Table 1_Predictive factors for referral to a peer support worker in psychosocial rehabilitation centers.docx

IntroductionPeer support workers are individuals with personal lived experience of mental health conditions, addictions, or neurodevelopmental disorders, and can be employed as professionals within mental health services. This study aims to identify predictive factors for patient referral to peer support intervention in psychosocial rehabilitation services.
MethodsUsing data from the French REHABase cohort, we compared variables between patients referred (n=134) and not referred (n=242) to peer support intervention. We evaluated an expert-based model (clinician-selected variables) against a machine-based model (algorithm-selected variables) for predictive accuracy.
ResultsThe machine-based model outperformed the expert-based model in the full dataset (AUC = 0.78 vs 0.71). However, the predictive performance of both models substantially declined after cross-validation, yielding modest AUC values (0.60 and 0.59), which constitutes a key limitation of the study.
DiscussionNeurodevelopmental disorder diagnosis, social isolation, and low treatment adherence predicted peer support referral. Poor model performance may be due to unmeasured factors like patient motivation or clinicians’ perceptions of peer support workers.

## 引言
同伴支持工作者（Peer Support Worker）指具备精神健康状况、成瘾障碍或神经发育障碍等个人亲身经历的个体，可作为专业人员受聘于精神卫生服务领域。本研究旨在识别社会心理康复服务中，患者被转介至同伴支持干预（Peer Support Intervention）的预测因素。

## 研究方法
本研究采用法国REHABase队列数据，对比了被转介至同伴支持干预的患者（n=134）与未被转介者（n=242）的各项变量。本研究分别构建基于专家的模型（由临床医师选定变量）与基于机器学习的模型（由算法选定变量），并对二者的预测准确率进行评估。

## 研究结果
在全数据集上，基于机器学习的模型表现优于基于专家的模型（曲线下面积[AUC]分别为0.78与0.71）。然而，经交叉验证后，两种模型的预测性能均出现显著下降，仅获得较低的AUC值（分别为0.60与0.59），这构成了本研究的关键局限。

## 讨论
神经发育障碍诊断、社会隔离与治疗依从性低下可预测患者被转介至同伴支持干预。模型性能不佳可能源于未被测量的因素，例如患者的治疗动机，或是临床医师对同伴支持工作者的认知态度。