基于机器学习构建社区老年人焦虑抑郁评估预测模型

Constructing a Prediction Model for Anxiety and Depression among Elderly People in the Community Based on Machine Learning

目的基于机器学习算法开发社区老年人焦虑抑郁预测模型。方法根据2019年山西省调查收集的15079例老年人社会人口统计学因素和疾病状况,利用广泛性焦虑障碍量表(GAD-7)和患者健康问卷(PHQ-9)进行焦虑抑郁评估,学习患有心理疾病的老年人群表征,使用随机森林、XGBoost和LightGBM算法分别建立预测模型,以受试者工作曲线(ROC)曲线下面积(AUC)、准确率、精度、召回率和F分数量化了预测性能并进行排序。基于Shapley值的特征归因框架分析老年人焦虑抑郁的高风险因素。结果我们构建的基于LightGBM的全特征预测模型AUC可达0.805(95%CI:0.794~0.811),与随机森林0.730(95%CI:0.702~0.741)和XGboost 0.802(95%CI:0.780~0.807)相比,LightGBM算法显示出了较高的准确率,从而使其成为较强的预测模型。简化后的模型确定的8个特征可以达到0.75左右的AUC。结论构建的全新的焦虑抑郁评估预测模型,可以应用于基层健康普查或者自查,对社区老年人焦虑抑郁进行快速预测。

Objective To develop a prediction model using machine learning to identify anxiety and depression in elderly individuals.MethodsThis study collected data from 15079 elderly individuals in Shanxi Province,including their social demographic factors and disease status.Anxiety and depression were evaluated using GAD-7 and PHQ-9 scales to understand the characteristics of mental illness in the elderly.The evaluation indexes included accuracy,recall,precision,F1 score,Receiver Operating Characteristic Curve(ROC),and area under the curve(AUC),which were derived from the confusion matrix and several models.ResultsThe output of our study clearly demonstrates that the full feature prediction based on LightGBM is highly accurate,with an AUC of 0.805[95%CI:0.794-0.811].This outperforms the Random Forest model,which achieved an AUC of 0.730[95%CI:0.702-0.741],and the XGboost model,which achieved an AUC of 0.802[95%CI:0.780-0.807].Therefore,LightGBM algorithm proves to be a strong prediction model.Our simplified model,based on eight selected features,also achieves a respectable AUC of approximately 0.75.ConclusionsThe new prediction model for anxiety and depression specifically designed for the elderly can be effectively utilized in grassroots health surveys or for self-examinations to efficiently predict anxiety and depression levels among the elderly population in the community.