家庭醫療費用的探討－以全民健保自負額為例

A Study of Family Medical Expenditures: An Example in National Health Insurance Self-Pay Medical Expenditures

林真真(Jen-Jen Lin) ； 張凱淵(Kai-Yuan Chang) ； 胡學穎(Shueh-Inn Hu)

《智慧科技與應用統計學報》 12卷2期 (2014/12) Pp. 61-74

摘要

本研究係探討家庭醫療費用的預測模型。採用的資料，來自於中央研究院調查研究專題中心的「學術調查研究資料庫」。主要針對民國86年到92年，共7年度的各項家庭收入支出調查資料，進行資料探勘研究。利用訂定的準則，篩選出具有代表性的有效樣本，並且採用分類迴歸樹（Classification And Regression Tree, CART）之分類方法，對全民健保自負額做預測；以Receiver Operating Characteristic（ROC）作為模型評估的準則。本研究之結果顯示：兩種家庭的全民健保高低自負額分類，各有準則：年輕族家庭的低自負額樣本比例為高自負額樣本比例的3倍，銀髮族家庭的高自負額樣本比例為低自負額樣本比例的3倍。影響兩種家庭健保自負額的6個收支變數：財產所得收入、健保保險受益、對私人之經常移轉支出、對政府之經常移轉支出、食品費合計、家事管理費。未來可考慮針對這些項目分別在兩種家庭進行CART模型分析，並採用相同節點數，以探討最主要的影響收支變數。

關鍵字

全民健保自負額 ； 資料探勘

並列摘要

The goal of this study is to build a prediction model for family medical expenditures. The database we used is the Survey Research Data Archive from the Center for Survey Research, Academia Sinica. The research data consists of family income and expenditures collected from 1997 to 2003 and are used for Data Mining. We used extracted criterion to find typical sufficient sample and the Classification and Regression Tree (CART) method to predict the National Health Insurance Self-Pay Medical Expenditures. The Receiver Operating Characteristic (ROC) is the evaluating criterion and Cross Validation was the method used in model selection for sample consistency. Our study results include two types of families where high and low self-pay expenditures are determined using different criteria. For young families, the proportion of low self-pay expenditures within the sample is three times that of the high self-pay expenditures. For senior families, the proportion of high self-pay expenditures within the sample is three times that of the low self-pay expenditures. There are 6 income and expenditure variables that affect the National Health Insurance Self-Pay Medical Expenditures for the two types of families, which are property income, health insurance benefits, private current transfer expenses, government current transfer expenses, food expenses, and housekeeping expenses. In order to investigate what the most influential income and expenditure variables are, CART model analyses with the same number of knots for the two types of families can be performed in the future.

並列關鍵字

National Health Insurance Expenditures ； Data Mining ； CART ； ROC