{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,9,22]],"date-time":"2026-09-22T13:12:49Z","timestamp":1790082769571,"version":"4.0.1"},"reference-count":24,"publisher":"JMIR Publications Inc.","content-domain":{"domain":[],"crossmark-restriction":false},"short-container-title":["J Med Internet Res"],"abstract":"Abstract<\/jats:title>\n \n Background<\/jats:title>\n Large language models (LLMs) exhibit extensive medical knowledge but are prone to hallucinations and show low fact-level explainability, limiting clinical adoption and regulatory compliance. Existing approaches, such as retrieval-augmented generation, partially address these issues by grounding answers in source documents; however, the aforementioned problems persist.<\/jats:p>\n <\/jats:sec>\n \n Objective<\/jats:title>\n We propose the application of an atomic fact-checking framework designed to enhance the reliability and explainability of LLMs in medical long-form question answering. By decomposing generated answers into discrete atomic facts and verifying each against an authoritative knowledge base of medical guidelines, this approach enables precise identification and correction of incorrect statements, alongside explicit linkage to supporting literature.<\/jats:p>\n <\/jats:sec>\n \n Methods<\/jats:title>\n The fact-checking algorithm operates within a retrieval-augmented generation framework: LLM-generated answers are decomposed into atomic facts (smallest and self-contained information units), each of which is assessed and corrected if FALSE. To determine an optimal strategy, the validation\u2013question and answer (Q&A) set on prostate cancer treatment was tested under varying instructions. An extensive evaluation, including multireader assessments by human medical experts and the automated open Q&A benchmark AMEGA (Autonomous Medical Evaluation for Guideline Adherence), was conducted for the final pipeline. In addition to another radiooncologic test\u2013Q&A set, anonymized real-world tumor board cases and an independent, established neurology-Q&A set were used. Given their transparency and accessibility advantages, we compared various open-source models in pairs of generalist models and their medical fine-tuned counterparts, with regard to performance and improvements by fact-checking.<\/jats:p>\n <\/jats:sec>\n \n Results<\/jats:title>\n The framework significantly reduced hallucinations and inaccuracies. Medical expert assessment and automated benchmarks demonstrated significant improvements in factual accuracy, achieving up to a 50% overall answer improvement and an 80% hallucination detection rate. Notably, the observed gain was strongest in real tumor-board questions\u2014the most challenging dataset. Additionally, the framework achieved high explainability by tracing each atomic fact back to the most relevant chunks from the database, providing a granular, transparent explanation of the generated responses.<\/jats:p>\n <\/jats:sec>\n \n Conclusions<\/jats:title>\n To conclude, we present the application of an atomic fact-checking algorithm to medical Q&A. It identifies factual inaccuracies and hallucinations in LLM-generated answers, achieving the greatest gains on clinically realistic, complex questions. Correction via fact-checking improves the overall answer quality while achieving fact-wise explainability, paving the way for more credible clinical use of LLMs.<\/jats:p>\n <\/jats:sec>","DOI":"10.2196\/92090","type":"journal-article","created":{"date-parts":[[2026,9,22]],"date-time":"2026-09-22T12:36:31Z","timestamp":1790080591000},"page":"e92090-e92090","source":"Crossref","is-referenced-by-count":0,"title":["Improving Reliability and Explainability of Medical Question Answering Through Atomic Fact-Checking in Retrieval-Augmented Large Language Models: Creation and Validation