Prediction of Medical Costs in a Health Insurance Carrier according to Risk Profiles and Uses by its Affiliates

Abstract

Objective: To find a model of prediction of the medical cost of a Health Benefits Management Company (EAPB) with adequate statistical criteria. Methods: A Cross-sectional study with retrospective follow-up of the use of health services in an EAPB during a one-year period. The sampling frame consisted of a population of 1,529,188 affiliates who were assigned to a primary care IPS group. By simple random sampling size was estimated at 190,917 users. The dependent variable was the cost of the services used deflated to the year 2013. As independent variables besides the traditional sociodemographic variables chosen in this type of prediction models, variables of the insurance were added; Variables of risk management (inclusion or not in promotion and prevention program) and of comorbidities. Results: Simple Linear Regression modeling showed errors of inappropriate statistical criteria such as violating the principle of normality in cost errors. The Generalized Linear Models, proposed to estimate POS average costs, have an appropriate goodness of fit and evaluated with small Deviations and minimum Akaike criterion (AIC) compared to other models of the exponential family Conclusions: The appropriate statistical model to predict medical costs was the Generalized Linear Model with two parts segmented by age groups and gender. This research suggests that to estimate the benefit premium of any EAPB, besides socio-demographic variables, insurance variables, membership or not in promotion programs and risk prevention and/or management and the burden of disease of that population should be used.

Keywords

• health care costs
• health programs and plans
• two-part model
• generalized linear model