Background: Suicide is the 10th leading cause of death. PHQ9 item 9 (which asks about suicidal thoughts) identifies those at risk of suicide attempt/death. Patients with scores of 2 or 3 on item 9 show a sustained increase in risk, with a cumulative hazard near 4 % over 12 months. Methods: Outpatients who score a 2 or 3 on item 9 of the PHQ9 are identified using EHR data at 3 MHRN sites: Group Health, HealthPartners and Kaiser Colorado. Using a modified Zelen design, patients are automatically assigned 1:1:1 to continue in usual care (i.e. no contact) or to be offered one of two population-based prevention programs meant to supplement usual care: (1) Care Management (systematic outreach to assess risk, EHR-based tools for risk-based pathways, and care management to facilitate and monitor recommended follow-up care), or (2) Skills Training (interactive online training in dialectical behavioral therapy skills supported by reminder and reinforcement messages). Randomization automatically occurs within each site’s sampling computer program, stratified by item 9 score. A computer-generated concealed allocation table provides randomly generated assignments in block sizes of either 6 or 9. The multi-site interventions are embedded in the EHR. Online patient-provider secure messaging via the EHR patient portal is used for patient invitation and outreach, as well as administration of suicide risk questionnaires. Secure provider-to-provider messaging is used to communicate with primary care and mental health providers. Population management and reporting tools are used to apply follow-up algorithms and deliver recommendations to care managers regarding outreach and follow-up. Nonfatal and fatal suicide attempts are identified using state vital statistics data and diagnoses of self-inflicted injury from EHR and claim records. Primary evaluation will compare risk of first suicide attempt over the 18 months following randomization. Groups will be compared according to initial treatment assignment, regardless of level of participation in either intervention. Results: To date, 4,869 outpatients out of a planned 18,000 have been randomized across the 3 sites. Conclusion: Our experience thus far illustrates the promise and challenges of implementing multi-site clinical trial recruitment and intervention delivery in electronic health records systems.