Clinical Decision Support to Prevent Suicide (CDS_PS)

A Pragmatic Randomized Controlled Trial of Machine Learning-driven Clinical Decision Support to Improve Suicide Risk Screening in Ambulatory Care Settings

Suicide kills 132 Americans every day. The first step of suicide prevention is risk identification and prognostication. Researchers like this study team have developed and validated predictive models that use routinely collected Electronic Health Record (EHR) data like past diagnoses and medications to predict future suicide attempt risk. The study team's model based in machine learning is known as the Vanderbilt Suicide Attempt and Ideation Likelihood (VSAIL). VSAIL has been validated prospectively and externally to predict suicide attempt risk with a number needed to screen (NNS) of 271 for suicide attempt and 23 for suicidal ideation. NNS is the number of people who need to receive a test result to prevent one outcome - lower NNS is better. This study will evaluate the effectiveness of a Clinical Decision Support System called Vanderbilt Safecourse using VSAIL to prompt a novel Best Practice Advisory (BPA) to prompt face-to-face screening with a validated suicide screening instrument like the Columbia Suicide Severity Rating Scale (CSSRS).

The investigators seek to study if identifying patients at high predicted risk of suicide in clinical settings where suicide risk screening only happens sporadically, if at all, will improve face-to-face screening rates and documentation of suicide risk assessment in their EHRs. The investigators will measure the VSAIL-prompted BPA's effectiveness in real-world clinical settings to increase rates of face-to-face suicide risk screening. VSAIL requires only data already collected in routine clinical encounters and is calculated in real- time (seconds) at the start of a clinical visit (inpatient or outpatient) at VUMC. VSAIL does not replace clinical judgment in treating suicidality, but the investigators seek to measure whether VSAIL increases the rates at which the important problem of suicide is addressed and screened effectively. The investigators seek to compare an active, Interruptive intervention, a VSAIL-prompted BPA pushed to providers, to a passive, non-interruptive visual prompt to determine if 1) CDS driven by automated risk modeling improves face-to-face screening rates and 2) whether or not that CDS needs to be interruptive or non-interruptive to be effective. In the latter case, effective non-interruptive CDS would improve care without worsening "alert fatigue." For equipoise, risk scores for all patients in the study sites would be made available in Epic flowsheets for review by providers if they choose to do so. In the first phase, The investigators will pilot this CDS in Neurology outpatient clinics for six months. If study goals are met, The investigators will scale the CDS intervention trial across non-mental health specialty settings at VUMC over the following 18 months.

Those randomized to intervention arm will have the intervention BPA prompt. Those randomized to the control arm will not have an associated BPA prompt. Risk predictions for all participants in study settings will be visible in eStar flowsheets for equipoise.

Adults at Elevated Predicted Risk (>=2% predicted risk, based on our study, DOI: 10.1001/jamanetworkopen.2021.1428) at visit registration (aka "check-in") will be randomized to either Interruptive Alert or Passive Prompt CDS interventions. In the Interruptive Alert, the physician who next opens that patient's chart will be prompted to review a BPA describing the patient's risk and asking the physician to choose from the BPA options (see Interventions). The BPA will need to be dismissed to the Epic Storyboard or completed as above to continue clinical workflow.

Adult patients in the highest predicted risk tier (>=2% predicted risk, based on our research study, DOI: 10.1001/jamanetworkopen.2021.1428) at the time of visit registration (aka "check-in") will be randomized to either the Interruptive Alert or Passive Prompt CDS intervention arms. In the Passive Prompt arm, the physician who next opens the patients' charts after check-in (e.g., "Chart Review") will see a Storyboard icon for Elevated Suicide Risk on the left side of the screen. Hovering over this icon will bring up a window with the BPA information in a view identical to the Interruptive Alert arm. Clicking on the window will bring up the BPA with full functionality as in the Interruptive Alert arm.

