列线图与CART决策树模型对老年脑卒中患者病耻感预测效能的比较

Efficacy comparison of nomogram and classification and regression tree in predicting stigma in elderly stroke patients

目的构建列线图与CART决策树模型,探讨2种模型预测老年脑卒中患者发生病耻感风险的应用价值,为早期识别和干预高危患者提供参考。方法基于课题组于2020年5~10月对保定市某三级甲等医院住院的252例老年脑卒中患者病耻感研究资料。将资料按照是否发生病耻感分为无病耻感组(n=125)和有病耻感组(n=127),建立列线图模型和CART决策树模型。内部验证采用Bootstrap重采样1000次的方法,通过AUC、灵敏度、特异度、阳性预测值、阴性预测值、预测准确率对2种模型的预测性能进行比较。结果CART决策树和列线图模型的AUC分别为0.903和0.880,灵敏度分别为90.6%和86.8%,特异度分别为86.0%和80.0%,阳性预测值分别为85.1%和77.8%,阴性预测值分别为90.7%和87.4%,预测准确率分别为88.0%和82.6%。2个模型AUC值相比,差异有统计学意义(t=19.814,P<0.001)。结论2种模型均具有较高的准确性和较好的指导价值,CART决策树预测性能略优于列线图,可为有效评估患者病耻感和采取针对性的干预措施提供指导。

Objective To construct a nomogram and classification and regression tree(CART),to explore the application value of the two models in predicting the risk of stigma in elderly stroke patients,and to provide reference for early identification and intervention in high-risk patients.Methods The data of 252 elderly stroke patients hospitalized in a tertiary grade-A hospital in Baoding from May to October 2020 were retrospectively analyzed.The patients were divided into non-stigma group(n=125)and stigma group(n=127)according to whether stigma occurred or not,and a nomogram and CART were established.Internal validation was performed by Bootstrap resampling 1,000 times,and the predictive performance of the 2 models was compared in terms of AUC,sensitivity,specificity,positive predictive value,negative predictive value,and predictive accuracy.Results The AUC,sensitivity,specificity,positive predictive value,negative predictive value and predictive accuracy of CART and nomogram was 0.903 and 0.880,90.6%and 86.8%,86.0%and 80.0%,85.1%and 77.8%,90.7%and 87.4%,and 88.0%and 82.6%respectively.The differences were statistically significant when comparing the AUC value of the two models(t=19.814,P<0.001).Conclusion Both models have high accuracy and better guidance value,and the prediction performance of the CART is slightly better than that of the nomogram,which could guide the assessment of patient stigma and targeted interventions.