Pith. sign in

REVIEW 3 major objections 5 minor 72 references

Unified text serialization matches or beats specialized multimodal clinical models

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-01 23:30 UTC pith:JDEWRNQM

load-bearing objection Consistent benchmark evidence that unified text serialization of multimodal EHR data works across three tasks, but the headline graft-failure win over the deployed GBDT is a 0.01 AUROC gap with no error bars. the 3 major comments →

arxiv 2607.15380 v1 pith:JDEWRNQM submitted 2026-07-16 cs.CL cs.AI

Large Language Models as Unified Multimodal Learners for Clinical Prediction

Ajay Madhavan Ravichandran , Bilgin Osmandoja , Klemens Budde , Klaus Netter , Tobias Strapatsas , Aljoscha Burchardt , Sebastian M\"oller , Roland Roller This is my paper
classification cs.CL cs.AI
keywords clinical predictionelectronic health recordsmultimodal fusiontext serializationlarge language modelsfine-tuninggraft failuretriage
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

This paper argues that the standard way of building clinical prediction systems — separate encoders for free-text notes and structured measurements, combined by learned fusion layers — is unnecessary. It claims that converting every patient variable into a single natural-language sequence and fine-tuning a pretrained language model end-to-end matches or exceeds those task-specific multimodal systems on three different clinical tasks. The tasks are ICU mortality, kidney transplant graft failure, and emergency triage. The strongest concrete claim: the serialized language model beats the gradient-boosting system currently used in the transplant clinic at predicting graft failure. If correct, the result simplifies clinical machine-learning infrastructure, since one pipeline replaces modality-specific preprocessing and fusion designs.

Core claim

The central claim is that a single serialization-based pipeline is sufficient for multimodal clinical prediction. All structured data—vital signs, labs, demographics, comorbidities—is written as key–value text and concatenated with the clinical narrative; a pretrained language model is then fine-tuned end-to-end with no architectural change for fusion. Across three tasks, the authors report that this unified text approach matches or exceeds dedicated fusion baselines, and that two of their models, an encoder and a decoder, outperform the deployed gradient-boosting model on graft failure.

What carries the argument

Unified multimodal text serialization: converting structured EHR variables into key–value textual pairs (e.g., 'Heart Rate: 76.09, 78.75') and concatenating them with the free-text note to form one input sequence. The language model's self-attention is what does the fusion; no gated attention, cross-modal transformer, or separate encoders are used. The claim is that attention alone is enough to learn cross-modal relationships.

Load-bearing premise

The specialized fusion baselines are strong, properly tuned representatives of their paradigm; if they are underpowered, the claim that serialization 'matches or exceeds' specialized designs has not actually been established.

What would settle it

Run the three tasks again with the same serialized pipeline against fusion baselines that have been given the same tuning budget and report repeated-run confidence intervals; if a well-tuned fusion model beats serialization on even one task by a margin larger than the run-to-run variation, the universality claim fails. For graft failure specifically, a prospective evaluation against the deployed gradient-boosting system on a new chronological test set would test the outperformance claim.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Clinical prediction systems can be built with a single language-model pipeline instead of task-specific fusion architectures, cutting engineering overhead.
  • Models that have never seen a modality such as time-series vitals can still use it once it is written as text, so new data types do not require new encoders.
  • On graft failure, the serialized LLM matched or surpassed a gradient-boosting model already in clinical use, suggesting the approach is viable for real deployment.
  • Across all three tasks, combining text and structured features improved over using either alone, so the two modalities carry complementary signal even in text form.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • If the parity result holds under stricter baselines, the bottleneck in clinical prediction shifts from architecture design to serialization format and context-window management.
  • The token overhead of serialization is likely to become the main practical limit on very long patient histories; efficient serialization and summarization are the natural next targets.
  • A testable extension is to apply the same pipeline to multimodal tasks outside medicine, like tabular-plus-text business or sensor data, to see whether the pattern is general.
  • The paper's lack of statistical significance tests leaves open that the reported differences could be noise; a re-analysis with repeated runs and confidence intervals would settle this.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper argues that a single text-serialization pipeline—encoding both free-text notes and structured variables as key–value text—can replace task-specific multimodal fusion architectures for clinical prediction. It fine-tunes ModernBERT and several decoder LLMs on three tasks (MIMIC-III mortality, transplant graft failure, triage), comparing against unimodal and fusion baselines and a gradient-boosting model used in post-transplant care. The central claim is that serialized LLMs match or exceed modality-specific fusion baselines on all three tasks and outperform the clinical GBDT on graft failure.

