In a recent article, researchers tested novel machine learning tools to detect Major Depressive Disorder (MDD) and Generalized Anxiety Disorder (GAD) using electronic health records (EHRs). EHRs comprise personal health information, including biometric markers such as height and weight, previous medical diagnoses, and demographic information such as access to health insurance and household income. MDD and GAD are prevalent mental health conditions that often go undetected, resulting in delayed treatment. The study utilized a dataset of 4,184 undergraduate students who underwent health screenings and psychiatric assessments. Notably, the authors excluded psychiatric information from their machine learning model; instead, they relied solely on biometric and demographic features as their predictors.

The machine learning model consisted of various algorithmic methods, including deep learning techniques. The model exhibited accuracy above chance for the detection of MDD and GAD, demonstrating a medium effect size (indicating a good, but not excellent performance). This accuracy is comparable to machine learning models developed in prior studies that used psychiatric features for their prediction.

The study also utilized an advanced machine learning classification technique to determine the impact of individual features on the predictions. Factors like satisfaction with living conditions, health insurance type, vaccination status, and marijuana use emerged as top predictors for MDD and GAD. Biometric markers such as high blood pressure and hypertension were also identified as predictors for GAD. Additionally, two-way interactions between features played a significant role in influencing predictions. For example, someone with high marijuana use was generally more likely to have GAD when they were also overweight. Limitations to the study included a homogeneous sample, as all participants were French undergraduate students. Therefore, the efficacy of the model may not extend to the general population.

Overall, the study demonstrates promising advances in the use of machine learning tools for detecting mental health issues through readily available clinical data. Therefore, it presents a potential tool for early detection of GAD and MDD, although is not a diagnostic tool in itself. If a similar (and perhaps better performing) model is deployed in the healthcare system at some point, individuals that are identified as “high risk” will still need to go through a mental health screening conducted by a professional before receiving a diagnosis.