Pith. sign in

REVIEW 4 major objections 6 minor 37 references

Time-Aware Attention for Enhanced Electronic Health Records Modeling

T0 review · 4 major / 6 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read TALE-EHR folds learnable continuous time gaps into Transformer attention and reports state-of-the-art results for EHR disease and code prediction on MIMIC-IV and PIC.

desk verdict An incremental but plausible EHR transformer whose headline AUC numbers cannot be verified until the disease-onset negative sampling is specified and the tables are fixed. read the letter →

arxiv 2507.14847 v1 pith:WU76CVLT submitted 2025-07-20 cs.LG

Junhan Yu , Zhunyi Feng , Junwei Lu , Tianxi Cai , Doudou Zhou This is my paper
classification cs.LG MSC 68T0762M30
keywords electronichealthrecordstime-awareattentiontemporalpointprocessTransformermedicalcodepredictiondiseaseonsetLLMembeddingsMIMIC-IV
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 claims that the main obstacle to accurate EHR forecasting is not the sequence of clinical codes but the irregular, continuous time between events, and that a Transformer can be made to respect that time by folding a learnable function of the time gap directly into each attention score. It introduces TALE-EHR, which pairs this time-aware attention with embeddings that a general-purpose text encoder produces from standard code descriptions, so the model starts with clinical semantics rather than random code vectors. On MIMIC-IV the authors report an average area under the ROC curve of 0.926 across nine diseases, top Acc@5 of 0.902 for next-code prediction, and gains on readmission and length-of-stay benchmarks; on the PIC pediatric database they report the best AUROC, AUPRC, and F1 across pneumonia, heart malformations, mortality, and ICU stay longer than seven days. If these results hold, the practical message is that explicit continuous time itself, not just richer code representations, carries much of the predictive signal in EHRs.

What carries the argument

The key object is the time-aware attention function $A(Q,K,t)=\mathrm{softmax}(Q^\top K/\sqrt{d}\, w(t))$ with a learnable polynomial-gated weighting $w(t)=\sigma(a_0+a_1 t+\dots+a_s t^s)$ evaluated on the continuous gap $|t_j-t_k|$ between events. A sigmoid keeps the temporal weight in $[0,1]$, so the mechanism can downweight distant events without erasing them; the polynomial basis gives a flexible, interpretable decay curve that is fitted jointly with the rest of the network. A hierarchical attention step then forms the patient history representation $h_t=\sum_j \alpha_j(t)E_{c_j}(t)$, where the same temporal weighting modulates how much each event contributes to the current state. This machinery lets the model treat time as part of the attention computation itself rather than as a preprocessing step, and it is trained jointly with a temporal point process loss for event timing and a focal loss for code prediction.

What would settle it

Re-run the nine disease tasks on MIMIC-IV with explicit negative-patient construction: each control is assigned the same index time as a matched case (or a random index from the control's own record) and the prediction window ends at that index; drop all cases with fewer than 1024 prior events exactly as the paper does. If TALE-EHR's AUROC advantage over the strongest baseline, Cehr_Bert, shrinks to within noise under this leakage-free labeling, then the reported superiority is an artifact of label construction rather than of time-aware attention.

Watch

Extended reading notes

Core claim

TALE-EHR's central claim is that explicitly modeling the continuous time difference between clinical events inside the attention mechanism is the decisive ingredient for EHR sequence modeling. Rather than relying on sequence order, separate time embeddings, or discretized time tokens, the model multiplies each softmax attention weight by a learnable temporal function $w(t)=\sigma(a_0+a_1 t+\dots+a_5 t^5)$ of the time gap, so that a past event's influence decays or persists depending on the condition being predicted. The learned weights are disease-specific: the paper shows rapid decay for acute kidney injury and gradual decay for chronic conditions such as type 2 diabetes and heart failure. The same time-aware weighting is used twice, first inside code-level attention and again to aggregate an entire patient history into a single vector used for downstream prediction. On top of this, the model freezes the embeddings that a pre-trained general-purpose text encoder produces from the textual descriptions of 12,232 medical codes, and the paper's ablation attributes a smaller but consistent accuracy gain to these embeddings, with the time-aware mechanism contributing the larger share.

Load-bearing premise

The claim depends on the disease-onset evaluation being built without temporal leakage: the paper says positive patients are scored on the 1024 events before their first diagnosis, but it never specifies how negative patients are defined, which index time they receive, or how censoring is handled, so if those choices differ between cases and controls the reported AUC gains could be inflated.

