Held Philip, Boland Ashby, Pridgen Sarah A, Smith Dale L
Department of Psychiatry and Behavioral Sciences, Rush University Medical Center, Chicago, IL, USA.
Department of Psychiatry, University of Illinois - Chicago, Chicago, IL, USA.
Eur J Psychotraumatol. 2025 Dec;16(1):2514885. doi: 10.1080/20008066.2025.2514885. Epub 2025 Jun 23.
This study examined whether baseline demographic and clinical variables could predict clinically significant reductions in insomnia symptoms among veterans receiving a 2-week Cognitive Processing Therapy (CPT)-based intensive PTSD treatment programme (ITP). A key aim was to identify individuals likely to benefit from additional, sleep-focused interventions. A total of 449 veterans completed the Insomnia Severity Index (ISI) at baseline, post-treatment, and at 1- and 3-month follow-up. Linear mixed models were used to analyse insomnia trajectories and clinical predictors (e.g. PTSD severity, depression, posttrauma cognitions, neurobehavioral symptoms). Machine learning models (neural net, random forest, elastic net, and ensemble) were trained to classify participants with clinically meaningful insomnia improvements. Veterans reported large average PTSD severity reductions ( = 0.96), whereas depression and insomnia symptoms reduced moderately ( = 0.57) and modestly ( = 0.34), respectively. Higher PTSD severity, depression severity, negative posttrauma cognitions, and neurobehavioral symptoms were linked to poorer insomnia outcomes. None of the demographic or military service factors significantly predicted insomnia trajectories. Machine learning models performed only slightly better than chance (AUC range = 0.52-0.62) in predicting insomnia severity reductions. Although insomnia symptoms improved during the ITP and improvements persisted up to 3-month follow-up, a substantial number of veterans continued to experience significant sleep problems post-treatment. Given the limited predictive accuracy of machine learning models using self-reported variables, incorporating additional biological or psychosocial factors may be necessary to identify veterans with PTSD who need more specialised sleep interventions.
本研究调查了基线人口统计学和临床变量是否能够预测接受为期2周的基于认知加工疗法（CPT）的创伤后应激障碍强化治疗项目（ITP）的退伍军人失眠症状临床上的显著减轻。一个关键目标是识别可能从额外的、以睡眠为重点的干预措施中获益的个体。共有449名退伍军人在基线、治疗后、1个月和3个月随访时完成了失眠严重程度指数（ISI）。使用线性混合模型分析失眠轨迹和临床预测因素（如创伤后应激障碍严重程度、抑郁、创伤后认知、神经行为症状）。训练机器学习模型（神经网络、随机森林、弹性网络和集成模型）对失眠症状有临床意义改善的参与者进行分类。退伍军人报告创伤后应激障碍严重程度平均大幅降低（=0.96），而抑郁和失眠症状分别适度降低（=0.57）和轻度降低（=0.34）。创伤后应激障碍严重程度较高、抑郁严重程度较高、创伤后消极认知和神经行为症状与较差的失眠结局相关。人口统计学或军事服务因素均未显著预测失眠轨迹。机器学习模型在预测失眠严重程度降低方面仅略优于随机水平（AUC范围=0.52 - 0.62）。尽管在ITP期间失眠症状有所改善，且改善持续到3个月随访，但仍有相当数量的退伍军人在治疗后继续经历严重的睡眠问题。鉴于使用自我报告变量的机器学习模型预测准确性有限，纳入额外的生物学或心理社会因素可能有必要，以识别需要更专业睡眠干预的创伤后应激障碍退伍军人。
