目的:本研究透過妥善的健康檢查,以達早期發現早期治療。本研究建立大腸息肉風險因子與大腸鏡異常之發現預測模式,以提供醫師作為臨床輔助診斷,減少侵入性檢測,降低檢查成本。方法:本研究收集西元2009年1月至西元2009年12月間,台灣中部個案醫院健康檢查中心做過大腸鏡篩檢之民眾資料,共計分析809筆。本研究以一般檢查、血液檢查、生化檢查—血脂肪、尿液檢查、癌胎抗原檢查、免疫法糞便檢查共28項變數,以大腸息肉與否為依變數,使用決策樹分類方法進行分析。結果:本研究結果發現,以免疫法糞便檢查的預測績效最好,其訓練資料的Az值為0.902,測試資料的Az值為0.879。癌胎抗原檢查次之,其訓練資料的Az值為0.897,測試資料的Az值為0.843。結論:本研究結果顯示,決策樹分類方法適用於醫學大腸息肉之健檢資料,可有效探勘其重要變數。本研究結果可提供醫院健康管理中心作為輔助決策。
Objective: The aim of this study was to establish a predictive model for risk factors for colon polyps to help physicians reduce invasive testing and the costs of examinations.Methods: Data were collected from a community hospital physical examination center located in Central Taiwan during the period of January 2009 to December 2009. We analyzed data from 809 patients who received colonoscopies. Risk factors associated with colon polyps were determined by using decision tree algorithms.Results: The results showed that the best predictor was the presence of fecal occult blood. The receiver operative characteristic curve (Az value) of training data was 0.902, and the Az value of the test data was 0.879. The second best predictor was the Carcinoembryonic Antigen the Az value of training data was 0.897, and the Az value of the test data was 0.843.Conclusions: Decision tree classification technology was an effective way to use physical examination data to make a decision index with regard to colon polyps. It was easy to determine and provided a highly accurate predictive model for the need for colonoscopy.