Mitigating Automation Bias in Physician-LLM Diagnostic Reasoning Using Behavioral Nudges

The goal of this randomized controlled trial is to evaluate whether behavioral nudges can reduce automation bias, the uncritical acceptance of automated output, in physicians using large language models (LLM) like ChatGPT-5.1 for clinical decision-making.

The main question it aims to answer is: Does a dual-mechanism behavioral nudge intervention (baseline accuracy anchoring plus case-specific color-coded confidence signals) reduce physicians' uncritical acceptance of incorrect LLM recommendations?

Researchers will compare physicians who receive LLM recommendations along with a behavioral nudge to those who receive LLM recommendations without the nudge to assess if the nudge reduces automation bias.

Participants will:

• Evaluate six clinical vignettes accompanied by LLM-generated recommendations (half containing deliberate, clinically significant errors).
• Control group: Be able to view LLM recommendations in standard format without the nudge.
• Treatment group: Be able to view ChatGPT's diagnostic accuracy on standard medical datasets as an initial anchor, then receive color-coded confidence signals alongside each recommendation (e.g., red for low confidence).
• Have their responses evaluated by blinded reviewers using an expert-developed assessment rubric to detect uncritical acceptance of erroneous information.

Study Overview

• Status: Recruiting
• Conditions: Diagnosis
• Intervention / Treatment: Other: Behavioral Nudge Intervention

Detailed Description

Automation bias represents a critical challenge in modern clinical practice, particularly as artificial intelligence (AI) tools become increasingly embedded in healthcare workflows. This cognitive phenomenon describes the tendency of clinicians to favor suggestions from automated decision-making systems, even when those suggestions are incorrect. As Large Language Models (LLM) such as ChatGPT-5.1 gain traction in medical settings, their potential to reduce errors and improve efficiency must be weighed against a significant concern: these models lack rigorous medical validation and may amplify existing cognitive biases through incorrect or misleading recommendations.

The emergence of automation bias in medical contexts reflects a complex interplay of environmental and psychological factors. Time constraints in high-volume clinical settings create pressure to accept AI-generated recommendations without adequate scrutiny. Financial incentives that prioritize efficiency over thoroughness may further discourage critical evaluation necessary for sound clinical judgment. Cognitive fatigue during extended shifts diminishes physicians' capacity for sustained analytical thinking. These pressures interact with psychological mechanisms including diffusion of responsibility, overconfidence in technological solutions, and cognitive offloading, collectively creating conditions where uncritical acceptance of AI-generated recommendations becomes more likely.

This randomized controlled trial evaluates the effectiveness of a behavioral nudge intervention designed to mitigate automation bias among medical doctors utilizing LLM-generated diagnostic recommendations. The primary objective is to determine whether this intervention improves diagnostic reasoning performance scores when evaluating clinical vignettes that include deliberately flawed LLM recommendations. Secondary objectives include assessing whether physician experience level, gender, and prior LLM experience moderate the intervention's effectiveness, determining differential effectiveness for vignettes across different confidence signals.

This study employs a single-blind, randomized controlled trial with two parallel arms. Participants will be randomly assigned 1:1 to either the intervention or control arm. To eliminate variability from differences in prompting skills, participants will not interact directly with a live LLM interface. Instead, all participants will use a custom-built web platform displaying clinical vignettes with pre-generated LLM recommendations, ensuring identical LLM-generated content for each vignette.

All participants will evaluate six clinical vignettes during a single, proctored session lasting approximately 75 minutes. Three vignettes will contain deliberately introduced clinical reasoning flaws in the LLM recommendations, while three will contain correct recommendations. Vignettes will be presented in randomized order to prevent pattern detection.

Control arm participants will evaluate clinical vignettes with LLM diagnostic recommendations generated by ChatGPT presented in standard, neutral text format without additional contextual information. Intervention arm participants will evaluate the same vignettes alongside a behavioral nudge. This intervention consists of two synchronized cognitive cues: (1) an anchoring cue displaying ChatGPT's baseline diagnostic accuracy on standard medical datasets at the top of the interface panel, explicitly anchoring expectations to the model's fallibility, and (2) a selective attention cue displaying the LLM recommendation alongside a color-coded confidence signal generated through an ensemble assessment: three independent state-of-the-art LLMs (Claude Sonnet 4.5, Gemini 2.5 Pro Thinking, and GPT-5.1) each provide confidence ratings for the recommendation, and the mean confidence determines the signal color to mitigate single-model miscalibration.