Editorial extensions

If this is right

  • Disease forecasting: TALE-EHR reports the highest AUROC on all nine MIMIC-IV diseases, including 0.961 for acute kidney injury, so disease-onset screening could move to a model that uses continuous event timing.
  • Next-code prediction: the reported Acc@5 of 0.902 and F1 of 0.850 across 12,232 codes suggest better auto-completion of the medical record, reducing missed future codes.
  • Benchmark tasks: reported 30-day readmission AUROC of 0.762 and next-visit length-of-stay over 7 days AUROC of 0.759 beat all baselines, indicating the representation transfers to outcomes beyond code prediction.
  • Pediatric generalization: on PIC, the best AUROC on pneumonia (0.945), heart malformations (0.906), mortality (0.934), and ICU stay over 7 days (0.897) suggests the method transfers across patient populations.
  • Component attribution: the ablation shows that removing time-awareness costs more accuracy than replacing LLM embeddings with random vectors, so continuous temporal modeling is the primary driver of the reported gains.

Reading between the lines

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

  • One testable extension not reported in the paper: the learned $w(t)$ curves could be read as per-disease forgetting functions, and plotting them across hospitals or across code systems would show whether the model is learning clinical time constants or dataset artifacts.
  • A practical diagnostic follows from the paper's own design: if $w(t)$ saturates to a constant near 1 for all diseases after training, the time-aware mechanism would collapse to standard self-attention, so routinely plotting the learned weight curve would reveal when the claimed mechanism is actually inert.
  • Because the embedding component contributes less than the temporal component, a cheaper variant that replaces the general-purpose text encoder with hand-curated code vectors might retain most of the gain; this is directly checkable and would matter for deployment in low-resource settings.
  • The unclear negative-patient definition implies that the same framework should be re-benchmarked under a public, standardized outcome-labeling protocol before the absolute AUC numbers are used to set clinical decision thresholds.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 6 minor

Summary. The manuscript proposes TALE-EHR, a Transformer-based marked point-process model for EHR sequences. Its core idea is to multiply the scaled dot-product attention score by a learned sigmoid-polynomial w(t) of the continuous inter-event time, and to embed medical codes with fixed BGE embeddings of their textual descriptions. The model is pre-trained with a joint least-squares intensity loss and a focal code-prediction loss, then fine-tuned for disease-onset and clinical benchmark tasks. On MIMIC-IV the paper reports an average disease-prediction AUC of 0.926 across nine conditions, gains on readmission and length-of-stay benchmarks, and Acc@5 of 0.902 on next-code prediction; on PIC it reports consistent gains across four pediatric tasks. Ablations attribute most of the improvement to the time-aware mechanism, with a smaller contribution from the LLM-derived embeddings.

Significance. The proposed mechanism is a reasonable and testable modification of Transformer attention: instead of adding time embeddings or using discrete time tokens, it directly modulates attention logits by a learned continuous function of the time gap. The architecture and losses are described in enough detail that the method could in principle be reimplemented, and I find no derivation loop: w(t) is learned on training labels and evaluated on held-out data, and the BGE embeddings are a fixed external encoder. The ablation design is sensible and the high-level claims are plausible. However, the empirical evidence as printed is not yet verifiable: the disease-onset label protocol is underspecified, Section 4.3 points to a nonexistent Appendix E, Tables 1 and 4 contain duplicate or malformed rows, no code or data are released, and the Acc@K numbers in Table 7 are not consistent with the standard cumulative definition. The contribution is worth considering, but the reported SOTA claims need stronger reporting and, where possible, re-analysis under a clearly defined evaluation protocol.

