Pith. sign in

REVIEW 4 major objections 4 minor 37 references

CKD-EHR:Clinical Knowledge Distillation for Electronic Health Records

T0 review · 4 major / 4 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 The paper claims that clinical knowledge can be distilled from a 7-billion-parameter LLM into a compact BERT model, improving MIMIC-III disease prediction while cutting inference time by 22.2 times.

desk verdict Plausible pipeline, but EADF likely leaks the next-visit diagnosis into the input, so the reported gains cannot be trusted. read the letter →

arxiv 2506.15118 v1 pith:OE6BRWYD submitted 2025-06-18 cs.CL

Junke Wang , Hongshun Ling , Li Zhang , Longqian Zhang , Fang Wang , Yuan Gao , Zhi Li This is my paper · ORCID
classification cs.CL
keywords clinicalknowledgedistillationelectronichealthrecordsdiseasepredictionMIMIC-IIIlargelanguagemodelsLoRAfine-tuningsoftlabelsmulti-labelclassification
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 tries to establish that a large medical language model can be turned into a small, fast, and accurate disease-prediction model through a three-stage distillation framework called CKD-EHR. A Qwen2.5-7B teacher is first fine-tuned with LoRA on patient records enriched by Efficacy-Aware Data Fusion, which ranks medication and procedure combinations by their observed effect on the next visit's diagnosis. The teacher's hidden states are projected through a multi-label head into 25 disease probabilities that serve as soft labels, and a BERT student is trained on a weighted mix of these soft labels and the true labels. On MIMIC-III the student reaches 94.76% accuracy and 80.25 F1, roughly 9% higher accuracy and 27% higher F1 than the retrieval-augmented baseline it is compared against, while running 22.2 times faster and using about 35 times less memory than the 7B teacher. The broader point the authors want a reader to accept is that distilled soft knowledge plus real clinical treatment signals can make lightweight models practical for clinical decision support.

What carries the argument

The load-bearing mechanism is the trio of EADF, LoRCKD, and MLAPH. EADF converts pairs of consecutive visits into natural-language records and attaches a rank q of the most effective treatment for the current diagnosis, computed statistically from the dataset, thereby injecting an efficacy signal. LoRCKD freezes Qwen2.5-7B's weights and trains only LoRA low-rank matrices in the attention modules, then extracts 25-dimensional soft-label probabilities via MLAPH, a linear projection of pooled hidden states. The BERT student minimizes BCEWithLogitsLoss on a weighted combination of hard and soft labels, so the knowledge transfer path runs from real outcomes to ranked efficacies to teacher hidden states to soft probabilities to student logits. MLAPH is what avoids the standard vocabulary-logit KL divergence, which would be expensive and noisy for generative models.

What would settle it

Re-run all models in Table 1 from a single MIMIC-III phenotyping split with identical preprocessing, and check whether CKD-EHR's ACC, F1, AUC, and AUPR gaps over RAM-EHR and ClinicalBERT persist; if the F1 advantage over the best clinical BERT baseline drops below a few points, the central claim of large-model knowledge transfer is not supported.

Watch

Extended reading notes

Core claim

The central discovery is that the distilled student can carry the teacher's clinical performance: CKD-EHR reaches ACC 94.76, F1 80.25, AUC 91.11, and AUPR 76.33 on the MIMIC-III 25-phenotype task, outperforming all listed baselines including ClinicalBERT, BioBERT, and Bio+ClinicalBERT. The largest gap is against RAM-EHR (ACC 85.54, F1 53.01), matching the abstract's stated 9% accuracy and 27% F1 improvements. The ablation shows that EADF alone gives the largest F1 lift (81.61 to 82.94 on the teacher), while adding LoRCKD trades a little F1 for large efficiency gains; the full student model, with hard-label weight α=0.9, sits at 80.25 F1. The authors interpret this as evidence that augmentation built from the target EHR's own visit outcomes beats retrieving generic text, and that soft labels produced from the teacher's hidden states transfer richer uncertainty and inter-label dependencies than hard labels alone.

