Predicting Depression Risk in Adolescents From Multimodal Data: Current Evidence and Future Directions.
Predicting Depression Risk in Adolescents From Multimodal Data: Current Evidence and Future Directions.
复制标题
DOI：
10.1016/j.bpsc.2021.12.006
复制
发表时间：
2022-04
期刊：
影响因子：
--
通讯作者：
T. Ho
中科院分区：
文献类型：
--
作者：
T. Ho
青春期是发展为重度抑郁症（MDD）的脆弱时期，也是识别这种使人衰弱的疾病的早期风险因素的好时机。尽管先前的研究已经记录了mdd发展的多种风险因素，包括人口统计学、临床、社会心理和神经生物学预测因素，但该领域的研究大多集中在相对有限的一组风险标记上，而不是在广泛的因素中优化预测。这种规模的表型分析是资源密集型的，通常需要多地点的倡议，以充分采样一个广阔的参数空间。使用数据驱动的统计方法，研究人员可以识别预测模型中的顶级特征，并有效地限制该参数空间，这对于指导未来更有限样本量的抑郁症研究具有巨大的潜力。在这一期的《生物精神病学：认知神经科学和神经影像学》中，Toenders等人的一项研究(1)通过利用IMAGEN联盟的数据，在识别青少年抑郁症的多层次风险因素方面取得了重大进展。IMAGEN联盟是一个欧洲八站合作组织，对2000多名14岁的青少年进行了评估，并随访到16岁、19岁和22岁。在他们的研究中，作者纳入了145个基线变量——包括临床、认知和环境因素，以及来自34个皮层和8个皮层下区域的灰质形态测量测量——在惩罚逻辑回归中识别19岁的抑郁症（定义为满足阈下或完全阈下的重度抑郁症，从一个良好验证的抑郁症状的自我报告测量）。作者发现，14岁时的基线抑郁严重程度、女性性别、神经质、有压力的生活事件和边缘上回的表面积是预测抑郁症发病模型的最重要因素（AUROC值：0.68-0.72）。关键的是，作者使用了不同程度的惩罚，并在一组独立的数据（n= 137）上评估了他们的模型性能，这些数据（n= 407）不用于训练他们的模型。有趣的是，同样的模型也预测了数据中一个独立子样本的风险酒精使用的开始，这强调了这些特征可能更普遍地预测青春期的心理健康结果。
Adolescence is a vulnerable period for developing Major Depressive Disorder (MDD) and represents an opportune time to identify early risk factors for this debilitating illness. Although previous studies have documented multiple risk factors for the development of MDD—including demographic, clinical, psychosocial, and neurobiological predictors—research in this area has mostly focused on a relatively limited set of risk markers rather than optimizing prediction across a broad set of factors. Phenotyping at this scale is resource intensive and often requires multi-site initiatives to sufficiently sample an expansive parameter space. Using data-driven statistical methods, researchers can then identify top features in a predictive model and effectively constrain this parameter space, which has enormous potential to guide future depression studies with more limited sample sizes.In this issue of Biological Psychiatry: Cognitive Neuroscience and Neuroimaging, a study by Toenders et al.(1) makes significant strides in identifying multi-level risk factors for adolescentonset depression by leveraging data from the IMAGEN consortium, an eight-site European collaborative where over 2,000 adolescents at age 14 were assessed and followed through ages 16, 19, and 22. In their investigation, the authors included 145 baseline variables—including clinical, cognitive, and environmental factors, as well as measures of gray matter morphometry from 34 cortical and 8 subcortical regions—in a penalized logistic regression to identify depression (defined as meeting subthreshold or full-threshold MDD from a well-validated selfreport measure of depression symptoms) by age 19. The authors found that baseline depression severity at age 14, female sex, neuroticism, stressful life events, and surface area of the supramarginal gyrus were the strongest contributors to a model that predicted onset of depression (AUROC values: 0.68–0.72). Critically, the authors used different levels of penalization and evaluated their model performance on an independent set of data (n= 137) that were not used to train their model (n= 407). Interestingly, the same model was also predictive of the onset of risky alcohol use in an independent subsample from the data, which highlights that these features are likely to be predictive more generally of mental health outcomes during adolescence.