AI System Crushes Manual Lab Data Review in Peer-Reviewed Pharmacology Study, Unlocking 20% More Patient Records

Artificial intelligence has demonstrated measurable superiority over manual methods in a critical but unglamorous corner of clinical research: cleaning and standardizing laboratory data for real-world evidence studies. Cornerstone AI will present peer-reviewed findings at the 42nd International Society for Pharmacoepidemiology (ISPE) Annual Meeting showing its AI-assisted system achieved 99.9% accuracy while identifying 20% more clinically relevant records than traditional manual workflows—a validation that could accelerate drug development timelines and reshape how pharmaceutical companies approach observational research.

The study, conducted in collaboration with Gilead Sciences researchers, analyzed approximately 4 million laboratory records from 19,415 patients with metabolic dysfunction-associated steatohepatitis (MASH) drawn from HealthVerity electronic health record data. Manual workflows achieved 93.5% accuracy at the LOINC component level, while Cornerstone's hybrid system—combining rule-based logic with natural language processing and hierarchical classification—reached 99.9% frequency-weighted accuracy.

The Completeness Advantage

Beyond accuracy improvements, the AI system demonstrated a significant completeness advantage that directly impacts research quality. The platform identified 801,874 MASH-related laboratory records compared to 667,857 surfaced by manual methods—a 20% overall increase. Across 14 prespecified MASH-related tests, the AI system achieved a mean 38% increase in record identification per test.

This completeness gap has direct implications for statistical power and disease characterization in observational studies. Missing laboratory data can lead to incomplete cohort definitions, biased exposure assessments, and reduced ability to detect safety signals or treatment effects. The system also achieved 99.8% standardization of units and 96.3% standardization of laboratory tests to LOINC coding, the universal standard for clinical laboratory observations.

Third-party review adjudicated discrepancies between AI and manual mappings to validate the findings, lending methodological rigor to the comparison.

Implications for Real-World Evidence

The validation establishes AI-assisted data cleaning as a potentially regulatory-aligned method for real-world evidence generation, addressing a historically labor-intensive bottleneck in pharmacoepidemiology. Laboratory data standardization is foundational to observational research but notoriously time-consuming, often requiring months of expert review for large-scale studies.

Mary Beth Ritchey, PhD, FISPE, Principal at Med Tech Epi, LLC and Associate Research Professor at Rutgers Center for Pharmacoepidemiology and Treatment Science, authored the study titled "AI-assisted real-world lab data cleaning to improve quality and efficiency of real-world evidence generation." The research will be presented August 29–September 2, 2026, in Milan, Italy.

For pharmaceutical companies and contract research organizations, the findings suggest AI-driven data curation could accelerate study timelines while improving cohort accuracy. The technology's applicability extends beyond metabolic disease to any condition requiring laboratory-based phenotyping, potentially reshaping competitive dynamics in the real-world evidence space where manual and semi-automated workflows have dominated.

The peer-reviewed results set a new accuracy benchmark for laboratory data standardization and signal maturation of clinical AI beyond predictive modeling into foundational data quality infrastructure.

Sources

1. Cornerstone AI to Present Research Findings at 42nd ISPE Annual Meeting: AI-Assisted Laboratory Data Cleaning Achieves 99.9% Accuracy, Surfaces 20% More Records Than Manual Workflows — Yahoo Finance