Author: Ishmam Khan, Class of 2025

Figure 1: MDD is a devastating, extremely common, and fast-growing disease in terms of suffering, mortality, and cost to society.

Since COVID-19, the rates of mental health disorders have increased significantly. One such disorder is Major Depressive Disorder (MDD), a serious disorder affecting more than 8% of the US population. As of 2024, the remission rates, or rate of mental health disorders being suppressed as a result of treatment, for people with MDD are disproportionately lower than those with other mood disorders. To address this statistical disparity, researchers have turned to multimodal approaches to address treatment outcomes and applications. Recently, efforts to forecast MDD remission using a combination of clinical data, neuroimaging, and demographic features have grown, but have achieved little consistent success in clinical practice.

A 2024 study from Stony Brook University done by Wang et. al set out on a similar task. They created an advanced machine learning algorithm that utilized data from two clinical trials to “predict” remission rates for MDD based on the Hamilton Depression Rating Scale (HDRS). These two clinical trials – APAT(Advanced Personalized Antidepressant Treatment) and EMBARC (Establishing Moderators and Biosignatures of Antidepressant Response for Clinical Care for Depression) – were represented in studies done in Pittsburgh and Pisa respectively. Participants were subject to personal questionnaires regarding MDD. Then, they were placed under an MRI machine to image the biological mechanisms of MDD in real-time. Afterward, all this data was integrated into the researchers’ algorithm. Since this trial built upon past studies, the algorithm was able to diagnose the participants with MDD more consistently.

 However, the researchers wondered if they could extrapolate these findings to the results of past studies. Since they had already used clinical data from two past research groups, they looked at the remission rates predicted by the respective algorithms in Pittsburgh and Pisa. Notably, Pittsburgh treatments had comparatively different remission rates represented in their findings compared to the non-significant differences from the group in Pisa. This indicated that the lack of high performance of past models was not due to one particular group of features. Instead, the issue lay with something more fundamental. With how current ML algorithms are structured for clinical trials, remission rates cannot be reliably portrayed in research findings. Consequently, those with MDD will continue to seek treatment without being accurately represented for remission by these methods.

While neuroimaging offers objective insights into brain function, integrating it with clinical features in a meaningful way to predict MDD remission remains a challenge. This study underlines the complexity of using multimodal data for predictive modeling in mental health treatment and emphasizes the need for more refined data processing and feature selection techniques. Until these improvements are made, machine-learning models will struggle to reliably forecast MDD remission outcomes in practical settings.

Works Cited:

[1]Wang, J., Wu, D. D., DeLorenzo, C., & Yang, J. (2024). Examining factors related to low performance of predicting remission in participants with major depressive disorder using neuroimaging data and other clinical features. Plos one, 19(3), e0299625.

[2] Image retrieved from: https://pixabay.com/illustrations/mind-brain-mindset-perception-544404/