The color-coded confidence signals are categorized into three distinct levels based on the ensemble's mean confidence relative to baseline diagnostic accuracy. Red signals are triggered when the mean confidence falls below ChatGPT's established baseline accuracy, explicitly flagging high-uncertainty cases that demand heightened critical scrutiny. Orange signals indicate that while the mean confidence exceeds the baseline average, it remains below 100%, signaling the need for continued clinical vigilance and the avoidance of complacency. Finally, green signals are reserved for instances of 100% ensemble consensus; however, even at this level of confidence, standard AI safety warnings remain present to guard against over-reliance on the system's output.

Participants will be presented with six clinical vignettes specifically designed to measure automation bias, sourced and modified from real cases representing a range of diagnostic difficulty and common medical specialties. Each vignette follows a standardized format including chief complaint, history of present illness, relevant past medical/social/family history, physical examination findings, and initial laboratory results.

The primary outcome is the Diagnostic Reasoning Performance Score, a composite percentage score based on a structured rubric evaluating: quality of differential diagnoses, supporting findings, opposing findings, final diagnosis accuracy, and appropriateness of next steps. Secondary outcomes include top-choice diagnosis accuracy (incorrect, partially correct, or correct). All responses will be evaluated by blinded reviewers using the assessment rubric.

• Study Type: Interventional
• Enrollment (Estimated): 50
• Phase: Not Applicable

Contacts and Locations

• Study Contact: Name: Ihsan Ayyub Qazi, PhD; Phone Number: 8368 +923233333766; Email: ihsan.qazi@lums.edu.pk
• Study Contact Backup: Name: Ayesha Ali, PhD; Phone Number: 8235 +923419494940; Email: ayeshaali@lums.edu.pk

Study Locations

• Pakistan
  • Punjab Province
    • Lahore, Punjab Province, Pakistan, 54792
      • Recruiting
      • Lahore University of Management Sciences
      • Principal Investigator: Ihsan Ayyub Qazi, PhD
      • Contact: Ayesha Ali, PhD; Phone Number: 8235 +923419494940; Email: ayeshaali@lums.edu.pk
      • Contact: Ihsan Ayyub Qazi, PhD; Phone Number: +923233333766; Email: ihsan.qazi@lums.edu.pk

Participation Criteria

Eligibility Criteria

• Ages Eligible for Study: Child; Adult; Older Adult
• Accepts Healthy Volunteers: Yes

Description

Inclusion Criteria:

• Full or Provisionally Registered Medical Practitioners with the Pakistan Medical and Dental Council (PMDC).
• Completed Bachelor of Medicine, Bachelor of Surgery (MBBS) Exam. The equivalent degree of MBBS in US and Canada is the Doctor of Medicine (MD).
• Participants must have completed a structured training program on the use of ChatGPT (or a comparable large language model), totaling at least 10 hours of instruction. The program must include hands-on practice related to LLM's key aspects, specifically prompt engineering and content evaluation.

Exclusion Criteria:

• Any other Registered Medical Practitioners (Full or Provisional) with PMDC (e.g., professionals with Bachelor of Dental Surgery or BDS).

Study Plan

How is the study designed?

Design Details

• Primary Purpose: Diagnostic
• Allocation: Randomized
• Interventional Model: Parallel Assignment
• Masking: Single

Number of Arms

2

Arms and Interventions

Participant Group / Arm	Intervention / Treatment
Active Comparator: ChatGPT Recommendations alongside a Behavioral Nudge; Participants will evaluate six clinical vignettes. During the trial, they will have access to clinical recommendations from a specific, commercially available LLM (ChatGPT) in addition to conventional diagnostic resources. LLM recommendations for three vignettes will contain deliberately flawed diagnostic information and for three vignettes it will contain accurate recommendations). The cases will be presented in random order. Participants in this arm will receive a behavioral nudge embedded in the LLM recommendations interface that presents two synchronized cognitive cues when the LLM panel is expanded: (1) an anchoring cue displaying ChatGPT's baseline diagnostic accuracy on standard medical datasets at the top of the panel to set realistic expectations before cue intervention located immediately below, which shows the LLM recommendations alongside a case-specific color-coded confidence signal.	Other: Behavioral Nudge Intervention; Participants in the treatment group will receive a behavioral nudge intervention embedded in the LLM recommendations interface that presents two synchronized cognitive cues when the LLM panel is expanded: (1) an anchoring cue displaying ChatGPT's baseline diagnostic accuracy on standard medical datasets at the top of the panel to set realistic expectations before viewing the specific recommendation, and (2) a selective attention cue located immediately below, which shows the LLM recommendation alongside a case-specific and color-coded confidence signal. This signal is categorized as red when the mean ensemble confidence falls below the established baseline accuracy, flagging high-uncertainty cases that demand critical evaluation; orange when confidence meets or exceeds the baseline but remains below 100%, intended to prevent complacency and maintain active clinical scrutiny; and green for a 100% ensemble consensus, though standard cautionary warnings still apply to guard against.
No Intervention: ChatGPT Recommendations without a Behavioral Nudge; Participants will evaluate six clinical vignettes. During the trial, they will have access to clinical recommendations from a specific, commercially available LLM (ChatGPT) in addition to conventional diagnostic resources. LLM recommendations for three vignettes will contain deliberately flawed diagnostic information. The cases will be presented in random order. Participants in this arm will not receive any behavioral nudge.

