Related Experiment Video
Updated: Aug 6, 2026
Virtual Agent for Real-Time Motivational Interviewing by Integrating Adaptive Nonverbal Behavior and Language Models
Published on: December 23, 2025
Bridging algorithmic prediction and clinical agency: an exploratory pilot study of AI-augmented physician
Akiva Kleinerman1, David Benrimoh2,3, Amit Yaniv-Rosenfeld4,5,6,7
1Department of Information Science and Applied AI, Bar-Ilan University, Ramat Gan, Israel.
Introduction:
Effective psychiatric decision-making requires balancing data-driven predictions with clinical agency. This challenge is particularly acute in the pharmacological management of Major Depressive Disorder (MDD), where clinicians must navigate complex, patient-specific trade-offs between remission probabilities and diverse side-effect risks. Although AI-driven Clinical Decision Support Systems (AI-CDSS) can support the prediction of individual treatment outcomes, the optimal mechanism for aggregating multiple clinical criteria remains an open research challenge.
Methods:
We investigated how the locus of control in the aggregation mechanism affects clinical utility and treatment decisions. Three weighting schemes were evaluated: (1) an Implicit Weighting baseline, in which raw probabilities were presented; (2) a Static Expert-Derived Weighting scheme, using linear aggregation with fixed expert-based weights; and (3) a Dynamic Clinician-Determined Weighting scheme, using linear aggregation with adjustable clinician-defined weights. These schemes were implemented within a prototype decision support system for antidepressant selection and evaluated in a user study with 22 physicians.
Results:
The Dynamic Clinician-Determined Weighting scheme significantly enhanced perceived clinical utility compared with the alternative approaches (p < 0.01). It also led to the most frequent data-informed revision of physicians' initial unassisted antidepressant choices, occurring in 33.3% of cases. This effect was observed among both psychiatrists and primary care physicians, suggesting that adjustable weighting can support more informed treatment decisions across clinical specialties.
Discussion:
These findings suggest that effective integration of AI into psychiatric practice requires flexible decision support systems that preserve clinical agency while incorporating data-driven predictions. By allowing clinicians to determine the relative importance of remission probabilities and side-effect risks, dynamic weighting may better reflect the nuanced and individualized nature of mental health care.
