Memorization in Large Language Models in Medicine: Prevalence, Characteristics, and Implications

• URL: http://arxiv.org/abs/2509.08604v2
• Date: Thu, 06 Nov 2025 18:15:30 GMT
• Title: Memorization in Large Language Models in Medicine: Prevalence, Characteristics, and Implications
• Abstract summary: Large Language Models (LLMs) have demonstrated significant potential in medicine.<n>We present the first comprehensive evaluation of memorization of LLMs in medicine.<n>The results demonstrate that memorization is prevalent across all adaptation scenarios and significantly higher than reported in the general domain.
• Score: 42.69954853399731
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Large Language Models (LLMs) have demonstrated significant potential in medicine. To date, LLMs have been widely applied to tasks such as diagnostic assistance, medical question answering, and clinical information synthesis. However, a key open question remains: to what extent do LLMs memorize medical training data. In this study, we present the first comprehensive evaluation of memorization of LLMs in medicine, assessing its prevalence (how frequently it occurs), characteristics (what is memorized), volume (how much content is memorized), and potential downstream impacts (how memorization may affect medical applications). We systematically analyze common adaptation scenarios: (1) continued pretraining on medical corpora, (2) fine-tuning on standard medical benchmarks, and (3) fine-tuning on real-world clinical data, including over 13,000 unique inpatient records from Yale New Haven Health System. The results demonstrate that memorization is prevalent across all adaptation scenarios and significantly higher than reported in the general domain. Memorization affects both the development and adoption of LLMs in medicine and can be categorized into three types: beneficial (e.g., accurate recall of clinical guidelines and biomedical references), uninformative (e.g., repeated disclaimers or templated medical document language), and harmful (e.g., regeneration of dataset-specific or sensitive clinical content). Based on these findings, we offer practical recommendations to facilitate beneficial memorization that enhances domain-specific reasoning and factual accuracy, minimize uninformative memorization to promote deeper learning beyond surface-level patterns, and mitigate harmful memorization to prevent the leakage of sensitive or identifiable patient information.

