AI-powered medical review assistant: automated clinical consistency checking for enhanced safety monitoring in clinical trials - medical data review is a cornerstone of clinical trial safety…

The National Center for Advancing Translational Sciences awarded Octozi Inc. $306,013 on September 1, 2026, under the NCATS program (CFDA 93.350) to develop the Medical Review Assistant, an AI system for automated clinical consistency checking in clinical trial data.

Octozi will develop and validate AI algorithms that combine large language models with clinical heuristics to automate consistency checks across adverse events, efficacy assessments, and safety endpoints. In Phase I, the company will expand clinical checks from 11 to 35 expert-defined rules, targeting 90% macro recall and 75% macro precision with higher thresholds for safety-critical data. The system will also deliver transparent, interpretable AI outputs that reference trial data, documents, and external knowledge, achieving 85% agreement with reviewer expectations in user studies. Phase II work will extend the platform to support protocol-specific customization, broader data management checks, and integration with investigator's brochures. Preliminary studies with 10 reviewers demonstrated that AI-assisted medical review increased throughput 6.03-fold, reduced cleaning errors 6.44-fold, and lowered false positive queries 15.5-fold.

The period of performance runs through August 31, 2027, with place of performance in Brooklyn Heights, New York.

AI-powered medical review assistant: automated clinical consistency checking for enhanced safety monitoring in clinical trials - medical data review is a cornerstone of clinical trial safety monitoring and efficacy assessment, yet IT remains a manual, error-prone, and time-consuming process. As modern trials generate increasingly complex datasets across adverse events, labs, imaging, and efficacy assessments, reviewers face growing pressure to identify safety signals and ensure data integrity without the aid of purpose-built tools. These inefficiencies create costly delays, increase site burden, and risk late detection of safety concerns. Octozi is developing the medical review assistant (MRA), an AI system that combines large language models with clinical heuristics to automate consistency checks in trial data. Unlike existing visualization and outlier-detection tools, MRA is designed to capture nuanced clinical logic and provide interpretable outputs for medical reviewers. In preliminary studies with 10 reviewers, AI-assisted review increased throughput by 6.03X, reduced cleaning errors by 6.44X, and lowered false positive queries by 15.5X, demonstrating the potential for both efficiency gains and improved accuracy. In phase i, we will develop and validate AI algorithms for automated clinical consistency checking, expanding from 11 to 35 expert-defined checks across adverse events, efficacy, and safety endpoints, and achieving ≥90% macro recall and ≥75% macro precision, with higher thresholds for safety-critical data. We will also build transparent, interpretable AI for medical data review, providing reasoning paths that reference trial data, documents, and external knowledge, and achieving ≥85% agreement with reviewer expectations in user studies. Phase II will expand the platform to support protocol-specific customization, broader data management checks, and integration with investigator's brochures. With a $9B oncology trial data management market as the initial target, MRA addresses a major bottleneck in clinical research by reducing review timelines, improving early safety signal detection, and accelerating database lock. Long-term, this work will establish the first benchmark for AI-assisted medical data review, supporting NIH's mission to enhance patient safety and accelerate therapeutic development.