中国麻风皮肤病杂志 ›› 2023, Vol. 39 ›› Issue (5): 338-343.doi: 10.12144/zgmfskin202305338

• 论著 • 上一篇    下一篇

基于随机森林模型的大疱性类天疱疮患者预后因素分析

赵丹丹1,甄莉2   

  1. 1山西医科大学，太原，030001；2山西医科大学第一医院皮肤科，太原，030001
  • 出版日期:2023-05-15 发布日期:2023-05-16

Analysis of the prognostic factors of bullous pemphigoid based on random forest model

ZHAO Dandan1, ZHEN Li2   

  1. 1 Shanxi Medical University，Taiyuan 030001,China；2 Department of Dermatology First Hospital of Shanxi Medical University，Taiyuan 030001, China
  • Online:2023-05-15 Published:2023-05-16

PDF (PC)

59

摘要： 目的：应用随机森林模型和Logistic回归模型，分析大疱性类天疱疮患者预后的影响因素，为治疗和预后判断提供依据。方法：以2015年1月至2021年4月山西医科大学第一医院皮肤科住院部诊断的大疱性类天疱疮患者为研究对象，构建随机森林和Logistic回归模型，分析大疱性类天疱疮患者预后的影响因素，并比较2种模型的预测效能。结果：随机森林模型结果显示，影响预后因素重要性的排序前五位分别是年龄、是否累及黏膜、是否合并神经系统疾病、血钙、是否伴有局部皮肤感染。Logistic回归模型显示高龄、血钙水平降低、合并神经系统疾病是BP患者预后不良的危险因素，病变累及黏膜的患者1年内死亡率更低。Logistic回归模型训练集和测试集的差异与随机森林模型相比较小，模型稳定性更好。两种模型并集后训练集准确率、灵敏度、特异度均为100%。测试集准确率均高于两模型单独预测。结论：年龄、是否累及黏膜、是否合并神经系统疾病、血钙、是否伴有局部皮肤感染是影响大疱性类天疱疮患者预后较为重要的因素。随机森林和Logistic回归两个模型取并集共同预测大疱性类天疱疮患者的预后更具实践意义。

关键词: 大疱性类天疱疮, 预后, Logistic回归, 随机森林

Abstract: Objective: To analyze the prognostic factors of patients with bullous pemphigoid using random forest model and Logistic regression model, so as to provide a basis for treatment and prognosis judgment. Methods: Inpatients who were diagnosed with bullous pemphigoid in our hospital from January 2015 to April 2021 were enrolled in this study. The factors on bullous pemphigoid were analyzed through  random forest model and Logistic regression model, and the predictive efficacy of the two models was compared. Results: The results of random forest model showed that the top five factors were age, mucous membrane involvement, neurological disease, blood calcium, and local skin infection. Logistic regression model showed that advanced age, decreased blood calcium level, and neurological diseases were risk factors for poor prognosis in BP patients and patients with mucosal involvement had lower mortality with 1 year. The difference between the training set and the test set of the Logistic regression model was smaller than that of the random forest model and the model is more stable. Conclusion: Age, mucous membrane involvement, neurological diseases, blood calcium and local skin infection are the most important factors affecting the prognosis of patients with bullous pemphigoid. The combination of random forest and logistic regression moolels to predict the prognosis of bullous pemphigoid patients is of practical significance.

Key words: bullous pemphigoid, prognosis, Logistic regression, Random forest