Pith. sign in

REVIEW 4 major objections 5 minor 15 references

MedFactEval and MedAgentBrief: A Framework and Workflow for Generating and Evaluating Factual Clinical Summaries

T0 review · 4 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read This paper claims that an LLM jury—ten language models voting on whether clinician-specified key facts appear in a draft summary—agrees with a seven-physician gold standard almost perfectly (Cohen's kappa 81%) and is statistically no…

desk verdict A solid, useful evaluation framework with a real validation result, but the jury's transfer to single-prompt summaries is unvalidated and one reported p-value looks inconsistent with its confidence interval. read the letter →

arxiv 2509.05878 v1 pith:TB2SDYXP submitted 2025-09-07 cs.CL

François Grolleau , Emily Alsentzer , Timothy Keyes , Philip Chung , Akshay Swaminathan , Asad Aali , Jason Hom , Tridu Huynh
show 12 more authors
Thomas Lew April S. Liang Weihan Chu Natasha Z. Steele Christina F. Lin Jingkun Yang Kameron C. Black Stephen P. Ma Fateme N. Haredasht Nigam H. Shah Kevin Schulman Jonathan H. Chen
This is my paper · ORCID
classification cs.CL
keywords clinicaltextsummarizationfactualityevaluationLLMjurydischargesummarieskeyfactextractionCohen'skappanon-inferioritytestingAIqualityassurance
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper introduces MedFactEval, a framework for fact-grounded evaluation of AI-generated clinical text, and MedAgentBrief, a multi-step workflow for generating discharge summaries. It claims that a ten-model LLM jury, taking a majority vote on whether clinician-defined key facts are present in a summary, agrees with a seven-physician gold standard with Cohen's kappa of 81%, statistically non-inferior to a single physician reviewer (kappa 67%, P<0.001). The same jury is then used to benchmark generation strategies, showing that MedAgentBrief consistently improves factual presence over a single-prompt baseline at higher inference cost. The intended payoff is a scalable, automated proxy for physician review in continuous quality assurance of clinical AI.

What carries the argument

The load-bearing mechanism is the LLM Jury: ten distinct language models each answer a binary 'is fact X present in this summary?' question, and the majority vote decides the verdict, smoothing the errors of individual judges. The other component is the clinician-defined key fact: instead of scoring a summary globally, physicians specify three high-salience facts per case, converting 'is this summary good?' into concrete, verifiable questions. Agreement is measured with Cohen's kappa against a seven-physician majority gold standard, with non-inferiority tested against a pre-specified margin of 10 percentage points.

What would settle it

Take a holdout set of single-prompt summaries, have a seven-physician panel label the same key facts, and compare the untuned LLM jury's majority vote to that gold standard; if the kappa falls below the pre-specified non-inferiority margin of 10 percentage points relative to a single physician, then the jury's benchmark scores cannot be trusted for the comparison that drives the paper's main practical claim.

Watch

Extended reading notes

Core claim

The central discovery is that an ensemble of ten LLMs, each giving a binary verdict on whether a clinician-specified key fact is present in a generated summary, can reproduce physician judgment reliably. On 60 discharge summaries produced by two MedAgentBrief systems, the jury's majority vote agreed with the seven-physician majority at kappa = 81% (95% CI 66–92%), and the authors establish statistical non-inferiority to the average single physician (kappa = 67%, P<0.001). The paper also reports that MedAgentBrief, which refines a draft note-by-note with provenance tagging and a verification pass, raises the fact-presence score of a representative foundation model from 48% to 65%, and that errors of omission dominate over contradictions across all systems tested.

Load-bearing premise

The jury's agreement with physicians was measured only on 60 MedAgentBrief-generated summaries, yet the same jury is then used to score single-prompt summaries, so the reported MedAgentBrief advantage depends on the jury's reliability transferring to a different generation strategy without direct validation on that distribution.

