Pith. sign in

REVIEW 3 major objections 4 minor 20 references

Latent Factor Point Processes for Patient Representation in Electronic Health Records

T0 review · 3 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 A few eigenvalues of the Fourier-transformed code co-occurrence matrix are a sufficient patient representation for classification and clustering under the paper's latent factor point process model.

desk verdict Useful-looking EHR embedding under a latent factor point process model, but the central eigenvalue linearity claim is false, so the advertised classification and clustering guarantees don't hold as stated. read the letter →

arxiv 2508.20327 v1 pith:H57SLIEB submitted 2025-08-28 stat.ME stat.ML

Parker Knight , Doudou Zhou , Zongqi Xia , Tianxi Cai , Junwei Lu This is my paper · ORCID
classification stat.MEstat.ML MSC 62M1562M3062H30
keywords latentfactorpointprocessFourier-Eigenembeddingspectraldensitymatrixelectronichealthrecordshigh-dimensionalpatientrepresentationclassificationclustering
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 patient-level temporal structure in electronic health records can be compressed into a small vector of spectral eigenvalues without losing the signal that separates clinical subgroups. It introduces a latent factor point process model in which thousands of observed medical codes are driven by a few unobserved Poisson processes, and defines the Fourier-Eigen embedding as the leading eigenvalues of the Fourier-transformed cross-covariance matrix of a patient's codes. Under this model, the paper proves that classifiers trained on these embeddings attain bounded risk and that spectral-initialized K-means recovers the true subgroup labels exactly when the latent intensity vectors are sufficiently separated. If correct, this gives EHR analysts a nonparametric, theoretically grounded alternative to count-based summaries and code-level embeddings, and simulations plus an Alzheimer's disease cohort analysis indicate the embeddings capture clinically meaningful heterogeneity.

What carries the argument

The load-bearing object is the spectral density matrix $F\{V^{N_i}\}(\xi) = W(\xi) D_i W(\xi)^\dagger$, where $W(\xi)$ collects the Fourier transforms of the shared transfer functions $\omega_{j\ell}$ and $D_i = \operatorname{diag}(\mu_i)$ collects the latent Poisson intensities of patient $i$. Its top $k$ eigenvalues are the Fourier-Eigen embedding used for downstream tasks. The mechanism runs through three steps: Proposition 2.1 shows the cross-covariance inherits the low rank of the latent process; Lemma 3.4 translates differences between latent intensity vectors into $\ell^2$ separation of the population embeddings; and Theorems 3.10 and 3.13 convert that separation into classifier risk bounds and exact clustering recovery. The kernel-smoothed cross-covariance estimator of Algorithm 4 supplies the finite-sample analogue of the population matrix.

What would settle it

Compute the eigenvalues of $W \operatorname{diag}(\mu) W^\dagger$ for $W = [[1,0],[1,1]]$ at $\mu = (1,1)$, $(1,2)$, $(2,1)$, and $(2,2)$: the eigenvalue shifts for the two unit directions do not combine linearly, e.g. the shift for $(1,1)$ is $(2.618, 0.382)$, not the sum of the shifts for $(1,0)$ and $(0,1)$, so no fixed $H$ can express the top eigenvalues as $H\mu$. This is a concrete case where the linearity premise after Eq. (2.10) fails.

Watch

Extended reading notes

Core claim

The central claim is that under the latent factor point process model (2.1)-(2.3), the cross-covariance matrix of the observed code process factors as $V^{N_i}(\tau) = (\omega \star D_i \star \omega^\top)(\tau)$, with $D_i = \operatorname{diag}(\mu_i)$ the latent intensity matrix, so that its Fourier transform is $W(\xi) D_i W(\xi)^\dagger$. The top $k$ eigenvalues of this spectral density matrix define the Fourier-Eigen embedding $f_i$, and the paper argues these eigenvalues inherit the subgroup differences in $\mu^{(g)}$. Theorems 3.10 and 3.13 formalize the guarantee: an empirical risk minimizer on the estimated embeddings has expected loss bounded by an approximation term, a dimension-dependent estimation term of order $d^2 T^{6/5} \exp(c T^{-1/5})$, and a Rademacher or VC complexity term, while spectral-initialized K-means recovers the latent class labels with high probability once the signal strength condition (3.12) holds.

Load-bearing premise

The proof assumes that the top eigenvalues of the spectral matrix change linearly with the latent disease-process intensities for any full-rank transfer matrix, but eigenvalue maps are generally nonlinear unless the transfer columns are orthogonal, so separation in latent intensities may not survive in the embeddings.

