應用資料採礦技術分析臺灣健保資料庫－以攝護腺癌病人為例

根據行政院衛生署統計資料顯示，2011年惡性腫瘤（癌症）已連續29年為十大死因之首，其中攝護腺癌更高居男性死亡原因排名的第七名，該疾病之預防工作顯得更加重要。攝護腺為男性生殖系統的一個腺體，其中若有細胞的基因因為突變導致增殖失控，就會變成癌症。攝護腺癌也可能會導致疼痛、排尿困難、性功能障礙等症狀，其中最常發生於50歲以上的人。本研究納入全民健康保險資料庫2000年1月1日至2009年12月31日期間，住院診斷出現「185」診斷碼的攝護腺癌病人，利用羅吉斯迴歸、決策樹、類神經、支援向量機及隨機森林判斷病人是否會死亡，並找出其危險因子，提供醫學研究方面參考。經本研究實證結果顯示，決策樹之整體平均鑑別率最高，顯示決策樹建構之模型在判斷病人死亡與否有較佳鑑別力。此外，決策樹模型篩選出3個重要變數，顯示此3個變數應為影響分類模型之重要變數。

關鍵字

資料採礦；攝護腺癌；臺灣健康保險資料庫

並列摘要

According to Department of Health, Executive Yuan, R.O.C statistics, in 2011, the cancer has consecutive top ten causes of death in 29 year, of which Prostate cancer is the highest ranked cause of death in man in the seventh. Therefore the prevention of disease is getting more and more important. The prostate gland is a gland in the male reproductive system, if cell gene mutation will lead to uncontrolled proliferation, it will become cancer. Prostate cancer may cause pain, difficulty urinating, sexual dysfunction and so on, which occurs most often in people over the age of 50. This study included patients who were documented with ICD9-CM coded of ”185” in non-sampled NHI claim database from 1 January 2000 through 31 December 2009. This study use logistic regression, Decision Tree, Neural Net, Support Vector Machine and Random Forests to determine whether patients will die and finding risk-factor to provide medical research reference. Analytic results demonstrated that Decision Tree outperforms the other analysis approaches in terms of classification accuracy. In addition, the Decision Tree model find out the three important variables to display the three variables should be important variables for the classification model.