基于随机森林算法的中青年妇科恶性肿瘤患者心理痛苦预测模型构建及验证

Construction and validation of risk prediction model of psychological distress in young and middle-aged patients with gynecologic malignancy based on random forest algorithm

目的基于随机森林算法构建中青年妇科恶性肿瘤患者心理痛苦风险预测模型并验证其预测效果,为医护人员早期发现患者心理痛苦提供工具。方法采用横断面研究,以整群抽样法选取2021年10月至2022年10月天津市6所三级甲等医院妇科和肿瘤科收治的中青年妇科恶性肿瘤患者385例,采用R-studio软件以7∶3比例将研究对象随机分为训练集270例和测试集115例。根据是否存在心理痛苦将训练集患者分为心理痛苦阳性151例和心理痛苦阴性119例,对各影响因素进行单因素分析。采用R-studio软件在训练集上建立中青年妇科恶性肿瘤患者心理痛苦预测的随机森林模型并在测试集上验证。结果模型预测准确度为94.78%,灵敏度为96.88%,特异度为92.16%,阳性预测值为93.94%,阴性预测值为95.92%,AUC为0.992(95%CI 0.982~1.000)。根据各影响因素在随机森林模型中的Gini系数平均下降量进行排序,得出前5位重要预测变量依次为:一般自我效能感量表得分、Herth希望量表得分、领悟社会支持量表得分、抑郁自评量表得分、焦虑自评量表得分。结论基于随机森林算法构建的中青年妇科恶性肿瘤患者心理痛苦预测模型有较高的预测效能,可为医护人员及早识别患者心理痛苦并制订干预措施提供参考。

Objective To construct a prediction model of psychological distress risk in young and middle-aged patients with gynecologic malignancy based on random forest algorithm and validate its prediction effect,which provided a tool for healthcare professionals to detect patients'psychological distress in early stage.Methods This was a cross-sectional study,a total of 385 cases of young and middle-aged patients with gynecologic malignancies admitted to the gynecology and oncology departments of six tertiary hospitals in Tianjin from October 2021 to October 2022 were consecutively included,the study subjects were randomly divided into 270 cases in the training set and 115 cases in the testing set according to 7:3 by R-studio software.After grouping the training set patients according to the presence or absence of psychological distress(positive psychological distress 151 cases and negative psychological distress 119 cases),univariate analysis was performed on each influencing factor.A random forest model for the prediction of psychological distress in young and middle-aged gynecological malignancy patients using R-studio software on the training set,and the prediction effect was verified on the testing set.Results The prediction accuracy was 94.78%,sensitivity was 96.88%,specificity was 92.16%,positive predictive value was 93.94%,negative predictive value was 95.92%,and AUC was 0.992(95%CI 0.982-1.000).The top 5 significant predictor variables were ranked according to the average decrease in the Gini coefficient of each influencing factor in the random forest model:General Self-Efficacy Scale score,Herth Hope Index score,Perceived Social Support Scale score,Self-Rating Depression Scale score,Self-Rating Anxiety Scale score.Conclusions In this study,the prediction model of psychological distress in young and middle-aged patients with gynecologic malignancy constructed by random forest algorithm has high predictive efficacy,which provides reference for healthcare professionals to identify patients'psychological distress early and formulate interventions.