针灸治疗中风恢复期临床预测模型的建立和分析

Development and analysis of clinical prediction model of acupuncture and moxibustion treatment for stroke at recovery stage

目的:建立针灸治疗不同状态中风恢复期患者疗效的临床预测模型,为预测针灸疗效提供工具。方法:纳入2012年至2019年于浙江中医药大学附属第三医院住院治疗的中风恢复期患者1410例,提取性别、年龄、发病时间、神经功能缺损评分和针灸疗法等相关信息,通过治疗前后神经功能缺损评分的差值评定患者疗效,采用SPSS26.0软件及CART决策树分析建立临床预测模型。结果:针灸治疗不同状态中风恢复期患者疗效预测模型的关键变量为年龄、发病时间、高血压、心脏病、糖尿病、中医诊断、血红蛋白(HB)、血清同型半胱氨酸(HCY)和针灸疗法。决策树模型生成的主要规则共有12条,预测疗效好转的有8条,预测疗效未好转(包括无变化及恶化)的有4条。模型训练集和检验集的准确率分别为80.0%和72.8%,ROC曲线下方面积(AUC)为0.797,模型判别及分类效果较好。结论:采用CART决策树分析建立的临床预测模型对不同状态中风恢复期患者疗效预测准确性较高,医生可在患者就诊时根据预测疗效而采取相应的针灸治疗方案。

Objective To develop the clinical prediction model of therapeutic effect in treatment with acupuncture and moxibustion for the patients with stroke at recovery stage under different conditions so as to provide a tool for predicting the therapeutic effect of acupuncture and moxibustion. Methods A total of 1410 patients with stroke at recovery stage were collected from the Third Affiliated Hospital of Zhejiang Chinese Medical University from 2012 to 2019. The relevant data were extracted, i.e. sex, age, time of onset, neurological functional deficit score(NFDS) and acupuncture and moxibustion therapy. The difference of NFDS before and after treatment was adopted to evaluate the therapeutic effect in the patients. Using SPSS26.0 software and CART decision tree analysis, the clinical prediction model was developed. Results The key variables in the prediction model of therapeutic effect in the patients with stroke at recovery stage under different conditions included age, time of onset, hypertension, cardiac disease, diabetes, TCM diagnosis, hemoglobin(HB), serum homocysteine(HCY) and acupuncture and moxibustion therapy. There were 12 main rules generated by the decision tree model, including 8 rules for predicting the improvements of therapeutic effect and 4 rules for predicting the absence of improvements(i.e. no change and deterioration). The accuracy rates of the model training set and test set were 80.0% and 72.8% respectively, the area under curve(AUC) of ROC was 0.797 and the model identification and classification results were satisfactory. Conclusion The clinical prediction model developed by CART decision tree analysis is high in accuracy for the prediction of the therapeutic effect in the patients with stroke at recovery stage under different conditions. Based on the therapeutic effect predicted in the hospital visit, the physicians may adopt the corresponding regimens of acupuncture and moxibustion therapy in patients.