Human-AI Collaboration for Decision Making

Overseen By Jeremy M Wolfe, PhD

Age: 18+

Sex: Any

Trial Phase: Academic

Sponsor: Brigham and Women's Hospital

Disqualifiers: Color blindness, Vision worse than 20/25

No Placebo Group All trial participants will receive the active study treatment (no placebo)

Compensation: Varies

Number of Visits: Unknown

What You Need to Know Before You Apply

What is the purpose of this trial?

This trial examines how people make decisions with AI assistance, particularly in tasks like determining the need for additional medical tests based on images. The main focus is to assess how varying the number of cases affects decision-making when AI aids in image review. Participants will encounter different scenarios to observe their interactions with AI suggestions, including the use of a Simulated Second Reader AI. This trial suits individuals who can pass a basic color vision test and have at least 20/25 vision with glasses or contacts. As an unphased trial, it offers participants the chance to contribute to groundbreaking research on AI's role in medical decision-making.

Do I need to stop my current medications for the trial?

The trial information does not specify whether you need to stop taking your current medications.

What prior data suggests that this method is safe for human decision-making?

Research has shown that using AI as a second reader in mammograms is safe and can enhance the process. Studies have found that AI detects breast cancer as effectively as human doctors. This capability allows AI to reduce doctors' workload without compromising safety. No major safety issues have emerged in these studies, indicating that AI is well-accepted in this role.

Regarding "target prevalence," specific safety data is not available, but the current trial phase "Not Applicable" typically indicates a focus on understanding functionality rather than testing safety. This suggests no major safety concerns related to this aspect of the trial.

Overall, evidence suggests that AI is a safe tool in medical screenings, with no significant risks identified in the reviewed trials. ^{1 2 3 4 5}

Why are researchers excited about this trial?

Researchers are excited about the Human-AI Collaboration for Decision Making because it explores how artificial intelligence can enhance medical decision-making processes. Unlike traditional methods that rely solely on human expertise, this approach incorporates a Simulated Second Reader AI, which acts as an additional reviewer to help interpret complex medical data. This collaboration aims to improve diagnostic accuracy and efficiency, potentially reducing misdiagnoses and optimizing patient outcomes. Additionally, by testing various conditions through Target Prevalence, the trial seeks to fine-tune AI models to better reflect real-world scenarios, making AI tools more reliable and applicable in medical settings.

What evidence suggests that this AI method is effective for decision making in mammography?

Research shows that using AI as a second reader in medical imaging, such as mammograms, improves diagnostic accuracy. AI increases the ability to correctly identify true positive results by 7.6%, helping to find more actual cases of concern. It also significantly reduces radiologists' workload, with up to 71% fewer cases needing human review.

In this trial, participants will experience both AI as a simulated second reader and adjustments in target prevalence. Studies indicate that target frequency affects decision-making. In visual searches, infrequent target appearances can lead to more missed detections. Adjusting target frequency can reduce these errors, improving overall accuracy. These findings suggest that using AI and adjusting target frequency can enhance the effectiveness and reliability of medical screenings. ^{1 2 4 6 7}

Who Is on the Research Team?

Jeremy M Wolfe, PhD

Principal Investigator

Brigham and Women's Hospital

Are You a Good Fit for This Trial?

This trial is open to everyone who can enroll online. However, participants must pass the Ishihara color vision test and have at least 20/25 vision with correction if needed.

Inclusion Criteria

- All welcome to enroll on line

Exclusion Criteria

My vision is at least 20/25 with glasses or contacts.

Must pass the Ishihara color vision screening test

Timeline for a Trial Participant

Screening

Participants are screened for eligibility to participate in the trial

1-2 weeks

Data Collection

Participants complete 8 blocks of tasks with varying AI input and prevalence rates

Up to one week

Multiple sessions

Follow-up

Participants are monitored for performance and subjective feedback after task completion

1 week

What Are the Treatments Tested in This Trial?

Interventions

• Simulated Second Reader AI
• Target Prevalence

Trial Overview The study tests how simulated AI as a Second Reader affects human decision-making in tasks like mammography screening when the frequency of cases (prevalence) changes.

How Is the Trial Designed?

Group I: Experiment Experimental Treatment 2 Interventions

All participants are tested in all conditions of this experiment.

Who Is Running the Clinical Trial?

Brigham and Women's Hospital

Lead Sponsor

Trials

1,694

Recruited

14,790,000+

Citations

1. pmc.ncbi.nlm.nih.gov pmc.ncbi.nlm.nih.gov/articles/PMC11605140/

AI as a Second Reader Can Reduce Radiologists ... AI can reduce radiologist workload by acting as a second reader or triage tool, while maintaining or improving diagnostic accuracy in ...

2. thelancet.com thelancet.com/journals/landig/article/PIIS2589-7500(25)00064-0/fulltext

AI as an independent second reader in detection of ... This study simulated different screening scenarios, evaluating the performance of population-based breast cancer screening when using an AI ...

3. nature.com nature.com/articles/s41467-024-51725-8

Comparison of AI-integrated pathways with human-AI ... Our simulation studies investigate possibilities for the positive, neutral and negative influence of AI integration on human reader performance ...

4. sciencedirect.com sciencedirect.com/science/article/pii/S0720048X25001871

AI-supported approaches for mammography single and ... AI support significantly enhanced sensitivity across all reading approaches, particularly benefiting worse performing radiologists.

5. academic.oup.com academic.oup.com/radadv/article/1/2/umae011/7667296

Artificial intelligence as a second reader for screening ... AI as a second reader in mammography may reduce healthcare burden, improve sensitivity by 7.6%, and reduce caseload by 71% compared to routine double read.

6. rsna.org rsna.org/news/2025/march/ai-as-a-second-reader-in-mammography

AI as a Second Reader in Mammography A study out of Denmark evaluating the use of AI in mammography screening demonstrated AI's ability to significantly reduce the workload of radiologists.

7. journals.plos.org journals.plos.org/digitalhealth/article?id=10.1371/journal.pdig.0000726

Safety of human-AI cooperative decision-making within ... Here, we share the results of an observational study of human-AI interaction in a high-fidelity simulation suite focusing on the influence of safe and unsafe AI ...