Load-bearing premise

The central comparison assumes every baseline in Table 1 was trained and evaluated on the same MIMIC-III train/test division under comparable settings, yet the paper (Sections 4.2.1 and 4.3.1) does not state that this equivalence was enforced.

Editorial extensions

If this is right

  • If the results hold, a hospital can deploy BERT-scale models that deliver most of a 7B model's disease-prediction accuracy on standard hardware.
  • The 22.2x inference speedup and 34.96x memory reduction make real-time screening on large EHR populations feasible without large GPU clusters.
  • Building the augmentation knowledge from the dataset's own visit outcomes, rather than web text, gives distillation access to treatment-effect signals that RAG-style retrieval does not provide.
  • Soft labels produced in the fixed disease-label space via MLAPH let the student learn inter-label dependencies and uncertainty, which is why the full method beats hard-label-only training in the α=0.9 setting.

Reading between the lines

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

  • Editorial inference: The headline gains are measured against published baseline numbers; if every baseline were retrained on the exact CKD-EHR split, the ranking could shift even if the method's own results stand.
  • Editorial inference: EADF's efficacy ranking is derived from observed visit transitions, so it is associational rather than causal; the pipeline would be a stronger clinical tool if the ranking were validated against randomized or adjusted treatment effects.
  • Editorial inference: The same teacher–projection-head–student recipe should transfer to other structured prediction tasks such as MIMIC-IV phenotyping or length-of-stay prediction; a cheap test is whether F1 gains persist when the teacher is a smaller instruction-tuned model.
  • Editorial inference: The paper itself limits the claim to one dataset and 25 phenotypes; extending to multimodal data or thousands of codes is not established and would face label-sparsity problems like the false negatives seen for low-support diseases.
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 / 4 minor

Summary. The paper proposes CKD-EHR, a knowledge-distillation framework for EHR-based disease prediction. A Qwen2.5-7B teacher is fine-tuned with LoRA on data augmented by the Efficacy-Aware Data Fusion (EADF) module, which uses visit pairs from MIMIC-III to generate efficacy-ranked treatment information; a Multi-Label Adaptive Projection Head (MLAPH) converts the teacher's hidden states into soft labels, and a BERT student is trained on a weighted combination of soft and hard labels. Experiments on MIMIC-III for 25 disease-prediction tasks report accuracy 94.76, F1 80.25, AUC 91.11, AUPR 76.33, a 22.2x inference speedup, and a 34.96x memory reduction relative to the teacher model. The abstract and conclusion claim a 9% accuracy increase and a 27% F1 improvement over baseline models with p<0.01.

Significance. If the reported results are valid, the framework would provide a practical approach to compressing large language models for clinical prediction while retaining accuracy. The paper has notable strengths: the authors make code and data available, provide concrete efficiency measurements, include ablation studies for the main components, and explore alternative soft-label calculation methods. The central claim, however, rests on three load-bearing issues that currently undermine confidence: the headline improvement numbers do not match Table 1, the EADF feature construction may leak future-visit information into the input, and the baseline comparisons are not documented as sharing a common evaluation protocol. The manuscript is technically interesting and potentially useful, but these issues must be resolved before the performance claims can be accepted.

