When Life Affects Health

Connecting the dots between social determinants and health care utilization Ksenia Whittal

Statements of fact and opinions expressed herein are those of the individual author(s) and are not necessarily those of the Society of Actuaries or the respective authors’ employers.


The health care financing sector is finally looking at something public health has known for decades: Funding upstream efforts to reduce the prevalence of chronic diseases is a way to reduce future health care spending. Growing recognition that social determinants are significant drivers of health and health care utilization patterns has increased the desire to better understand and identify these issues, as well as to develop actionable steps at both the population and member levels.1

There has been a growing focus on developing the ability to identify the presence of social vulnerabilities among population health entities, Medicaid state agencies, risk-taking provider organizations such as accountable care organizations (ACOs), and any entity with a vested interest in the reduction of health care spending. Therefore, it is not unusual for health actuaries to get involved in this discussion. You may ask, why actuaries? The first reason is that actuaries are experts in quantifying risks related to health care, costs and utilization patterns. If another set of factors or characteristics affects these risks, then actuaries should (at the very least) understand these factors and be involved in evaluating the impact of efforts aimed at addressing them.2

Second, health actuaries play a supporting role in a system where the primary function is to produce health (or at least treat disease). Because actuarial training is rooted in the financial aspects of health care funding, provider reimbursement arrangements and resource use, actuaries typically pay less attention to the root causes of disease and health care utilization. Let us step back from our typical day-to-day thinking and take a look at what is currently happening in the industry with regard to social determinants of health and how actuaries can play a greater role.

Social Determinants of Health

Social determinants of health are the conditions in which people are born, grow up, live and work that shape health outcomes.3 These conditions include a wide spectrum of life factors—income, housing, education, food access, transportation, social support and stress, just to name a few. Social determinants come to life in the stories of patients unable to take their prescribed medications due to lack of food, as taking these medications without food results in nausea. Similarly, patients with diabetes who are not able to keep their insulin sufficiently cold due to housing or income instability and patients who cannot get to their appointments due to lack of transportation provide additional examples of intersections between social determinants and health care. Simply stated, life affects health. Systems such as education historically have incorporated social elements into their processes such as addressing food insecurity through reduced and free lunch programs. The notion has been around for years in the public health and population health domains, and it has finally become a focus for the financial side of the health care system.

Research4 indicates social determinant factors can be linked to each of the top 10 causes of premature mortality in the United States, which include diseases such as heart and cardiovascular diseases, cancer, respiratory diseases, diabetes, kidney disease, injuries and suicide. These researchers used mortality as a proxy for morbidity and made the case that addressing these social determinants is key for improving health outcomes and reducing disparities in health. There are two reasons this is important:

  1. The conversation around social determinants frequently involves these figures. While it is not unreasonable to use mortality as the ultimate health outcome, most health care utilization (and thus cost) occurs before this outcome takes place. More research is needed to understand the association among social determinants and the prevalence of chronic conditions (although there is some evidence already5,6), health care service utilization and clinical outcomes.
  2. When it comes to implementing programs aimed at addressing social determinants, the costs are immediate and the benefits (e.g., reduced mortality) are deferred, sometimes decades into the future. This notion alone should temper expectations for a return from these programs in the short term. Just as one does not expect to see a reduction in next year’s medical claim costs for a member who quit smoking this year, this does not mean society should stop investing in smoking cessation programs. Research argues: “The cost-effectiveness of various interventions to improve population health is less clear. In a vexing example of double standards, public investments in health promotion seem to require evidence that future savings in health and other social costs will offset the investments in prevention. Medical treatments do not need to measure up to this standard; all that is required here is evidence of safety and effectiveness. The cost- effectiveness challenge often is made tougher by a sense that the benefits need to accrue directly and in the short term to the payer making the investments. Neither of these two conditions applies in many interventions for health promotion.”7

A Look at Data

While individual-level data on social determinants of health is currently sparse and difficult to collect,8 the availability of detailed codes within the International Classification of Diseases, 10th Revision (ICD-10) is a place to start. Although still not frequently used, the subset of codes that most closely aligns with the relevant social determinants can be the best source of this information, particularly if providers make a consistent effort to use these codes in the years to come. Armed with a database of 2016 medical claims experience containing ICD-10 diagnosis coding, we set out to test some of the existing hypotheses and discover new ones about health care utilization among claimants who experience such vulnerabilities.

