Machine learning for mental health and psychiatry research has emerged as a powerful tool set to harness the increased computing power to analyze relationships in large, complex data sets. These findings are ultimately positioned to inform clinical recommendations for diagnosis and prediction of psychopathology and treatment of psychopathology.
The story of machine learning (ML) begins with several possible concepts. IBM computer scientist Arthur Samuel coined the phrase “machine learning” in 1959, and described his pioneering working programming machine to essentially “teach” by applying mathematical and statistical models. His breakthrough was building the first computer-based program of Strategy Game Checker to use the scoring feature to function as a rudimentary algorithm to predict the probability of victory.
Credits: Marina Spence and Jackie Num / Istock / Getty Images Plus
However, Samuel was not alone in cutting out ML as the starting point for what is considered a central element of artificial intelligence. Donald Heb's 1949 theory described the coordinated or associative activity of neurons as a function of synaptic plasticity that occurs during learning, and was placed elegantly as “neurons, wires that fire together,” and remained the basis for influencing the development of artificial neural networks. The fact that ML essentially emerged from emulating brain mechanisms — identifying relationships, inducible reasoning, and predicting the behavior of others — makes it a unique and powerful tool for investigating human behavior and brain health.
Like many journals that feature mental health and psychiatry research, we have seen commentary on the influx of primary research using ML technology and potential clinical applications. Envelope estimates using databases such as PubMed show a surge in ML and psychiatric publications from about 200 in 2023 to over 1,200 in 2024. This is a likely trend to continue. However, with its rapid rise, there is a risk that technical and computational applications will outweigh the general mental health leadership ability to analyze these developments. Our team's editors are familiar with ML, and we have come to understand that these evolving applications can provide powerful insights and improve our understanding of mental health conditions, but these studies, and their findings and implications, are aware of how important it is for researchers from other disciplines to grasp and use in their work.
January 2025 issue of Natural Mental Health It includes several papers that advance the agenda of making ML more accessible for mental health and psychiatric research. Reviews by Lucasius and Coauthors provide a comprehensive overview of the specific processes involved in ML, but are framed to capture the entire lifecycle of an ML project. This is an important distinction from previous reviews and perspectives on ML applications in psychiatry in that it aims to promote collaboration between clinicians, mental health researchers and data scientists to utilize their respective expertise. The paper also considers some basic issues that ML must compete with, such as acceptance and adoption of ML models of diagnostic predictions to assist clinicians, as well as gaps in data sources, including electronic health records that can be used in training models. The author also includes interactive Jupyter notes that allow readers to practice concepts using open datasets.
Previously, standard systematic reviews and meta-analytical techniques were required to synthesize and analyze methods, definitions, and assessments used in large-scale studies. ML technology offers additional means to enhance the assessment of more traditional or established evidence. In the analysis, Blekic and colleagues conducted a systematic review of 30 studies using ML techniques to predict post-traumatic stress disorder. Findings from the systematic review show that ML approaches can improve risk stratification in people with post-traumatic stress disorder, as they can identify different risk profiles and untapped predictors beyond those identified by standard analyses. The authors argue that an integrated model of data-driven components and theory-based models could be a promising new tool for investigating this complex obstacle.
The issue also includes articles by Vannucci and colleagues, showing how to use ML techniques to overcome issues such as replication and generalizability unique to some mental health studies. The authors investigated ML applied to care-related early adversity. This constitutes a wide range of environmental risk factors for later psychopathology, and may be influenced by individual differences and different developmental trajectories. Findings suggest that previous caregiver-related adversity and longer duration of adversity were associated with the development of mental health disorders. Furthermore, the overall presence or absence of these adversities helped to estimate diagnostic risk.
These three papers offer a glimpse into some of the exciting, novel and practical ways in which researchers can incorporate ML into their research. It is noteworthy that each emphasizes the need to use ML as a complement or enhancement to other existing methodologies. This is an important note given that ML technology is in a near constant state of development and improvement. The value of applying ML within a more established research program is its ability to bring to the surface relationships and factors that may not be observed with traditional data analysis. However, a unique limitation of ML applications is that without standard statistical methods for comparison, the interpretability and impact of these models will be impeded. Together, these studies highlight that it is a ripe time for stakeholders in mental health and psychiatric research to create vast hypotheses and collaborate.
