運用全民健康保險資料比較不同風險校正診斷因子對個人門診醫療費用之預測

Abstract

背景與目的：台灣全民健康保險自2002年起已全面實施前瞻性的總額預算支付制度，大部分引入前瞻式預算之先進國家，為避免前瞻式預算之預算額度決定，存在無效率及缺乏公平性問題，因此近年來導入風險校正(risk adjustment)機制，以更具精確性的設定預算，而目前國際研究最普遍採行的風險校正模式以診斷基礎模型最為成熟，而國內風險校正之研究尚處於起步階段，本研究將運用三年門診診斷資料，探討目前國際上最普遍採行的ACGs模式與DCGs模式，運用在台灣全民健康保險制度之可行性，且進一步與本土發展之診斷基礎模型進行比較，以提供未來發展本土或修正既有的系統之建議。 材料與方法：本研究資料為國家衛生研究院全民健康保險研究資料庫承保抽樣歸人20萬人資料檔，並選取承保資料中保險對象2000與2002年的全年納保之保險對象，研究樣本為164,248人，並擷取其門診及住院醫療費用申報等相關資料。診斷群組建構之風險計價模式，採用目前國際上廣泛被採用之ACGs 模式、DCGs 模式，以及由本研究所發展之門診診斷群組TASGs(Taiwan Ambulatory Spending Groups)，並預估各模式對2001年及2002年之醫療費用費用預測力；為避免以相同資料進行建構模式與評估模式預測力可能產生之過度估計問題，本研究將研究樣本隨機分割為二子樣本，以估計子樣本建立風險計價模式，而以預測子樣本驗證所建立之風險計價模式的準確性。 結果： ACGs模式對未來門診醫療費用預測力，PR2為9.44%~10.25%，DCGs模式對未來門診醫療費用預測力，PR2為37.36%~ 32.71%，而本土發展之TASGs模式對未來門診醫療費用預測力，PR2為36.78%~ 36.05%。 結論：本土發展的診斷群組分類系統，其預測力與穩定性均較國外模式為佳，而未來若考量將ACGs、DCGs等國外診斷群組分類系統導入台灣全民健康保險制度時，需要考量台灣本土執業及疾病型態加以修正，以提昇於台灣全民健保制度之可適用性。

Background and purposes: Taiwan’s National Health Insurance has completely adopted a global budget payment system since 2002. In order to avoid the persistence of inefficiency and inequality, most developed countries adopting a prospective payment system has introduced some forms of risk-adjusted mechanism to refine the budget setting. This trend can be followed by Taiwan, especially setting the budget of outpatient care, which accounts for two thirds of total NHI’s expenditure. This study intends to compare the predictability of different diagnosis-based risk adjusters classified by ACGs, DCGs, and TASGs to individual outpatient expenditure. The results should provide insight into future development of indigenous risk adjusters and necessary modifications when foreign systems are introduced. Material and methods: Using the panel data 200,000 individuals compiled by the National Health Research Institutes (NHRI), 164,248 individuals who had a full three years of eligibility in 2000 to 2002 were selected as the study sample. All outpatient claimed data were collected. Diagnosis-based risk adjusters were classified through ACGs, DCGs, and TASGs using 2000 data, and these adjusters were employed to establish risk adjustment models and to predict the outpatient expenditures of 2001 and 2002. Least squares regression models were built with an estimation sample (one half of the total), and were then cross-validated with the validation sample (the other half of the total). Results: While the predictability of the ACG model are 9.44% and 10.25% for 2001 and 2002, the predictability of the DCG model are 37.36% and 32.71% for 2001 and 2002. The predictability of the TASG model are 36.78% and 36.05% for 2001 and 2002. Conclusion: Between the two foreign systems, DCGs outperforms ACGs in terms of predictability. The indigenous risk adjusters have similar predictability to that of DCGs. Nevertheless, the predictability of the indigenous risk adjusters is more stable than that of DCGs.