Exploring the Perspectives of Pediatric Health Care Providers, Youth Patients, and Caregivers on Machine Learning Suicide Risk Classification: Mixed Methods Study.

Background: Suicide was the second leading cause of death for youth aged between 10 and 24 years in 2023, necessitating improved risk identification to better identify those in need of support. While machine learning (ML) applied to electronic health records shows promise in improving risk identification, further research on the perspectives of these tools is needed to better inform implementation strategies. Objective: These findings incorporate 2 studies aimed to explore patient, caregiver, and pediatric health care provider perspectives on suicide risk models and associated clinical practices. We sought to use these findings to inform the design and implementation of a suicide risk model and associated clinical workflow to improve the quality of care provided to at-risk youth. Methods: We conducted a convergent mixed methods study to evaluate pediatric provider perspectives by quantitatively surveying and qualitatively interviewing provider participants. The provider study was guided by the Consolidated Framework for Implementation Research, and data were analyzed descriptively and using a template analysis for quantitative and qualitative data, respectively. Qualitative interviews conducted among youth patients and caregivers as part of a sequential mixed methods study, guided by the Theoretical Framework of Acceptability, were analyzed using template analysis as well. The integration of quantitative and qualitative data was achieved through a joint display, and results were interpreted through a narrative review. Results: Forty-five providers completed the first section of the survey (risk model clinical preferences), while 44 completed the second section (risk model usability perspectives) and 38 completed the third section (Consolidated Framework for Implementation Research barriers and facilitators). Eight pediatric providers were interviewed. Ten semistructured qualitative interviews were conducted among 9 patient participants aged between 8 and 25 years and 1 caregiver. Overall, providers, patients, and caregivers expressed interest in applying ML methods to improve suicide risk identification. Providers felt that these tools could address current challenges in suicide-related care, such as inadequacies in manual suicide screeners or communication across care teams. Patients and caregivers saw potential for these tools to facilitate discussions regarding suicide and promote early intervention for patients who might otherwise be missed by current care practices. However, providers also expressed concerns about increased demand for mental health services, implications for patient confidentiality, coercive care, inaccurate diagnosis and response, and medical-legal liability. Ethical concerns shared by patients and caregivers spanned data safety practices and privacy regulations, respect for patient autonomy and informed consent, and decreased future health care engagement due to poor implementation. Conclusions: There is conditional acceptability and enthusiasm among providers, patients, and caregivers for ML-based suicide risk models. Successful implementation requires the incorporation of provider perspectives in a user-led fashion to build trust and empower clinicians to respond appropriately to risk flags, while upholding youth and caregiver perspectives to adequately accommodate patient needs.