醫院間平均住院日之比較：Rasch分析法之應用

Assessing the efficiency of DRGs based length of stay in hospital among Taiwanese medical centers

行政院衛生署中央健康保險局以DRGs作為醫院住院醫療費用給付之依據，並發 展出台灣版的969項DRGs，住院天數是DRGs實施後各醫院的重要控管主題。本研究以 2004年住院申報資料，用Rasch（1960）分析當時全國17家醫學中心的DRGs，以其平 均住院天數除以DRGs的相對權重作為「平均住院天數」的評價依據。 將全部17家醫學中心住院案件編譯成663,017筆有效的DRGs碼，以1至12月為試 題（欄），各醫院（17家）的各個不同的DRGs件數當作受試者（列），建構成一具 13,0730×12矩陣的答題反應成績（即校正後住院日=該醫院在該DRGs平均住院日除以 DRGs的相對權重），作為分析各醫院各DRG「住院天數」優（天數短）劣（天數長） 的比較依據。利用WINSTEPS的Rasch分析軟體，（1）檢驗該測量符合單向度的測量 特性、（2）比較各類醫學中心DRG「住院天數」之差異、（3）建立醫院間與醫院內 的DRGs住院日數評估與比較。 研究結果發現：各月份的難度以2月及5月最易，3月最難，該二個月份間的困難 度呈顯著性差異（p <.05）。Infit及Outfit的MNSQ皆介於1.06及0.94間，資料符合Rasch 模式的預期。整體上醫學中心的DRG住院日管理效率尚佳（p <.05），內部一致性信 度為0.73。各類醫學中心的住院日管理效率排名依序為財團法人、榮總體系、私立醫 學院及公立醫學院，其間呈統計顯著性（p <.05）的差異。財團法人有二家落在離群值 （outlier），公立醫學院醫院在住院日管理效率未呈全部相等。 本研究以健保2004年全國醫學中心資料做探討，採用Rasch分析克服傳統較難解決 的（1）等距量尺；（2）遺漏值；（3）樣本獨立及試題獨立；（4）試題（指標）權 重比值不等的問題。現代測驗理論提供較多的訊息，對健保台灣版DRGs的住院日數習 性，提供更深層的認識，並作為醫院管理決策分析上的參考。

Taiwan’s government has implemented DRGs-based payment scheme applied to hospitals’ inpatient services for each discharge under the prospective payment system (PPS). Comparing length of stay (LOS) for each hospital to improve efficiency of hospital management is urgently required. Reimbursement data from medical centers in Taiwan 2004 were analyzed using Rasch analysis with Winsteps computer program to attain following three goals: 1) whether data construct a unidimensionality fitting to the Rasch model’s expectation; 2) whether LOS management efficiencies among types of medical centers in Taiwan are equivalent; 3) whether an evaluation mechanism can be formed in comparison of LOS management efficiencies within or between hospitals. The results showed that: 1) the rarest to reach for LOS in month was March, the easiest were February and May; 2) mean square errors of Infit and Outfit indices were within 1.06 and 0.94 fitting to the Rasch model fairly well; 3) each medical centers in LOS management efficiency were shown rather good comparing to the average item difficulty (odd ratio = 0); 4) performances in LOS management efficiency were statistically significantly different (p <.05) in types of medical centers, followed in rank by no-profit hospitals, veterans system, and both of private and public medical schools. Rasch analysis is recommended to use for comparing DRGs based LOS management efficiency within and between hospitals across months. Future studies related DRGs impacts and changes across different levels of hospitals are required to carry out.