CoChat

Large language models encode clinical knowledge

Open access
Singhal, Karan, Azizi, Shekoofeh, Tu, Tao, Mahdavi, S. Sara, Wei, Jason, Chung, Hyung Won, Scales, Nathan, Tanwani, Ajay, Cole-Lewis, Heather, Pfohl, Stephen, Payne, Perry, Seneviratne, Martin, Gamble, Paul, Kelly, Chris, Babiker, Abubakr, Schärli, Nathanael, Chowdhery, Aakanksha, Mansfield, Philip, Demner-Fushman, Dina, Agüera y Arcas, Blaise, Webster, Dale, Corrado, Greg S., Matias, Yossi, Chou, Katherine, Gottweis, Juraj, Tomasev, Nenad, Liu, Yun, Rajkomar, Alvin, Barral, Joelle, Semturs, Christopher, Karthikesalingam, Alan, Natarajan, Vivek · 2023 · Nature

The DOI matches its Crossref record; 115 of the 116 references it cites were found in a registry or an index.

Abstract

Large language models (LLMs) have demonstrated impressive capabilities, but the bar for clinical applications is high. Attempts to assess the clinical knowledge of models typically rely on automated evaluations based on limited benchmarks. Here, to address these limitations, we present MultiMedQA, a benchmark combining six existing medical question answering datasets spanning professional medicine, research and consumer queries and a new dataset of medical questions searched online, HealthSearchQA. We propose a human evaluation framework for model answers along multiple axes including factuality, comprehension, reasoning, possible harm and bias. In addition, we evaluate Pathways Language Model 1 (PaLM, a 540-billion parameter LLM) and its instruction-tuned variant, Flan-PaLM 2 on MultiMedQA. Using a combination of prompting strategies, Flan-PaLM achieves state-of-the-art accuracy on every MultiMedQA multiple-choice dataset (MedQA 3 , MedMCQA 4 , PubMedQA 5 and Measuring Massive Multitask Language Understanding (MMLU) clinical topics 6 ), including 67.6% accuracy on MedQA (US Medical Licensing Exam-style questions), surpassing the prior state of the art by more than 17%. However, human evaluation reveals key gaps. To resolve this, we introduce instruction prompt tuning, a parameter-efficient approach for aligning LLMs to new domains using a few exemplars. The resulting model, Med-PaLM, performs encouragingly, but remains inferior to clinicians. We show that comprehension, knowledge recall and reasoning improve with model scale and instruction prompt tuning, suggesting the potential utility of LLMs in medicine. Our human evaluations reveal limitations of today’s models, reinforcing the importance of both evaluation frameworks and method development in creating safe, helpful LLMs for clinical applications.