The ICD-10 codes that represent social determinants are those starting with “Z,” referenced as Z-codes. In particular, we focused on nine categories of codes in the current classification:

  • Z55XXX. Educational problems.
  • Z56XXX. Employment problems.
  • Z57XXX. Occupational hazard exposure.
  • Z59XXX. Housing problems.
  • Z60XXX. Various social problems.
  • Z62XXX. Child/parent problems.
  • Z63XXX. Family problems.
  • Z64XXX. Unwanted pregnancy.
  • Z65XXX. Criminal problems.

We posed one main question we were hoping to answer with this analysis: After controlling for differences in morbidity and demographics, are there significant differences in cost and utilization among members affected by these social circumstances? If there are differences, are they statistically valid (and not just random)?

We conducted a matched cohort analysis where we drew a matched sample of claimants without a claim with a Z-code for each group of claimants in the nine Z-cohort categories. The matched samples of members were similarly distributed by age, gender, type of coverage9 and morbidity represented by a risk score10 range. We then compared risk-adjusted11 costs and utilization metrics for various medical services between the Z-cohorts and the matched cohorts. The source database is Milliman’s proprietary research database containing 49 million lives with experience nationwide.

Our findings include a mix of expected and unexpected results. The total risk-adjusted health care costs12 per member per month (PMPM) between Z-cohorts and the matched cohorts were similar, on average within $8 PMPM (or 2.4 percent) of each other (ranging from –$32 PMPM to +$40 PMPM), as shown in Figures 1 and 2. The Z-cohort categories with the greatest disparity in total costs were cohorts dealing with child/parent problems, occupational hazard and family problems. Assuming that the utilization of medical care by the matched cohorts is appropriate, the observed differences in costs among these groups of patients imply either underutilization or overutilization of services for a given level of morbidity, and neither is ideal. Chronic underutilization of needed medical care typically leads to unnecessary complications or worsening of potentially treatable conditions, resulting in deferred higher-cost treatment. Medication nonadherence alone was estimated to cost the U.S. health care system between $100 billion and $289 billion annually.13 On the other hand, overutilization of health care services may be indicative of a significantly higher prevalence of medical conditions treated with these services, inefficiencies, or lack of integration and coordination of services among providers, or delivering unnecessary or duplicate services. According to a report by the Institute of Medicine, the estimated value of unnecessary health care services was as large as $210 billion in 2010.14

Figure 1: Comparison of Total Risk-adjusted Cost PMPM, by Z-cohort

Hover Over Image for Specific Data

Source: Milliman Proprietary Research Claim Database, CY2016. Analysis conducted by Whittal, K., Milliman. September 2018.

Figure 2: Comparison of Differences in Total Risk-adjusted Cost PMPM, by Z-cohort

Hover Over Image for Specific Data

Source: Milliman Proprietary Research Claim Database, CY2016. Analysis conducted by Whittal, K., Milliman. September 2018.

The results of the analysis also show that patients with social vulnerabilities and their matched counterparts utilize health care services differently. After breaking down the total cost of care into its component services, we saw a number of noteworthy differences (see Figure 3). While emergency room (ER) visits and inpatient hospital visits typically get most of the attention when it comes to cost containment through addressing social determinants of health, the data show that the greatest disparity in costs arise from utilization (and hence prevalence) of psychiatric mental health (MH) and substance abuse (SA) services. Cost differences on a PMPM basis for these services are more than four and three times greater, respectively.

Figure 3: Comparison of Risk-adjusted Cost PMPM Between Z-cohorts and Matched Cohorts, for Selected Services

Hover Over Image for Specific Data

Source: Milliman Proprietary Research Claim Database, CY2016. Analysis conducted by Whittal, K., Milliman. September 2018.

Additionally, we saw significant potential overutilization of inpatient hospital, ambulance and emergency services across the majority of Z-cohorts, as shown in Figure 4. (Figure 5 compares these costs on a percentage basis.) The most notable difference is $97 PMPM in psychiatric services for members with child/parent problems. If we compare these costs on a percentage basis, differences for MH/SA services range from 91 percent to 734 percent higher for patients with social vulnerabilities as compared to their matched counterparts for all cohorts, with the exception of the unwanted pregnancy and occupational hazard cohorts. This finding is consistent with existing research on this topic, which has found positive associations between food insecurity and MH problems in children and adults.15,16,17,18,19,20 Certainly, we are not aiming to imply causality between these events in either direction (i.e., does prevalence of MH/SA develop or increase because of life difficulties, or is it the other way around?).