Significance. If the empirical claims hold, the result is practically significant: a single pretrained language model, fine-tuned on serialized text, would remove the need for bespoke fusion encoders across different EHR prediction tasks. The evaluation is broad in task diversity and model coverage, and the authors include a production-like clinical baseline. However, the headline comparisons are reported as point estimates without confidence intervals, significance tests, or repeated-seed variation, and the graft-failure baseline comparison is under-specified. These omissions directly affect the strongest contribution claim, so the significance of the result is currently conditional.

major comments (3)
  1. [Section 6.3, Table 4; Abstract] The statement that serialized LLMs 'outperform the clinically deployed gradient boosting system' is supported by a 0.01 AUROC difference (0.90 vs. 0.89) and AUPR differences of 0.45/0.47 vs. 0.43. No confidence intervals, bootstrap resampling, DeLong tests, or repeated-seed standard deviations are reported anywhere in the paper. A 0.01 AUROC gap is well within typical model variability for clinical data, so the superiority claim is not statistically established as reported. Please add uncertainty quantification (e.g., bootstrap CIs and significance tests on the test set, and seed variance for fine-tuned models).
  2. [Section 5.1, Graft Failure baseline] The GBDT is described as 'used already in previous work (Roller et al., 2022)' and 'currently evaluated within a clinical study,' but the paper does not state whether it was retrained on the same chronological split, the same feature preprocessing, and the same 360-day outcome definition used for the LLMs. If the 0.89 AUROC is inherited from a prior evaluation on a different data partition or feature set, the comparison is not apples-to-apples. Please specify the retraining protocol or retrain the baseline on the identical train/validation/test split.
  3. [Section 5.1, Tables 2–4] The baseline models are described in one sentence each, with no tuning budget, number of random seeds, or best-model selection procedure. The triage baseline is 'introduced in more deapth in previous work' and the fusion baseline is a gated fusion implementation, but no hyperparameter search is reported. Since the paper's central claim is that serialization 'matches or exceeds' task-specific multimodal baselines, the strength of these baselines is load-bearing. Please provide full baseline tuning details, repeated runs, and statistical comparisons; otherwise the claim reduces to beating particular point-estimate implementations.
minor comments (5)
  1. [Section 4.3 vs Tables 2–4] Model names/sizes are inconsistent. Section 4.3 lists Gemma 2B and DeepSeek-R1-Qwen 8B, while Tables 2 and 4 report 'Gemma-2-9B' and Table 3 reports 'Llama3.2-3B'. Please align model identifiers and parameter counts throughout.
  2. [After Section 8] The 'Limitations' heading after the Conclusion is empty; either remove it or fold the discussion in Section 7 into a formal limitations subsection.
  3. [Appendix A] The instruction-prompt appendix appears after the references and is not referenced from the main text. Add a reference and move the appendix before the bibliography or cite it in the text.
  4. [Reproducibility] No data or code availability statement is included. For the public MIMIC-III results, releasing preprocessing and serialization code would aid reproducibility; for the private datasets, at least a data-sharing statement is needed.
  5. [Section 5.2] 'Self-supervised fine-tuning (SFT)' is a misnomer for supervised fine-tuning; the text elsewhere correctly calls it 'supervised instruction-style training' (Section 4.3). Please correct this.

