基于Lasso回归构建精神分裂症患者暴力风险预测模型

Lasso Regression-Driven Model Construction for Violence Risk Stratification in Schizophrenia

中文关键词:  Lasso回归  精神分裂症  暴力风险  模型

英文关键词:Lasso regression  Schizophrenia  Risk of violence  model

中文摘要:

      背景 精神分裂症患者的暴力攻击行为其突发性强、难以预测，且暴力危害程度重、防范难度大，实现对精神分裂症患者暴力攻击风险的早期识别和评估具有重要的临床价值。方法 以2022年3月-2024年9月在太原市精神病医院住院治疗的200例精神分裂症患者为建模队列。基于Lasso回归算法对特征变量进行筛选和降维处理，选取机器学习的支持向量进行模型训练与预测。结果 通过对16个关键特征变量进行Pearson相关分析，结果显示：既往暴力史与临床症状严重程度呈中度正相关（r = 0.58，p < 0.001）；治疗依从性与病情稳定性负相关（r = -0.41，p = 0.003）；文化程度与家庭经济水平显著关联（r = 0.57，p < 0.001）；Lasso回归建模得到曲线下面积（AUC）为0.85，说明支持向量机算法模型可以应用于暴力风险的早期识别和分类。结论 支持向量机算法模型在精神分裂症暴力风险评估中的适用水平和整体预测效能较好。

英文摘要:

      Background The violent and aggressive behaviors of patients with schizophrenia are highly sudden, difficult to predict, and characterized by severe harmful consequences and great challenges in prevention. Therefore, achieving early identification and assessment of the risk of violent aggression in schizophrenia patients holds significant clinical value. Method A modeling cohort of 200 schizophrenia patients admitted to Taiyuan psychiatric hospital for treatment from March 2022 to September 2024 was used. Based on Lasso regression algorithm, feature variables are screened and dimensionality reduced, and machine learning support vectors are selected for model training and prediction. Result Pearson correlation analysis of 16 key feature variables revealed: a moderate positive correlation between history of violence and severity of clinical symptoms (r = 0.58, p < 0.001); a negative correlation between treatment compliance and disease stability (r = -0.41, p = 0.003); and a significant association between educational level and family economic status (r = 0.57, p < 0.001). The Lasso regression model achieved an Area Under the Curve (AUC) of 0.85, indicating good performance of the subsequent SVM model. Conclusion The support vector machine algorithm model has good applicability and overall predictive performance in assessing the risk of violence in schizophrenia, and is suitable for clinical promotion.