Patient perspectives on acceptability of, and implementation preferences for, use of electronic health records and machine learning to identify suicide risk

Objective: Assess patient understanding of, potential concerns with, and implementation preferences related to automated suicide risk identification using electronic health record data and machine learning. Method: Focus groups (n = 23 participants) informed a web-based survey sent to 11,486 Kaiser Permanente Northwest members in April 2020. Survey items assessed patient preferences using Likert and visual analog scales (means scored from 50 to 50). Descriptive statistics summarized findings. Results: 1357 (12%) participants responded. Most (84%) found machine learning-derived suicide risk identification an acceptable use of electronic health record data; however, 67% objected to use of externally sourced data. Participants felt consent (or opt-out) should be required (mean = 14). The majority (69%) supported outreach to at-risk individuals by a trusted clinician through care messages (57%) or telephone calls (47-54%). Highest endorsements were for psychiatrists/therapists (99%) or a primary care clinician (75-96%); less than half (42%) supported outreach by any clinician and participants generally felt only trusted clinicians should have access to risk information (mean = 16). Conclusion: Patients generally support use of EHR data (not externally sourced risk information) to inform automated suicide risk identification models but prefer to consent or opt-out; trusted clinicians should outreach by telephone or care message to at risk individuals.