Circularity Check

0 steps flagged

No significant circularity: the paper is an empirical benchmark with held-out evaluation; self-citations to baseline systems are contextual, not load-bearing.

full rationale

This paper is an empirical benchmark rather than a derivation chain. The proposed method is a serialization procedure, and the prediction targets are labels from held-out test sets: MIMIC-III uses a 15% held-out test split, graft failure uses a chronological 20% test split, and triage uses a stratified 10% test split. There is no equation in which a fitted parameter is defined as the target quantity, no ansatz imported via citation, and no uniqueness claim. The main comparisons use existing baselines, some from co-authored prior work (Maschhur et al. 2024; Roller et al. 2022; Osmanodja et al. 2024), but these citations support the existence and clinical deployment of the baselines, not the outcome of the comparison; the performance numbers are reported from the current experiments. The absence of confidence intervals and significance tests around the 0.01 AUROC difference with the GBDT is a statistical robustness issue, not a circularity issue, because the result does not reduce to a fit or to the self-citations by construction. Therefore no significant circularity is present.

Axiom & Free-Parameter Ledger

4 free parameters · 4 axioms · 0 invented entities

No new physical or conceptual entities are introduced; the ledger instead records hand-designed representation choices and unvalidated domain assumptions.

free parameters (4)
  • Serialization format and feature ordering
    Key-value templates and the order/grouping of structured variables (Figures 1-3) are hand-designed; no ablation shows their effect on results.
  • Temporal aggregation window and number of serialized time-points
    Section 4.2 says values are 'aggregated over the observation window', but Figure 2 shows four raw values per vital sign; the choice affects token length and information content.
  • Fine-tuning hyperparameters
    Learning rates, batch sizes, epochs, LoRA settings, and decoding details are not reported; results depend on these choices.
  • Model selection
    Tables report many model variants; the abstract emphasizes the best configurations, but the selection rule across models/checkpoints is not specified.
axioms (4)
  • domain assumption Textual serialization of structured variables preserves the clinically relevant information (no information loss beyond what preprocessing already discards).
    Central to the method; asserted in Section 4.2 but no controlled comparison against a lossless numeric representation is provided.
  • domain assumption The fusion baselines are representative of the task-specific multimodal paradigm.
    Section 5.1 lists only one fusion baseline per task; no evidence they are state-of-the-art or tuned comparably.
  • domain assumption Labels in the private datasets (graft failure, MTS triage) are reliable and follow the stated clinical definitions.
    No inter-annotator agreement, label audit, or external validation is reported for the two German datasets.
  • standard math Standard supervised fine-tuning of a decoder LM to output class tokens is a valid classification procedure.
    Section 4.4 uses next-token prediction; this is common practice but the paper does not ablate prompt/answer formatting.

pith-pipeline@v1.3.0-alltime-deepseek · 12006 in / 12393 out tokens · 111159 ms · 2026-08-01T23:30:43.691265+00:00 · methodology

0 comments
read the original abstract

Electronic health records combine free-text clinical narratives with structured measurements such as vital signs, laboratory values, and comorbidities. Yet most clinical prediction systems still rely on task-specific fusion architectures, pairing dedicated encoders for each modality with learned combination mechanisms that must be re-engineered for every new task and clinical setting. We propose a simpler alternative: convert all patient data, regardless of modality, into a single natural language sequence and fine-tune a pretrained language model end-to-end, with no architectural modification for fusion. We evaluate this approach across three clinically distinct prediction tasks: in-hospital mortality on MIMIC-III, graft failure prediction using longitudinal data from a German transplant center, and emergency triage classification from ambulance records - comparing encoder-based (ModernBERT) and decoder-based (Llama 3.1, Gemma, DeepSeek-R1-Qwen, Qwen3) fine-tuning against established multimodal baselines and, for graft failure, a gradient boosting model currently used in clinical practice for post-transplant patient management. Across all three tasks, unified textual serialization matches or exceeds task-specific multimodal baselines, and outperforms the clinically deployed gradient boosting system on graft failure prediction. These results indicate that a single serialization-based paradigm, without bespoke fusion architectures, is sufficient for multimodal clinical prediction - substantially reducing system complexity while matching or exceeding specialized designs.

