分析醫院品質指標：使用健康保險資料和先進統計方法

Analyzing Hospital Quality Indicators: A Comparative Study Using Health Insurance Data and Advanced Statistical Methods

健康保險資料開放服務平台，提供公務機關與學術研究機構使用。然而，該平台只有數據，卻缺少分析與視覺圖的資訊。本研究從公開的醫院156項醫療品質指標中，選取三指標（A、B及C：出院後三天內同一醫院再急診率、處方箋藥品超過十項的案件比率、急診後住院超過48小時的案件比率），分析其自2023年第二季起至前54季的資料。利用R語言製作動態直方圖，比較醫學中心間的差異，指標比率越高表示其品質越低。再由集群分析、Kendall's W（資料一致性係數，以卡方檢定ｐ<0.05為判準分數的可加性）、和羅序模式的殘差分析，揭示動態直方圖的限制。研究發現：（1）使用R語言能輕易製作動態直方圖，比較各醫學中心間的品質指標；（2）集群分析顯示醫學中心間指標結果的相似性；（3）三項指標的Kendall's W分別為0.64、0.001和0.0，表示不適於將指標B與C的54季資料加總，以比較醫學中心間的品質差異；（4）利用羅序模式分析近10季資料，顯示殘差分析的必要性。本研究透過視覺圖，比較醫學中心間的3項品質指標。未來可針對健康保險資料開放服務平台提供的指標，進行其它不同層級醫院間的比較。

The Health Insurance Data Open Service Platform (HIDOSP) enhances the interaction between the public sector and stakeholders. Although HIDOSP makes data available, it lacks visual information for the public. This study analyzed three selected medical quality indicators from hospitals: (A) the rate of emergency department visits within three days of discharge at the same hospital, (B) the proportion of cases with more than ten prescription drugs, and (C) the proportion of cases staying in the hospital for more than 48 hours after emergency treatment. Using data spanning from the second quarter of 2023 back to the previous 54 quarters, dynamic histograms were generated in R to visualize comparisons among medical centers, where a higher indicator ratio suggests lower quality. The analysis included cluster analysis, Kendall's W, and residual analysis of the Rasch model. Findings indicate that while R can effectively create dynamic histograms for comparison, Kendall's W values (0.64, 0.001, and 0.0) suggest that aggregating data across 54 quarters may not always be appropriate for quality performance comparisons in indicators B and C. Recent data from the last ten quarters analyzed through the Rasch model highlighted the importance of residual analysis. The study demonstrates the utility of dynamic histograms in comparing quality indicators over time and highlights areas for future research, including comparisons between different levels of hospitals based on HIDOSP indicators.