How different ways of assigning responsibility affect doctors when using artificial intelligence to make diagnoses

Impact of responsibility allocation structures on diagnostic quality in AI-assisted diagnosis: an individually randomised four-arm parallel-group controlled experiment

Status

Active, not recruiting

Phases

Unknown

Study type

Interventional

Source

ISRCTN

Registry ID

ISRCTN16943519

Enrollment

105

Registered

2026-06-20

Start date

2025-12-01

Completion date

Unknown

Last updated

2026-06-29

For informational purposes only — not medical advice. Sourced from public registries and may not reflect the latest updates. Terms

Conditions

Diagnostic decision-making quality in AI-assisted diagnosis, with a focus on how different responsibility allocation structures affect doctors’ diagnostic accuracy and confidence calibration. Other

Interventions

Participants are practising hospital doctors randomly assigned to one of four parallel groups: dynamic responsibility, full responsibility, shared responsibility, or control. All participants complete

Eligibility

Sex/Gender

All

Age

18 Years to 65 Years

Inclusion criteria

Inclusion criteria: 1. Verified hospital doctors with current or recent clinical practice experience 2. Aged 18 years or above 3. Able to read and understand the study materials and provide informed consent 4. Able to complete the online experimental tasks independently 5. Willing to participate in an AI-assisted diagnostic decision-making study

Exclusion criteria

Exclusion criteria: 1. Individuals who are not licensed hospital doctors or practising physicians 2. Individuals younger than 18 years of age 3. Inability to read or understand the study materials or provide informed consent 4. Inability to complete the online experimental tasks independently 5. Previous participation in the same experiment 6. Failure to complete the assigned diagnostic vignette tasks or provision of unusable data

Design outcomes

Primary

Measure	Time frame
Diagnostic accuracy of final clinical diagnosis measured using Proportion of correct final diagnoses across AI-assisted clinical vignette diagnostic tasks, derived from comparison of each participant’s final diagnosis with a predefined reference diagnosis in the study vignette dataset at Immediately after completion of the AI-assisted clinical vignette diagnostic tasks during the study session;Confidence calibration in diagnostic decision making measured using Calibration between participant-reported confidence ratings following each clinical vignette and diagnostic correctness, calculated using confidence ratings collected via a study-specific post-vignette confidence question and compared with final diagnostic accuracy across all tasks at Confidence ratings collected immediately after each final diagnostic decision; calibration calculated after completion of all AI-assisted clinical vignette diagnostic tasks during the study session	—

Measure

Time frame

Diagnostic accuracy of final clinical diagnosis measured using Proportion of correct final diagnoses across AI-assisted clinical vignette diagnostic tasks, derived from comparison of each participant’s final diagnosis with a predefined reference diagnosis in the study vignette dataset at Immediately after completion of the AI-assisted clinical vignette diagnostic tasks during the study session;Confidence calibration in diagnostic decision making measured using Calibration between participant-reported confidence ratings following each clinical vignette and diagnostic correctness, calculated using confidence ratings collected via a study-specific post-vignette confidence question and compared with final diagnostic accuracy across all tasks at Confidence ratings collected immediately after each final diagnostic decision; calibration calculated after completion of all AI-assisted clinical vignette diagnostic tasks during the study session

—

Countries

China

Contacts

Public ContactTianya Liu

liuty73@mail2.sysu.edu.cn+86 15702996366

Outcome results

None listed

Source: ISRCTN (via WHO ICTRP) · Data processed: Jul 3, 2026