Editorial extensions

If this is right

  • Clinical teams could run continuous quality assurance on LLM-generated drafts by defining a few key facts per case and letting an LLM jury flag omissions and contradictions, rather than paying physicians to review every summary.
  • The benchmarking result implies that multi-step generation workflows with provenance tagging and verification improve factuality over single-prompt generation for every foundation model tested, at a predictable cost in inference dollars and latency.
  • Because omissions dominate over contradictions, future development effort should focus on recall of high-salience facts rather than hallucination reduction alone.
  • The open-source code and benchmark let other institutions build their own local key-fact benchmarks; the authors state the framework generalizes to other clinical information synthesis tasks once key facts are redefined.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If the jury's reliability transfers beyond the 60 MedAgentBrief summaries on which it was validated, the method could replace single-reviewer spot checks in deployment; the paper itself does not yet demonstrate that transfer for single-prompt outputs.
  • The key-fact framing suggests a testable extension: feed omission reports back to the generation model as few-shot corrections and measure whether factual presence rises across iterations, closing the improvement loop the paper implies but does not run.
  • A likely boundary condition is that the jury's agreement depends on the clarity of key facts and the summary distribution; institutions with different note styles or patient populations should revalidate locally before relying on the reported kappa.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The paper introduces MedFactEval, a framework in which clinicians define high-salience key facts and a ten-LLM jury votes on whether those facts are present or contradicted in AI-generated discharge summaries. It also introduces MedAgentBrief, a multi-step, model-agnostic workflow with iterative refinement and provenance tagging, and benchmarks it against a single-prompt baseline. The central validation is a meta-evaluation on 60 summaries generated by two MedAgentBrief systems (GPT-4o and DeepSeek-R1), where the LLM Jury's fact-presence judgments are compared with a seven-physician majority-vote gold standard. The jury achieves Cohen's kappa of 81% (95% CI 66–92%) on the GPT-4o summaries and is statistically non-inferior to a single physician (kappa difference 15%, 95% CI -1% to 29%, P<0.001). The same jury is then used in Section 4.2 to score single-prompt summaries across multiple models, leading to the claim that MedAgentBrief improves factual presence (e.g., GPT-4o from 48% to 65%).

Significance. If the reported meta-evaluation holds, the paper makes a useful practical contribution: an LLM jury with pre-specified clinician facts can approximate physician fact-checking at scale, and the leave-one-out human baseline, bootstrap confidence intervals, and pre-specified non-inferiority margin are appropriate methodological choices. The open-source code and the concrete demonstration of a performance-cost trade-off are also strengths. The central risk is that the jury is validated only on MedAgentBrief-style outputs and then applied without direct validation to single-prompt outputs, so the headline benchmarking result could reflect distribution-specific jury behavior. The contradiction-assessment component, also part of MedFactEval, receives no meta-evaluation at all.

