Working Papers

What Saves More: Coinsurance or Copayment? (Job Market Paper with Xiaoxi Zhao)

We use large-scale health insurance claims data to examine whether consumers with cost sharing through coinsurance or copayment are more price sensitive in their health care choices. To estimate demand elasticities across a broad array of services under different cost sharing designs, we instrument individual monthly price levels by a full set of interactions of plan indicators and month indicators, capturing within year variation in prices at the plan level. We assume that consumers are myopic and backward-looking when deciding how much health care to consume, and interpret our elasticity estimates as the short-run spot price elasticities. We find that consumers are more price sensitive when charged a fixed percent coinsurance rate instead of a fixed dollar copayment, suggesting that coinsurance is more effective for cost containment purposes.

Diagnostic Category Prevalence in 3 Classification Systems Across the Transition to the International Classification of Diseases, Tenth Revision, Clinical Modification (with Randall P Ellis, Heather E Hsu, Tzu-Chun Kuo, Bruno Martins, Jeffrey J Siracuse, Ying Liu, Arlene S Ash), JAMA Open Network, April 2020

We use regression discontinuity analysis on monthly time series data to assess changes in diagnostic category prevalence associated with the International Statistical Classification of Diseases, Tenth Revision (ICD-10) transition. IBM MarketScan commercial insurance claims from 2010–2017 are mapped into three widely used diagnostic classification systems: the World Health Organization’s disease chapters (WHO); the Department of Health and Human Services Hierarchical Condition Categories (HHS-HCC); and the Agency for Healthcare Research and Quality (AHRQ) Clinical Classification System (CCS). This study covers over 20 million privately-insured enrollees under age 65. In all three classification systems, ICD-10 implementation was associated with statistically significant changes in monthly prevalence of any magnitude among 58–59% of diagnostic categories. Clinical review suggested that these patterns were largely due to the omission or addition of diagnoses from the affected diagnostic categories following the ICD-10 transition. Previously developed predictive models and diagnostic classification systems for payment and quality reporting should be used with caution until refined for ICD-10 data.

Diagnostic Items: A New Framework for Disease Surveillance, Prediction and Risk Adjustment (with Randall P Ellis, Cori Andriola, Arlene S Ash, Victoria Fan, Summer Hawkins, Heather E Hsu, Brian C Jacobson, Tzu-Chun Kuo, Karen E Lasser, Bindu Kalesan, Jeffrey J Siracuse, Ying Liu, Allan Walkey)

We create a new organizational framework of multiple dimensions based on Diagnostic Items (DXIs) that can be used for disease surveillance, prediction of spending, and estimation of risk-adjusted payments, taking full advantage of the fivefold increase in diagnostic details of ICD-10-CM. IBM MarketScan commercial insurance claims from 2016–2017 are first organized using the Agency for Healthcare Research and Quality (AHRQ) Clinical Classification System (CCS), and then clinically refined to create approximately 1,700 DXIs. We use linear regression with step-by-step iterative selection of variables to identify new predictors for risk adjustment and utilization prediction, and demonstrate that there is a meaningful improvement of 10% above models using only coarser diagnostic categories. These refinements meaningfully reduce potential profit to health plans from selecting enrollees whose expected payments exceed actual costs.

Work in Progress

High Speed Rail and Demographic Patterns - Evidence from China