有限混合模型在肝硬化住院患者医疗费用研究中的应用

The Application of Finite Mixture Model in the Study of Medical Expenditures for Liver Cirrhosis Inpatients

目的针对医疗费用的偏峰、厚尾分布特征,探讨有限混合模型(finite mixture model,FMM)在识别肝硬化患者住院费用异质性、提高医疗费用预测精度等方面的可行性,为准确估计和预测医疗费用提供统计方法学支持。方法介绍FMM原理,并将其应用于广州市第八人民医院肝硬化住院患者医疗费用研究,识别群体异质性,分析异质性来源,并与单成分广义线性模型的预测效果进行比较。结果对2 760名肝硬化患者住院费用进行FMM分析,最优模型为包括低、中等和高费用3个成分,拟合优度与预测效果均高于广义线性模型,异质性来源分析进一步验证了FMM识别各类患者的能力。结论 FMM能够有效地识别医疗费用的异质性,解决医疗费用偏峰和厚尾分布问题,提高医疗费用预测精度。

Objective To explore the finite mixture model（FMM） in identification of heterogeneity of medical cost of liver cirrhosis inpatients and feasibility of improving the prediction precision, we provide statistical methodology support for accurate estimation and forecast in terms of the skewed and heavy tail distribution characteristics of medical expenditures. Methods The principle of FMM is introduced and applied to medical expenditures of liver cirrhosis inpatients from the eighth peopleg hospital of Guangzhou to identify population heterogeneity, and then we analyze sources of heterogeneity, and compare the prediction results with single component of generalized linear model. Results After modeling the medical expenditures of 2 760 liver cirrhosis inpatients by FMM, the three gamma distribution components of FMM is fitted, including the low expenditures, median expenditures, high expenditures. The goodness-of-fit and predictive effect of FMM are better than the generalized linear model, and the source of heterogeneity analysis further verifies identification ability of FMM. Conclusion The finite mixture model has a good effect on identifying heterogeneity of hospitalization medical expenditures, solving the problem of the skewed and heavy tail distribution characteristics of medical expenditures, finally improve the predictive accuracy.