major comments (4)
  1. [Section 3.4; Algorithm 1 (line 19)] The disease-onset evaluation is the load-bearing support for the paper's headline result (average AUC 0.926, Table 1), but the label construction is not specified. The text says only that the model uses 1024 clinical events before the target event, 'e.g., the first occurrence of a disease,' and Algorithm 1 says 'Extract histories before disease onset time.' It never defines negative patients, an index time for patients who never develop the disease, or how censoring is handled. Appendix F's description of negatives as patients who 'did not' develop the disease suggests labels are assigned from the full record. If positive histories are truncated at first onset while negative histories are drawn from the entire record, then sequence length/composition and post-index events become available to the classifier, which can inflate AUROC independently of the time-aware attention. Please specify the negative-label definition and a single index time for all patients, add a censoring-aware evaluation (e.g., fixed prediction horizon or time-matched negative sampling), and state whether the 1024-event window is padded or truncated; the same protocol applies to the PIC results in Table 3.
  2. [Tables 1 and 4] The empirical tables contain errors that prevent verification: in Table 1 the HiTANet row for Hyperlipidemia is identical (up to reporting precision) to its Arteriosclerosis row, and in Table 4 the Hyperlipidemia row appears twice, with malformed entries such as '0.942±.0060.937±.006' and '0.964±.0050.957±.004'. Combined with the absence of released code or data, this means the reported baseline strengths and ablation comparisons cannot be checked from the manuscript. Please correct the tables, verify that every row corresponds to a distinct disease and run, and release code/data or detailed evaluation scripts so that the comparisons are reproducible.
  3. [Section 4.3; Table 7] Section 4.3 refers to Appendix E for the definitions of Acc@K, macro F1, and Recall, but no Appendix E exists in the manuscript. This is not merely a missing cross-reference: in Table 7 every model's Acc@K decreases as K grows (e.g., TALE-EHR: 0.902, 0.896, 0.862), which is incompatible with the standard cumulative top-K recall definition. Please supply Appendix E or define the metric precisely, and state whether Acc@K is cumulative recall, precision at K, or an exact-set match; the same clarification is needed for the multi-label targets and the meaning of 'next visit.'
  4. [Section 6; Table 4] The polynomial order (Poly=5) is selected after inspecting the ablation results reported in Table 4. If these AUROC values are computed on the test set, then the choice of w(t)'s degree is made using test information, and the reported advantage of Poly=5 over the alternatives may be optimistic. Please state explicitly whether the ablation comparisons were made on a validation set, and if not, provide validation-based model selection or report the selection procedure.
minor comments (6)
  1. [Section 3.3] The symbol γ is used both for the focal-loss exponent (γ=2.0) and for the weight of L_code in the combined loss L = L_time + γ L_code; please use different symbols for these two quantities.
  2. [Section 3.1] BGE-M3 is described as a 'pre-trained LLM,' but it is a text embedding model rather than an autoregressive or generative language model; please adjust the terminology to 'pre-trained text encoder' or similar.
  3. [Algorithm 1] The pseudocode does not specify how patient histories with fewer than 1024 events are handled, nor how the 'current time t' is chosen for each training example; please make these details explicit.
  4. [Appendix D] The discussion of disease-specific decay rates (acute vs. chronic) is qualitative and visual; a quantitative summary, such as fitted decay constants or a comparison of w(t) across diseases, would strengthen the claim.
  5. [References] The reference list contains duplicate and inconsistent entries (e.g., two HiTANet entries with different venues, and multiple Retain entries); please consolidate and correct the bibliography.
  6. [Table 6] The median frequency for CCS codes is reported as 1.587, while surrounding entries use comma separators; this is likely a formatting error and should be corrected.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular dependency: TALE-EHR's temporal weighting is learned from data and evaluated on held-out test sets; no claim reduces to its inputs by construction.

full rationale

The paper's central novelty is a learnable temporal weighting function w(t)=sigma(a0+a1*t+...+as*t^s) inserted into softmax attention. This function is not defined in terms of any target outcome; its coefficients are trained on the pre-training and fine-tuning objectives, and all reported AUC/Acc values are computed on a held-out 20% test partition (Section 4.1). The BGE code embeddings are an external, fixed pretrained encoder (Chen et al., 2024), not derived from the paper's own outputs. The temporal point process loss is a standard least-squares intensity estimator, and the statement that it is minimized by the true intensity is a known Doob-Meyer property rather than a self-referential definition. No load-bearing premise is justified by a self-citation: the authors do not invoke their own prior work for any uniqueness or ansatz claim. The main weaknesses are evaluation-protocol omissions (negative-patient and index-time definitions for disease-onset prediction, Section 3.4 and Algorithm 1) and apparent duplicate rows in Tables 1 and 4; these are correctness and reproducibility risks, not circularity. Selecting polynomial order 5 after ablation on the same benchmark is standard model selection. Therefore no circular step is present.

