Pathology News Roundup: January 2026

Bias in Pathology AI. A new study from Harvard Medical School, published in Cell Reports Medicine, found that artificial intelligence systems used to diagnose cancer from pathology slides do not perform equally across all patient populations.

Researchers discovered that diagnostic accuracy varied by race, gender, and age, with meaningful performance gaps appearing in nearly 30 percent of the cancer diagnostic tasks analyzed. Notably, the AI models were able to infer demographic information directly from tissue images—something human pathologists cannot do—introducing unexpected sources of bias into what is typically considered an objective diagnostic process.

The research team identified three main drivers behind these disparities. First, imbalanced training datasets can leave certain demographic groups underrepresented, reducing model accuracy for those populations. Second, differences in disease prevalence across demographic groups can cause models to become more optimized for some populations than others. Third, the AI systems were found to detect subtle molecular and biological patterns linked to specific demographics, using these signals as shortcuts for diagnosis. Over time, this can cause models to rely on demographic-associated features rather than disease-specific characteristics, further widening performance gaps.

To address the issue, the researchers introduced a new framework called FAIR-Path, designed to refocus AI models on disease-relevant features while minimizing reliance on demographic signals. When applied to multiple pathology AI systems, FAIR-Path reduced diagnostic disparities by approximately 88 percent. The findings highlight the importance of routine bias testing, local validation, and ongoing performance monitoring in medical AI, reinforcing that fairness, accuracy, and clinical reliability must be evaluated continuously to ensure AI tools support equitable cancer care for all patients.

CAP Updates Lab Accreditation Checklists. The College of American Pathologists (CAP) has released the 2025 edition of its Laboratory Accreditation Program Checklists, reinforcing its focus on patient safety, quality, and regulatory clarity. The updated checklists reflect continued advances in laboratory medicine, with new provisions designed to support emerging technologies and evolving care models. CAP leaders emphasized that adherence to these standards helps ensure accurate testing, reduce errors, and maintain trust in laboratory results that guide diagnosis and treatment.

Key updates in the 2025 edition include expanded guidance for digital pathology and remote data assessment, addressing image quality, data integrity, and new clinical use cases across laboratory disciplines. The checklists also introduce new requirements for patient specimen self-collection, including written instructions to support specimen quality and reliability. In addition, regulatory updates aligned with recent CMS clarifications simplify compliance and modernize personnel requirements—helping laboratories confidently adopt new workflows while maintaining rigorous accreditation and CLIA standards.

Recommended Reading: Validating AI. As AI adoption accelerates in pathology and laboratory medicine, the College of American Pathologists' Artificial Intelligence Committee is offering guidance to help labs navigate the full AI lifecycle, from validation and deployment to ongoing monitoring.

This recent CAP TODAY article offers practical, real-world insights on why rigorous local validation is essential (even for FDA-cleared tools), how to address bias and population drift, and what successful clinical implementation looks like in practice. It's a valuable read for any lab evaluating or expanding AI to ensure safe, effective, and sustainable use.

The article also highlights how thoughtful clinical validation and workflow integration can translate AI from theory into measurable operational impact. Case studies show how pathologists are using AI as a decision-support and triage tool to prioritize high-risk cases, reduce turnaround times, and improve consistency, while maintaining full physician oversight. The piece further emphasizes the importance of ongoing performance monitoring, human factors, and workflow alignment to prevent model drift and ensure AI delivers lasting clinical and operational value.