神经科人群中患者健康问卷-9的替代评分：一种基于从单个项目得分得出的预测算法的方法。

Alternative scoring of the Patient Health Questionnaire-9 in neurological populations: an approach based on a predictive algorithm deriving from individual item scores.

作者信息

Hong Zachary M, Williams Jeanne, Bulloch Andrew, Patten Scott B

机构信息

Cumming School of Medicine, University of Calgary, 3330 Hospital Drive NW, Calgary, AB T2N 4N1, Canada.

Cumming School of Medicine, University of Calgary, 3330 Hospital Drive NW, Calgary, AB T2N 4N1, Canada; Department of Community Health Sciences, University of Calgary, 3280 Hospital Drive NW, Calgary, AB T2N 4Z6, Canada.

出版信息

Gen Hosp Psychiatry. 2022 Jul-Aug;77:37-39. doi: 10.1016/j.genhosppsych.2022.04.011. Epub 2022 Apr 30.

DOI:10.1016/j.genhosppsych.2022.04.011

PMID:35526426

Abstract

OBJECTIVE

The study objective was to assess whether machine learning methods could improve predictive performance of the PHQ-9 for depression in patients with neurological disease. Specifically, we assessed whether a predictive algorithm deriving from all nine items could outperform the tradition of summing the items and applying a cut-point.

METHOD

Data from the NEEDS Study was used (n = 825). Demographic data, PHQ-9 scores, and MDD diagnoses (via the SCID) were obtained. Logistic LASSO, logistic regression, and non-parametric ROC analyses were performed. The ROC curve was used to identify the optimal cut-point for regression-derived predictive algorithms using the Youden method.

RESULTS

The traditional approach to PHQ-9 scoring had a classification accuracy of 85.1% (sensitivity: 84.5%; specificity: 85.2%). The logistic LASSO regression model had a classification accuracy of 85.6% (sensitivity: 83.3%; specificity: 86.1%). The logistic regression model had a classification accuracy of 85.8% (sensitivity: 91.4%; specificity: 84.8%). Both models had similar areas under the curve values (logistic LASSO: 0.9097; logistic regression: 0.9026).

CONCLUSIONS

The current cut-off threshold approach to PHQ-9 scoring and interpretation remains clinically appropriate.

摘要

目的

本研究旨在评估机器学习方法是否能提高PHQ-9对神经疾病患者抑郁症的预测性能。具体而言，我们评估了基于所有九个项目得出的预测算法是否优于传统的将各项目相加并应用切点的方法。

方法

使用了NEEDS研究的数据（n = 825）。获取了人口统计学数据、PHQ-9评分和重度抑郁症诊断（通过SCID）。进行了逻辑LASSO回归、逻辑回归和非参数ROC分析。使用尤登方法通过ROC曲线确定回归衍生预测算法的最佳切点。

结果

PHQ-9评分的传统方法分类准确率为85.1%（敏感性：84.5%；特异性：85.2%）。逻辑LASSO回归模型的分类准确率为85.6%（敏感性：83.3%；特异性：86.1%）。逻辑回归模型的分类准确率为85.8%（敏感性：91.4%；特异性：84.8%）。两个模型的曲线下面积值相似（逻辑LASSO：0.9097；逻辑回归：0.9026）。