Assumptions & free parameters 7 free parameters · 5 assumptions · 0 invented entities

The central empirical claim rests on many trained parameters in the neural network, several hand-set hyperparameters, and domain assumptions about EHR semantics and temporal encoding. There are no invented physical entities. The model is a data-driven method; the point process formalism is used to justify the time loss but does not constrain the architecture.

free parameters (7)
  • w(t) polynomial coefficients a0..a5 = not reported
    The temporal weighting function in Eq. 4 is learned from data; curves are shown in Appendix D but coefficients are not listed.
  • Q/K/V projection MLPs and Q_base query = not reported
    All attention projections are trained (Section 3.1); sizes and layer counts are not reported.
  • intensity network g and code network f = not reported
    The joint objective trains these MLPs (Section 3.3), but their hidden sizes are unspecified.
  • task-specific classifiers MLP_k and weights w_k = not reported
    Disease prediction heads are trained during fine-tuning (Section 3.4).
  • downstream sequence length = 1024 events
    Fixed history length for disease tasks (Section 3.4) without ablation.
  • focal loss parameters = alpha=0.25, gamma=2.0
    Hand-set standard values in Section 3.3.0.2.
  • combined loss weight gamma = not reported
    Balances L_time and L_code in Section 3.3; value and tuning procedure are not given.
assumptions (5)
  • domain assumption Event histories are realizations of a marked temporal point process with conditional intensity lambda(t) determined by history and covariates.
    Section 3 defines the marked point process and Eq. 1; this is an untested modeling assumption.
  • standard math The Doob-Meyer decomposition justifies the squared-error intensity loss.
    Invoked in Section 3.3.0.1 to claim L_time is minimized at the true intensity; not proved in the paper.
  • domain assumption BGE embeddings of standardized code descriptions capture sufficient clinical semantics.
    Section 3.1 fixes BGE embeddings; no alternative encoder or description source is ablated.
  • domain assumption Log transformation of relative timestamps preserves relevant temporal structure.
    Section 4.1 states timestamps are log-transformed; only forms of w(t) are ablated, not the transform itself.
  • domain assumption Pre-training on next-event time and code prediction transfers to disease onset prediction.
    Section 3.4 fine-tunes from pre-trained representations to disease labels; transfer benefit is not separately measured.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Time-Aware Attention for Enhanced Electronic Health Records Modeling." pith.science (2026). https://pith.science/paper/WU76CVLT

@misc{pith2026250714847,
  author       = {Pith},
  title        = {Pith review of: Time-Aware Attention for Enhanced Electronic Health Records Modeling},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/WU76CVLT}},
  note         = {Machine review of arXiv:2507.14847}
}
read the original abstract

Electronic Health Records (EHR) contain valuable clinical information for predicting patient outcomes and guiding healthcare decisions. However, effectively modeling Electronic Health Records (EHRs) requires addressing data heterogeneity and complex temporal patterns. Standard approaches often struggle with irregular time intervals between clinical events. We propose TALE-EHR, a Transformer-based framework featuring a novel time-aware attention mechanism that explicitly models continuous temporal gaps to capture fine-grained sequence dynamics. To complement this temporal modeling with robust semantics, TALE-EHR leverages embeddings derived from standardized code descriptions using a pre-trained Large Language Model (LLM), providing a strong foundation for understanding clinical concepts. Experiments on the MIMIC-IV and PIC dataset demonstrate that our approach outperforms state-of-the-art baselines on tasks such as disease progression forecasting. TALE-EHR underscores the benefit of integrating explicit, continuous temporal modeling with strong semantic representations provides a powerful solution for advancing EHR analysis.

Figures

Figures reproduced from arXiv: 2507.14847 by the authors.

Figure 1
Figure 1. Overview of the TALE-EHR framework. where (cj , tj ) ∈ Ht . To generate robust semantic embeddings (vc) for medical codes from their textual descriptions, we utilize BGE (Chen et al., 2024), a powerful general-purpose text encoder pre-trained on diverse, broad-domain corpora. This approach helps mitigate potential biases from medical-specific pre-training and enhances generalization. The query (Qc), key (Kc), and va… view at source ↗
Figure 2
Figure 2. Learned temporal weighting function w(∆t) for downstream disease prediction tasks. The x-axis represents the time interval between clinical events (in days), and the y-axis shows the attention weight. The curve demonstrates how the model modulates attention across different temporal scales, with a gradual decay indicating that both recent and historical events contribute to disease prediction, though with decreasing… view at source ↗
Figure 3
Figure 3. UMAP visualization of disease representations learned by TALE-EHR for nine [PITH_FULL_IMAGE:figures/full_fig_p021_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: UMAP visualization of disease representations learned by LSTM for the same [PITH_FULL_IMAGE:figures/full_fig_p022_4.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

37 extracted references · 36 canonical work pages

  1. [1]

    Bartlett, P. L. and S. Mendelson (2006, Jul). Empirical minimization. Probability Theory and Related Fields\/ 135\/ (3), 311--334

    work page 2006

  2. [2]