Figures

Figures reproduced from arXiv: 2607.15380 by Ajay Madhavan Ravichandran, Aljoscha Burchardt, Bilgin Osmandoja, Klaus Netter, Klemens Budde, Roland Roller, Sebastian M\"oller, Tobias Strapatsas.

Figure 2
Figure 2. Figure 2: Example of a serialized input with text, time [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 1
Figure 1. Figure 1: Example of a serialized input with text and [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 3
Figure 3. Figure 3: German prompt use for instruction tuning to predict graft failure. [PITH_FULL_IMAGE:figures/full_fig_p011_3.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

72 extracted references · 3 canonical work pages

  1. [1]

    Emily Alsentzer, John R Murphy, Willie Boag, et al. 2019. Publicly available clinical BERT embeddings. In Proceedings of the 2nd Clinical Natural Language Processing Workshop (ACL), pages 72--78

    2019

  2. [2]

    Shebesh, Jing Su, Paul R

    Zina Ben-Miled, Jacob A. Shebesh, Jing Su, Paul R. Dexter, Randall W. Grout, and Malaz A. Boustani. 2025. https://doi.org/10.3390/info16010054 Multi-modal fusion of routine care electronic health records (ehr): A scoping review . Information, 16(1)

    work page doi:10.3390/info16010054 2025

  3. [3]

    Ofir Ben Shoham and Nadav Rappoport. 2024. https://doi.org/10.1371/journal.pdig.0000680 CPLLM : Clinical prediction with large language models . PLOS Digital Health, 3(12):e0000680

    doi:10.1371/journal.pdig.0000680 2024

  4. [5]

    Hejie Cui, Alyssa Unell, Bowen Chen, Jason Alan Fries, Emily Alsentzer, Sanmi Koyejo, and Nigam Shah. 2025. TIMER : Temporal instruction modeling and evaluation for longitudinal clinical records. arXiv preprint

    2025

  5. [7]

    Tuan Dinh, Yuchen Zeng, Ruisu Zhang, et al. 2022. LIFT : Language-interfaced fine-tuning with non-language labels. In Advances in Neural Information Processing Systems (NeurIPS)

    2022

  6. [10]

    Jifan Gao, Mahmudur Rahman, John Caskey, Madeline Oguss, Ann O'Rourke, Randy Brown, Anne Stey, Anoop Mayampurath, Matthew M Churpek, Guanhua Chen, and Majid Afshar. 2025. https://doi.org/10.1038/s41746-025-02219-4 MoMA : A mixture-of-multimodal-agents architecture for enhancing clinical prediction modelling . npj Digital Medicine. ArXiv:2508.05492, publis...

    Pith/arXiv arXiv 2025

  7. [12]

    Stefan Hegselmann, Alejandro Buendia, Hunter Lang, Monica Agrawal, Xiaoyi Jiang, and David Sontag. 2023. TabLLM : Few-shot classification of tabular data with large language models. In International Conference on Artificial Intelligence and Statistics (AISTATS)

    2023

  8. [14]

    Alistair E W Johnson, Tom J Pollard, Lu Shen, Li-wei H Lehman, Mengling Feng, Mohammad Ghassemi, Benjamin Moody, Peter Szolovits, Leo Anthony Celi, and Roger G Mark. 2016. MIMIC-III , a freely accessible critical care database. Scientific Data, 3:160035

    2016

  9. [15]

    Swaraj Khadanga, Karan Aggarwal, Shafiq Joty, and Jaideep Srivastava. 2019. Using clinical notes with time series data for ICU management. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 6432--6437

    2019

  10. [17]

    Jinhyuk Lee, Wonjin Yoon, Sungdong Kim, et al. 2020. BioBERT : a pre-trained biomedical language representation model for biomedical text mining. Bioinformatics, 36(4):1234--1240

    2020

  11. [19]