Overview

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.1038/s41586-023-06291-2

Cites

115 of 116 references verified0 citation errors · 1 not checkable

Statistics

Figures

Fig. 1 Overview of our contributions. We curate MultiMedQA, a benchmark for answering medical questions spanning medical exam, medical research and consumer medical questions. We evaluate PaLM and its instructed-tuned variant, Flan-PaLM, on MultiMedQA. Using a combination of prompting strategies, Flan-PaLM exceeds state-of-the-art performance on MedQA (US Medical Licensing Examination (USMLE)), MedMCQA, PubMedQA and MMLU clinical topics. In particular, it improves over the previous state of the art on MedQA (USMLE) by over 17%. We next propose instruction prompt tuning to further align Flan-PaLM to the medical domain, producing Med-PaLM. Med-PaLM’s answers to consumer medical questions compare favourably with answers given by clinicians under our human evaluation framework, demonstrating the effectiveness of instruction prompt tuning.
Fig. 1 Overview of our contributions. We curate MultiMedQA, a benchmark for answering medical questions spanning medical exam, medical research and consumer medical questions. We evaluate PaLM and its instructed-tuned variant, Flan-PaLM, on MultiMedQA. Using a combination of prompting strategies, Flan-PaLM exceeds state-of-the-art performance on MedQA (US Medical Licensing Examination (USMLE)), MedMCQA, PubMedQA and MMLU clinical topics. In particular, it improves over the previous state of the art on MedQA (USMLE) by over 17%. We next propose instruction prompt tuning to further align Flan-PaLM to the medical domain, producing Med-PaLM. Med-PaLM’s answers to consumer medical questions compare favourably with answers given by clinicians under our human evaluation framework, demonstrating the effectiveness of instruction prompt tuning.CC BY · PubMed Central
Fig. 2 Comparison of our method and prior state of the art. Our Flan-PaLM 540B model exceeds the previous state-of-the-art performance (SOTA) on MedQA (four options), MedMCQA and PubMedQA datasets. The previous state-of-the-art results are from Galactica20 (MedMCQA), PubMedGPT19 (MedQA) and BioGPT21 (PubMedQA). The percentage accuracy is shown above each column.
Fig. 2 Comparison of our method and prior state of the art. Our Flan-PaLM 540B model exceeds the previous state-of-the-art performance (SOTA) on MedQA (four options), MedMCQA and PubMedQA datasets. The previous state-of-the-art results are from Galactica20 (MedMCQA), PubMedGPT19 (MedQA) and BioGPT21 (PubMedQA). The percentage accuracy is shown above each column.CC BY · PubMed Central
Fig. 3 Selective prediction analysis. Analysis of deferral behaviour of the Flan-PaLM 540B model with self-consistency. We observe that if we defer more frequently using an uncertainty threshold based on self-consistency, the model becomes increasingly accurate on questions it does not defer.
Fig. 3 Selective prediction analysis. Analysis of deferral behaviour of the Flan-PaLM 540B model with self-consistency. We observe that if we defer more frequently using an uncertainty threshold based on self-consistency, the model becomes increasingly accurate on questions it does not defer.CC BY · PubMed Central
Fig. 4 Clinician evaluation of answers. a–f, Clinicians were asked to rate answers to questions in the HealthSearchQA, LiveQA and MedicationQA datasets for agreement with scientific and clinical consensus (a), the presence of incorrect content (b), the omission of content (c), the extent of possible harm (d), the likelihood of harm (e) and possible bias in answers (f). We compare answers from Flan-PaLM, Med-PaLM and clinicians. Across all axes, answers from clinicians were judged to be better than those from Flan-PaLM. Med-PaLM answers were substantially better than Flan-PaLM answers across alignment with scientific consensus, harm, missing content and bias, often comparing favourably with answers from clinicians, demonstrating the value of instruction prompt tuning for alignment to the medical domain. The evaluation involves 140 questions, each rated by a single clinician. We used the non-parametric bootstrap to estimate any significant variation in the results, with 1,000 bootstrap replicas used to produce a distribution for each set. We used the 95% bootstrap percentile interval to assess variations. Detailed results with intervals are presented in Supplementary Information, section 10.
Fig. 4 Clinician evaluation of answers. a–f, Clinicians were asked to rate answers to questions in the HealthSearchQA, LiveQA and MedicationQA datasets for agreement with scientific and clinical consensus (a), the presence of incorrect content (b), the omission of content (c), the extent of possible harm (d), the likelihood of harm (e) and possible bias in answers (f). We compare answers from Flan-PaLM, Med-PaLM and clinicians. Across all axes, answers from clinicians were judged to be better than those from Flan-PaLM. Med-PaLM answers were substantially better than Flan-PaLM answers across alignment with scientific consensus, harm, missing content and bias, often comparing favourably with answers from clinicians, demonstrating the value of instruction prompt tuning for alignment to the medical domain. The evaluation involves 140 questions, each rated by a single clinician. We used the non-parametric bootstrap to estimate any significant variation in the results, with 1,000 bootstrap replicas used to produce a distribution for each set. We used the 95% bootstrap percentile interval to assess variations. Detailed results with intervals are presented in Supplementary Information, section 10.CC BY · PubMed Central
Fig. 5 Evaluation of comprehension, retrieval and reasoning capabilities by clinicians. a,b, Evaluation of correctness (a) and incorrectness (b) of reading comprehension, recall of knowledge and reasoning steps. The results indicate a gap between Flan-PaLM and clinicians, and show that Med-PaLM is able to substantially reduce the gap. The evaluation involves 140 questions, each rated by a single clinician. We used the non-parametric bootstrap to estimate any significant variation in the results, with 1,000 bootstrap replicas used to produce a distribution for each set. We used the 95% bootstrap percentile interval to assess variations.
Fig. 5 Evaluation of comprehension, retrieval and reasoning capabilities by clinicians. a,b, Evaluation of correctness (a) and incorrectness (b) of reading comprehension, recall of knowledge and reasoning steps. The results indicate a gap between Flan-PaLM and clinicians, and show that Med-PaLM is able to substantially reduce the gap. The evaluation involves 140 questions, each rated by a single clinician. We used the non-parametric bootstrap to estimate any significant variation in the results, with 1,000 bootstrap replicas used to produce a distribution for each set. We used the 95% bootstrap percentile interval to assess variations.CC BY · PubMed Central
Fig. 6 Lay user assessment of answers. a,b, Lay user assessment of answers, addressing relevance to the intent of the query (a) and helpfulness (b). Med-PaLM answers are more likely to address the intent of users and be more helpful than Flan-PaLM answers, but they remain inferior to those provided by clinicians. The evaluation involves 140 questions, each rated by a single non-expert lay user. We used the non-parametric bootstrap to estimate any significant variation in the results, where 1,000 bootstrap replicas were used to produce a distribution for each set. We used the 95% bootstrap percentile interval to assess variations.
Fig. 6 Lay user assessment of answers. a,b, Lay user assessment of answers, addressing relevance to the intent of the query (a) and helpfulness (b). Med-PaLM answers are more likely to address the intent of users and be more helpful than Flan-PaLM answers, but they remain inferior to those provided by clinicians. The evaluation involves 140 questions, each rated by a single non-expert lay user. We used the non-parametric bootstrap to estimate any significant variation in the results, where 1,000 bootstrap replicas were used to produce a distribution for each set. We used the 95% bootstrap percentile interval to assess variations.CC BY · PubMed Central
Extended Data Fig. 1 Instruction prompt tuning for Med-PaLM. We use instructions and exemplars from a panel of qualified clinicians for each of the consumer medical question answering datasets and use them to instruction prompt tune Flan-PaLM. Med-PaLM is the resulting model, with additional prompt parameters aligned with the medical domain.
Extended Data Fig. 1 Instruction prompt tuning for Med-PaLM. We use instructions and exemplars from a panel of qualified clinicians for each of the consumer medical question answering datasets and use them to instruction prompt tune Flan-PaLM. Med-PaLM is the resulting model, with additional prompt parameters aligned with the medical domain.CC BY · PubMed Central
Extended Data Fig. 2 Comparison of SOTA LLMs on MMLU clinical topics. Flan-PaLM achieves state-of-the-art performance on MMLU clinical topics.
Extended Data Fig. 2 Comparison of SOTA LLMs on MMLU clinical topics. Flan-PaLM achieves state-of-the-art performance on MMLU clinical topics.CC BY · PubMed Central

