Ellis, Randall P.Song, Chenlu2022-03-042022-03-042021https://hdl.handle.net/2144/43962This dissertation examines various aspects of U.S. health care markets using the claim and enrollment files from a large set of employment-based insurance plans containing detailed records of service utilizations by individual consumers and their corresponding costs. The first chapter, joint with Xiaoxi Zhao, studies the impact of two different types of cost sharing: coinsurance, in which the consumer out-of-pocket cost is calculated as a fraction of total fees, and copayments, in which the consumer cost is a fixed dollar amount regardless of the fee level charged. The paper’s focus is on how these two types of cost sharing affects consumer demand and health care spending given estimated price elasticities for categories of health care services. It is well documented in the literature that health care consumption decreases with consumer out-of-pocket costs and yet remarkably little is known about whether coinsurance and copayments affect consumer demand differently. Using a dataset in which we have no information about the plan policies, we first infer the type of the observed consumer out-of-pocket costs, i.e., a coinsurance or a copayment, for a given insurance policy and a given type of service from the claims and enrollment files. We then estimate the price elasticity for this given type of service paired with the inferred type of out-of-pocket costs using a set of novel instruments and fixed effect regressions. The results show that consumption decreases with both coinsurance and copayments. Specifically, consumer demand is found to be more elastic by 0.2 to 0.5 percentage points when coinsurance is used for cost sharing instead of copayments. Our model is among the first to quantify in monetary terms the savings generated by different types of cost sharing that are widely adopted in insurance policies. The second chapter, joint with Randall P. Ellis, Heather E. Hsu, Tzu-Chun Kuo, Bruno Martins, Jeffery J. Siracuse, Ying Liu and Arlene S. Ash, uses piecewise linear regression models on monthly time series data to assess changes in diagnostic category prevalence associated with the transition from International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) to the Tenth Revision (ICD-10-CM) in October 2015. Private insurance claims from 2010 to 2017 are mapped into three widely used diagnostic categories: the Department of Health and Human Services Hierarchical Condition Categories (HHS-HCC); the Agency for Healthcare Research and Quality (AHRQ) Clinical Classification System (CCS); and the World Health Organization’s disease chapters (WHO). The analytic sample contains information on 2.1 billion enrollee person-months with 3.4 billion clinically assigned diagnosis. In all three classification systems, the ICD-10-CM implementation is associated with statistically significant changes in monthly prevalence among 58–59% of diagnostic categories. This interrupted time series analysis and cross-sectional study finds increases or decreases of 20% or more associated with the ICD-10-CM transition for nearly 1 in 6 (16%) diagnostic categories in 2 of 3 influential diagnostic classification systems, suggesting that diagnostic classification systems developed with ICD-9-CM data may need to be refined for use with ICD-10-CM data for disease surveillance, performance assessment, or risk-adjusted payment. The third chapter, joint with Corinne Andriola, examines the performance of three risk adjustment frameworks at predicting the health care spending by people with rare diseases, i.e., diseases that affect fewer than 0.05% of the population. Three risk adjustment models are considered: the Health and Human Services Hierarchical Condition Categories (HHS-HCC), the Agency for Healthcare Research and Quality Clinical Classification System Refined (CCSR), and the Diagnostic Items (DXIs) introduced in Ellis et al, (2021). Due to their low prevalence rate, rare conditions are largely excluded from HHS-HCC and CCSR risk adjustment formulas, resulting in health insurance plans and providers having incentives to undertreat rare disease patients. The more informative and flexible DXIs model, however, is likely to give more attention to rare diseases. To evaluate their predictive power, the three risk-adjustment models are estimated on the same development sample (N=59.2 million) using both OLS and stepwise regressions, and then validated on a validation sample (N=6.6 million) to test for overfitting. The regression results show that, compared to other disease classification systems, the DXIs lower the average residual spending for people with rare diseases by at least 25% across all the regression models considered.en-USEconomicsThesis/Dissertation2022-03-04