    Nikita Makarov, Maria Bordukova, Papichaya Quengdaeng, Daniel Garger, Raul Rodriguez-Esteban, Fabian Schmich, and Michael P Menden. 2025. https://doi.org/10.1038/s41746-025-02004-3 Large language models forecast patient health trajectories enabling digital twins . npj Digital Medicine, 8:588

    doi:10.1038/s41746-025-02004-3 2025

  12. [21]

    Harsha Nori, Nicholas King, Scott Mayer McKinney, Dean Carignan, and Eric Horvitz. 2023. Capabilities of GPT-4 on medical challenge problems. arXiv preprint arXiv:2303.13375

    Pith/arXiv arXiv 2023

  13. [22]

    Bilgin Osmanodja, Zeineb Sassi, Sascha Eickmann, Carla Maria Hansen, Roland Roller, Aljoscha Burchardt, David Samhammer, Peter Dabrock, Sebastian M \"o ller, Klemens Budde, and Anne Herrmann. 2024. Investigating the impact of ai on shared decision-making in post-kidney transplant care (prima-ai): Protocol for a randomized controlled trial. JMIR Research P...

    2024

  14. [23]

    Chantal Pellegrini, Ege Özsoy, David Bani-Harouni, Matthias Keicher, and Nassir Navab. 2026. https://arxiv.org/abs/2506.04831 Ehr2path: Scalable modeling of longitudinal patient pathways from multimodal electronic health records . Preprint, arXiv:2506.04831

    arXiv 2026

  15. [24]

    Alvin Rajkomar, Eyal Oren, Kai Chen, et al. 2018. Scalable and accurate deep learning with electronic health records. npj Digital Medicine, 1(1):18

    2018

  16. [25]

    Roland Roller, Manuel Mayrdorfer, Wiebke Duettmann, Marcel G Naik, Danilo Schmidt, Fabian Halleck, Patrik Hummel, Aljoscha Burchardt, Sebastian M \"o ller, Peter Dabrock, Bilgin Osmanodja, and Klemens Budde. 2022. https://doi.org/10.3389/fpubh.2022.979448 Evaluation of a clinical decision support system for detection of patients at risk after kidney trans...

    arXiv 2022

  17. [27]

    Karan Singhal, Shekoofeh Azizi, Tao Tu, et al. 2023. Large language models encode clinical knowledge. Nature, 620:172--180

    2023

  18. [28]

    Luis R Soenksen, Yu Ma, Cynthia Zeng, et al. 2022. Integrated multimodal artificial intelligence framework for healthcare applications. npj Digital Medicine, 5(1):149

    2022

  19. [30]

    Qwen Team. 2025. https://qwenlm.github.io/blog/qwen3/ Qwen3

    2025

  20. [31]

    Benjamin Warner, Antoine Chaffin, Benjamin Clavi \'e , Orion Weller, Oskar Hallstr \"o m, Said Taghadouini, Alexis Gallagher, Raja Biswas, Faisal Ladhak, Tom Aarsen, et al. 2025. Smarter, better, faster, longer: A modern bidirectional encoder for fast, memory efficient, and long context finetuning and inference. In Proceedings of the 63rd Annual Meeting o...

    2025

  21. [32]

    Zhenbang Wu, Anant Dadu, Mike Nalls, Faraz Faghri, and Jimeng Sun. 2024. Instruction tuning large language models to understand electronic health records. In Proceedings of the 38th International Conference on Neural Information Processing Systems, NIPS '24, Red Hook, NY, USA. Curran Associates Inc

    2024

  22. [34]

    Yusuf Yildiz, Goran Nenadic, Meghna Jani, and David A Jenkins. 2025. https://doi.org/10.1186/s41512-025-00211-w Will large language models transform clinical prediction? Diagnostic and Prognostic Research, 9:28

    work page doi:10.1186/s41512-025-00211-w 2025

  23. [35]

    Frontiers in Public Health , volume=

    Roller, Roland and Mayrdorfer, Manuel and Duettmann, Wiebke and Naik, Marcel G and Schmidt, Danilo and Halleck, Fabian and Hummel, Patrik and Burchardt, Aljoscha and M. Frontiers in Public Health , volume=. 2022 , publisher=