major comments (4)
  1. [Abstract, Section 4.3.1 (Table 1), Section 5] The abstract and conclusion state that CKD-EHR increases diagnostic accuracy by 9% and F1-score by 27% compared with the baseline. These numbers are not derivable from Table 1: the accuracy improvement over the strongest published baseline (ClinicalBERT) is 0.14 absolute points (about 0.15% relative), and the F1 improvement is 3.47 absolute points (about 4.5% relative). The paper does not specify which baseline the 9% and 27% refer to, nor whether these are relative or absolute changes. Please report the exact baseline used, the calculation method, and align the abstract and conclusion with the numbers in Table 1.
  2. [Section 3.2 (EADF) and Section 4.3.1] The EADF module constructs q, a ranked treatment-efficacy list, by pairing each visit with the next visit and analyzing changes in disease state. The augmented input includes q, which is derived from d(i+1), the next-visit diagnosis. If Q is estimated on the full MIMIC-III cohort rather than only on training folds, then test-set labels leak into the test inputs, making the reported performance gains invalid. The manuscript never states that Q is computed strictly from training data. Even if Q is training-only, the comparison lacks a baseline that also receives q as an input feature, so the observed gains cannot be attributed specifically to knowledge distillation rather than to the additional feature. Please clarify the splitting procedure for Q and add a control baseline that receives the same augmented input without distillation.
  3. [Section 4.2 and Section 4.3.1] The paper claims statistical significance (p<0.01) in the abstract, but Section 4.2 describes no statistical tests, no confidence intervals, no error bars, and no cross-validation or repeated-run variability. In addition, Table 1 appears to compare against published baseline numbers without stating whether all models were trained and evaluated on the same MIMIC-III train/validation/test split, the same set of 25 disease labels, and the same aggregation rule for multi-label metrics. Without a shared evaluation protocol and variance estimates, the comparison to baselines is not meaningful. Please provide per-fold or repeated-run results, statistical significance tests, and a description of the common data split and metric aggregation used for all models, including the baselines.
  4. [Section 4.4.1 (Table 3)] The comprehensive ablation shows that the full CKD-EHR model (EADF + LoRCKD) achieves F1=80.25, whereas the teacher with EADF only achieves F1=82.94 and the teacher without either component achieves F1=81.61. Thus, the distillation step reduces F1 by 2.69 points relative to the teacher with EADF, and the only metric where the full model is competitive is accuracy (94.76 vs. 94.75). This pattern weakens the claim that LoRCKD improves prediction quality; the paper's own text acknowledges that F1, AUC, and AUPR are lower for the full model than for the teacher. The contribution of distillation should be reframed as an efficiency-accuracy trade-off, and the ablation discussion should quantify the cost of the F1 drop rather than presenting the result as a clear success.
minor comments (4)
  1. [Section 4.2] The phrase '10 iterations' should be clarified as 10 training epochs; as written it is ambiguous whether the student model is trained for 10 iterations over the whole dataset or 10 update steps.
  2. [Table 1] The baseline name 'GrphCare' appears to be a typo for 'GraphCare', which is spelled correctly in Section 4.2.1 and in the reference list.
  3. [Section 3.3, Eq. (6)] The text states that alpha and beta are weighting parameters with alpha + beta = 1, but the range of alpha tested in Section 4.5 is not formally stated. In addition, 'soft target contribution rates' in the caption of Fig. 2 should be defined as the beta value or as a complement of alpha.
  4. [Section 4.6] The confusion-matrix figure captions do not state whether the matrices are computed on the test set or on a validation set; please specify the data split used for visualization.

Circularity Check

1 steps flagged · score 6.0 of 10

EADF's efficacy feature q is derived from the next-visit outcome d(i+1) and is inserted into the input, so the model is predicting a label that has already contributed to its input feature.

  1. self definitional [Section 3.2 (Efficacy-Aware Data Fusion), equations x∈[d(i),m(j),p(k),d(i+1)] and x∈[d(i),m(j),p(k),q,d(i+1)]; Section 4.3.1, Table 1]
    "Based on the temporal order T, we pair each patient’s i-th visit with their subsequent( i + 1)-th visit to form a Visit Pair... d(i+1) reflects the disease status at the next visit, indicating potential treatment outcomes. Subsequently, we apply statistical methods to analyze the changes in disease state across visit pairs, quantifying the efficacy of different combinations of medications and procedures... The samples are then augmented with efficacy information as follows: x∈[d (i),m (j),p (k),q,d (i+1)]... Meanwhile, the subsequent diagnosisd(i+1) is used as a supervision label..."