    Chen, J., C. Liu, J. Sun, S. Saria, and H. Xue (2020). Modeling inter-dependence between time and mark in multivariate temporal point processes. Journal of Machine Learning Research\/ 21\/ (171), 1--38

    work page 2020

  3. [3]

    Chen, J., S. Xiao, P. Zhang, K. Luo, D. Lian, and Z. Liu (2024). Bge m3-embedding: Multi-lingual, multi-functionality, multi-granularity text embeddings through self-knowledge distillation. arXiv preprint arXiv:2402.03216\/

    arXiv 2024

  4. [4]

    Choi, E., M. T. Bahadori, J. A. Kulas, A. Schuetz, W. F. Stewart, and J. Sun (2017). Retain: An interpretable predictive model for healthcare using reverse time attention mechanism

    work page 2017

  5. [5]

    Choi, E., M. T. Bahadori, A. Schuetz, W. F. Stewart, and J. Sun (2016). Doctor ai: Predicting clinical events via recurrent neural networks. In Machine Learning for Healthcare (MLHC)

    work page 2016

  6. [6]

    Choi, E., M. T. Bahadori, D. Sontag, and J. Li (2016). Learning the joint representation of heterogeneous temporal events for clinical endpoint prediction. In Proceedings of the AAAI Conference on Artificial Intelligence , Volume 30

    work page 2016

  7. [7]

    Choi, E., M. T. Bahadori, J. Sun, J. Kulas, A. Schuetz, and W. Stewart (2016). Retain: An interpretable predictive model for healthcare using reverse time attention mechanism. Advances in neural information processing systems\/ 29

    work page 2016

  8. [8]

    Choi, S., H. J. Kim, J. Lee, and et al. (2020). Using transformer models to predict clinical events in electronic health records. Proceedings of the 2020 IEEE International Conference on Healthcare Informatics (ICHI)\/ , 290--295

    work page 2020

Show all 37 references
  1. [9]

    Daley, D. J. and D. Vere-Jones (2003). An introduction to the theory of point processes: volume I: elementary theory and methods . Springer

    2003

  2. [10]

    Du, N., H. Dai, R. Trivedi, U. Upadhyay, M. Gomez-Rodriguez, and L. Song (2016). Recurrent marked temporal point processes: Embedding event history to vector. In Proceedings of the 22nd ACM SIGKDD international conference on Knowledge Discovery and Data Mining , pp.\ 1555--1564

    2016

  3. [11]

    Geer, S. (2000). Empirical Processes in M-Estimation . Cambridge Series in Statistical and Probabilistic Mathematics. Cambridge University Press

    2000

  4. [12]

    Hochreiter, S. and J. Schmidhuber (1997). Long short-term memory. Neural Computation\/ 9\/ (8), 1735--1780

    1997

  5. [13]

    Hur, K., J. Lee, J. Oh, W. Price, Y.-H. Kim, and E. Choi (2022). Unifying heterogeneous electronic health records systems via text-based code embedding

    2022

  6. [14]

    Johnson, A. E., L. Bulgarelli, L. Shen, A. Gayles, A. Shammout, S. Horng, T. J. Pollard, S. Hao, B. Moody, B. Gow, et al. (2023). Mimic-iv, a freely accessible electronic health record dataset. Scientific data\/ 10\/ (1), 1

    2023

  7. [15]

    Kim, J., C. Shim, B. S. K. Yang, C. Im, S. Y. Lim, H.-G. Jeong, and E. Choi (2024). General-purpose retrieval-enhanced medical prediction model using near-infinite history

    2024

  8. [16]

    Koltchinskii, V. (2011). Oracle Inequalities in Empirical Risk Minimization and Sparse Recovery Problems: \'E cole d' \'E t \'e de Probabilit \'e s de Saint-Flour XXXVIII -2008 . Lecture Notes in Mathematics. Springer Berlin, Heidelberg

    2011

  9. [17]

    Kwon, B. C., H. Park, J. Soman, J. Sun, and H. Park (2018). Retainex: Interpretable ehr document representation for clinical predictions. In Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining (KDD '18) , pp.\ 1245--1254. ACM

    2018

  10. [18]

    Conneau, and et al

    Lample, G., A. Conneau, and et al. (2019). Neural architectures for named entity recognition. Proceedings of the 2019 International Conference on Natural Language Processing\/ , 292--300

    2019

  11. [19]

    Lee, J. W., S. Amin, A. Amini, and et al. (2020). Transformers for patient data: A deep learning approach to time-series medical records. arXiv preprint arXiv:2004.03873\/

