Related Experiment Video
Updated: Jan 17, 2026
Implementation of a Real-Time Psychosis Risk Detection and Alerting System Based on Electronic Health Records using CogStack
Published on: May 15, 2020
Developing a Suicide Risk Prediction Algorithm Using Electronic Health Record Data in Mental Health Care: Real-World
Linda Hummel1,2, Karin C A G Lorenz-Artz1,2, Joyce J P A Bierbooms1
1Tranzo Scientific Center for Care and Wellbeing, Tilburg School of Social and Behavioral Sciences, Tilburg University, Prof. Cobbenhagenlaan 125, Reitse Poort, Room RP 204, Tilburg, 5037 DB, The Netherlands, 31683662495.
Developing artificial intelligence (AI) for mental health care faces challenges in data and implementation. Addressing these requires integrating clinical and technical views for effective, data-driven mental health services.
Area of Science:
- Mental Health Technology
- Artificial Intelligence in Healthcare
- Clinical Informatics
Background:
- Artificial intelligence (AI) presents solutions for strained mental health systems, yet few AI tools are implemented in clinical practice.
- The algorithm development phase is crucial for bridging innovation and practical application, influencing future implementation success.
- AI development must consider how design choices impact clinical adoption and usability.
Purpose of the Study:
- To examine the development process of a suicide risk prediction algorithm using electronic health record (EHR) data.
- To identify challenges encountered during algorithm development and the strategies used to address them.
- To derive key considerations for integrating technical and clinical perspectives in AI for data-driven mental health care.
Main Methods:
- An exploratory, multimethod qualitative case study.
- Data collection via desk research, team meetings, and feedback sessions.
- Thematic analysis to identify development challenges and responses, informing future algorithm development.
Main Results:
- Challenges included defining suicide incidents in EHRs due to data quality issues and operationalizing psychosocial variables.
- Natural language processing of unstructured data enabled sentiment analysis, but model complexity impacted explainability.
- Bias risk arose from unequally distributed questionnaire data, necessitating careful input selection.
Conclusions:
- Advancing data-driven mental health care requires robust data governance, quality control, and cultural shifts in documentation.
- Mitigating bias, balancing predictive accuracy with explainability, and maintaining a clinician-in-the-loop approach are critical.
- Future research should focus on sociotechnical aspects of AI development, implementation, and use in mental health care.
