Related Experiment Video
Updated: May 2, 2026
07:31
Implementation of a Real-Time Psychosis Risk Detection and Alerting System Based on Electronic Health Records using CogStack
Published on: May 15, 2020
8.3K
Evaluating model generalizability for suicide attempt risk prediction: traditional machine vs deep learning.
Nicholas Josselyn1,2, Sahil Sawant1,2,3, Rachel E Davis-Martin2
1Data Science, Worcester Polytechnic Institute, Worcester, MA, USA.
Npj Mental Health Research
|April 30, 2026
Summary
Existing AI suicide attempt risk prediction models need improvement. Machine learning outperformed deep learning, but neither generalized well, highlighting the need for robust models in diverse healthcare settings.
Area of Science:
- Public Health
- Artificial Intelligence in Healthcare
- Clinical Informatics
Background:
- Suicide is a major public health issue in the US.
- Most suicide decedents have recent healthcare visits, offering intervention opportunities.
- Existing AI models for suicide attempt risk prediction (SARP) lack external validation and comparison between machine learning (ML) and deep learning (DL).
Purpose of the Study:
- To externally validate a state-of-the-art SARP model.
- To compare the performance of ML versus DL for tabular SARP.
- To assess the generalizability of SARP models across different healthcare settings.
Main Methods:
- External validation of a Mental Health Research Network SARP model using over 750,000 patient encounters from UMass Memorial Health.
- Comparative analysis of ML and DL algorithms for SARP.
- Evaluation of cross-setting healthcare generalizability.
Main Results:
- The existing SARP model did not generalize well to the new dataset.
- Machine learning (ML) significantly outperformed deep learning (DL) on most performance metrics.
- Deep learning (DL) achieved higher sensitivity in predicting suicide attempts.
Conclusions:
- There is a critical need for developing more robust and generalizable SARP models.
- Current AI models require improvement to effectively identify at-risk individuals across diverse healthcare contexts.
- Further research is needed to optimize ML and DL approaches for suicide prevention in healthcare.