Artificial Intelligence (AI)-Enhanced Pretreatment Peer-review Process to Improve Patient Safety in Radiation Oncology

Purpose

This prospective study will test artificial intelligence (AI) and machine learning (ML) decision support tools. This tool is designed to help doctors, physicists and other staff during pre-treatment peer review, a step where treatment plans are checked before a patient begins care. The system highlights summaries showing how different providers may vary in their treatment planning (provider-variability summaries) and points out the best signals or warning signs to look for (optimal cues). By drawing attention to these patterns and cues, the tool aims to help reviewers spot possible treatment-planning mistakes earlier, reduce the chance of errors, and improve overall patient safety.

Conditions

  • Cancer
  • Prostate Cancer

Eligibility

Eligible Ages
Over 18 Years
Eligible Sex
All
Accepts Healthy Volunteers
Yes

Inclusion Criteria

Providers only - ≥18 years - Peer-review attendees at participating clinics Patients only - ≥18 years - All patients with prostate cancer radiation therapy cases treated at participating sites (no intervention delivered to patients)

Exclusion Criteria

Providers only • Providers unwilling/unable to comply with study procedures; sites unable to implement the workflow or provide required outcomes. Patients and Providers • Has dementia, altered mental status, or any psychiatric or co-morbid condition prohibiting the understanding or rendering of informed consent

Study Design

Phase
N/A
Study Type
Interventional
Allocation
Non-Randomized
Intervention Model
Parallel Assignment
Primary Purpose
Health Services Research
Masking
None (Open Label)

Arm Groups

ArmDescriptionAssigned Intervention
Other
Providers 		Radiation oncology providers engaged in peer-review at participating clinics.
  • Device: The Artificial Intelligence (AI)/ Machine Learning (ML) contribution to treatment planning
    All treatment planning and clinical monitoring are conducted in accordance with institutional standards and established departmental policies. Peer review activities proceed as they would in routine clinical practice, with the addition of optional Artificial Intelligence (AI) generated analytics available for clinician review. AI / Machine Learning (ML) system is embedded in scheduled departmental peer review meetings and presents analytic summaries and visualizations through a dashboard that is integrated into the existing clinical workflow. The system functions solely as a decision support aid and does not perform or initiate any autonomous treatment planning actions, dose delivery changes, or clinical interventions. During simulation (SIM) review, physician generated target and organ at risk contours are reviewed first, consistent with standard practice. Only after this initial review may the treating physician optionally access the AI generated contours for comparative purposes.
    Other names:
    • Clinical decision support / workflow support
No Intervention
Patients 		Prostate cancer patients who receive radiation therapy contribute de-identified safety outcomes.

Recruiting Locations

University of North Carolina at Chapel Hill, Department of Radiation Oncology
Chapel Hill, North Carolina 27599
Contact:
Olivia Morton
984-974-8441
olivia_roberts@med.unc.edu

More Details

Status
Recruiting
Sponsor
UNC Lineberger Comprehensive Cancer Center

Study Contact

Olivia Morton
(984) 974-8441
olivia_roberts@med.unc.edu

Detailed Description

As radiation therapy (RT) becomes more complex, the number of possible error pathways increases. AI-supported peer review can help catch errors that might otherwise go unnoticed and promote consistent, equitable safety standards across both rural and urban clinics. Radiation therapy (RT) is used in about 50% of cancer patients and usually given in outpatient clinics. Newer technologies such as intensity-modulated radiation therapy (IMRT), Volumetric Modulated Arc Therapy (VMAT), and Image-guided radiation therapy (IGRT), improve treatment by better protecting normal tissue and higher dose in target areas. However, they are more complex and require very precise definition of tumor targets and normal tissues. Even small errors in outlining these areas can lead to under-treating the tumor or over-treating healthy tissue. Studies show that errors in defining target areas have increased in modern radiation oncology. Because these treatments are more cognitively demanding, the risk of planning errors has increased and, in some cases, errors can cause serious harm. Pre-treatment peer review is where a multidisciplinary team reviews the treatment plan before therapy begins is an important safety step and is strongly recommended. It is most effective when done before treatment starts, since making corrections later can cause treatment delays, rushed changes, and added The potential impact on patient safety is substantial. Because of the growing complexity and workload, there is a need to strengthen and partially automate pre-treatment peer review. AI/ML decision-support tools can help by summarizing key information, highlighting unusual plan features, and drawing attention to areas of potential risk. These tools do not make treatment decisions. Instead, they provide analytics and visual summaries to support clinicians and reduce cognitive burden. Because the tool also highlights differences in how providers plan treatments, it may help identify variation in care and bring attention to potential health disparities, supporting future efforts to improve equity in radiation oncology.