Scan_2016-06-01-FINAL.pdf
9. Howell EM, Pettit KLS, Ormond BA, Kingsley GT. Using the National Neighborhood Indicators Partnership to improve public health. J Public Health Manag Pract. 2003;9(3):235-242. doi:10.1097/00124784-200305000-00009
10. Erwin PC, Myers CR, Myers GM, Daugherty LM. State responses to America’s health rankings: the search for meaning, utility, and value. J Public Health Manag Pract. 2014;20(5):472-480. doi:10.1097/PHH.0b013e318211b49f
11. Rohan AMK, Booske BC, Remington PL. Using the Wisconsin County Health Rankings to catalyze community health improvement. J Public Health Manag Pract. 2009;15(1):24-32. doi:10.1097/PHH.0b013e3181903bf8
12. Kreiger N, Waterman PD, Spasojevic J, Li W, Maduro G, Van Wye G. Public health monitoring of privilege and deprivation with the index of concentration at the extremes. Am J Public Health. 2016;106(2):256-263. doi:10.2105/AJPH.2015.302955
13. Remington PL, Catlin BB, Gennuso KP. The County Health Rankings: rationale and methods. Popul Health Metr. 2015;13:11. doi:10.1186/s12963-015-0044-2
14. 3M Clinical Risk Groups: measuring risk, managing care. 3M. 2016. Accessed November 11, 2019. https://multimedia.3m.com/mws/media/765833O/3m-crgs-measuring-risk-managing-care-white-paper.pdf
15. Hendryx M, Ahern MM, Zulig KJ. Improving the environmental quality component of the County Health Rankings model. Am J Public Health. 2013;103(4):727-732. doi:10.2105/AJPH.2012.301016
16. Flanagan BE, Hallisey EJ, Adams E, Lavery A. Measuring community vulnerability to natural and anthropogenic hazards: the Centers for Disease Control and Prevention’s Social Vulnerability Index. J Environ Health. 2018;80(10):34-36.
17. Institute of Medicine. Capturing Social and Behavioral Domains and Measures in Electronic Health Records: Phase 2. The National Academies Press; 2014. Accessed January 15, 2020. https://www.ncbi.nlm.nih.gov/books/NBK269334/
18. Park H, Roubal AM, Jovaag A, Gennuso KP, Catlin B. Relative contributions of a set of health factors to selected health outcomes. Am J Prev Med. 2015;49(6):961-969. doi:10.1016/j.amepre.2015.07.016
19. Athens JK, Catlin BB, Remington PL, Gangnon RE. Using empirical Bayes methods to rank counties on population health measures. Prev Chronic Dis. 2013;10:E129. doi:10.5888/PCD10.130028
20. Hood CM, Gennuso KP, Swain GR, Catlin BB. County Health Rankings: relationships between determinant factors and health outcomes. Am J Prev Med. 2016;50(2):129-135. doi:10.1016/j.amepre.2015.08.024
21. Doran KM, Raven MC, Rosenheck RA. What drives frequent emergency department use in an integrated health system? national data from the Veterans Health Administration. Ann Emerg Med. 2013;62(2):151-159. doi:10.1016/j.annemergmed.2013.02.016
22. Mandelberg JH, Kuhn RE, Kohn MA. Epidemiologic analysis of an urban, public emergency department’s frequent users. Acad Emerg Med. 2000;7(6):637-646. doi:10.1111/j.1553-2712.2000.tb02037.x
23. Meddings J, Reichert H, Smith SN, et al. The impact of disability and social determinants of health on condition-specific readmissions beyond Medicare risk adjustments: a cohort study. J Gen Intern Med. 2017;32(1):71-80. doi:10.1007/s11606-016-3869-x
24. Improving price and quality transparency in American healthcare to put patients first. Fed Regist. 2020;84(124):30849-30852.
25. Predmore Z, Hatef E, Weiner JP. Integrating social and behavioral determinants of health into population health analytics: a conceptual framework and suggested road map. Popul Health Manag. 2019;22(6):488-494. doi:10.1089/pop.2018.0151