major comments (4)
  1. [§3.4, §4.2, Figure 1] The jury's agreement with the seven-physician gold standard is measured only on 60 MedAgentBrief-generated summaries (from GPT-4o and DeepSeek-R1), but the same jury is then used to score Single-Prompt summaries in §4.2 and Figure 1. This transfer is unvalidated: single-prompt outputs differ in format, length, and error profile, and the jury's false-positive or false-negative rates could shift on that distribution. The reported advantage of MedAgentBrief (e.g., GPT-4o from 48% to 65%) is therefore not directly supported by the meta-evaluation. The authors should either provide human labels (or at least a sample-based reliability check) for single-prompt summaries, or explicitly re-frame the §4.2 results as exploratory and conditional on the unverified transfer assumption.
  2. [§3.3, Step 2; §4.2, paragraph 'Analysis of Factual Errors'] The contradiction-assessment component of the LLM Jury is never meta-evaluated against human judgment. The manuscript reports contradiction rates of 10–25% and claims high sensitivity in detecting subtle inconsistencies, but no gold-standard comparison or kappa is reported for the contradiction task. Since contradictions are a central safety concern and a stated part of the framework, the absence of validation should be acknowledged as a limitation, or the contradiction results should be presented as purely illustrative rather than as validated framework output.
  3. [§3.3, Step 1; §4.1] Key facts were optionally suggested by a Task-Assisting LLM (GPT-4o or Gemini 2.0 Flash) and then reviewed by physicians. Because the same or overlapping model families later serve as jury members, there is a potential information leak: a key fact phrased by GPT-4o may be easier for a GPT-4o-based judge to recognize than facts independently authored by physicians. This does not invalidate the external-gold-standard design, but a sensitivity analysis separating LLM-suggested from physician-authored key facts would strengthen the claim that the jury is measuring clinician-defined facts rather than reproducing its own phrasing.
  4. [§3.4, Statistical Analysis] The non-inferiority test uses a pre-specified margin of 10% on the kappa difference, but the manuscript does not state whether the one-sided P-value was computed against this margin with a bootstrap percentile, a normal approximation, or another method. The reported values (P<0.001 for a difference of 15% with a 95% CI of -1% to 29%) are plausible, but the exact test should be described to allow replication. This is a local methodological clarity issue rather than a fatal flaw.
minor comments (5)
  1. [Figure 2] The caption states that the blue shaded area represents the 95% CI for a single physician's agreement, but the figure panels are not all referenced with sufficient precision in the text; please ensure each panel (A–D) is explicitly tied to the GPT-4o or DeepSeek-R1 workflow in the caption.
  2. [§2, Related Work] The discussion of automated fact-checking would benefit from citing a few more recent LLM-as-judge and clinical factuality evaluations beyond VeriFact, MedHallBench, and FactEHR, particularly work that validates LLM judges against clinician panels for generation tasks, to better position the contribution.
  3. [§3.1] The cohort description states that 30 patients were randomly sampled, but the sampling frame and inclusion criteria (LOS 2–14 days, Division of Hospital Medicine) could be described more precisely, including whether the 30 cases are all from a single site and how seasonality was handled.
  4. [§4.1, Table 1] The key-fact categorization sums to 90, but the category percentages (51% + 40% + 6% + 3% = 100%) are rounded such that the counts (46+36+5+3=90) are not exactly proportional to the displayed percentages; this is a trivial rounding issue but should be reconciled for a journal version.
  5. [Supplementary Data links] The URLs for Supplementary Data 1 and 2 are not embedded as clickable anchors in the text; please format them as standard hyperlinks and verify that the anonymized samples load without authentication.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the MedFactEval jury is validated against an external seven-physician gold standard, not against its own inputs.

full rationale

The central validation claim is empirical and externally grounded. In Section 3.4, the gold standard is built from a seven-physician majority vote, and in Section 4.3 the LLM Jury's Cohen's kappa is measured directly against that gold standard (e.g., kappa=81%, 95% CI 66-92%). This is not a construction in which X is defined in terms of Y, nor is a fitted parameter renamed as a prediction. The key facts are clinician-defined from human-written discharge summaries (Section 3.3, Step 1), so the jury's target labels are independent of the LLM evaluators; the single-human baseline is also computed with a leave-one-out design, avoiding self-inclusion. The subsequent application of the same jury to Single-Prompt summaries in Section 4.2 is a distribution-shift / external-validity concern: the meta-evaluation used only 60 MedAgentBrief summaries, and no human labels are reported for Single-Prompt outputs. That is a legitimate correctness caveat about transfer of the jury's reliability, but it does not make the MedAgentBrief-vs-Single-Prompt comparison circular: the jury was not fitted to the Single-Prompt data, and no equation or definition reduces to the paper's conclusion. Related-work citations to prior work by the same group (e.g., VeriFact, FactEHR) are contextual and not load-bearing for the framework's validity. I therefore find no self-definitional, fitted-input, self-citation, or uniqueness-imported circularity.

Assumptions & free parameters 0 free parameters · 5 assumptions · 0 invented entities

The core validation compares the LLM Jury to an external human gold standard, which is appropriate. The main added presumptions are that the seven-physician majority is a valid gold standard, that the clinician-chosen key facts are the right facts to check, that the jury's performance on MedAgentBrief summaries transfers to single-prompt summaries, and that averaging ten LLM votes is a stable ensemble. No free parameters are fitted and no new scientific entities are introduced.