Verification

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

40 more claims about this paper

Create a free account to see every claim other papers make about this one.

See all claims — free account

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Citations

Cites

Show all 116 references

Research works printed but not linked: 2

  1. Abacha, A. B. et al. in Studies in Health Technology and Informatics (eds Ohno-Machado, L. & Séroussi, B.) 25–29 (IOS Press, 2019).
  2. Matheny, M., Israni, S. T., Ahmed, M. & Whicher, D. eds. Artificial Intelligence in Health Care: The Hope, the Hype, the Promise, the Peril (National Academy of Medicine, 2022). 39146448

Cited by

See all 4,083

Metadata

Authors
Singhal, K., Azizi, S., Tu, T., Mahdavi, S. S., Wei, J., Chung, H. W., Scales, N., Tanwani, A., Cole-Lewis, H., Pfohl, S., Payne, P., Seneviratne, M., Gamble, P., Kelly, C., Babiker, A., Schärli, N., Chowdhery, A., Mansfield, P., Demner-Fushman, D., Agüera y Arcas, B., Webster, D., Corrado, G. S., Matias, Y., Chou, K., Gottweis, J., Tomasev, N., Liu, Y., Rajkomar, A., Barral, J., Semturs, C., Karthikesalingam, A., & Natarajan, V.
Affiliations
Google (United States) (US)
Corresponding author
  • Singhal, Karan
  • Azizi, Shekoofeh
  • Karthikesalingam, Alan
  • Natarajan, Vivek
Venue
Nature
Published
2023-07-12
Volume
620
Issue
7972
Pages
172-180
ISSN
  • 0028-0836 (ISSN-L)
  • 1476-4687
OpenAlex
W4384071683
Cited by
4,083
References
85
Citations by year
2026: 1627 · 2025: 1275 · 2024: 933 · 2023: 247
FWCI
162.53
Type
journal article
Language
en
Open access
Licence
Locations
  • Nature (publishedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • arXiv (Cornell University) (submittedVersion)
  • RePEc: Research Papers in Economics (submittedVersion)
  • arXiv (Cornell University) (cc-by)
Funders
  • DeepMind
  • U.S. National Library of Medicine
Topics
  • Health Sciences › Medicine › Health Informatics › Artificial Intelligence in Healthcare and Education
  • Physical Sciences › Computer Science › Artificial Intelligence › Topic Modeling
  • Physical Sciences › Computer Science › Artificial Intelligence › Machine Learning in Healthcare
Keywords
large language models, clinical knowledge, medical question answering, human evaluation, Arecaceae, MedQA, MedMCQA, PubMedQA, MMLU, clinical reasoning
MeSH terms
Clinical Competence, Computer Simulation, Humans, Licensure, Medicine, Natural Language Processing, Physicians, Bias, Knowledge, Benchmarking, Comprehension, Patient Safety, Datasets as Topic

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: Large language models encode clinical knowledge. https://app.cochat.ai/papers/10.1038/s41586-023-06291-2
The paper
Singhal, K., Azizi, S., Tu, T., Mahdavi, S. S., Wei, J., Chung, H. W., Scales, N., Tanwani, A., Cole-Lewis, H., Pfohl, S., Payne, P., Seneviratne, M., Gamble, P., Kelly, C., Babiker, A., Schärli, N., Chowdhery, A., Mansfield, P., Demner-Fushman, D., . . . Natarajan, V. (2023). Large language models encode clinical knowledge. Nature. https://doi.org/10.1038/s41586-023-06291-2

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.