Optimizing psychiatric hospitalization decisions: staff perspectives on a machine learning clinical decision support tool for suicidal veterans in the emergency department

Abstract Background Although hospitalization is the standard of care for emergency department (ED) patients judged to have high risk of suicide, experimental and non-experimental studies have failed to find evidence that hospitalization of suicidal ED patients is associated with reduced prevalence of subsequent suicides. However, a recent machine learning analysis of Veterans Health Administration (VHA) data found evidence that this failure might be due to hospitalization helping some suicidal veterans (i.e., reducing their risk of subsequent suicide) but hurting others (i.e., increasing risk). If made readily available to ED staff, this information creates an opportunity to optimize hospitalization decisions, thereby improving clinical outcomes and decreasing costs associated with low-value hospitalization. Similar to other acute issues seen in the ED (e.g., concern for heart attack), this can be accomplished through use of a clinical decision support (CDS) tool available through the medical record that uses VHA data to provide real-time risk classification and hospitalization recommendations. In this qualitative study, we sought to conduct an initial implementation assessment to understand ED suicide prevention-related workflows and ED-based staff perspectives on barriers and facilitators to using a prototype CDS tool. Methods Two phases of individual interviews ( n = 18) were conducted with VHA ED-based staff ( n = 14) using an interview guide informed by the Consolidated Framework for Implementation Research, a framework which seeks to explain implementation barriers and facilitators across domains such as where the implementation is occurring (e.g., inner setting, outer setting), who is implementing (e.g., characteristics of individuals), what is being implemented (e.g., innovation), and how the implementation is happening (e.g., implementation process). The interviews focused on the best way to present CDS recommendations to clinicians. Interview data were analyzed using rapid qualitative analysis and themes were identified through use of templated summaries, matrices, and discussion to reach consensus. Results Interviewees are generally open to integrating a simple, easy to access CDS tool to assist with hospitalization decisions. Most interviewees found the refined CDS tool prototype to be user-friendly, with some caveats. Namely, (1) workflow integration would need to be individualized to maximize uptake, (2) comprehensive staff training on use of the tool would be necessary, (3) strong evidence that the tool is effective would be needed to support use in clinical practice. Interviewees maintained that the tool would be most useful in situations where clinicians are undecided about whether to hospitalize a veteran expressing suicidal ideation and other behaviors. Conclusions Feedback from ED staff facilitated the refinement of a user-friendly CDS tool to be used in a future pilot implementation study.

Authors

Institutions

Publication Details

Journal
BMC Medical Informatics and Decision Making
Published
2026-11-13
DOI
https://doi.org/10.1186/s12911-026-03932-0
Primary Topic
Electronic Health Records Systems
Type
article
Field-Weighted Citation Impact
0.00

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Optimizing psychiatric hospitalization decisions: staff perspectives on a machine learning clinical decision support tool for suicidal veterans in the emergency department

Brian P. Marx, Claire Houtsma, Ronald C. Kessler, Brian Fuehrlein et al.
BMC Medical Informatics and Decision Making
Electronic Health Records Systems
article

Optimizing psychiatric hospitalization decisions: staff perspectives on a machine learning clinical decision support tool for suicidal veterans in the emergency department

Brian P. Marx, Claire Houtsma, Ronald C. Kessler, Brian Fuehrlein, Neil Patel, Jaclyn C. Kearns, Kate H. Bentley
article en

Abstract

Abstract Background Although hospitalization is the standard of care for emergency department (ED) patients judged to have high risk of suicide, experimental and non-experimental studies have failed to find evidence that hospitalization of suicidal ED patients is associated with reduced prevalence of subsequent suicides. However, a recent machine learning analysis of Veterans Health Administration (VHA) data found evidence that this failure might be due to hospitalization helping some suicidal veterans (i.e., reducing their risk of subsequent suicide) but hurting others (i.e., increasing risk). If made readily available to ED staff, this information creates an opportunity to optimize hospitalization decisions, thereby improving clinical outcomes and decreasing costs associated with low-value hospitalization. Similar to other acute issues seen in the ED (e.g., concern for heart attack), this can be accomplished through use of a clinical decision support (CDS) tool available through the medical record that uses VHA data to provide real-time risk classification and hospitalization recommendations. In this qualitative study, we sought to conduct an initial implementation assessment to understand ED suicide prevention-related workflows and ED-based staff perspectives on barriers and facilitators to using a prototype CDS tool. Methods Two phases of individual interviews ( n = 18) were conducted with VHA ED-based staff ( n = 14) using an interview guide informed by the Consolidated Framework for Implementation Research, a framework which seeks to explain implementation barriers and facilitators across domains such as where the implementation is occurring (e.g., inner setting, outer setting), who is implementing (e.g., characteristics of individuals), what is being implemented (e.g., innovation), and how the implementation is happening (e.g., implementation process). The interviews focused on the best way to present CDS recommendations to clinicians. Interview data were analyzed using rapid qualitative analysis and themes were identified through use of templated summaries, matrices, and discussion to reach consensus. Results Interviewees are generally open to integrating a simple, easy to access CDS tool to assist with hospitalization decisions. Most interviewees found the refined CDS tool prototype to be user-friendly, with some caveats. Namely, (1) workflow integration would need to be individualized to maximize uptake, (2) comprehensive staff training on use of the tool would be necessary, (3) strong evidence that the tool is effective would be needed to support use in clinical practice. Interviewees maintained that the tool would be most useful in situations where clinicians are undecided about whether to hospitalize a veteran expressing suicidal ideation and other behaviors. Conclusions Feedback from ED staff facilitated the refinement of a user-friendly CDS tool to be used in a future pilot implementation study.

BMC Medical Informatics and Decision Making
Boston University (US), United States Department of Veterans Affairs (US), Harvard University (US), University of California, Los Angeles (US), Veterans Health Administration (US), VA Boston Healthcare System (US), Yale University (US), VA Connecticut Healthcare System (US), Massachusetts General Hospital (US), Southeast Louisiana Veterans Health Care System (US)
U.S. Department of Veterans Affairs
Good health and well-being
Electronic Health Records Systems
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon