MHRN III Pilot Project 1: Stakeholder Views on Implementation of Suicide Risk Prediction Models

Grant Details

Funder: NIMH

Grant Number: U19MH121738

Grant Period: 09/24/2019 – 6/30/2021

Narrative: Age-adjusted suicide rates have been increasing in the U.S. over the past two decades. In 2017, more than 47,000 Americans died of suicide. Health care visits represent opportunities for suicide prevention because most individuals make an outpatient health care visit within a year of their suicide death and almost half have a visit within a month of their death. However, suicide risk is not always easily recognizable to clinicians—traditional clinical prediction is hardly better than chance. Predictive modeling that identifies patterns in “big data” from administrative and electronic health records has proven superior to clinical suicide risk prediction and routinely used suicide screening instruments. While predictive modeling holds promise for suicide prevention, how models should be implemented in routine clinical practice and the contextual factors that influence their use are understudied. The potential benefits of any risk prediction model, including those designed to identify suicide risks, are dependent on making sure that the models are deployed in a manner that does not harm patients, supports clinical care management, and is sustainable for health care delivery systems. We propose a pre-implementation pilot study in three settings, using one-on-one, in-depth interviews to explore health system administrators’, clinicians’, and patients’ expectations, experiences with, concerns, and suggestions for the early use of suicide risk prediction models. In the first setting, health system administrators are still considering what might be the best implementation approach. Interviews will help us understand how various stakeholder expectations match what is actually occurring in the two other settings where small pilot studies will be in process. One of these settings is planning outreach to high-risk patients independent of health care visits while the other is planning delivery of risk scores at the point of care. By studying different implementation strategies, we can compare relative advantages and disadvantages. We are particularly interested in effects on clinical workflows, clinician-patient relationships, and patient experiences. While there is an emerging literature supporting the promise of predictive models in health care, implementation factors and patient impacts have been largely ignored. Yet decisions regarding design and modeling methods and implementation processes should be driven by stakeholder requirements. Results of this pilot study will have important clinical implications and will not only inform large-scale implementation of suicide risk prediction models in health systems across the country but will also inform development of future risk prediction models and associated care processes tailored to stakeholders needs more generally (not limited to suicide risk). The long-term goals of this pilot project are to inform ongoing health system-level efforts to reduce suicide prevalence and prevent suicides by optimizing the use of suicide risk prediction tools.

  • Lead Site:
    • Overall PI: KPNW (Bobbi Jo Yarborough)
  • Participating Sites/Subcontractors:
    • HPI (site project lead Rebecca Rossom)
    • KPWA (site project lead Julie Richards; site PI Greg Simon)
  • Funder Contacts
    • Science Officer: Susan Azrin
    • Program Official: Michael Freed
    • Grants Management Official: Julie Bergerud


Funding Announcement

Notice of Award

Personnel Contact List

Current Status

We have completed and analyzed interviews with 10 health care administrators, 30 clinicians in behavioral health departments, and 62 patients across three health systems.

Summary of Findings

Administrators and clinicians

  • Use of a suicide risk prediction model and two differing implementation approaches were acceptable.
  • Clinicians desired opportunities for input on implementation decision-making.
  • They wanted to know how this manner of risk identification enhanced existing suicide prevention efforts.
  • They wanted additional training on how the models determined risk and why some patients appeared at risk while others do not.
  • Clinicians were concerned about lack of suicide prevention resources for newly identified patients.
  • They wanted clear procedures for situations when they could not reach patients or when patients remained at-risk over a sustained period.
  • They would like consolidated suicide risk information in a dedicated module in the EHR to increase efficiency.


  • Patients were generally supportive of suicide risk prediction models derived from EHR data.
  • Concerns included: 1) apprehension about inducing anxiety and suicidal thoughts, or 2) triggering coercive treatment, particularly among those who reported prior negative experiences seeking mental health care.
  • Participants engaged in mental health care or case management expected to be asked about suicide risk and largely appreciated suicide risk conversations
  • Patients preferred conversations to come from clinicians comfortable discussing suicidality.


Yarborough BJH, Stumbo SP. Patient perspectives on acceptability of, and implementation preferences for, use of electronic health records and machine learning to identify suicide risk. Gen Hosp Psychiatry. 2021 May-Jun;70:31-37. doi: 10.1016/j.genhosppsych.2021.02.008.

Yarborough BJH, Stumbo SP, Schneider JL, Richards JE, Hooker SA, Rossom RC . Patient expectations of and experiences with a suicide risk identification algorithm in clinical practice. BMC Psychiatry. 2022 Jul 23;22(1):494. doi: 10.1186/s12888-022-04129-1 .