Clinical Scorecard: An Overview of Demographic Subgroup Reporting in Commercial Radiology AI: A Scoping Review

At a Glance

Key Highlights

• Algorithmic bias in AI poses risks to patient subgroups
• Limited evidence exists on subgroup bias in commercial medical AI, impacting patient safety
• Current regulations do not mandate peer-reviewed evidence for AI approval, raising concerns
• Best practices for estimating algorithmic bias remain unclear; specific strategies needed
• Regulatory groups are preparing solutions for identifying algorithmic bias, but challenges persist

Guideline-Based Recommendations

Diagnosis

• Identify studies validating commercially available AI-based products
• Assess reporting of age, sex, and race/ethnicity demographics with specific metrics

Management

• Provide practical recommendations to improve demographic reporting, including examples

Monitoring & Follow-up

• Monitor performance discrepancies between patient demographics and report findings

Risks

• Potential harmful biases in AI products due to lack of data and regulations; emphasize need for oversight

Patient & Prescribing Data

Patients in external cohorts differing demographically from testing cohorts

AI may use demographic subgroups as shortcuts instead of performing intended tasks; examples needed

Clinical Best Practices

• Follow PRISMA-ScR guidelines for scoping reviews
• Utilize publicly available information for data collection
• Ensure comprehensive reporting of demographic ratios in studies; provide specific examples

Related Resources & Content

• Source Article