modelinganddetectingstreaksinhealthcareriskadjus.docVIP

Modeling and Detecting Streaks in Health Care Risk Adjustment Joseph S. Koopmeiners1, Bryan E. Dowd2, and Bradley P. Carlin1 1Division of Biostatistics, School of Public Health, University of Minnesota 2Division of Health Services Research, Policy and Administration, School of Public Health, University of Minnesota Correspondence author: Bradley P. Carlin Telephone: (612) 624-6646 Fax: (612) 626-0660 Email: brad@ Address: Division of Biostatistics, MMC 303, School of Public Health, University of Minnesota Minneapolis, Minnesota 55455-0392, U.S.A. October 2, 2004 Abstract To account for differential expected health care costs of enrollees in different health plans, employers often risk adjust the out-of-pocket premiums paid by employees for observable cost factors (e.g., age and sex) that are exogenous to the health plan. While this risk adjustment can level the playing field among competing plans in the long run, random fluctuations in the risk can lead to “streaks” of consecutive financial losses that can threaten the solvency of a plan. If there is dependency across time (say, due to the plans’ tendencies to attract the same group of enrollees year after year), the likelihood of such streaks is even higher. In this paper we use probability modeling and simulation to study the likelihood of streaks in both the independent and autocorrelated cases for a particular two-plan model. We then use Bayesian statistical methods to address the problem of estimating when data from such a model exhibit evidence of departure from the assumptions of accurate risk adjustment and independence over time. Our Bayesian approach exhibits good bias, Type I error, and power properties under a variety of scenarios. Finally, we investigate a sequential version of our setting, where we check the stopping behavior of our procedure for various levels of underlying true bias and autocorrelation. Introduction Limited Cost-Plus Contracts We cons