The Impact of AI-Assisted Diagnosis and Treatment on Physicians' Potential Legal Liability: An Expert Perspective
Trial statusNot yet recruiting
Trial phaseNot applicable
Trial typeInterventional
Biological sexAll
Age18+
SponsorXie Xiaoyun
About this trial
The goal of this randomized vignette-based experiment is to investigate how AI-related contextual factors influence expert assessment of physician negligence in AI-assisted diagnosis and treatment scenarios among experts in medical malpractice assessment.
The main questions it aims to answer are:
Does physician acceptance or rejection of AI recommendations, compared with non-AI use, influence the likelihood of being assessed as negligent?
Researchers will compare a no-AI assistance control group with four AI-assistance groups under different AI use conditions to examine how AI involvement and contextual factors affect assessments of physician negligence.
Participants will:
Complete a web-based questionnaire after eligibility confirmation. Evaluate a series of clinical vignettes describing adverse patient outcomes following physician decision-making under different AI-use conditions.
Eligibility criteria
This trial accepts healthy volunteersQualifiers
Physicians qualified to participate in medical negligence assessments, generally holding the title of associate chief physician or above and having relevant clinical experience; or forensic medical experts affiliated with an officially recognized forensic assessment institution and qualified to conduct medical negligence assessments independently.
Able to read and understand the clinical vignettes and complete the online survey independently.
Disqualifiers
Physicians who are not qualified to participate in medical negligence assessments.
Forensic medical experts who are not qualified to conduct medical negligence assessments independently or are not affiliated with an officially recognized forensic assessment institution.
Unable to understand or complete the vignette-based online survey.
Trial design
Design model
Parallel
Treatments tested in this trial
General AI With a Standard-of-Care Recommendation
Other interventionParticipants will evaluate four clinical vignettes involving a general AI system whose historical performance is comparable to that of a township hospital attending physician. The AI provides a recommendation consistent with the standard of care. Physician capability and acceptance or rejection of the AI recommendation vary across the vignettes.
General AI With a Non-Standard-of-Care Recommendation
Other interventionParticipants will evaluate four clinical vignettes involving a general AI system whose historical performance is comparable to that of a township hospital attending physician. The AI provides a recommendation inconsistent with the standard of care. Physician capability and acceptance or rejection of the AI recommendation vary across the vignettes.
Expert AI With a Standard-of-Care Recommendation
Other interventionParticipants will evaluate four clinical vignettes involving an expert AI system whose historical performance is comparable to that of a tertiary hospital chief physician. The AI provides a recommendation consistent with the standard of care. Physician capability and acceptance or rejection of the AI recommendation vary across the vignettes.
Expert AI With a Non-Standard-of-Care Recommendation
Other interventionParticipants will evaluate four clinical vignettes involving an expert AI system whose historical performance is comparable to that of a tertiary hospital chief physician. The AI provides a recommendation inconsistent with the standard of care. Physician capability and acceptance or rejection of the AI recommendation vary across the vignettes.
Treatment groups
440 Participants
are divided into 5 treatment groupsGroup A: No AI Assistance (Control Group)No intervention 0 interventions
Group B: Intervention Group 1Active comparator 1 intervention
Group C: Intervention Group 2Active comparator 1 intervention
Group D: Intervention Group 3Active comparator 1 intervention
Group E: Intervention Group 4Active comparator 1 intervention
Trial outcomes
Primary outcomes
1
Physician Negligence Assessment
Participants will assess whether the physician fulfilled the duty of care in each clinical vignette. Responses are recorded as "Yes" (no negligence) or "No" (negligence).
Time frame
Immediately after each clinical vignette during the single online survey session.Secondary outcomes
1
Reasonableness of the Physician's Decision
Participants will rate the extent to which a reasonable physician would make the same decision under similar circumstances. Scores range from 1 (definitely would not) to 7 (definitely would), with higher scores indicating greater perceived reasonableness.
Time frame
Immediately after each clinical vignette during the single online survey session.2
Magnitude of Physician Fault
Participants will rate the magnitude of the physician's fault using a visual analogue scale ranging from 0 (completely faultless) to 100 (completely at fault). Higher scores indicate greater perceived fault.
Time frame
Immediately after each clinical vignette during the single online survey session.3
Allocation of Responsibility
Participants will allocate responsibility for the adverse outcome among the physician, healthcare institution, AI developer and others(e.g. patient, disease factors). The percentages assigned to the three parties must total 100%.
Time frame
Immediately after each clinical vignette during the single online survey session.4
Confidence in the Assessment
Participants will rate their confidence in their assessments on a 7-point Likert scale ranging from 1 (very unconfident) to 7 (very confident). Higher scores indicate greater confidence.
Time frame
Immediately after each clinical vignette during the single online survey session.Locations
Sponsors and contacts
Click on the lead sponsor to view all of their trials.
Xie Xiaoyun
Lead sponsor
Huazhong University of Science and Technology
Sponsor institution
Source ClinicalTrials.gov
This trial is not recruiting at the moment. You can still explore other options: