Enhancing diagnostic and prognostic prediction of child psychopathology via multivariate predictive modeling: Where are we and where do we need to go?

Advances in computational science and machine learning provide an opportunity to improve access and quality of mental health care in the near-term. Applications of these tools to child dysregulatory psychopathology and attention-deficit/hyperactivity disorder are of particular interest because they are common and place high burden on healthcare systems; early identification and effective treatment can alter life-long trajectories. The successful development and validation of predictive tools would allow clinicians to 1) efficiently and accurately screen for diagnoses, even in settings with limited behavioral or mental health training; 2) allocate scarce behavioral health resources; 3) work more effectively with children and families around risk-benefit decisions; and 4) initiate tailored treatment plans before crises emerge. Yet this work is challenging. Here, we outline some of what we see as the most pressing conceptual, practical, and computational challenges. We propose potential paths to move the field forward by leaning into individual variation not as noise, but as predictive information; considering novel phenotypic features that have traditionally been outside the scope of diagnostic practice, such as cognitive and temperament-based emotional features; and enhancing cross-disciplinary collaboration, including with patient and provider stakeholders, to ensure robust model development and evaluation that directly addresses clinical goals. We provide specific recommendations for rigorously testing internal and external validation and addressing clinical utility to enhance clinical translation. As the field undertakes this work, careful thought about who is being helped and who is being potentially left behind because of limitations of existing data sources and limits on generalizability will be critical.