JMIR Publications

ABSTRACT

Background:

he integration of artificial intelligence (AI) into clinical workflows through decision support systems has transformed diagnostic and treatment pathways by providing real-time recommendations . However, the rapid adoption of these technologies introduces significant challenges regarding accountability, particularly when AI-assisted decision results in patient harm. Clinical decision-making in the age of AI is no longer a linear process; it is a complex interaction involving algorithm developers, healthcare organizations, and clinicians. Consequently, responsibility can no longer be viewed as a simple causal chain assigned to a single actor but must be understood as a distributed responsibility chain spanning multiple phases and stakeholders

Objective:

To investigate how different physicians–AI responsibility allocation structures affect diagnostic accuracy and calibration during AI-assisted diagnosis.

Methods:

This individually randomised, four-arm, parallel-group controlled experiment was conducted in a simulated clinical environment on the Credamo platform. 96 physicians were randomly assigned to dynamic responsibility, full responsibility, equal responsibility, or a control group. The only between-group difference was the responsibility allocation structure. Primary outcomes were final diagnostic accuracy and confidence calibration; secondary outcomes included agreement rates and post-task subjective evaluations.

Results:

Responsibility allocation structures significantly modulated diagnostic quality. Compared to the control group, the full responsibility structure yield no significant improvement in accuracy or confidence calibration, while equal responsibility resulted in poorer confidence calibration Conversely, the dynamic responsibility structure demonstrated superior performance, significantly enhancing both diagnostic accuracy and confidence calibration

Conclusions:

The dynamic responsibility structure may enable healthcare organizations to use AI more fully and appropriately without compromising clinicians’ diagnostic performance, thereby improving the safety and quality of AI-assisted diagnosis. Clinical Trial: AsPredicted preregistration record. URL: https://aspredicted.org/b5ib3v.pdf [Accessed 2025-12-27]