Experimental: CDS Intervention Adult patients in the highest predicted risk tier (>=2% predicted risk) at the time visit registration (aka "check-in"). The Intervention would then be described as: Physicians review a Best Practice Advisory describing the patient's risk and choose from the following options: Document face-to-face suicide risk screening with the Columbia Suicide Severity Rating Scale (CDS complete once the CSSRS is complete) Agree to screen using an alternative method chosen at the clinician's discretion (CDS complete) Disagree with alert and provide a rationale for disagreement (CDS complete) Indicate suicide risk face-to-face screening has already occurred (CDS complete) Dismiss the alert. The alert will be available in Epic "Storyboard" persistently until acted on further. The Storyboard functionality after dismissal is identical to the Passive Prompt, CDS intervention (below)

Adult patients in the highest predicted risk tier (>=2% predicted risk, based on our research study, DOI: 10.1001/jamanetworkopen.2021.1428) at the time of visit registration (aka "check-in") will be randomized to either the Interruptive Alert or Passive Prompt CDS intervention arms. In the Passive Prompt arm, the physician who next opens the patients' charts after check-in (e.g., "Chart Review") will see a Storyboard icon for Elevated Suicide Risk on the left side of the screen. Hovering over this icon will bring up a window with the BPA information in a view identical to the Interruptive Alert arm. Clicking on the window will bring up the BPA with full functionality as in the Interruptive Alert arm.

Documentation of the validated, brief-version of the Columbia Suicide Severity Rating Scale (https://www.cms.gov/files/document/cssrs-screen-version-instrument.pdf) or recording a binary response "Will screen with alternative method" input by providers into the intervention Best Practice Advisory using a button in the Acknowledge Reason section of the BPA.

Documented clinical encounter anywhere in health system with documented diagnostic code (ICD10CM) for suicidal ideation The Electronic Health Record (EHR) will be queried every month and all ICD10CM codes documented for patients seen in the study sites will be analyzed for the presence of any ICD10CM codes in a reference list for suicidal ideation. In ICD10CM, a single code, R45.81, is used to document suicidal ideation. If any ICD10CM codes that match that list are found within sixty days of the encounter date in study settings, the outcome will be marked as a case of subsequent suicidal ideation. This outcome is measured on a binary scale, 1= presence of encounter for suicidal ideation and 0 = absence

Documented clinical encounter anywhere in health system with documented diagnostic code (ICD10CM) for suicide attempt The EHR will be queried every month and all ICD10CM codes documented for patients seen in the study sites will be analyzed for the presence of any ICD10CM codes in a reference list for suicide attempt. In ICD10CM, the National Center for Health Statistics provides a full list of acceptable codes that will be used for reference (https://www.cdc.gov/nchs/data/nhsr/nhsr108.pdf). If any ICD10CM codes that match that list are found within sixty days of the encounter date in study settings, the outcome will be marked as a case of subsequent suicidal ideation. This outcome is measured on a binary scale, 1= presence of encounter for suicide attempt and 0 = absence

Any Emergency Department Encounter within sixty days of study encounter and reason for Emergency Department Encounter The EHR will be queried every month and all clinical encounters (aka visits) will be analyzed for location, which is recorded in both Visit Type fields (type "E") in our records and by location (e.g., "Emergency Department"). The presence of a recorded visit occurring within sixty days of a clinical encounter in the study settings will be marked as a case of Emergency Department Utilization for trial purposes. The reason for visit will be analyzed as the ICD10CM codes billed for that encounter, also stored in the EHR. This outcome is measured on a binary scale, 1= presence of Emergency Department encounter and 0 = absence. The reason for encounter will be analyzed secondarily and descriptive statistics used to count the most common reasons related to subsequent Emergency Department encounters.

Inclusion Criteria: > 18 years of age A visit in Neurology (first phase) or other non-mental health specialty setting at VUMC (second phase) Exclusion Criteria: 1) A CSSRS conducted within one week of the visit in another care setting at VUMC (recent screening)

Walsh CG, Johnson KB, Ripperger M, Sperry S, Harris J, Clark N, Fielstein E, Novak L, Robinson K, Stead WW. Prospective Validation of an Electronic Health Record-Based, Real-Time Suicide Risk Model. JAMA Netw Open. 2021 Mar 1;4(3):e211428. doi: 10.1001/jamanetworkopen.2021.1428.