    2020 arXiv

  12. [20]

    Lee, S. A., S. Jain, A. Chen, K. Ono, J. Fang, A. Rudas, and J. N. Chiang (2024). Emergency department decision support using clinical pseudo-notes

    2024

  13. [21]

    Li, B., C. Zhou, Z. Wei, and et al. (2019). Knowledge graph embedding for predicting healthcare events. Proceedings of the 2019 International Joint Conference on Artificial Intelligence\/ , 3587--3594

    2019

  14. [22]

    Li, S., J. Jin, W. Song, L. Zhang, and B. Wu (2022). Hitanet: Hierarchical time-aware attention networks for risk prediction on electronic health records. IEEE Journal of Biomedical and Health Informatics\/ 26\/ (4), 1680--1691

    2022

  15. [23]

    Liu, L. J., V. Ortiz-Soriano, J. A. Neyra, and J. Chen (2022, December). KIT-LSTM : Knowledge-guided time-aware LSTM for continuous clinical risk prediction. In 2022 IEEE International Conference on Bioinformatics and Biomedicine ( BIBM ) . IEEE

    2022

  16. [24]

    Shang, F

    Liu, Y., J. Shang, F. Wang, and J. Sun (2019). Point processes for competing observations with recurrent networks (popcorn): A generative model of ehr data. In Proceedings of the 28th International Joint Conference on Artificial Intelligence (IJCAI)

    2019

  17. [25]

    Luo, J., M. Ye, C. Xiao, and F. Ma (2020, August). Hitanet: Hierarchical time-aware attention networks for risk prediction on electronic health records. In Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining , KDD ’20. ACM

    2020

  18. [26]

    Chitta, J

    Ma, F., R. Chitta, J. Zhou, Q. You, T. Sun, and J. Gao (2017, August). Dipole: Diagnosis prediction in healthcare via attention-based bidirectional recurrent neural networks. In Proceedings of the 23rd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining ...

    2017

  19. [27]

    Massart, P. (2007). Concentration Inequalities and Model Selection: Ecole d'Et \'e de Probabilit \'e s de Saint-Flour XXXIII -2003 . Lecture Notes in Mathematics. Springer Berlin, Heidelberg

    2007

  20. [28]

    Jiang, K

    Pang, C., X. Jiang, K. S. Kalluri, M. Spotnitz, R. Chen, A. Perotte, and K. Natarajan (2021). Cehr-bert: Incorporating temporal information from structured ehr data to improve prediction tasks

    2021

  21. [29]

    Shang, J., Y. Chen, X. Wang, Y. Liu, and J. Sun (2019). Event-based contrastive learning for medical time series. In Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining , pp.\ 2199--2207

    2019

  22. [30]

    Shang, J., Y. Chen, X. Wang, Y. Liu, and J. Sun (2021). Interpretable neural temporal point processes for modeling electronic health records. Artificial Intelligence in Medicine\/ 116 , 102075

    2021

  23. [31]

    Steinberg, E., J. A. Fries, Y. Xu, and N. Shah (2024). MOTOR : A time-to-event foundation model for structured medical records. In The Twelfth International Conference on Learning Representations

    2024

  24. [32]

    Messica, Y

    Su, X., S. Messica, Y. Huang, R. Johnson, L. Fesser, S. Gao, F. Sahneh, and M. Zitnik (2025). Multimodal medical code tokenizer

    2025

  25. [33]

    Wang, S., X. Li, W. Wang, and et al. (2020). Multi-relational knowledge graph embedding for recommendation and predictive tasks in healthcare. IEEE Transactions on Knowledge and Data Engineering\/ 32\/ (10), 1913--1925

    2020

  26. [34]

    Wang, Z., C. Gao, C. Xiao, and J. Sun (2024). Meditab: Scaling medical tabular data predictors via data consolidation, enrichment, and refinement

    2024

  27. [35]

    Wang, Z. and J. Sun (2022). Transtab: Learning transferable tabular transformers across tables

    2022

  28. [36]

    Zeng, X., G. Yu, Y. Lu, L. Tan, X. Wu, S. Shi, H. Duan, Q. Shu, and H. Li (2020, January). Pic, a paediatric-specific intensive care database. Scientific Data\/ 7\/ (1)

    2020

  29. [37]

    Zhang, X., W. Wei, H. Li, and et al. (2019). Learning knowledge graph embeddings for predicting drug-target interactions. IEEE Access\/ 7 , 52677--52686

    2019

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.