Clinical Scorecard: Equity and Applicability of AI Implementation for Predicting Lung Cancer Biomarkers

At a Glance

Key Highlights

• AI models like EAGLE show performance variability based on ancestry and specimen context.
• Significant performance degradation observed in Asian patients (AUC, 0.68) and pleural samples (AUC, 0.66).
• Need for ancestry-aware and tissue-context aware validation frameworks for AI models.
• Proposed solutions include diverse benchmarking datasets and continuous monitoring post-deployment.
• Emphasis on equitable care to prevent exacerbation of healthcare disparities.

Guideline-Based Recommendations

Diagnosis

• Implement AI models with consideration of ancestry and specimen type.

Management

• Ensure subgroup-specific validation before clinical deployment of AI tools.

Monitoring & Follow-up

• Continuous monitoring and calibration of AI models post-deployment.

Risks

• Potential exacerbation of healthcare disparities if AI tools are not validated across diverse populations.

Patient & Prescribing Data

Patients with lung adenocarcinoma, particularly those of Asian descent.

AI tools should complement molecular testing while addressing variability in performance.

Clinical Best Practices

• Mandate performance reporting across major ancestral groups in validation studies.
• Develop and share large, diverse, multiancestry benchmarking datasets.

Related Resources & Content

• Rakaee et al study