    2022

  24. [36]

    doi:10.18653/v1/2024.bionlp-1.46 , year=

    Maschhur, Faraz and Netter, Klaus and Schmeier, Sven and Ostermann, Katrin and Palunis, Rimantas and Strapatsas, Tobias and Roller, Roland , booktitle=. doi:10.18653/v1/2024.bionlp-1.46 , year=

    work page doi:10.18653/v1/2024.bionlp-1.46 2024

  25. [37]

    JMIR Research Protocols , volume=

    Investigating the Impact of AI on Shared Decision-Making in Post-Kidney Transplant Care (PRIMA-AI): Protocol for a Randomized Controlled Trial , author=. JMIR Research Protocols , volume=. 2024 , publisher=

    2024

  26. [38]

    Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages=

    Smarter, better, faster, longer: A modern bidirectional encoder for fast, memory efficient, and long context finetuning and inference , author=. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages=

  27. [39]

    arXiv preprint arXiv:2407.21783 , year=

    The llama 3 herd of models , author=. arXiv preprint arXiv:2407.21783 , year=

    Pith/arXiv arXiv

  28. [40]

    arXiv preprint arXiv:2408.00118 , year=

    Gemma 2: Improving open language models at a practical size , author=. arXiv preprint arXiv:2408.00118 , year=

    Pith/arXiv arXiv

  29. [41]

    arXiv preprint arXiv:2501.12948 , year =

    Pith/arXiv arXiv

  30. [42]

    Qwen3 , url =

    Qwen Team , month =. Qwen3 , url =

  31. [43]

    2016 , publisher =

    Johnson, Alistair E W and Pollard, Tom J and Shen, Lu and Lehman, Li-wei H and Feng, Mengling and Ghassemi, Mohammad and Moody, Benjamin and Szolovits, Peter and Celi, Leo Anthony and Mark, Roger G , journal =. 2016 , publisher =

    2016

  32. [44]

    npj Digital Medicine , volume =

    Scalable and accurate deep learning with electronic health records , author =. npj Digital Medicine , volume =. 2018 , publisher =

    2018

  33. [45]

    npj Digital Medicine , volume =

    Integrated multimodal artificial intelligence framework for healthcare applications , author =. npj Digital Medicine , volume =. 2022 , publisher =

    2022

  34. [46]

    Using Clinical Notes with Time Series Data for

    Khadanga, Swaraj and Aggarwal, Karan and Joty, Shafiq and Srivastava, Jaideep , booktitle =. Using Clinical Notes with Time Series Data for

  35. [47]

    Proceedings of the 30th ACM International Conference on Multimedia , year =

    Multi-Modal Masked Autoencoders for Medical Vision-and-Language Pre-Training , author =. Proceedings of the 30th ACM International Conference on Multimedia , year =

  36. [48]

    Learning to Diagnose with

    Lipton, Zachary C and Kale, David C and Elkan, Charles and Wetzel, Randall , journal =. Learning to Diagnose with

  37. [49]

    Devlin, Jacob and Chang, Ming-Wei and Lee, Kenton and Toutanova, Kristina , journal =

  38. [50]

    Advances in Neural Information Processing Systems (NeurIPS) , volume =

    Language Models are Few-Shot Learners , author =. Advances in Neural Information Processing Systems (NeurIPS) , volume =

  39. [51]

    International Conference on Learning Representations (ICLR) , year =

    Finetuned Language Models Are Zero-Shot Learners , author =. International Conference on Learning Representations (ICLR) , year =

  40. [52]

    Hu, Edward J and Shen, Yelong and Wallis, Phillip and others , booktitle =

  41. [53]

    Publicly Available Clinical

    Alsentzer, Emily and Murphy, John R and Boag, Willie and others , booktitle =. Publicly Available Clinical

  42. [54]

