Health Insurance Claim Risk Stratification Using Zero-Inflated Poisson Regression Models

For the usual Poisson regression model, the annual number of health insurance claims is assumed to have the equidispersion and positive-support properties, but they do not hold for the data in this study: a large percentage of policyholders (48.4% in this study) experience zero claims in a year, resulting in excess zeros that are systematically underestimated by the standard Poisson regression, and that cause the variance of the count distribution to exceed its expected mean. The Zero-Inflated Poisson (ZIP) regression model accounts for this by taking the data generating process to consist of two parts: a logistic regression for the probability of being a ‘structural zero’—that is, a non-claimant, who is not related to their risk exposure-and a standard Poisson regression for the expected number of claims for the rest of the population. In this paper, we analyze how the ZIP model can be used to identify the risks of health insurance claims, using 284,600 policyholder records from three health insurance companies in two

Hali tarjima qilinmagan