利用判定係數及Rasch模式探討醫院層級間的護病比

Using determination coefficient and Rasch model to investigate nurse-patient ratios among Taiwan hospital levels

目的：利用資料的判定係數及Rasch模式，探討醫院層級間的護病比。方法：下載中央健康保險署2015年各層級479家醫院各月全日平均護病比資訊公開的資料，利用作者於微軟Excel下開發出來的試題反應理論Rasch模式的資訊模組，(1)先挑出護病比資料的判定係數最小者；(2)再決定符合與不符合護病比的醫院分類；(3)最後排序護病比的穩定性指標，示範提供比傳統的統計學提供更多訊息，挑出資料的異常，做為進一步分析及檢討改善的依據。結果：顯示研究結果：(1)挑出護病比資料判定係數最小的層級醫院為醫學中心(=0.71)；(2)決定符合與不符合護病比的醫院，各有21家及1家；(3)2家醫學中心呈現偏高(>2.0)的護病比不穩定性。結論：由作者設計的資訊模組可以用來比較各層級醫院全日平均護病比的回饋資料，也驗證醫學中心護病比資料的判定係數並非最高，資料分析方法，提供醫院管理者的參考。

Objectives: To investigate nurse-patient ratio among Taiwan hospital levels using coefficient of determination and Rasch model. Methods: We downloaded data of 2015 nurse-patient ratio of 479 hospitals from the website of Taiwan National Health Insurance Administration and used the author-made computer module on Miscro-soft Excel based on Rasch model of item response theory to (i) screen out the minimum of determination coefficient on data of nurse-patient ratio among hospital levels for improvement consideration, again to (ii) separate two groups from the criterion of hospital accreditation on nurse-patient ratio, and to (iii) rank the stability indicators for hospitals as well as demonstrating a method compared to the traditional statistical approach to provide more meaningful information to readers for further improvement made in future. Results: We found that (i) the minimal determination coefficient (=0.71) among hospital levels is attributable to medical center, (ii) one hospital in medical center is under the criterion (i.e., 9 patients per nurse) set by hospital accreditation regulation and other 21 hospital beyond the criterion, and (iii) two hospitals in medical center are shown most significantly unstable (i.e., Outfit mean square error greater than 2.0). Conclusions: A novel module designed by authors was demonstrated and used for getting data from nurse-patient ratio among Taiwan hospital levels. We confirmed that medical centers suffer the minimal determination coefficient beyond our expectation in 2016 and the data in nurse-patient ratios of medical center examined by Rasch model can be addible and construct a unidimensional scale. The author-made Rasch continuous model is recommended to readers analyzing their data in future.