    The prediction target is d(i+1). The ranked efficacy list Q is computed from visit pairs whose second element is d(i+1), and q is selected from Q and added to the input x alongside d(i+1). Thus the input feature q is a function of the outcome the model is asked to predict. The paper describes Q as built from the MIMIC-III database without stating that Q is estimated only on training folds, and Section 4.3.1 credits the gains to 'information associated with the dataset,' so the standard reading is that q encodes next-visit outcomes. The Table 1 gains (ACC 94.76 vs 94.62, F1 80.25 vs 76.78) and the abstract's 9%/27% claims therefore reduce, at least in part, to the model being given information derived from its own supervision label.

full rationale

The only load-bearing circularity is the EADF feature construction: q is derived from visit-pair outcomes that include d(i+1), and the same d(i+1) is the supervised prediction target. Unless Q is built strictly from training folds and never touches the evaluation visits, the main accuracy/F1 improvements are contaminated by target leakage; the paper never documents such a split. The distillation chain itself (Qwen teacher fine-tuned on training data, soft labels distilled into BERT with hard-label supervision) is standard and not circular, and the inference-speed comparison is architectural. There are no load-bearing self-citations or imported uniqueness claims. However, because the headline quantitative gains rest on the EADF input that is a function of the next-visit label, the central reported improvement is partially circular by construction; a corrected evaluation with train-only Q and a baseline that also receives q is needed before the claimed 9%/27% can be credited to the method.

Assumptions & free parameters 2 free parameters · 3 assumptions · 0 invented entities

The framework rests on assumptions about label reliability, temporal generalization of efficacy statistics, and comparability of baselines. The only tuned free parameter explicitly varied is the loss weight alpha.

free parameters (2)
  • alpha (hard label loss weight) = 0.9
    Chosen based on Fig 2; controls balance between hard and soft label losses.
  • LoRA rank (r) = not reported
    Rank of low-rank adaptation matrices; affects capacity of teacher fine-tuning.
assumptions (3)
  • domain assumption MIMIC-III diagnosis codes are accurate enough to serve as ground truth for next-visit disease prediction.
    Labels are derived from coded diagnoses; no manual verification described.
  • domain assumption Treatment efficacy statistics computed from the full MIMIC-III dataset generalize to the test period without temporal leakage.
    EADF computes Q from visit pairs; the paper does not describe a time-based split to prevent future data from informing training features.
  • domain assumption The published baseline numbers in Table 1 were obtained under the same evaluation protocol as CKD-EHR.
    The paper does not document the data split or re-implementation details for baselines.

how reviews work

0 comments
Cite this review

Pith. "Pith review of CKD-EHR:Clinical Knowledge Distillation for Electronic Health Records." pith.science (2026). https://pith.science/paper/OE6BRWYD

@misc{pith2026250615118,
  author       = {Pith},
  title        = {Pith review of: CKD-EHR:Clinical Knowledge Distillation for Electronic Health Records},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/OE6BRWYD}},
  note         = {Machine review of arXiv:2506.15118}
}
read the original abstract

Electronic Health Records (EHR)-based disease prediction models have demonstrated significant clinical value in promoting precision medicine and enabling early intervention. However, existing large language models face two major challenges: insufficient representation of medical knowledge and low efficiency in clinical deployment. To address these challenges, this study proposes the CKD-EHR (Clinical Knowledge Distillation for EHR) framework, which achieves efficient and accurate disease risk prediction through knowledge distillation techniques. Specifically, the large language model Qwen2.5-7B is first fine-tuned on medical knowledge-enhanced data to serve as the teacher model.It then generates interpretable soft labels through a multi-granularity attention distillation mechanism. Finally, the distilled knowledge is transferred to a lightweight BERT student model. Experimental results show that on the MIMIC-III dataset, CKD-EHR significantly outperforms the baseline model:diagnostic accuracy is increased by 9%, F1-score is improved by 27%, and a 22.2 times inference speedup is achieved. This innovative solution not only greatly improves resource utilization efficiency but also significantly enhances the accuracy and timeliness of diagnosis, providing a practical technical approach for resource optimization in clinical settings. The code and data for this research are available athttps://github.com/209506702/CKD_EHR.