Figure 4: Differences in Risk-adjusted Cost PMPM Between Z-cohorts and Matched Cohorts, for Potentially Overutilized Services

Hover Over Image for Specific Data

Source: Milliman Proprietary Research Claim Database, CY2016. Analysis conducted by Whittal, K., Milliman. September 2018.

Figure 5: Differences in Risk-adjusted Cost (as a Percentage) Between Z-cohorts and Matched Cohorts, for Potentially Overutilized Services

Hover Over Image for Specific Data

Source: Milliman Proprietary Research Claim Database, CY2016. Analysis conducted by Whittal, K., Milliman. September 2018.

We also noted a significant potential underutilization of surgical services, pharmacy, maternity and radiology, as displayed in Figure 6. Interestingly, there were no consistent disparities in the preventive services category, as we originally hypothesized could be the case for populations with social vulnerabilities. The reduction in pharmacy cost was expected—it is probable that individuals dealing with stressful life circumstances would be less able to adhere to their treatment plans, keep up with refills of maintenance medications and so on.

Figure 6: Differences in Risk-adjusted Cost PMPM Between Z-cohorts and Matched Cohorts, for Potentially Underutilized Services

Hover Over Image for Specific Data

Source: Milliman Proprietary Research Claim Database, CY2016. Analysis conducted by Whittal, K., Milliman. September 2018.

One unexpected finding was the reduction in surgical services. Upon further investigation of the types of surgical services that were reduced, the most common less-utilized procedures were elective (non-acute) procedures such as biopsies, and laparoscopic procedures such as appendectomy, colonoscopies, esophagogastroduodenoscopy, angioplasty/stent placement, cataract removal, knee arthroscopy, tissue debridement and others.

As a final test of the validity of the observed differences, we performed a regression analysis in which we modeled cost of the four service categories with the greatest differences (MH/SA, maternity, surgical and pharmacy) using age, gender, risk score, type of coverage and presence of Z-code (one indicator for each type of Z category) as independent predictors. The resulting p-values for the presence of Z-code binary indicator variables, with the exception of a few cases, are statistically significant at the 1 percent level.

Limitations to Consider

There is a bias inherent in this type of analysis. In this approach, a member can be identified with a particular social issue only if he or she has health coverage, has a health issue and seeks medical care from a health care provider, who in turn becomes aware of and codes the social issues using the ICD-10 coding methodology and files a health insurance claim. Many things need to “go right” for the social determinants data to make its way into an administrative claim database. This by itself significantly limits the population of interest, as any individual facing the same social determinant issues who does not meet any one of these criteria will not be identified using these codes. In addition, this analysis considered a 12-month snapshot of experience in time, rather than tracking these members longitudinally over several years. Hence, it is less clear how the presence of such codes on a medical claim relates to the actual timing of the underlying event or circumstances identified with the code during the 2016 calendar year. Certainly, some life events have effects long into the future, or take a longer time to manifest in a medical issue. Because we are looking at a pooled group of members with different coverage types, there are inherent differences in the data quality and coding levels that could materially affect risk scores. Finally, there is a lack of data on potentially important confounders not present in the administrative claim data such as income, educational attainment, family composition and other demographic variables, including ethnicity. While we recognize these limitations of the study, we believe there is still knowledge to be gained from analysis on these emerging data.

In conclusion, what do all these findings suggest and what is the takeaway for a reader? Actuaries tend to root their decision making in data-driven evidence. This analytical exploration of the connection between social determinants and the cost of health care, even in this admittedly limited capacity, does reveal an undeniable variation in utilization of health care services by individuals facing social vulnerabilities. There is much motivational discourse and evidence around the importance of addressing these complex social issues. Still, very little gets done on a large scale beyond local pilot projects, as it is difficult for any single stakeholder to take complete ownership and coordinate efforts of multiple entities that need to be involved. The systemic multisector policy changes to education, the food industry and the health sector, along with redesigned financial incentives for all parties that would be required to achieve a desired level of population wellness—including social, mental and physical well-being—can seem too far out of reach. Actuaries are in a unique position to bring together perspectives on both cost and health outcomes, and ground the motivation to recognize the impact of the social determinants (or life conditions) in solid actuarial analysis and data-driven evidence.

Ksenia Whittal, FSA, MAAA, is a consulting actuary with Milliman’s Denver Health practice.