Editorial extensions

If this is right

  • Count-based EHR summaries are provably discarding temporal signal that the Fourier-Eigen embeddings retain, and the embeddings' classification performance improves with observation time $T$ while counts do not.
  • Only $k$ spectral eigenvalues are needed as features, so downstream classifiers can work in dimension $k$ instead of code dimension $d$, shrinking VC dimension and Rademacher complexity accordingly.
  • Unsupervised subgroup discovery is exact under a stronger signal condition than classification, so the gap between the two theorems quantifies the price of not observing labels.
  • In the Alzheimer's disease cohort, the two clusters formed from the embeddings have significantly different nursing-home admission trajectories, and the cluster membership carries a significant Cox hazard ratio, unlike the competing PMI and transformer-based embeddings.

Reading between the lines

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

  • The single-frequency choice $\xi_0$ may be more consequential than the paper suggests: averaging embeddings across a band of frequencies would test whether the guarantees are robust to misspecification of the transfer functions and would likely stabilize the estimator.
  • A direct test of the latent Poisson-driver assumption would be to apply the embedding to EHR data with known self-exciting code patterns; if the cross-covariance factorization fails, the separation guarantees should degrade in a measurable way.
  • If the linearity premise behind Lemma 3.4 fails for non-orthogonal transfer matrices, a natural repair is to replace raw eigenvalues by singular values or to orthogonalize $W(\xi)$; this would preserve the low-rank separation argument while changing the embedding's definition.
  • The Alzheimer's analysis uses nursing-home admission as a proxy for severity; validating the clusters against chart-reviewed severity labels would clarify whether the embedding discovers disease progression itself or merely correlates with care utilization.
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

3 major / 4 minor

Summary. The paper proposes a latent factor point process model for high-dimensional EHR event streams, in which an observed code process has conditional intensity driven by an unobserved low-dimensional homogeneous Poisson process. The authors define Fourier-Eigen embeddings as the top k eigenvalues of the Fourier-transformed cross-covariance matrix of the observed process, and claim theoretical guarantees for downstream classification by empirical risk minimization and clustering by K-means with spectral initialization. The theoretical results are supplemented by simulation studies and an application to clustering Alzheimer's disease patients at UPMC, with nursing home admission as a progression outcome.

Significance. If the theoretical claims were correct, the paper would contribute a principled spectral method for patient-level temporal EHR representations, with explicit generalization and clustering guarantees. The modeling idea is appealing, the algorithmic pipeline is clearly stated, and the empirical section is extensive, including a real-data analysis with interpretable cluster characterizations. However, the central theoretical bridge from latent-class separation to embedding separation rests on an eigenvalue linearity claim that is false under the paper's own assumptions, and the main risk bound in Theorem 3.10 grows with observation time rather than vanishing. These are load-bearing issues for the advertised guarantees, so the paper's core theoretical contribution is not currently established.

major comments (3)
  1. [Eq. (2.10) and Lemma 3.4] The claim after Eq. (2.10) that, for full-column-rank W(ξ), the top k eigenvalues of W(ξ) D_i W(ξ)^† can be written as H D_i for a fixed invertible H is false for non-orthogonal W. The nonzero eigenvalues of W D W^† coincide with those of D^{1/2} G D^{1/2} with G = W^† W, which is a nonlinear function of D unless G is diagonal. For k=2 and G = [[1,c],[c,1]] with c≠0, D=diag(1,0) gives sorted eigenvalues (1,0), D=diag(0,1) gives (0,1), and D=diag(1,1) gives (1+c,1-c); no fixed H can reproduce all three because any H mapping (1,0) and (0,1) to the first two cases is the identity. Since Assumption 3.3 permits arbitrary full-column-rank W, the separation between population embeddings asserted in Lemma 3.4, and hence the signal conditions (3.9) and (3.12) for Theorems 3.10 and 3.13, are unsupported for the matrices allowed by the paper's own assumptions.
  2. [Theorem 3.10, Eq. (3.10)] The second term on the right-hand side of Eq. (3.10), 2 c2 d^2 T^{6/5} exp(c3 T^{-1/5}) diam(G), grows without bound as T increases because the exponential tends to 1. Thus the claimed generalization bound does not improve with longer observation windows and is vacuous for large T. This is not a harmless presentation issue: it contradicts the paper's narrative that longer trajectories improve the embeddings, and it stands in contrast to Theorem 3.13, whose probability bound has the complementary factor exp(-c3 T^{1/5}_{min}) and tends to 1 as T_min grows. The scaling appears to be a sign or bandwidth error and must be corrected before the theorem can be interpreted as a valid risk bound.
  3. [Lemma 3.4 and Appendix] Lemma 3.4 is the sole bridge translating separation in the latent intensity vectors μ(g) into separation of the Fourier-Eigen embeddings, but its proof is not included in the manuscript; the text refers to a supplement that is not available in the submitted version. Given the false linearity premise identified above, the proof cannot be reconstructed from the stated assumptions, so the conditions (3.9) and (3.12) are unverified. This missing support is load-bearing because both main theorems depend on the lemma.
minor comments (4)
  1. [Eq. (2.10)] The factorization in Eq. (2.10) uses the transpose (W(ξ))^⊤, but for complex-valued Fourier transforms the Hermitian property stated in Lemma 3.2 requires the conjugate transpose W(ξ)^†. The notation should be corrected for consistency with the definition of A† in Section 1.2.
  2. [Assumption 3.6 vs. Section 4] The theory assumes compactly supported transfer functions (Assumption 3.6), but the main simulation uses ω_jℓ(t) = a_jℓ e^{-t^2/2}, which is not compactly supported on R_+; only the robustness study uses compactly supported kernels. The primary simulations therefore do not operate in the setting covered by the theorems.
  3. [Section 2.1, Eq. (2.2)] The sentence defining M_i as a homogeneous Poisson process 'on R^k with intensity vector μ' conflates the time domain of the point process with the dimension k of its components; the process is k-dimensional over time [0,T_i]. This wording should be clarified.
  4. [Throughout] There are several minor typographical issues, including 'has beget' in the first sentence of the Introduction, a missing comma in the sentence before Eq. (2.7), and an apparent paragraph break after Eq. (2.9). These do not affect the technical content but should be cleaned up.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the central derivation is self-contained; the main gap is an unsupported algebraic premise, not a recycled input.

