Recent studies have provided promising evidence that neuroimaging data can predict treatment outcomes for patients with major depressive disorder (MDD). As most of these studies had small sample sizes, a meta-analysis is warranted to identify the most robust findings and imaging modalities, and to compare predictive outcomes obtained in magnetic resonance imaging (MRI) and studies using clinical and demographic features. We conducted a literature search from database inception to July 22, 2023, to identify studies using pretreatment clinical or brain MRI features to predict treatment outcomes in patients with MDD. Two meta-analyses were conducted on clinical and MRI studies, respectively. The meta-regression was employed to explore the effects of covariates and compare the predictive performance between clinical and MRI groups, as well as across MRI modalities and intervention subgroups. Meta-analysis of 13 clinical studies yielded an area under the curve (AUC) of 0.73, while in 44 MRI studies, the AUC was 0.89. MRI studies showed a higher sensitivity than clinical studies (0.78 vs. 0.62, Z = 3.42, P = 0.001). In MRI studies, resting-state functional MRI (rsfMRI) exhibited a higher specificity than task-based fMRI (tbfMRI) (0.79 vs. 0.69, Z = −2.86, P = 0.004). No significant differences in predictive performance were found between structural and functional MRI, nor between different interventions. Of note, predictive MRI features for treatment outcomes in studies using antidepressants were predominantly located in the limbic and default mode networks, while studies of electroconvulsive therapy (ECT) were restricted mainly to the limbic network. Our findings suggest a promise for pretreatment brain MRI features to predict MDD treatment outcomes, outperforming clinical features. While tasks in tbfMRI studies differed, those studies overall had less predictive utility than rsfMRI data. Overlapping but distinct network-level measures predicted antidepressants and ECT outcomes. Future studies are needed to predict outcomes using multiple MRI features, and to clarify whether imaging features predict outcomes generally or differ depending on treatments.

最近的研究提供了有希望的证据，表明神经影像数据可以预测重度抑郁症（MDD）患者的治疗结果。由于大多数这些研究的样本量较小，因此有必要进行荟萃分析来确定最可靠的发现和成像方式，并比较磁共振成像 (MRI) 和使用临床和人口统计特征的研究中获得的预测结果。我们对从数据库建立到 2023 年 7 月 22 日的文献进行了检索，以确定使用治疗前临床或脑 MRI 特征来预测 MDD 患者治疗结果的研究。分别对临床和 MRI 研究进行了两项荟萃分析。采用荟萃回归来探索协变量的影响，并比较临床组和 MRI 组之间以及 MRI 模式和干预亚组之间的预测性能。 13 项临床研究的荟萃分析得出的曲线下面积 (AUC) 为 0.73，而 44 项 MRI 研究的 AUC 为 0.89。 MRI 研究显示出比临床研究更高的敏感性（0.78 vs. 0.62，Z = 3.42， P = 0.001）。在 MRI 研究中，静息态功能 MRI (rsfMRI) 表现出比基于任务的 fMRI (tbfMRI) 更高的特异性（0.79 vs. 0.69，Z = -2.86， P = 0.004）。结构性 MRI 和功能性 MRI 之间以及不同干预措施之间的预测性能均未发现显着差异。值得注意的是，在使用抗抑郁药物的研究中，预测治疗结果的 MRI 特征主要位于边缘系统和默认模式网络，而电休克治疗 (ECT) 的研究则主要局限于边缘系统网络。 我们的研究结果表明，治疗前脑 MRI 特征有望预测 MDD 治疗结果，优于临床特征。虽然 tbfMRI 研究的任务有所不同，但这些研究总体上的预测效用低于 rsfMRI 数据。重叠但不同的网络水平测量预测了抗抑郁药物和 ECT 的结果。未来的研究需要使用多种 MRI 特征来预测结果，并阐明成像特征是否可以普遍预测结果或根据治疗而有所不同。