探討初次全髖關節置換術術後5年死亡率之預測模式

研究目的　　全人工髖關節置換術(Total Hip Replacement)是用於治療嚴重髖關節病變的重要手術，也是骨科的常規手術，可以預期未來高齡化社會，人工髖關節置換術的利用將會持續成長。預測死亡率必須運用長期資料進行研究較為準確，本研究使用長達十年以上全國性資料庫進行探討，評估三種統計方式準確性：資料探勘(Artificial Neural Network, ANN)與傳統統計(邏輯斯迴歸、Cox存活分析)，並預測初次全人工髖關節術後5年死亡率與重要預測因子之權重差異，期望能發展出最佳之預測模型，以作為臨床醫師在醫療過程中決策之參考。研究方法　　本研究使用1996年至2010年之全國性資料庫進行探討，研究樣本共74,452位，針對病患人口學特質、疾病特質、醫院特質等三個構面進行死亡率預測，以類神經網路、邏輯斯迴歸分析和Cox存活分析所建立之預測模型進行準確度比較，進而運用全域敏感度分析(Global Sensitivity Analysis)找出影響術後五年內死亡率的重要預測因子。研究結果一、 THR手術病患術後五年內死亡率之影響因素。髖關節術後5年死亡率之影響因子經單變量分析，篩選出9個有意義變項計20個影響因子：「年齡」、「性別」、「疾病嚴重度-CCI指數」、「退化性關節炎」、「風濕性關節炎」、「股骨頭缺血性壞死」、「醫院層級」、「醫院服務量」及「醫師服務量」。二、利用效益指標（Performance Indices）比較THR術後五年存活三種預測模式（Forecasting Models）之準確度。類神經網路(ANN)與邏輯斯迴歸(LR)及存活分析(COX)之預測效能方面，THR術後5年存活預測模式，準確度為ANN=0.89、LR=0.10、COX=0.54；曲線下的面積(AUROC)方面，ANN=0.70、LR=0.65，COX=0.22。三、 THR術後五年內死亡率各個重要預測因子之權重。經由全域敏感度分析， ANN預測模式分析結果，術後5年死亡率重要影響因子依序為疾病嚴重程度、類風濕性關節炎及退化性關節炎。結論與建議　　整體而言，類神經網路(ANN)在預測THR術後5年存活模式之準確度較邏輯斯迴歸(LR)及存活分析(COX)為佳，本研究提出影響病患術後死亡率之顯著因素及比較驗證不同預測模式的優缺點，期使相關建議可作為各方臨床決策之參考，以提昇醫療品質。

關鍵字

全髖關節置換手術；預測模式；存活分析；全域敏感度分析

並列摘要

Purpose Total hip replacement (THR) is important for treatment of serious lesions in the hip joint surgery. This study used a nationwide population-based data for up to ten years to explore accuracy of the prediction models. The study purposes to validate the use of artificial neural network (ANN) model for predicting 5-year mortality after THR in Taiwan and to compare the predictive accuracy of ANN with that of logistic regression (LR) model and Cox proportional hazards model (COX). Research Methods A total of 74,452 THR patients were extracted from nationwide National Health Insurance Database from 1996 to 2010. For each pair of ANN, LR and COX models, the performance indices were calculated and compared using T-tests. Global sensitivity analysis and sensitivity score approach were also performed to assess the relative significance of input parameters in the system model and the relative importance of variables. Results 一、Iimpact within 5-year mortality after THR relative significance factor. 5 -years mortality after THR relative significance factor by univariate analysis, screening nine significant variables influencing factors into account 20: "age", "sex", "disease severity-CCI Index", "degenerative joint inflammation "," rheumatoid arthritis "," avascular necrosis "," hospital level "," hospital volume "and" physician volume. " 二、The use of efficiency indicators (Performance Indices) comparing five-year mortality after THR three prediction modes (Forecasting Models) of accuracy. The performance indices of ANN, LR and COX models of 5-years mortality rate are: accuracy of ANN = 0.89, LR = 0.11, and COX = 0.54; area under the receiver operating characteristic (AUROC) of ANN = 0.7, LR = 0.65, COX = 0.22. 三、THR right after an important predictor of mortality weight within five years. Which shows that ANN model is the best predictive model. Through ANN, global sensitivity analysis reveals that significant predictors of 5-years mortality are Charlson comorbidity index score, following by rheumatoid arthritis and osteoarthritis. Conclusions and Recommendations In comparison with the LR and COX models, the ANN model was more accurate in predicting 5-year mortality and had higher overall performance indices.