Abstract B017: Check: A hybrid continuous-learning framework for enhancing factual reliability in clinical language models
Bibliographic record
Abstract
Abstract Large Language Models (LLMs) possess significant potential across diverse clinical applications. However, their tendency to generate factual inaccuracies, termed hallucinations, presents a major obstacle to their dependable use in healthcare settings. To tackle this crucial issue, we introduce CHECK, a novel hybrid continuous-learning framework that synergistically combines meticulously curated, domain-specific databases of clinical knowledge with a comprehensive first of its type, database-free classifier for detecting hallucinated content. We evaluated CHECK on a comprehensive dataset of clinically relevant questions derived from 100 pivotal clinical trials, representing a critical domain within evidence-based medicine. Our results demonstrate a substantial improvement in factual accuracy, with LLMs achieving 97\% factuality and a minimal 0.3\% hallucination rate when utilizing our structured databases, compared to considerably lower accuracy - 38\% factuality, 31\% hallucinations - when relying solely on minimal context (trial titles). Furthermore, our first of its kind comprehensive ensemble-based classifier, that leverage information theory first principles, attained a strong Area Under the Curve (AUC) of 0.95 in distinguishing factual and hallucinated paragraphs in clinical trial questions. This clasifier generalized effectively to a synthetic UMLS disorder dataset (AUC = 0.96), indicating broader applicability across medical knowledge domains and generalization of our factuality phenotype. CHECK's continuous learning feedback loop, incorporating expert verification and classifier retraining, offers a pathway for ongoing enhancement of LLM trustworthiness and safety in various clinical contexts by proactively mitigating hallucinations to levels below other serious and accepted errors in the industry (e.g drug administration). Therefore, paving the way for a broad LLM clinical adoption. Citation Format: Carlos Garcia Fernandez, Luis Felipe, Gilmer Valdes. Check: A hybrid continuous-learning framework for enhancing factual reliability in clinical language models [abstract]. In: Proceedings of the AACR Special Conference in Cancer Research: Artificial Intelligence and Machine Learning; 2025 Jul 10-12; Montreal, QC, Canada. Philadelphia (PA): AACR; Clin Cancer Res 2025;31(13_Suppl):Abstract nr B017.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.007 | 0.019 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.002 | 0.001 |
| Science and technology studies | 0.000 | 0.001 |
| Scholarly communication | 0.002 | 0.002 |
| Open science | 0.003 | 0.002 |
| Research integrity | 0.002 | 0.002 |
| Insufficient payload (model declined to judge) | 0.003 | 0.001 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".