Related papers

• Clinically Grounded Privacy Evaluation of Medical LMs [14.192902006563045]
  We introduce a clinically grounded framework that evaluates leakage along a graded axis of adversarial access.<n>Applying the framework to an LM pretrained on 378k clinical notes, we find that routine encounter metadata elicits high rates of verbatim memorization.
  arXiv  Detail & Related papers  (2026-06-08T15:02:19Z)
• MedKGEval: A Knowledge Graph-Based Multi-Turn Evaluation Framework for Open-Ended Patient Interactions with Clinical LLMs [19.12790150016383]
  We present MedKGEval, a novel multi-turn evaluation framework for clinical large language models.<n>A knowledge graph-driven patient simulation mechanism retrieves relevant medical facts from a curated knowledge graph.<n>A turn-level evaluation framework assesses each model response for clinical appropriateness, factual correctness, and safety.
  arXiv  Detail & Related papers  (2025-10-14T07:22:26Z)
• Simulating Viva Voce Examinations to Evaluate Clinical Reasoning in Large Language Models [51.91760712805404]
  We introduce VivaBench, a benchmark for evaluating sequential clinical reasoning in large language models (LLMs)<n>Our dataset consists of 1762 physician-curated clinical vignettes structured as interactive scenarios that simulate a (oral) examination in medical training.<n>Our analysis identified several failure modes that mirror common cognitive errors in clinical practice.
  arXiv  Detail & Related papers  (2025-10-11T16:24:35Z)
• BRIDGE: Benchmarking Large Language Models for Understanding Real-world Clinical Practice Text [14.409097921305134]
  BRIDGE is a comprehensive benchmark comprising 87 tasks sourced from real-world clinical data sources across nine languages.<n>It covers eight major task types spanning the entire continuum of patient care across six clinical stages and 20 representative applications.<n>Our results reveal substantial performance variation across model sizes, languages, natural language processing tasks, and clinical specialties.
  arXiv  Detail & Related papers  (2025-04-28T04:13:18Z)
• Medical Hallucinations in Foundation Models and Their Impact on Healthcare [53.97060824532454]
  Foundation Models that are capable of processing and generating multi-modal data have transformed AI's role in medicine.<n>We define medical hallucination as any instance in which a model generates misleading medical content.<n>Our results reveal that inference techniques such as Chain-of-Thought (CoT) and Search Augmented Generation can effectively reduce hallucination rates.<n>These findings underscore the ethical and practical imperative for robust detection and mitigation strategies.
  arXiv  Detail & Related papers  (2025-02-26T02:30:44Z)
• Fact or Guesswork? Evaluating Large Language Models' Medical Knowledge with Structured One-Hop Judgments [108.55277188617035]
  Large language models (LLMs) have been widely adopted in various downstream task domains, but their abilities to directly recall and apply factual medical knowledge remains under-explored.<n>We introduce the Medical Knowledge Judgment dataset (MKJ), a dataset derived from the Unified Medical Language System (UMLS), a comprehensive repository of standardized vocabularies and knowledge graphs.<n>Through a binary classification framework, MKJ evaluates LLMs' grasp of fundamental medical facts by having them assess the validity of concise, one-hop statements.
  arXiv  Detail & Related papers  (2025-02-20T05:27:51Z)
• DIRI: Adversarial Patient Reidentification with Large Language Models for Evaluating Clinical Text Anonymization [13.038800602897354]
  We develop an adversarial approach using a large language model to re-identify the patient corresponding to a redacted clinical note. Our method uses a large language model to reidentify the patient corresponding to a redacted clinical note. Although ClinicalBERT was the most effective, masking all identified PII, our tool still reidentified 9% of clinical notes.
  arXiv  Detail & Related papers  (2024-10-22T14:06:31Z)
• Reasoning-Enhanced Healthcare Predictions with Knowledge Graph Community Retrieval [61.70489848327436]
  KARE is a novel framework that integrates knowledge graph (KG) community-level retrieval with large language models (LLMs) reasoning.<n>Extensive experiments demonstrate that KARE outperforms leading models by up to 10.8-15.0% on MIMIC-III and 12.6-12.7% on MIMIC-IV for mortality and readmission predictions.
  arXiv  Detail & Related papers  (2024-10-06T18:46:28Z)
• SemioLLM: Evaluating Large Language Models for Diagnostic Reasoning from Unstructured Clinical Narratives in Epilepsy [45.2233252981348]
  Large Language Models (LLMs) have been shown to encode clinical knowledge.<n>We present SemioLLM, an evaluation framework that benchmarks 6 state-of-the-art models.<n>We show that most LLMs are able to accurately and confidently generate probabilistic predictions of seizure onset zones in the brain.
  arXiv  Detail & Related papers  (2024-07-03T11:02:12Z)
• Adapted Large Language Models Can Outperform Medical Experts in Clinical Text Summarization [8.456700096020601]
  Large language models (LLMs) have shown promise in natural language processing (NLP), but their effectiveness on a diverse range of clinical summarization tasks remains unproven. In this study, we apply adaptation methods to eight LLMs, spanning four distinct clinical summarization tasks. A clinical reader study with ten physicians evaluates summary, completeness, correctness, and conciseness; in a majority of cases, summaries from our best adapted LLMs are either equivalent (45%) or superior (36%) compared to summaries from medical experts.
  arXiv  Detail & Related papers  (2023-09-14T05:15:01Z)
• Self-Verification Improves Few-Shot Clinical Information Extraction [73.6905567014859]
  Large language models (LLMs) have shown the potential to accelerate clinical curation via few-shot in-context learning. They still struggle with issues regarding accuracy and interpretability, especially in mission-critical domains such as health. Here, we explore a general mitigation framework using self-verification, which leverages the LLM to provide provenance for its own extraction and check its own outputs.
  arXiv  Detail & Related papers  (2023-05-30T22:05:11Z)
• Retrieval-Augmented and Knowledge-Grounded Language Models for Faithful Clinical Medicine [68.7814360102644]
  We propose the Re$3$Writer method with retrieval-augmented generation and knowledge-grounded reasoning. We demonstrate the effectiveness of our method in generating patient discharge instructions.
  arXiv  Detail & Related papers  (2022-10-23T16:34:39Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.