Risk Estimation Intervention Reveals Hurdles for Suicide Prevention

A mixed-methods quality improvement study published in JAMA Network Open found that using suicide risk estimation analytics did not augment existing prevention practices as intended during routine mental health specialty encounters, revealing important considerations for encounter-based identification of suicide risk.

The study noted that the development and testing of suicide risk estimation analytics models, which leverage sociodemographic and clinical characteristics to quantify risk for defined patient populations, have shown promise.

However, the researchers highlighted that there is little evidence to support the routine use of estimation models during clinical encounters. Existing efforts have focused primarily on patient outreach, such as the implementation of ReachVet, which successfully used suicide risk estimation analytics to identify at-risk veterans and address care needs within the Veterans Health Administration network.

To address this gap in the literature, the researchers designed a mixed-methods quality improvement study to describe the encounter-based implementation of suicide risk estimation analytics. The study evaluated the use of an estimation analytics tool to augment existing suicide prevention practices during routine mental health encounters at Kaiser Permanente Washington.

During the three-and-a-half-month observation period following implementation, researchers assessed 4,789 encounters for 1,939 patients. Prior to all encounters, the Patient Health Questionnaire-9 (PHQ-9) was administered to patients. An EHR prompt to complete the Columbia-Suicide Severity Rating Scale (C-SSRS) appeared for patients reporting frequent thoughts of self-harm, as indicated by a score of two or three on question nine of the PHQ-9.

The suicide risk estimation model, which had been previously validated, was implemented into existing clinical workflows through a schedule-based flag to prompt additional C-SSRS administration among patients identified at high risk of a suicide attempt.

As part of the implementation, clinicians were instructed to display the schedule-based flag and scripting to help normalize the new risk assessment workflow for patients.

From there, the researchers obtained EHR data during the postimplementation observation period between Dec. 1, 2019, and March 15, 2020, to assess how often the C-SSRS was completed among flagged encounters. Additionally, a sample of 16 participating clinicians and 30 flagged patients were invited to participate in semi-structured audio-recorded virtual interviews to gain additional insights into implementation.

During the observation period, 161 patient encounters were newly flagged via the suicide risk estimation analytics-enabled clinical workflow. Despite this, the researchers found that the encounter-based risk flag did not consistently prompt additional suicide risk assessment as intended.

The research team observed that during newly flagged encounters, 57 patients reported frequent suicidal ideation, characterized by a PHQ-9 question nine score greater than or equal to two, and 95 percent completed a C-SSRS under the pre-existing workflow.

Under the new workflow, 75 patients reported no or infrequent suicidal ideation, characterized by a PHQ-9 question nine score from zero to one, and only 13 percent completed a C-SSRS.

Further, 29 patients did not answer the PHQ-9, and only one patient, or 3 percent, completed a C-SSRS.

Eight of the interviewed clinicians identified implementation concerns, which included lack of follow-up, EHR-related inefficiencies, and reliability and accuracy of the flag. Some also expressed concerns about access to care and potential liability associated with known suicide risk.

Twenty of the interviewed patients echoed similar concerns about the reliability and accuracy of the tool and access to care. Many also expressed fears about the identification of suicide risk resulting in coercive care.

The researchers concluded that these findings highlight important implications for healthcare organizations considering the implementation of suicide risk estimation analytics to support encounter-based prevention efforts.

“First, while the schedule-based flag was simple and inexpensive to implement, it was not effective—a user-centered approach to clinical decision support design is key for prompting intended actions. Second, bidirectional leadership/clinician communication is critical for addressing implementation outcomes and concerns… Finally, findings underscore tensions between clinician concerns about responsibility for patient safety and patient concerns about coercive care,” they explained in the study.

With these in mind, the authors state that the use of suicide prevention tools and practices, including estimation analytics, should support therapeutic alliance rather than reinforcing a ‘culture of blame.’