full rationale

I walked the paper's derivation chain: the latent factor point process model (2.1)-(2.3) implies the cross-covariance factorization V^{N_i}(tau) = omega * D_i * omega^T and, after Fourier transformation, F{V^{N_i}}(xi) = W(xi) D_i W(xi)^dagger (Eq. 2.10). The Fourier-Eigen embedding is defined as the top k eigenvalues of this population or estimated matrix (Eqs. 2.11-2.14), and Theorems 3.10 and 3.13 state conditional guarantees: if the latent intensity vectors mu(g) are sufficiently separated, then ERM on the estimated embeddings has bounded risk and spectral-initialized K-means recovers the latent labels. This is a meaningful transfer from model parameters to downstream performance, not an identity: the embedding is computed from the observed point process, not from the mu(g) values, and the theoretical conditions are stated in terms of mu(g) separation. No fitted parameter is recycled into a 'prediction' of the same quantity, and the simulations use the proposed generative model as a stress test while the real EHR analysis provides an external benchmark. The paper's many self-citations (Lu et al. 2023, Xu et al. 2023, Zhou et al. 2023, Gan et al. 2025, Yu et al. 2025, Knight et al. 2025, Wang et al. 2025) appear in the related-literature review and are not load-bearing for the theorems; Proposition 2.1 cites Tang and Li (2023) and is proved in the supplement, so it is independent support rather than a self-citation chain. No 'uniqueness theorem' from the authors is invoked to force the embedding choice. The genuine weakness is the unsupported claim after Eq. (2.10) that the top k eigenvalues of W(xi) D_i W(xi)^dagger equal H D_i for a fixed invertible H whenever W has full column rank; that claim is false for non-orthogonal W, and Lemma 3.4 and the signal-strength conditions (3.9) and (3.12) depend on it. That is a correctness risk requiring repair, but it is an algebraic gap, not a circular reduction of the paper's conclusions to its inputs.

Assumptions & free parameters 5 free parameters · 8 assumptions · 1 invented entities

The central method rests on a specific generative model: observed codes are driven by an unobserved k-dimensional Poisson process with class-dependent intensity and shared transfer functions. The additional regularity assumptions are standard for kernel-smoothed cross-covariance estimation. The most fragile item is the unproven linearity of the eigenvalue map, which is specific to this paper and appears necessary for the theorems. Several tuning parameters, including h, Ctr, xi0, and k, are user-chosen or assumed known.