    Lee, Jinhyuk and Yoon, Wonjin and Kim, Sungdong and others , journal =

  43. [55]

    ACM Transactions on Computing for Healthcare , volume =

    Domain-Specific Language Model Pretraining for Biomedical Natural Language Processing , author =. ACM Transactions on Computing for Healthcare , volume =

  44. [56]

    Nature , volume =

    Large language models encode clinical knowledge , author =. Nature , volume =

  45. [57]

    Capabilities of

    Nori, Harsha and King, Nicholas and McKinney, Scott Mayer and Carignan, Dean and Horvitz, Eric , journal =. Capabilities of

  46. [58]

    Hegselmann, Stefan and Buendia, Alejandro and Lang, Hunter and Agrawal, Monica and Jiang, Xiaoyi and Sontag, David , booktitle =

  47. [59]

    Dinh, Tuan and Zeng, Yuchen and Zhang, Ruisu and others , booktitle =

  48. [60]

    Slack, Dylan and Krishna, Satyapriya and Lakkaraju, Himabindu and Singh, Sameer , booktitle =

  49. [61]

    BioNLP Workshop at ACL , year =

    Transfer Learning in Biomedical Natural Language Processing , author =. BioNLP Workshop at ACL , year =

  50. [62]

    Health Data Science , year =

    Large Language Models in Health Care: Development, Applications, and Challenges , author =. Health Data Science , year =

  51. [63]

    arXiv preprint arXiv:2403.08818 , year =

    Cui, Hejie and Fang, Xueting and Xu, Ran and Kan, Xuan and Ho, Joyce C and Yang, Carl , title =. arXiv preprint arXiv:2403.08818 , year =

    Pith/arXiv arXiv

  52. [64]

    arXiv preprint arXiv:2501.12275 , year =

    Ruan, Yinghao and Tan, Darren J and Huang, Luning and Feng, Mengling , title =. arXiv preprint arXiv:2501.12275 , year =

    Pith/arXiv arXiv

  53. [65]

    npj Digital Medicine , year =

    Gao, Jifan and Rahman, Mahmudur and Caskey, John and Oguss, Madeline and O'Rourke, Ann and Brown, Randy and Stey, Anne and Mayampurath, Anoop and Churpek, Matthew M and Chen, Guanhua and Afshar, Majid , title =. npj Digital Medicine , year =

  54. [66]

    arXiv preprint arXiv:2601.10092 , year =

    Kim, Jongseok and Kang, Seongae and Shin, Jonghwan and Lee, Yuhan and Jo, Ohyun , title =. arXiv preprint arXiv:2601.10092 , year =

    arXiv

  55. [67]

    npj Digital Medicine , volume =

    Wornow, Michael and Xu, Yizhe and Thapa, Rahul and Patel, Birju and Steinberg, Ethan and Fleming, Scott and Pfeffer, Michael A and Fries, Jason and Shah, Nigam H , title =. npj Digital Medicine , volume =. 2023 , doi =

    2023

  56. [68]

    and Su, Jing and Dexter, Paul R

    Ben-Miled, Zina and Shebesh, Jacob A. and Su, Jing and Dexter, Paul R. and Grout, Randall W. and Boustani, Malaz A. , TITLE =. Information , VOLUME =. 2025 , NUMBER =

    2025

  57. [69]

    arXiv preprint arXiv:2203.03540 , year =

    Yang, Xi and Chen, Aokun and PourNejatian, Nima and Shin, Hoo Chang and Smith, Kaleb E and Parisien, Christopher and Compas, Colin and Martin, Cheryl and Flores, Mona G and Zhang, Ying and others , title =. arXiv preprint arXiv:2203.03540 , year =

    Pith/arXiv arXiv

  58. [70]

    PLOS Digital Health , volume =

    Ben Shoham, Ofir and Rappoport, Nadav , title =. PLOS Digital Health , volume =. 2024 , doi =

    2024

  59. [71]

    Proceedings of the 38th International Conference on Neural Information Processing Systems , articleno =