References:

  1. 1. United States Department of Health and Human Services Office of the Assistant Secretary for Planning and Evaluation. Report to Congress: Social Risk Factors and Performance Under Medicare’s Value-based Purchasing Programs. U.S. Department of Health and Human Services, December 2016, (accessed September 11, 2018).
  2. 2. Task Force to Revise ASOP No. 12 of the General Committee of the Actuarial Standards Board. Actuarial Standard of Practice No.12. Risk Classification (for All ​Practice Areas). Actuarial Standards Board, December 2005, (accessed September 9, 2018).
  3. 3. World Health Organization. About Social Determinants of Health. World Health Organization: Social Determinants of Health, (accessed September 11, 2018).
  4. 4. McGinnis, J. Michael, Pamela Williams-Russo, and James R. Knickman. 2002. The Case for More Active Policy Attention to Health Promotion. Health Affairs 21, no 2.
  5. 5. Seligman, Hilary K., Barbara A. Laraia, and Margot B. Kushel. 2010. Food Insecurity is Associated With Chronic Disease Among Low-income NHANES Participants. The Journal of Nutrition 140, no 2:304–310.
  6. 6. Jones, Andrew D. 2017. Food Insecurity and Mental Health Status: A Global Analysis of 149 Countries. American Journal of Preventive Medicine 53, no 2:264–273.
  7. 7. Supra note 4.
  8. 8. While there are a number of screening tools designed specifically to screen for social determinants of health, the actual implementation of such screenings is time-consuming, potentially expensive and not always fruitful.
  9. 9. The types of insurance coverage present in our population of interest included employer-sponsored (large and small groups), individual Patient Protection and Affordable Care Act (ACA) policies, Medicaid, Medicare Advantage, Medicare Supplement and some unknown. The type of coverage affects provider reimbursement levels and allowed costs, so it is important to ensure there are no large differences in distribution of claimants by the source of coverage that are distorting the cost comparison.
  10. 10. A risk score is a common measure of morbidity for an individual, developed using medical diagnoses and assigned coefficients. Risk scores are calibrated such that in a population with average demographics and morbidity level, the average risk score is 1.00. In this analysis, we relied on the Milliman Advanced Risk Adjusters (MARA), commercial CxAdjuster concurrent model for all populations.
  11. 11. Risk adjustment refers to a process of normalizing a metric of interest (most commonly cost or utilization) between two populations for inherent differences in morbidity and demographic profile to make them comparable. Mechanically, this is done at a population level by dividing the average metric of interest by the population’s average risk score. Although we were already working with matched samples, we still risk-adjusted our costs and utilization metrics in order to ensure any observed differences were not driven by differences in morbidity and demographic profile.
  12. 12. The total cost represents allowed claim cost levels of medical and pharmacy benefits.
  13. 13. Viswanathan, M., C.E. Golin, C.D. Jones, M. Ashok, S.J. Blalock, R.C. Wines, et al. 2012. Interventions to Improve Adherence to Self-administered Medications for Chronic Diseases in the United States. Annals of Internal Medicine 157, no. 11:785–795.
  14. 14. Institute of Medicine. 2013. Best Care at Lower Cost: The Path to Continuously ​Learning Health Care in America. Washington, D.C.: The National Academies Press, (accessed September 11, 2018).
  15. 15. Supra note 6.
  16. 16. Cook, J.T., and D.A. Frank. 2008. Food Security, Poverty and Human Development in the United States. Annals of the New York Academy of Sciences 1136, no. 1:193­–209.
  17. 17. Mammen, S., J.W. Bauer, and L. Richards. 2009. Understanding Persistent Food Insecurity: A Paradox of Place and Circumstance. Social Indicators Research 92, no. 1:151–168.
  18. 18. Alaimo, K., C.M. Olson, and E.A. Frongillo. 2001. Food Insufficiency and American School-aged Children’s Cognitive, Academic and Psychosocial Development. Pediatrics 108, no. 1:44–53.
  19. 19. Jyoti, D.F., E.A. Frongillo, and S.J. Jones. 2005. Food Insecurity Affects School Children’s Academic Performance, Weight Gain and Social Skills. The Journal of Nutrition 135, no. 12:2831.
  20. 20. Compton, Michael T., and Ruth S. Shim (eds.). 2015. The Social Determinants of Mental Health. Washington, D.C.: American Psychiatric Publishing.

Copyright © 2018 by the Society of Actuaries, Schaumburg, Illinois.