Division of General Internal Medicine, Department of Medicine, St. Luke's International Hospital, Tokyo, Japan.
Am J Geriatr Psychiatry. 2013 Oct;21(10):957-62. doi: 10.1016/j.jagp.2012.08.009. Epub 2013 Feb 6.
To predict development of delirium among patients in medical wards by a Chi-Square Automatic Interaction Detector (CHAID) decision tree model.
This was a retrospective cohort study of all adult patients admitted to medical wards at a large community hospital. The subject patients were randomly assigned to either a derivation or validation group (2:1) by computed random number generation. Baseline data and clinically relevant factors were collected from the electronic chart. Primary outcome was the development of delirium during hospitalization. All potential predictors were included in a forward stepwise logistic regression model. CHAID decision tree analysis was also performed to make another prediction model with the same group of patients. Receiver operating characteristic curves were drawn, and the area under the curves (AUCs) were calculated for both models. In the validation group, these receiver operating characteristic curves and AUCs were calculated based on the rules from derivation.
A total of 3,570 patients were admitted: 2,400 patients assigned to the derivation group and 1,170 to the validation group. A total of 91 and 51 patients, respectively, developed delirium. Statistically significant predictors were delirium history, age, underlying malignancy, and activities of daily living impairment in CHAID decision tree model, resulting in six distinctive groups by the level of risk. AUC was 0.82 in derivation and 0.82 in validation with CHAID model and 0.78 in derivation and 0.79 in validation with logistic model.
We propose a validated CHAID decision tree prediction model to predict the development of delirium among medical patients.
通过卡方自动交互检测（CHAID）决策树模型预测内科病房患者发生谵妄的情况。
这是一项回顾性队列研究，纳入了一家大型社区医院内科病房的所有成年患者。通过计算机随机数生成，将研究对象患者随机分配到推导组或验证组（2:1）。从电子病历中收集基线数据和临床相关因素。主要结局为住院期间发生谵妄。将所有潜在预测因素纳入向前逐步逻辑回归模型。还对同一组患者进行了 CHAID 决策树分析，以建立另一个预测模型。绘制了受试者工作特征曲线，并计算了两个模型的曲线下面积（AUC）。在验证组中，基于推导组的规则计算这些受试者工作特征曲线和 AUC。
共纳入 3570 例患者：2400 例患者被分配到推导组，1170 例患者被分配到验证组。分别有 91 例和 51 例患者发生谵妄。CHAID 决策树模型中具有统计学意义的预测因素为谵妄史、年龄、基础恶性肿瘤和日常生活活动受损，导致根据风险水平分为六个不同组。CHAID 模型在推导组和验证组中的 AUC 分别为 0.82，在推导组和验证组中的逻辑模型分别为 0.78。
我们提出了一个经过验证的 CHAID 决策树预测模型，以预测内科患者发生谵妄的情况。
