基于机器学习的非药物因素预测精神分裂症疗效的研究

Study on predicting the therapeutic effect of schizophrenia by non drug factors based on machine learning

目的探讨基于机器学习的非药物因素在预测精神分裂症疗效的研究。方法选择2021年6月—2022年12月本院门诊收集符合ICD-10精神分裂症诊断标准的200例患者作为研究对象,随机分配到独立训练集和独立测试集,经过2周的治疗,根据PANSS减分率的高低,将其分为实验组(160例)和对照组(40例)两组,以评估两组患者的疗效。同时在治疗过程中单变量分析与多变量Logistic回归分析影响两组患者康复的因素,并采用随机森林算法建立机器学习模型和引入独立测试集,采用受试者操作特征曲线、校准曲线分析交叉验证机器学习模型的准确度与稳健性。结果机器学习模型具有较好的区分度,其中独立训练集AUC为0.877,而独立测试集AUC为0.853。经过对该模型的深入分析,我们发现,患者的非药物因素,如性别、民族、家庭收入、家庭人口数、婚姻状况、教育水平、职业、吸烟习惯、发病年龄、首次诊断时间、首次治疗时间、总发病次数、停药次数以及发病形式,都会显著影响精神分裂患者的治疗效果(P<0.05)。结论利用轻量梯度提升机机器学习方法开发并建立了一个非药物因素在预测精神分裂症疗效的预测模型,具有较好的预测效能,能够辅助临床医护人员对精神分裂症的非药物因素开展针对性的干预,从而降低精神分裂症的发生率并改善预后。

Objective To explore the study on predicting the efficacy of schizophrenia by non drug factors based on machine learning.Methods 200 patients who met the ICD-10 diagnostic criteria for schizophrenia those who visited outpatient department of our hospital from June 2021 to December 2022 were selected as study objects.All patient cohorts were randomly assigned to an independent training set and an independent testing set.After 2 weeks of treatment,the efficacy was evaluated using the PANSS reduction rate.The patients were divided into an experimental group(n=160)and a control group(n=40)based on the PANSS scores.At the same time,during the treatment,univariate analysis and multivariate logistic regression analysis were used to analyze the factors affecting the rehabilitation of the two groups of patients,and random forest algorithm was used to establish the machine learning model and introduce independent test sets.The accuracy and robustness of the machine learning model were cross verified by the analysis of the receiver operating characteristic curve and calibration curve.Results The machine learning model has good discrimination,AUC of independent training set was 0.877 and an AUC of independent testing set was 0.853.The interpretation and analysis results of this model indicate that the non-drug factors of patients,such as gender,ethnicity,family income,family population,marriage,education level,occupation,smoking,age of onset,age of first diagnosis,age of first treatment,total number of onset,number of drug discontinuations,and form of onset,significantly affect the treatment effectiveness of schizophrenia patients(P<0.05).Conclusions A predictive model for predicting the efficacy of non-drug factors in schizophrenia has been developed and established using a lightweight gradient lifting machine learning method.It has good predictive performance and can assist clinical healthcare professionals in targeted intervention of non-drug factors in schizophrenia,thereby reducing the incidence of schizophrenia and improving prognosis.