What is the study measuring?

Primary Outcome Measures

Outcome Measure	Measure Description	Time Frame
Diagnostic reasoning accuracy score	The primary outcome will be the percent correct for each case, ranging from 0 to 100%, where higher scores indicate better diagnostic performance. For each case, participants will be asked for their three leading diagnoses, findings that support each diagnosis, and findings that oppose each diagnosis. For each plausible diagnosis, participants will receive 1 point. Findings supporting the diagnosis and findings opposing the diagnosis will also be graded based on correctness, with 1 point for each correct response. Participants will then be asked to name their top diagnosis they believe is most likely, earning 9 points for a reasonable response and 18 points for the most accurate response. Finally participants will be asked to name up to 3 next steps to further evaluate the patient with 0.5 point awarded for a partially correct response and 1 point for a completely correct response. The primary outcome will be compared at the case-level between the randomized groups.	Assessed at a single time point for each case, during the scheduled diagnostic reasoning evaluation session, which takes place between 0-5 days after participant enrollment.

Secondary Outcome Measures

Outcome Measure	Measure Description	Time Frame
Top choice diagnosis accuracy score	The secondary outcome will measure participants' performance in identifying the most likely diagnosis for each clinical vignette. After evaluating each case, participants will select their single most likely diagnosis, which will be scored on a pre-specified Three-Tier Diagnostic Accuracy Scale: 18 points for the most accurate diagnosis, 9 points for a clinically reasonable alternative, and 0 points for an incorrect diagnosis. For each participant, a Top Choice Diagnosis Accuracy Score is calculated as (total points earned ÷ maximum possible points) × 100, yielding a 0-100 % range in which higher scores indicate greater diagnostic accuracy. This percentage score will be compared at the case-level between randomized groups to quantify the impact of automation bias on diagnostic decision-making.	Assessed at a single time point for each case, during the scheduled diagnostic reasoning evaluation session, which takes place between 0-5 days after participant enrollment.

Collaborators and Investigators

• Sponsor: Lahore University of Management Sciences
• Investigators: Principal Investigator: Muhammad Asadullah Khawaja, MBBS, King Edward Medical University; Principal Investigator: Ihsan Ayyub Qazi, PhD, Lahore University of Management Sciences (LUMS); Principal Investigator: Ali Zafar Sheikh, MBBS, Lahore General Hospital; Principal Investigator: Muhammad Junaid Akhtar, MBBS, Children's Hospital, Lahore; Principal Investigator: Muhammad Hamad Alizai, PhD, Lahore University of Management Sciences (LUMS)

Study record dates

Study Major Dates

• Study Start (Actual): January 17, 2026
• Primary Completion (Estimated): July 1, 2026
• Study Completion (Estimated): August 1, 2026

Study Registration Dates

• First Submitted: December 26, 2025
• First Submitted That Met QC Criteria: December 26, 2025
• First Posted (Actual): January 9, 2026

Study Record Updates

• Last Update Posted (Actual): March 31, 2026
• Last Update Submitted That Met QC Criteria: March 27, 2026
• Last Verified: March 1, 2026

More Information

Terms related to this study

• Keywords: clinical reasoning; large language models; computer-assisted diagnosis; automation bias; behavioral nudges
• Additional Relevant MeSH Terms: Pathologic Processes; Pathological Conditions, Signs and Symptoms; Disease
• Other Study ID Numbers: LUMS-IRB-0412/12192025/IAQ-FWA

Plan for Individual participant data (IPD)

• Plan to Share Individual Participant Data (IPD)?: NO

Drug and device information, study documents

• Studies a U.S. FDA-regulated drug product: No
• Studies a U.S. FDA-regulated device product: No