Figures

Figures reproduced from arXiv: 2506.15118 by the authors.

Figure 1
Figure 1. The CKD-EHR framework includes: (a) The basic principle of LoRA fine-tuning, [PITH_FULL_IMAGE:figures/full_fig_p007_1.png] view at source ↗
Figure 2
Figure 2. Changes in indicators under different pre-soft target contribution rates [PITH_FULL_IMAGE:figures/full_fig_p018_2.png] view at source ↗
Figure 3
Figure 3. Comparison of prediction results for the same data by CKD-EHR and BERT models. [PITH_FULL_IMAGE:figures/full_fig_p019_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Confusion experiments for different disease categories, with subplots showing TN, [PITH_FULL_IMAGE:figures/full_fig_p020_4.png]

Review history (2 revisions) →

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

37 extracted references · 26 canonical work pages

  1. [1]

    H. Xiao, F. Zhou, X. Liu, T. Liu, Z. Li, X. Liu, X. Huang, A compre- hensive survey of large language models and multimodal large language models in medicine, Information Fusion (2024) 102888

    work page 2024

  2. [2]

    T. A. D’Antonoli, A. Stanzione, C. Bluethgen, F. Vernuccio, L. Ugga, M. E. Klontzas, R. Cuocolo, R. Cannella, B. Koçak, Large language models in radiology: fundamentals, applications, ethical considerations, risks, and future directions, Diagnostic and Interventional Radiology 30 (2024) 80

    work page 2024

  3. [3]

    M. Moor, O. Banerjee, Z. S. H. Abad, H. M. Krumholz, J. Leskovec, E. J. Topol, P. Rajpurkar, Foundation models for generalist medical artificial intelligence, Nature 616 (2023) 259–265

    work page 2023

  4. [4]

    Lewis, E

    P. Lewis, E. Perez, A. Piktus, F. Petroni, V. Karpukhin, N. Goyal, H. Küttler, M. Lewis, W.-t. Yih, T. Rocktäschel, et al., Retrieval- augmented generation for knowledge-intensive nlp tasks, Advances in neural information processing systems 33 (2020) 9459–9474. 23

    work page 2020

  5. [5]

    Kresevic, M

    S. Kresevic, M. Giuffrè, M. Ajcevic, A. Accardo, L. S. Crocè, D. L. Shung, Optimization of hepatological clinical guidelines interpretation by large language models: a retrieval augmented generation-based framework, NPJ digital medicine 7 (2024) 102

    work page 2024

  6. [6]

    D. Edge, H. Trinh, N. Cheng, J. Bradley, A. Chao, A. Mody, S. Truitt, D. Metropolitansky, R. O. Ness, J. Larson, From local to global: A graph rag approach to query-focused summarization, arXiv preprint arXiv:2404.16130 (2024)

    arXiv 2024

  7. [7]

    S. Yao, J. Zhao, D. Yu, N. Du, I. Shafran, K. Narasimhan, Y. Cao, React: Synergizing reasoning and acting in language models, in: International Conference on Learning Representations (ICLR), 2023

    2023

  8. [8]

    Samsi, D

    S. Samsi, D. Zhao, J. McDonald, B. Li, A. Michaleas, M. Jones, W. Berg- eron, J. Kepner, D. Tiwari, V. Gadepally, From words to watts: Bench- marking the energy costs of large language model inference, in: 2023 IEEE High Performance Extreme Computing Conference (HPEC), IEEE, 2023, pp. 1–9

    work page 2023

Show all 37 references
  1. [9]

    Rabgay, G

    T. Rabgay, G. Kidman, Multiple iterations and messiness in the imple- mentation of action research by bhutanese secondary science teachers, Discover Education 2 (2023) 49

    2023

  2. [10]

    R. Xu, W. Shi, Y. Yu, Y. Zhuang, B. Jin, M. D. Wang, J. C. Ho, C. Yang, Ram-ehr: Retrieval augmentation meets clinical predictions on electronic health records, arXiv preprint arXiv:2403.00815 (2024)

