Reducing Medical Costs of Health Insurance: The COVID-19 Stress Testing and Portfolio Effects

Abstract

Reducing medical costs is one of the three strategic aims of U.S. health care reform. This research quantifies potential expense reductions of U.S. health insurers and examines portfolio effects on expenses of differential health insurance mixes. We use the coronavirus disease 2019 (COVID-19) pandemic as a natural stress test for potential savings in medical services. We employ a two-stage residual inclusion generalized linear model and uncover differential expense reductions for four major health insurance markets: individual, group, Medicaid managed care, and Medicare Advantage. Our results could serve as a benchmark for reducing medical costs through redesigning insurance coverages following the most effective group insurance model. The results also provide policy implications on restructuring health insurance markets to increase efficiency. We empirically examine mutual impacts among major health insurance markets, and document explicitly three optimal health insurance portfolios in reducing expenses of health care: Medicaid managed care and individual plans, exclusively group plans, and Medicaid managed care and Medicare Advantage plans.

ACKNOWLEDGMENTS

The authors gratefully acknowledge the valuable comments and assistance from co-editor Brian Hartman and the anonymous referees.

DISCLOSURE STATEMENT

The authors report there are no competing interests to declare.

Notes

1 Medicaid expansion may have significant impacts on a state’s overall Medicaid enrollment, and spillover effects on the enrollment of individual and other markets. This research is not designed to examine the impact of Medicaid expansion on enrollment, but to estimate potential savings for various insurance markets and portfolio effects on medical costs (the impact of one market’s enrollment on medical costs of other markets). Dependent variables are expenses of the target market.

2 Our data come from the following NAIC statements: summary of transactions with providers, enrollment by product type for health business only, exhibit of premiums, enrollment and utilization, supplemental health care exhibit, and demographics file for health companies. California (CA) health insurers are not required to report to the NAIC. However, some CA insurers who operate in both CA and other states may report their CA businesses, so our samples include some (but not all) CA insurers. Health insurers who only have Medicaid business are not required to file the supplemental health care exhibit.

3 In some NAIC financial statements, small-group and large-group details are listed separately, but they are aggregated in others. There is multicollinearity between small-group and large-group enrollments in our samples, so we do not separate them in our analyses.

4 Some insurers have no business in one or more markets (zero enrollments). We replace all the zero enrollments with one member month for the logarithmic transformation.

5 Total enrollments and total premiums of the insurers were highly correlated: The correlation coefficients were 0.97, 0.98, 0.97, and 0.82 for the samples of individual, group, Medicaid, and Medicare insurers, respectively. Since the correlation was relatively lower for the Medicare insurer sample, for a robustness check, we replaced total enrollments with total premiums, and the regression results are presented in Table C.5 in Appendix C. There were no significant differences, and our conclusions still held.

6 The size of the target market is at the insurer-state-year level. There is multicollinearity at the insurer-year level. Other control variables are all at the insurer-year level unless indicated otherwise.

7 Enrollments of other markets and ASO are at the insurer-state-year level. There is multicollinearity between “other markets” and ASO at the insurer-year level.

8 Due to multicolliearity, not all the HHIs are included in all the regression models.

9 The MLR requirement is 80% for individual insurance and 85% for Medicare Advantage plans, which is not a variable for our samples of individual and Medicare insurers. There is no mandatory national MLR requirement for Medicaid managed care plans, for which we control for the potential effects of state-level MLR requirements using state-fixed effects.

10 While they are not supposed to explain the dependent variables (expenses) of this research, Medicaid expansion-fixed effects (through the Medicaid expansion dummy) are subsumed in the included state-fixed effects.

11 The difference-in-differences (DiD) approach may also be applied for causal inference.

12 For portfolio effects, we analyze the impact of one market’s enrollment on medical costs of other markets. Dependent variables are expenses of the target market, while independent variables are enrollments in the markets of interest. Reverse causality exists when expenses of the target market are determinants of the selection of specific markets of interest.

13 The link functions are square root (individual), identity (group), log (Medicaid), and identity (Medicare) for medical expenses, and log (individual), square root (group), log (Medicaid), and log (Medicare) for other expenses.

14 We tried different numbers of replications. The estimates stabilize at 2000 replications. In some models, no residuals are significant (that is, the variables are not endogenous). To be consistent, we still use bootstrap standard errors.

15 The residuals are the same using different family distributions.

16 Some states had stricter and lengthier COVID lock-down measures (while some states were more lenient). Our models estimate average effects, with extra negative/positive state-level differences being controlled by including state-fixed effects.

17 This research aims to find the most efficient market model through estimating potential savings. We refrain from making speculations regarding the underlying explanations for potential expense reductions. This omission does not affect our conclusions. However, it should be interesting to explore efficient practices of health insurers in future research.

18 Besides empirical results and conclusions, we also make proposals regarding how to allocate potential savings, if realized. These proposals are our opinions or legitimate preferences. More discussions and inputs are needed from stakeholders on how to implement these proposals, especially the pass-throughs across different health insurance markets.

19 Marginal effect (–$10) × [log(759,646) – log(1)] = –$59, and –$59/$849 = –6.9%. We replace zero enrollment with one member month for log transformation.

20 The OLS model cannot address endogeneity issues, and it assumes linear link function and Gaussian distribution, which might not be the case. Therefore, the OLS results may not be valid and are presented only for the comparison purpose. It is not expected that the OLS and generalized linear model results are consistent.

21 This portfolio (as well as the Medicaid and Medicare portfolio in the insurer-year alternative analysis) can be referred to as suboptimal in that more enrollment in one market does not significantly increase or decrease medical costs of the other.