Study"],"prefix":"10.2196","volume":"28","author":[{"given":"Juraj","family":"Vladika","sequence":"first","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0009-0009-8822-2620","authenticated-orcid":false,"given":"Annika","family":"Domres","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Mai","family":"Nguyen","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Rebecca","family":"Moser","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Jana","family":"Nano","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0001-9770-8555","authenticated-orcid":false,"given":"Felix","family":"Busch","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0001-5836-4542","authenticated-orcid":false,"given":"Lisa","family":"Adams","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0001-9249-8624","authenticated-orcid":false,"given":"Keno K","family":"Bressem","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Denise","family":"Bernhardt","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Stephanie E","family":"Combs","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Kai","family":"Borm","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0002-6667-5452","authenticated-orcid":false,"given":"Florian","family":"Matthes","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0003-2679-9853","authenticated-orcid":false,"given":"Jan C","family":"Peeken","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"1010","published-online":{"date-parts":[[2026,9,21]]},"reference":[{"issue":"15","key":"R1","doi-asserted-by":"publisher","first-page":"1320","DOI":"10.1001\/jama.2023.27861","article-title":"Comparative analysis of multimodal large language model performance on clinical vignette questions","volume":"331","author":"Han","journal-title":"JAMA"},{"issue":"1","key":"R2","doi-asserted-by":"publisher","DOI":"10.1038\/s41746-025-01536-y","article-title":"Evaluating base and retrieval augmented LLMs with document or online support for evidence based neurology","volume":"8","author":"Masanneck","journal-title":"NPJ Digit Med"},{"issue":"9","key":"R3","doi-asserted-by":"publisher","first-page":"e662","DOI":"10.1016\/S2589-7500(24)00124-9","article-title":"A future role for health applications of large language models depends on regulators enforcing safety standards","volume":"6","author":"Freyer","journal-title":"Lancet Digit Health"},{"issue":"1","key":"R4","doi-asserted-by":"publisher","DOI":"10.1056\/AIra2400380","article-title":"RAG in health care: a novel framework for improving communication and decision-making by addressing LLM limitations","volume":"2","author":"Ng","journal-title":"NEJM AI"},{"issue":"6","key":"R5","doi-asserted-by":"publisher","DOI":"10.1056\/AIcs2300235","article-title":"GPT-4 for information retrieval and comparison of medical oncology guidelines","volume":"1","author":"Ferber","journal-title":"NEJM AI"},{"key":"R6","doi-asserted-by":"publisher","DOI":"10.18653\/v1\/2020.coling-main.474","article-title":"Explainable automated fact-checking: a survey","author":"Kotonya","journal-title":"Proc 28th Int Conf Comput Linguist"},{"issue":"1","key":"R7","doi-asserted-by":"publisher","DOI":"10.1038\/s44172-025-00453-y","article-title":"Explainability in the age of large language models for healthcare","volume":"4","author":"Mesinovic","journal-title":"Commun Eng"},{"key":"R8","doi-asserted-by":"publisher","DOI":"10.1162\/tacl_a_00713","article-title":"When can LLMs actually correct their own mistakes? A critical survey of self-correction of LLMs","volume":"12","author":"Kamoi","journal-title":"Trans Assoc Comput Linguist"},{"issue":"8","key":"R9","doi-asserted-by":"publisher","first-page":"852","DOI":"10.1038\/s42256-024-00881-z","article-title":"Factuality challenges in the era of large language models and opportunities for fact-checking","volume":"6","author":"Augenstein","journal-title":"Nat Mach Intell"},{"key":"R10","doi-asserted-by":"publisher","DOI":"10.18653\/v1\/2023.emnlp-main.741","article-title":"FActScore: fine-grained atomic evaluation of factual precision in long form text generation","author":"Min","journal-title":"Proc 2023 Conf Empir Methods Nat Lang Process"},{"key":"R11","doi-asserted-by":"publisher","DOI":"10.18653\/v1\/2023.findings-acl.387","article-title":"Scientific fact-checking: a survey of resources and approaches","author":"Vladika","journal-title":"Findings Assoc Comput Linguist"},{"issue":"1","key":"R12","doi-asserted-by":"publisher","DOI":"10.1038\/s41746-024-01239-w","article-title":"Closing