    2024 arXiv

  3. [11]

    J. Ko, S. Kim, T. Chen, S.-Y. Yun, Distillm: Towards streamlined distillation for large language models, arXiv preprint arXiv:2402.03898 (2024)

    2024 arXiv

  4. [12]

    Ridnik, G

    T. Ridnik, G. Sharir, A. Ben-Cohen, E. Ben-Baruch, A. Noy, Ml- decoder: Scalable and versatile classification head, in: Proceedings of the IEEE/CVF winter conference on applications of computer vision, 2023, pp. 32–41

    2023

  5. [13]

    Khanna, G

    P. Khanna, G. Dhillon, V. Buddhavarapu, R. Verma, R. Kashyap, H. Gre- wal, Artificial intelligence in multilingual interpretation and radiology assessment for clinical language evaluation (ai-miracle), Journal of Per- sonalized Medicine 14 (2024) 923. 24

    2024

  6. [14]

    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. Seehofnerová, et al., Adapted large language models can outperform medical experts in clinical text summarization, Nature medicine 30 (2024) 1134–1142

    2024

  7. [15]

    A. E. Kornblith, C. Singh, J. C. Innes, T. P. Chang, K. M. Adelgais, M. Holsti, J. Kim, B. McClain, D. K. Nishijima, S. Rodgers, et al., Analyzing patient perspectives with large language models: a cross- sectional study of sentiment and thematic classification on exception fr...

    2025

  8. [16]

    X. Liu, H. Lai, H. Yu, Y. Xu, A. Zeng, Z. Du, P. Zhang, Y. Dong, J. Tang, Webglm: towards an efficient web-enhanced question answering system with human preferences, in: Proceedings of the 29th ACM SIGKDD conference on knowledge discovery and data mining, 2023, pp. 4549–4560

    2023

  9. [17]

    T. Zada, N. Tam, F. Barnard, M. Van Sittert, V. Bhat, S. Rambhatla, et al., Medical misinformation in ai-assisted self-diagnosis: Development of a method (evalprompt) for analyzing large language models, JMIR Formative Research 9 (2025) e66207

    2025

  10. [18]

    S. Xu, L. Pang, M. Yu, F. Meng, H. Shen, X. Cheng, J. Zhou, Unsu- pervised information refinement training of large language models for retrieval-augmented generation, arXiv preprint arXiv:2402.18150 (2024)

    2024 arXiv

  11. [19]

    X. Xu, M. Li, C. Tao, T. Shen, R. Cheng, J. Li, C. Xu, D. Tao, T. Zhou, A survey on knowledge distillation of large language models, arXiv preprint arXiv:2402.13116 (2024)

    2024 arXiv

  12. [20]

    Jiang, Z

    C. Jiang, Z. Li, J. Yang, Y. Wu, S. Li, Effective and efficient conditional contrast for data-free knowledge distillation with low memory, The Journal of Supercomputing 81 (2025) 1–21

    2025

  13. [21]

    Z. Dong, W. Xu, X. Xu, Z. Zhang, Modality imbalance? dynamic multi- modal knowledge distillation in automatic alzheimer’s disease recognition, IEEE Journal of Biomedical and Health Informatics (2025)

    2025

  14. [22]

    C. Wang, J. Zhong, Q. Dai, Q. Yu, Y. Qi, B. Fang, X. Li, Mted: multiple teachers ensemble distillation for compact semantic segmentation, Neural Computing and Applications 35 (2023) 11789–11806. 25

    2023

  15. [23]

    Sclar, P

    M. Sclar, P. West, S. Kumar, Y. Tsvetkov, Y. Choi, Referee: Reference- free sentence summarization with sharper controllability through sym- bolic knowledge distillation, arXiv preprint arXiv:2210.13800 (2022)

    2022 arXiv

  16. [24]

    Shleifer, A

    S. Shleifer, A. M. Rush, Pre-trained summarization distillation, arXiv preprint arXiv:2010.13002 (2020)

    2020 arXiv

  17. [25]

