Kravchenko Olly, Bäckman Julia, Mataix-Cols David, Crowley James J, Halvorsen Matthew, Sullivan Patrick F, Wallert John, Rück Christian
Centre for Psychiatry Research, Department of Clinical Neuroscience, Karolinska Institutet & Stockholm Health Care Services, Region Stockholm, Sweden.
Karolinska University Hospital Huddinge, Region Stockholm, M48, SE-14186, Sweden.
BMC Psychiatry. 2025 May 30;25(1):555. doi: 10.1186/s12888-025-07012-x.
Internet-delivered cognitive behavioural therapy (ICBT) is an effective and accessible treatment for mild to moderate depression and anxiety disorders. However, up to 50% of patients do not achieve sufficient symptom relief. Identifying patient characteristics predictive of higher post-treatment symptom severity is crucial for devising personalized interventions to avoid treatment failures and reduce healthcare costs.
Using the Swedish multimodal database MULTI-PSYCH, we evaluated novel and established predictors associated with treatment outcome and assessed the added benefit of polygenic risk scores (PRS) and nationwide register data in a sample of 2668 patients treated with ICBT for major depressive disorder, panic disorder, and social anxiety disorder. Two linear regression models were compared: a baseline model employing six well-established predictors and a full model incorporating six clinic-based, 32 register-based predictors, and PRS for seven psychiatric disorders and traits. Predictor importance was assessed through bivariate associations, and models were compared by the variance explained in post-treatment symptom scores.
Our analysis identified several novel predictors of higher post-treatment severity, including comorbid ASD and ADHD, receipt of financial benefits, and prior use of psychotropic medications. The baseline model explained 27%, while the full model accounted for 34% of the variance.
The findings suggest that a model incorporating a broad array of multimodal data offered a modest improvement in explanatory power compared to one using a limited set of easily accessible measures. Employing machine learning algorithms capable of capturing complex non-linear associations and interactions is a viable next step to improve prediction of post-ICBT symptom severity.
Not applicable.
互联网认知行为疗法（ICBT）是治疗轻至中度抑郁和焦虑症的一种有效且可及的疗法。然而，高达50%的患者症状缓解不充分。识别可预测治疗后症状严重程度较高的患者特征，对于设计个性化干预措施以避免治疗失败和降低医疗成本至关重要。
利用瑞典多模式数据库MULTI-PSYCH，我们评估了与治疗结果相关的新的和已确立的预测因素，并在2668例接受ICBT治疗的重度抑郁症、惊恐障碍和社交焦虑症患者样本中评估了多基因风险评分（PRS）和全国登记数据的附加益处。比较了两个线性回归模型：一个采用六个已确立的预测因素的基线模型，以及一个纳入六个基于临床、32个基于登记的预测因素以及七种精神疾病和特征的PRS的完整模型。通过双变量关联评估预测因素的重要性，并通过治疗后症状评分中解释的方差比较模型。
我们的分析确定了几个治疗后严重程度较高的新预测因素，包括共病的自闭症谱系障碍和注意力缺陷多动障碍、领取经济福利以及先前使用精神药物。基线模型解释了27%的方差，而完整模型解释了34%的方差。
研究结果表明，与使用一组有限的易于获取的测量方法的模型相比，纳入广泛多模式数据的模型在解释力方面有适度提高。采用能够捕捉复杂非线性关联和相互作用的机器学习算法是下一步提高ICBT后症状严重程度预测的可行方法。
不适用。