    Wu, Zhenbang and Dadu, Anant and Nalls, Mike and Faghri, Faraz and Sun, Jimeng , title =. Proceedings of the 38th International Conference on Neural Information Processing Systems , articleno =. 2024 , isbn =

    2024

  60. [72]

    arXiv preprint arXiv:2502.17403 , year =

    Hegselmann, Stefan and von Arnim, Georg and Rheude, Tillmann and Kronenberg, Noel and Sontag, David and Hindricks, Gerhard and Eils, Roland and Wild, Benjamin , title =. arXiv preprint arXiv:2502.17403 , year =

    Pith/arXiv arXiv

  61. [73]

    npj Digital Medicine , volume =

    Makarov, Nikita and Bordukova, Maria and Quengdaeng, Papichaya and Garger, Daniel and Rodriguez-Esteban, Raul and Schmich, Fabian and Menden, Michael P , title =. npj Digital Medicine , volume =. 2025 , doi =

    2025

  62. [74]

    arXiv preprint , year =

    Cui, Hejie and Unell, Alyssa and Chen, Bowen and Fries, Jason Alan and Alsentzer, Emily and Koyejo, Sanmi and Shah, Nigam , title =. arXiv preprint , year =

  63. [75]

    Diagnostic and Prognostic Research , volume =

    Yildiz, Yusuf and Nenadic, Goran and Jani, Meghna and Jenkins, David A , title =. Diagnostic and Prognostic Research , volume =. 2025 , doi =

    2025

  64. [76]

    arXiv preprint arXiv:2402.17944 , year =

    Fang, Xi and Xu, Weijie and Tan, Fiona Anting and Zhang, Jiani and Hu, Ziqing and Qi, Yanjun and Nickleach, Scott and Socolinsky, Diego and Sengamedu, Srinivasan and Faloutsos, Christos , title =. arXiv preprint arXiv:2402.17944 , year =

    Pith/arXiv arXiv

  65. [77]

    2026 , eprint=

    EHR2Path: Scalable Modeling of Longitudinal Patient Pathways from Multimodal Electronic Health Records , author=. 2026 , eprint=

    2026

  66. [78]

    arXiv preprint arXiv:2509.25591 , year =

    others , title =. arXiv preprint arXiv:2509.25591 , year =

    arXiv

  67. [79]

    arXiv preprint arXiv:2402.00160 , year =

    Lee, Siena A and others , title =. arXiv preprint arXiv:2402.00160 , year =

    Pith/arXiv arXiv

  68. [80]

    Advances in Neural Information Processing Systems (NeurIPS) , volume =

    Brown, Tom B and Mann, Benjamin and Ryder, Nick and Subbiah, Melanie and Kaplan, Jared and Dhariwal, Prafulla and Neelakantan, Arvind and Shyam, Pranav and Sastry, Girish and Askell, Amanda and others , title =. Advances in Neural Information Processing Systems (NeurIPS) , volume =

  69. [81]

    International Conference on Learning Representations (ICLR) , year =

    Wei, Jason and Bosma, Maarten and Zhao, Vincent Y and Guu, Kelvin and Yu, Adams Wei and Lester, Brian and Du, Nan and Dai, Andrew M and Le, Quoc V , title =. International Conference on Learning Representations (ICLR) , year =

  70. [82]

    International Conference on Learning Representations (ICLR) , year =

    Hu, Edward J and Shen, Yelong and Wallis, Phillip and Allen-Zhu, Zeyuan and Li, Yuanzhi and Wang, Shean and Wang, Lu and Chen, Weizhu , title =. International Conference on Learning Representations (ICLR) , year =

  71. [84]

    Diagnostic and Prognostic Research , publisher =

    Will Large Language Models Transform Clinical Prediction? , author =. Diagnostic and Prognostic Research , publisher =. 2025 , note =

    2025

  72. [85]

    arXiv preprint arXiv:2604.01924 , year =

    El Khettari, Oumaima and others , title =. arXiv preprint arXiv:2604.01924 , year =

    arXiv