the gap between open source and commercial large language models for medical evidence summarization","volume":"7","author":"Zhang","journal-title":"NPJ Digit Med"},{"issue":"4","key":"R13","doi-asserted-by":"publisher","first-page":"474","DOI":"10.26421\/JDI3.4-5","article-title":"Improved methods to aid unsupervised evidence-based fact checking for online health news","volume":"3","author":"Deka","journal-title":"J Data Intell"},{"key":"R14","unstructured":"sebischair\/ImprovingReliabilityMedicalQA. GitHub. URL: https:\/\/github.com\/sebischair\/ImprovingReliabilityMedicalQA [Accessed 26-08-2026]"},{"issue":"1","key":"R15","doi-asserted-by":"publisher","DOI":"10.1038\/s41746-024-01356-6","article-title":"Autonomous medical evaluation for guideline adherence of large language models","volume":"7","author":"Fast","journal-title":"NPJ Digit Med"},{"key":"R16","doi-asserted-by":"publisher","DOI":"10.18653\/v1\/2023.emnlp-main.153","article-title":"G-eval: NLG evaluation using GPT-4 with better human alignment","author":"Liu","journal-title":"Proc Conf Empir Methods Nat Lang Process"},{"key":"R17","doi-asserted-by":"crossref","unstructured":"Madaan A Tandon N Gupta P et al. SELF-REFINE: iterative refinement with SELF-feedback. In: Oh A Naumann T Globerson A Saenko K Hardt M Levine S , editors. NIPS \u201923: Proceedings of the 37th International Conference on Neural Information Processing Systems. Curran Associates Inc; 2023:46534-46594. URL: https:\/\/dl.acm.org\/doi\/10.5555\/3666122.3668141 [Accessed 26-08-2026]","DOI":"10.52202\/075280-2019"},{"key":"R18","unstructured":"Bayerisches Krankenhausgesetz (BayKrG): art 27 Datenschutz [Article in German]. BAYERN.RECHT. Mar 28, 2007. URL: https:\/\/www.gesetze-bayern.de\/Content\/Document\/BayKrG-27 [Accessed 07-09-2026]"},{"key":"R19","unstructured":"Gesetz \u00fcber die Universit\u00e4tsklinika des Freistaates Bayern (Bayerisches Universit\u00e4tsklinikagesetz \u2013 BayUniKlinG) Art 16Anwendung hochschul- und krankenhausrechtlicher Vorschriften [Article in German]. BAYERN.RECHT. May 23, 2006. URL: https:\/\/www.gesetze-bayern.de\/Content\/Document\/BayUniKlinG-16 [Accessed 07-09-2026]"},{"issue":"9","key":"R20","doi-asserted-by":"publisher","first-page":"1","DOI":"10.1145\/3555803","article-title":"Trustworthy AI: from principles to practices","volume":"55","author":"Li","journal-title":"ACM Comput Surv"},{"key":"R21","doi-asserted-by":"publisher","DOI":"10.1145\/3644815.3644945","article-title":"Seven failure points when engineering a retrieval augmented generation system","author":"Barnett","journal-title":"CAIN 2024"},{"key":"R22","doi-asserted-by":"publisher","DOI":"10.18653\/v1\/2025.acl-long.1062","article-title":"FaithfulRAG: fact-level conflict modeling for context-faithful retrieval-augmented generation","author":"Zhang","journal-title":"Proc Annu Meet Assoc Comput Linguist"},{"issue":"2","key":"R23","doi-asserted-by":"publisher","first-page":"70","DOI":"10.1007\/s10462-025-11454-w","article-title":"Hallucination to truth: a review of fact-checking and factuality evaluation in large language models","volume":"59","author":"Rahman","journal-title":"Artif Intell Rev"},{"key":"R24","doi-asserted-by":"crossref","unstructured":"Adams LC Marx L Orberg ET et al. Atomic fact-checking increases clinician trust in large language model recommendations for oncology decision support: a randomized controlled trial. arXiv. Preprint posted online on May 5, 2026. [doi: 10.48550\/ARXIV.2605.03916]","DOI":"10.21203\/rs.3.rs-9227402\/v1"}],"container-title":["Journal of Medical Internet Research"],"original-title":[],"language":"en","deposited":{"date-parts":[[2026,9,22]],"date-time":"2026-09-22T12:36:33Z","timestamp":1790080593000},"score":1,"resource":{"primary":{"URL":"https:\/\/www.jmir.org\/2026\/1\/e92090"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,9,21]]},"references-count":24,"URL":"https:\/\/doi.org\/10.2196\/92090","relation":{},"ISSN":["1438-8871"],"issn-type":[{"value":"1438-8871","type":"electronic"}],"subject":[],"published":{"date-parts":[[2026,9,21]]},"article-number":"v28i15e92090"}}