assumptions (5)
  • domain assumption The seven-physician majority vote is a valid gold standard for key-fact presence.
    Section 3.4 defines the gold standard as the panel majority without reporting inter-rater reliability among the seven physicians; the gold standard is assumed to be correct by construction.
  • domain assumption The three key facts per case extracted from the human-written discharge summary are the high-salience facts that matter for evaluating AI summaries.
    Section 3.3 Step 1: annotators review the original discharge summary, optionally starting from LLM-suggested candidates; the limitations section acknowledges that key fact selection is inherently subjective.
  • ad hoc to paper The LLM Jury's agreement with the gold standard, measured on MedAgentBrief summaries, generalizes to Single-Prompt summaries scored in Figure 1.
    The meta-evaluation in Section 3.4 uses only 60 summaries from two MedAgentBrief systems; Figure 1 then scores Single-Prompt outputs with the same jury without any validation on that distribution.
  • domain assumption Majority voting over ten LLM judges yields a stable and meaningful ensemble verdict for binary fact-presence questions.
    Section 3.3 Step 2 fixes the jury size and voting rule; the paper does not report variability across different random subsets of judges or across prompt temperature settings.
  • standard math Cohen's kappa and the pre-specified 10% non-inferiority margin are appropriate statistical tools for this binary agreement comparison.
    Section 3.4 specifies the kappa-based one-sided non-inferiority test, following Landis and Koch; this is standard practice but the choice of margin affects the strength of the claim.

how reviews work

0 comments
Cite this review

Pith. "Pith review of MedFactEval and MedAgentBrief: A Framework and Workflow for Generating and Evaluating Factual Clinical Summaries." pith.science (2026). https://pith.science/paper/TB2SDYXP

@misc{pith2026250905878,
  author       = {Pith},
  title        = {Pith review of: MedFactEval and MedAgentBrief: A Framework and Workflow for Generating and Evaluating Factual Clinical Summaries},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/TB2SDYXP}},
  note         = {Machine review of arXiv:2509.05878}
}
read the original abstract

Evaluating factual accuracy in Large Language Model (LLM)-generated clinical text is a critical barrier to adoption, as expert review is unscalable for the continuous quality assurance these systems require. We address this challenge with two complementary contributions. First, we introduce MedFactEval, a framework for scalable, fact-grounded evaluation where clinicians define high-salience key facts and an "LLM Jury"--a multi-LLM majority vote--assesses their inclusion in generated summaries. Second, we present MedAgentBrief, a model-agnostic, multi-step workflow designed to generate high-quality, factual discharge summaries. To validate our evaluation framework, we established a gold-standard reference using a seven-physician majority vote on clinician-defined key facts from inpatient cases. The MedFactEval LLM Jury achieved almost perfect agreement with this panel (Cohen's kappa=81%), a performance statistically non-inferior to that of a single human expert (kappa=67%, P < 0.001). Our work provides both a robust evaluation framework (MedFactEval) and a high-performing generation workflow (MedAgentBrief), offering a comprehensive approach to advance the responsible deployment of generative AI in clinical workflows.

Figures

Figures reproduced from arXiv: 2509.05878 by the authors.

Figure 1
Figure 1. Performance-Cost Trade-off of AI Summary Generation, Measured by Med￾FactEval. The y-axis represents the MedFactEval score for factual presence (higher is better), while the x-axis shows inference cost per patient on a logarithmic scale. For each foundation model, the MedAgentBrief workflow (cyan) consistently yields a higher factuality score than the baseline Single Prompt approach (brown), demonstrating its effect… view at source ↗
Figure 2
Figure 2. Agreement of LLM Judges and Juries with the Physician Gold Standard. Each point represents the Cohen’s kappa agreement for an evaluator against the seven-physician majority vote. (A, C) Agreement versus evaluation time for summaries from the GPT-4o and Deepseek-R1 workflows, respectively. (B, D) Agreement versus evaluation cost for the same summaries. The full 10-member LLM Jury (red diamond) consistently achieves h… view at source ↗
Figure 3
Figure 3. Non-Inferiority of LLM Judges Compared to a Single Physician Expert. The plot shows the difference in Cohen’s kappa between each LLM judge and the average single physician, for summaries generated by the GPT-4o-based MedAgentBrief. Points to the right of the vertical line favor the LLM judge. The dashed line indicates the pre-specified non-inferiority margin. The full LLM Jury (bottom diamond) and the Small LLM Jury… view at source ↗

Review history (2 revisions) →

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

15 extracted references · 12 canonical work pages

  1. [1]

    A. J. Moy, J. M. Schwartz, R. Chen, S. Sadri, E. Lucas, K. D. Cato and S. C. Rossetti, Mea- surement of clinical documentation burden among physicians and nurses using electronic health records: a scoping review,Journal of the American Medical Informatics Association28, 998 (2021)

    work page 2021

  2. [2]

