建構所有診斷資訊群組及其風險預測模式

Abstract

為改善醫療費用高漲問題，許多國家的健康照護體系多採用前瞻性預算的方式支付醫療費用。在前瞻性支付體系下，為顧及健康保險的效率與公平性，最精緻的保費計算方法便是導入風險校正機制。風險校正成功與否的關鍵在於風險計價模模式的預測力，依據國際研究顯示，於計價模式中加入以診斷為基礎的風險校正因子為較能避免風險選擇，且能保有不錯模式預測力的做法之一。目前國內學者引用或自行發展的診斷群組，皆僅單獨使用門診或住院診斷建構，並無一整合所有診斷資訊的診斷群組，因此本研究欲發展結合所有診資訊斷的診斷群組，以及利用此診斷群組建構風險預測模式，並比較不同預測模式對未來醫療費用的預測能力，以提供國內健保前瞻式預算分配之參考。 本研究利用2002年和2003年的健保承保抽樣歸人檔進行分析，使用的診斷因子有國外發展的HCCs群組以及本研究利用臨床分類軟體（CCS）自行發展的CCS層級群組與CCS疾病組合群組。研究結果顯示，以HCCs或CCS層級群組加上人口統計因子的風險計價模式預測力較為良好，PR2在20%左右，預測比（PR）除少部分疾病外，值都接近於1，表示能準確估計各疾病分群的醫療費用，且本研究自行發展的CCS層級群組在疾病分群的整體預測比表現上又比HCCs更加準確，因此適合以其估算各疾病分群的預期醫療費用。 然本研究自行發展的CCS層級群組之PR2值相較於國外的HCCs層級群組稍低，原因可能為自行發展的群組其層級臨床性不如HCCs強烈，因此在未來的研究當中可再進一步加強診斷群組的臨床性，如加入醫生團隊之意見，修正原有的層級成更具有我國人口疾病嚴重度代表性的層級，並且加入用藥處方資訊，加強精神疾病和慢性病於診斷資訊使用上的不足，成為一更具有臨床性的所有診斷群組，以更精煉風險計價模式的預測力。

In order to contain the escalation of health care expenditure, many countries have adopted prospective budgets paying health care expenditure. Under prospective payment system, introducing risk adjustment mechanism in health insurance is considered to be able to maintain equity and efficiency. According to previous studies, the development of risk adjustment models with diagnostic information is one of most prominent approach to prevent risk selection and possess good predictability. However, only diagnostic groups based on outpatient and principal inpatient diagnoses respectively have been developed in Taiwan, and no diagnostic groups combining outpatient and inpatient diagnostic information is under development in this country. This study intends to develop risk adjusters based on all diagnostic information and compare the predictabilities of various risk adjustment models. The result can provide a reference of assigning prospective budget in Taiwan’s National Health Insurance. 2002 and 2003 NHI panel data was used to analyze in this study. The diagnostic-based risk adjusters are HCCs, CCS hierarchy groups, and CCS case-mix groups, and the later two systems were developed based on Clinical Classification Software. The results show that the PR2 and PR values of risk adjustment models with HCCs and CCS hierarchy groups perform well. The PR2s are about 20% and most PR values are close to 1, except for few specific disease subgroups. It depicts that the models can predict the expenditure of different disease subgroups accurately. And the performance of subgroup predictability in CCS hierarchy groups developing by this study is better than others, so it is suitable for calculating the expected medical expenditure of disease subgroups. However, the PR2 of CCS hierarchy groups is less then HCCs slightly. The reason may be that the clinical characteristic of CCS hierarchy was not as strong as that of HCCs. It is encouraged that the clinical characteristic should be strengthened in future study, such as consulting physicians’ opinion. Besides, we can add prescription information in risk assessments to improve the defect of only using diagnostic information in metal and chronic disease.