Bentley2022 - Implementing Machine Learning Models for Suicide Risk Prediction in Clinical Practice, Focus Group Study With Hospital Providers
- Date read: 2023-02-07
- Subject: Suicide machine learning
- Bibtex: @bentley2022
- Bibliography: Bentley, K. H., Zuromski, K. L., Fortgang, R. G., Madsen, E. M., Kessler, D., Lee, H., Nock, M. K., Reis, B. Y., Castro, V. M., & Smoller, J. W. (2022). Implementing Machine Learning Models for Suicide Risk Prediction in Clinical Practice: Focus Group Study With Hospital Providers. JMIR Formative Research, 6(3), e30946. https://doi.org/10.2196/30946
- 10 focus groups, 40 hospital providers in various disciplines of medicine. At MGH in Boston.
- Qualitative analysis of audio recordings
- (1) Current suicide risk assessment practices
- (2) Current suicide risk intervention practices
- (3) Attitudes about current assessment and interventions
- (4) General reactions and attitudes toward using ML for suicide risk prediction in clinical practice
- (5) Factors to consider when developing or implementing such models
- (6) Barriers and concerns to using such systems
- (7) Recommendations about system content or format
- (8) Recommendations about placement of systems within the electronic health record
- 53% of the participants felt that they received too many alerts per week from the electronic health record. Is this true in Sweden? My experience is that we don’t really have alerts in this way. Alert fatigue
- Future direction: include ideas from patients as well?
Notably, developing a safety plan or other brief, suicide-focused interventions with empirical support  was only mentioned in 20% (2/10) of the groups.
The single most common barrier to the potential use of automated suicide risk–prediction models in routine care was the implication for liability. For example, many were concerned about being held legally responsible if they decided not to hospitalize a patient who was categorized as high risk and then went on to attempt or die by suicide.
Is this an American-specific concern? I would be surprised if this was the number one concern in Sweden, for example.