應用決策樹預測壞死性筋膜炎病人的死亡率
中文摘要
目的 : 試圖從入院臨床特徵或檢查找出簡單的規則，以求能預測死亡 率及辨識出高死亡率的壞死性筋膜炎病人。
方法 : 此回溯性的病例回顧研究收錄了兩個醫院，出院診斷為壞死性 筋膜炎的急診成人病患。運用 chi-square 檢驗 ( 或 Mann-Whitney-U 檢 驗 ) 和 C4.5 決策樹來分析 23 項不同的入院臨床特徵或檢查變項。預 測的結果變項為院內死亡。
結果 : 共收錄了 272 名病患。死亡率為 17% 。在單變量分析方面，肝硬化、癌症、慢性腎病、腎上腺不足、低血壓、白血球數、白血球 band form 與血紅素，都和死亡有顯著相關。而 C4.5 決策樹則找出低 血壓、白血球數、與白血球 band form 3 個獨立變項，並衍生出六條 規則將壞死性筋膜炎病人分為高死亡率與低死亡率兩群。經交叉驗證 後， C4.5 決策樹的準確度為 84.2% (95% 信賴區間， 80.3%-88.1%)
。
結論 : 藉著常規的量血壓與檢驗白血球，急診醫師可快速地找出高死 亡率的壞死性筋膜炎病人。Predicting mortality in patients with necrotizing fasciitis by using decision tree algorithm
英文摘要
Objective: To identify simple admission clinical characteristics or laboratory tests to predict m ortality but also differentiate between high and low mortality risk groups in patients with necr otizing fasciitis.
Methods: This retrospective chart-review study included adult patients who were admitted to t wo hospitals through the emergency department with the discharge diagnosis of necrotizing fa sciitis. Both the chi-square test (or the Mann-Whitney-U test) and C4.5 decision tree were utili zed to analyze 23 variables among clinical characteristics and laboratory tests. The main outco me measure was in-hospital mortality.
Results: 272 patients were included and the overall mortality rate was 17%. On univariate anal ysis, significant variables associated with mortality included liver cirrhosis, cancer, chronic ki dney disease, adrenal insufficiency, hypotension, white blood cell (WBC) count, WBC band f orm, and hemoglobin. Three independent predictors of mortality - WBC count, WBC band for m, and hypotension- were determined by means of the C4.5 decision tree. From these predicto rs, six decision rules were produced to classify patients with necrotizing fasciitis into high and low mortality risk groups. The accuracy of C4.5 decision tree with cross-validation was 84.2%
(95% confidence interval, 80.3%-88.1%).
Conclusions: By using routine blood pressure measurement and simple laboratory test, WBC c ount and differential, emergency physicians may rapidly identify patients with high mortality.