    A. E. Johnson, T. J. Pollard, L. Shen, L.-w. H. Lehman, M. Feng, M. Ghassemi, B. Moody, P. Szolovits, L. Anthony Celi, R. G. Mark, Mimic-iii, a freely accessible critical care database, Scientific data 3 (2016) 1–9

    2016

  18. [26]

    Harutyunyan, H

    H. Harutyunyan, H. Khachatrian, D. C. Kale, G. Ver Steeg, A. Galstyan, Multitask learning and benchmarking with clinical time series data, Scientific data 6 (2019) 96

    2019

  19. [27]

    Y. Li, S. Rao, J. R. A. Solares, A. Hassaine, R. Ramakrishnan, D. Canoy, Y. Zhu, K.Rahimi, G. Salimi-Khorshidi, Behrt: transformer for electronic health records, Scientific reports 10 (2020) 7155

    2020

  20. [28]

    E. Choi, Z. Xu, Y. Li, M. Dusenberry, G. Flores, E. Xue, A. Dai, Learning the graphical structure of electronic health records with graph convolutional transformer, in: Proceedings of the AAAI conference on artificial intelligence, volume 34, 2020, pp. 606–613

    2020

  21. [29]

    R. Xu, M. K. Ali, J. C. Ho, C. Yang, Hypergraph transformers for ehr-based clinical predictions, AMIA Summits on Translational Science Proceedings 2023 (2023) 582

    2023

  22. [30]

    M. Ye, S. Cui, Y. Wang, J. Luo, C. Xiao, F. Ma, Medretriever: Target- driven interpretable health risk prediction via retrieving unstructured medical text, in: Proceedings of the 30th ACM International Conference on Information & Knowledge Management, 2021, pp. 2414–2423

    2021

  23. [31]

    Y. Xu, X. Chu, K. Yang, Z. Wang, P. Zou, H. Ding, J. Zhao, Y. Wang, B. Xie, Seqcare: Sequential training with external medical knowledge graph for diagnosis prediction in healthcare data, in: Proceedings of the ACM Web Conference 2023, 2023, pp. 2819–2830. 26

    2023

  24. [32]

    Jiang, C

    P. Jiang, C. Xiao, A. Cross, J. Sun, Graphcare: Enhancing health- care predictions with personalized knowledge graphs, arXiv preprint arXiv:2305.12788 (2023)

    2023 arXiv

  25. [33]

    Van Aken, J.-M

    B. Van Aken, J.-M. Papaioannou, M. Mayrdorfer, K. Budde, F. A. Gers, A. Loeser, Clinical outcome prediction from admission notes using self-supervised knowledge integration, arXiv preprint arXiv:2102.04110 (2021)

    2021 arXiv

  26. [34]

    A. Naik, S. Parasa, S. Feldman, L. L. Wang, T. Hope, Literature- augmented clinical outcome prediction, arXiv preprint arXiv:2111.08374 (2021)

    2021 arXiv

  27. [35]

    Huang, J

    K. Huang, J. Li, R. Ranganath, Clinicalbert: Modeling clinical notes and predicting hospital readmission, 2019. doi:10.48550/arXiv.1904.05342

    2019 doi arXiv

  28. [36]

    J. Lee, W. Yoon, S. Kim, D. Kim, S. Kim, C. H. So, J. Kang, Biobert: a pre-trained biomedical language representation model for biomedi- cal text mining, Bioinformatics 36 (2019) 1234–1240. doi: 10.1093/ bioinformatics/btz682

    2019

  29. [37]

    Alsentzer, J

    E. Alsentzer, J. R. Murphy, W. Boag, W.-H. Weng, D. Jin, T. Nau- mann, M. B. A. McDermott, Publicly available clinical bert embeddings, ArXiv abs/1904.03323 (2019). URL: https://api.semanticscholar. org/CorpusID:102352093. 27

    2019 arXiv

Pith tools

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