Análise discriminante aplicada para predição dos custos assistenciais no setor de saúde suplementar : um estudo de caso

Abstract

Supplementary health in Brazil plays an essential role in health care. In 2016, health plan operators were responsible for providing services to 47.8 million beneficiaries. This number denotes the degree of relevance of the supplementary health, consequently the costs presented by the operators are high, leading to several financial discussions, making management of expenses relevant. In this scenario, several authors presented papers regarding the importance of managing the risks from expenses, as well as the prediction of care costs. In this way, this work seeks, through the discriminant analysis, to generate a prediction model that contributes to the classification of the risk of the beneficiary, making possible to identify and analyze the degree of individual risk, allowing the company to direct strategic efforts to manage the costs of these people. As a result, the beneficiaries of the health plan operators were classified into six types of risk: without risk factor, low risk, medium risk, high risk, very high risk and catastrophic risk. From the risk classification, the multivariate analysis was applied to obtain a prediction model capable of indicating the classification of the individual. The discriminant analysis was applied using SPSS software version 19.0 in order to estimate the discriminant function. In the present study, the public of interest are the individuals with the highest expenses, that is, above the high cost, being these people the target of strategies to reduce future care costs. The realization of the discriminant analysis assumptions revealed that 55.04% of the beneficiaries were classified correctly. This way, as an additional analysis, the first three categories were grouped into the low-risk class and the last three were at high risk. In the first class it was observed that 99.08% of the individuals were classified correctly, however, in the second, only 19.71% of the observations were classified correctly. The results addressed in the model did not present expressive predictive validity for the proposed problem, but this fact does not imply that the discriminant analysis cannot be used in risk management operations in the health plan operators.