- 學位論文
膀胱癌手術之住院日與醫療費用評估研究
A Study on Length of Stay and Expenditure on Surgery of Bladder Cancer
摘要
膀胱癌近年來在台灣是第二大泌尿道癌症，死亡率達2.4%。尤其在烏腳病流行地區，居民因長期飲用井水造成烏腳病盛行及肺癌與膀胱癌死亡率偏高。膀胱癌依病理組織共分為零到四期，在第零期與第一期可以由簡單的經尿道膀胱腫瘤切除術進行治療，而第二期與第三期則是多以膀胱根除術進行治療。本研究為探討此兩種手術，申請健保局2006~2009年的資料作為研究對象，以人工智慧方法建立住院日以及醫療費用的預測模型進行資料探勘之評估研究，評估的項目有：膀胱根除術之超長期住院日、經尿道內視鏡腫瘤切除術之平均住院日與兩種手術的醫療費用。 經由實證分析後，在住院日上由人工智慧方法共建立了17個預測模型，9個為膀胱根除術之超長期住院日的預測模型；8個為TUR-BT之平均住院日的預測模型。在膀胱根除術之超長期住院日預測模型方面，以單一SVM預測模型優於其它模型，該模型驗證後平均準確率為80.7%，平均ROC曲線面積評估為0.71；經尿道膀胱腫瘤切除術之平均住院日研究上，則經由BPN驗證後測試準確率最佳之結果導入CBR做經尿道膀胱腫瘤切除術之平均住院日的分析較佳，其準確率為85.23%，ROC曲線下面積為0.78。醫療費用上，本研究分別對兩個手術各建立一個CBR的醫療費用預測系統，相似度分別為95.405、98.85，在誤差費用分析下，膀胱根除術的費用較不易預測；TUR-BT則是在一萬元以下即有八成以上的準確率。
並列摘要
The bladder cancer is the secondly common prevalent urinary tract cancer, leading to a death rate of 2.4% in average. Especially in the area of high Blackfoot prevalence, there also demonstrates a tendency of higher lung cancer and bladder cancer occurrence rates arising from residents’ drinking well water. When in classified by histopathology, the bladder cancer can be separate into from Phase 0 to Phase 4. In the preliminary Phase 0 and Phase 1, the effective medical treatment can be achieved by uncomplicated Transurethral Resection of Bladder Tumor (TUR-BT). For the Phase 2 and Phase 3, the mostly adopted medical treatment method is the Radical Cystectomy. Both the Radical Cystectomy and the TUR-BT are already adopted into the insurance coverage of fulfillment items of DRG (Diagnosis Related Groups) of the national healthcare insurance. For intending to explore those aforesaid two surgeries under such prerequisites, it has been submitted an application by the research to the Bureau of National Healthcare Insurance for getting access to database of between years 2006 and 2009 to be adopted as the research objects in conjunction the evaluation items including: conducting comparisons on between the average medical treatment costs for excessive length of stay for the Radical Cystectomy, and that for the average length of stay for TUR-BT. After the empirical study analysis is conducted, it has been established 17 prediction models belonging to Artificial Intelligence Model for estimating the length of stay, and 9 models for predicting the excessive length of stay for Radical Cystectomy; meanwhile, another 8 models for predicting the average length of stay for the TUR-BT. In regard of the prediction models for excessive length of stay for the Radical Cystectomy, the One-side SVM prediction model demonstrates better analysis results than other models with an average precision rate of up to 80.7% and a value of 0.71 in the average distribution area under the ROC Curve graph. Meanwhile, after the BPN is conducted, the searched best precision is applied in the CBT for prediction the length of stay of the TUR-BT can deliver a better prediction and precision function. Its precision rate is 85.23% and the value of the distribution area under the ROC curve graph is up to 0.78. In case of conducing comparisons on medical costs, this research has created a CBR-orientation medical cost prediction system for each of the two surgeries respectively. Each of their similarity levels is 95.405、98.85. In term of conducting standard deviation analysis for medical costs, both two can be judged by the professional physicians by perpetually increment or revision on different cases for enhancement of the system database for enabling such the system’s prediction concerning medical costs more compliable with the actual situations.