    Van Veen, C

    D. Van Veen, C. Van Uden, L. Blankemeier, J.-B. Delbrouck, A. Aali, C. Bluethgen, A. Pareek, M. Polacin, E. P. Reis, A. Seehofnerov´ aet al., Adapted large language models can outperform medical experts in clinical text summarization,Nature Medicine30, 1134 (2024)

    work page 2024

  3. [3]

    C. Y. Williams, C. R. Subramanian, S. S. Aliet al., Physician-and large language model- generated hospital discharge summaries,JAMA Internal Medicine(May 2025)

    work page 2025

  4. [4]

    Mesk´ o and E

    B. Mesk´ o and E. J. Topol, The imperative for regulatory oversight of large language models (or generative AI) in healthcare,NPJ Digital Medicine6, p. 120 (2023)

    work page 2023

  5. [5]

    D. J. Morgan, W. Branch-Elliman and K. E. Goodman, Time to study implementation of AI- generated discharge summaries,JAMA Internal Medicine(2025)

    work page 2025

  6. [6]

    C. Hong, A. Chowdhury, A. D. Sorrentino, H. Wang, M. Agrawal, A. Bedoya, S. Bessias, N. J. Economou-Zavlanos, I. Wong, C. Peanet al., Application of unified health large language model evaluation framework to in-basket message replies: bridging qualitative and quantitative assess- ments,Journal of the American Medical Informatics Association32, 626 (2025)

    work page 2025

  7. [7]

    Asgari, N

    E. Asgari, N. Monta˜ na-Brown, M. Dubois, S. Khalil, J. Balloch, J. A. Yeung and D. Pimenta, A framework to assess clinical safety and hallucination rates of LLMs for medical text summari- sation,NPJ Digital Medicine8, p. 274 (2025)

    work page 2025

  8. [8]

    Zuo and Y

    K. Zuo and Y. Jiang, MedHallBench: A new benchmark for assessing hallucination in medical large language models,arXiv preprint arXiv:2412.18947(2024)

    arXiv 2024

Show all 15 references
  1. [9]

    Chung, A

    P. Chung, A. Swaminathan, A. J. Goodell, Y. Kim, S. M. Reincke, L. Han, B. Deverett, M. A. Sadeghi, A.-B. Ariss, M. Ghanemet al., Verifact: Verifying facts in LLM-generated clinical text with electronic health records,arXiv preprint arXiv:2501.16672(2025)

    2025 arXiv

  2. [10]

    Munnangi, A

    M. Munnangi, A. Swaminathan, J. A. Fries, J. Jindal, S. Narayanan, I. Lopez, L. Tu, P. Chung, J. A. Omiye, M. Kashyapet al., Assessing the limitations of large language models in clinical fact decomposition,arXiv preprint arXiv:2412.12422(2024)

    2024

  3. [11]

    L. Tang, Z. Sun, B. Idnay, J. G. Nestor, A. Soroush, P. A. Elias, Z. Xu, Y. Ding, G. Durrett, J. F. Rousseauet al., Evaluating large language models on medical evidence summarization, NPJ Digital Medicine6, p. 158 (2023)

    2023

  4. [12]

    Zheng, W.-L

    L. Zheng, W.-L. Chiang, Y. Sheng, S. Zhuang, Z. Wu, Y. Zhuang, Z. Lin, Z. Li, D. Li, E. Xing et al., Judging LLM-as-a-judge with MT-bench and chatbot arena,Advances in Neural Infor- mation Processing Systems36, 46595 (2023)

    2023

  5. [13]

    M. Y. Ng, J. Helzer, M. A. Pfeffer, T. Seto and T. Hernandez-Boussard, Development of secure infrastructure for advancing generative artificial intelligence research in healthcare at an academic medical center,Journal of the American Medical Informatics Association32, 586 (2025)

    2025

  6. [14]

    J. R. Landis and G. G. Koch, The measurement of observer agreement for categorical data, Biometrics, 159 (1977)

    1977

  7. [15]

    S. Bedi, H. Cui, M. Fuentes, A. Unell, M. Wornow, J. M. Banda, N. Kotecha, T. Keyes, Y. Mai, M. Oezet al., MedHELM: Holistic evaluation of large language models for medical tasks,arXiv preprint arXiv:2505.23802(2025)

    2025 arXiv

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.