free parameters (5)
  • Bandwidth h of kernel smoothing estimator = h = c1 T^{-1/5}; h = 1 in simulations; half-week in real data
    Algorithm 4 needs a bandwidth, the constant c1 is unspecified, and the choice affects the cross-covariance estimate and downstream embeddings.
  • Truncation threshold Ctr = Ctr = 5 in simulations; constrained by Ctr <= 2 b0 in theory
    Algorithm 4 truncates lags beyond Ctr, and the value is user-chosen.
  • Frequency xi0 = xi0 = 1 in simulations and real data
    The paper says the choice does not matter in principle, but admits low and high frequencies are noisier. In practice it is a tuning parameter.
  • Latent dimension k = k = 2 in simulations; k = 5 in real data
    Remark 2.3 says k is assumed known and not estimated in this paper. The dimension of the embedding is the number of eigenvalues used.
  • Signal strength threshold constants C and C' = unspecified
    Theorems 3.10 and 3.13 require ||mu(g)-mu(r)|| above a level with unspecified constants, so the practical size of the required separation is not quantified.
assumptions (8)
  • domain assumption Assumption 3.1: Ni and Mi are stationary, Mi_l are independent across l, transfer functions are integrable, and data are drawn i.i.d. from model (2.1)-(2.3).
    This is the generative model; if real EHR data do not fit it, the theoretical guarantees do not apply.
  • domain assumption Assumption 3.3: there exists a frequency xi* at which W(xi*) is full column rank.
    Needed for the embeddings to inherit the latent signal. The paper notes it is satisfied under a proportional-kernel example.
  • domain assumption Assumption 3.6: transfer functions have support within [-b0, b0].
    Needed to apply the concentration bound from Chen et al. (2017); the paper states this is stronger than the cited assumption.
  • domain assumption Assumption 3.7: the L1 norm of each transfer function is bounded by gamma < 1.
    Stability condition for the point process; it limits the strength of excitation.
  • domain assumption Assumption 3.8: the cross-covariance matrix is theta-Lipschitz.
    Regularity condition for kernel smoothing.
  • ad hoc to paper The top k eigenvalues of W D W^dagger are an invertible linear transform H D of the latent intensity vector.
    Appears after Eq. (2.10) and supports Lemma 3.4. It is false for general non-orthogonal W, so the separation guarantee is not established.
  • domain assumption The latent process is homogeneous Poisson with class-specific intensity vector and transfer functions shared across patients.
    Model (2.2)-(2.3) restricts the temporal dynamics to be driven by latent Poisson events.
  • standard math Oracle classifier and Rademacher complexity framework.
    Standard learning theory used in Theorem 3.10.
invented entities (1)
  • Latent factor Poisson process M_i
    purpose: Unobserved low-dimensional process that drives the conditional intensity of all observed codes and carries the subgroup signal mu(g).
    No falsifiable measurement outside the paper is provided. It is a modeling device, and the paper does not predict a new observable that would distinguish this latent process from other latent drivers.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Latent Factor Point Processes for Patient Representation in Electronic Health Records." pith.science (2026). https://pith.science/paper/H57SLIEB

@misc{pith2026250820327,
  author       = {Pith},
  title        = {Pith review of: Latent Factor Point Processes for Patient Representation in Electronic Health Records},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/H57SLIEB}},
  note         = {Machine review of arXiv:2508.20327}
}
read the original abstract

Electronic health records (EHR) contain valuable longitudinal patient-level information, yet most statistical methods reduce the irregular timing of EHR codes into simple counts, thereby discarding rich temporal structure. Existing temporal models often impose restrictive parametric assumptions or are tailored to code level rather than patient-level tasks. We propose the latent factor point process model, which represents code occurrences as a high-dimensional point process whose conditional intensity is driven by a low dimensional latent Poisson process. This low-rank structure reflects the clinical reality that thousands of codes are governed by a small number of underlying disease processes, while enabling statistically efficient estimation in high dimensions. Building on this model, we introduce the Fourier-Eigen embedding, a patient representation constructed from the spectral density matrix of the observed process. We establish theoretical guarantees showing that these embeddings efficiently capture subgroup-specific temporal patterns for downstream classification and clustering. Simulations and an application to an Alzheimer's disease EHR cohort demonstrate the practical advantages of our approach in uncovering clinically meaningful heterogeneity.

Figures

Figures reproduced from arXiv: 2508.20327 by the authors.

Figure 1
Figure 1. Average AUC over 100 replications. The x axis denotes the signal strength ∥µ (0) −µ (1)∥2. kernel function β(t) in the set β(t) ∈ | sin(t)|1(t<π) t+1 ,(1− √ t)1(t<1),(1−t)1(t<1), 4 −t 1(t<2) [PITH_FULL_IMAGE:figures/full_fig_p018_1.png] view at source ↗
Figure 2
Figure 2. Average adjusted Rand index over 100 replications. The [PITH_FULL_IMAGE:figures/full_fig_p019_2.png] view at source ↗
Figure 3
Figure 3. Kaplan-Meier curves for time of nursing home admission, colored by the cluster assigned [PITH_FULL_IMAGE:figures/full_fig_p021_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Word clouds of EHR code descriptions for the slow and fast progression clusters obtained [PITH_FULL_IMAGE:figures/full_fig_p022_4.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

20 extracted references · 16 canonical work pages

  1. [1]

    Abul-Husn, N. S. and Kenny, E. E. (2019). Personalized medicine and the power of electronic health records. Cell 177 58–69. Agrawal, M., Hegselmann, S., Lang, H., Kim, Y. and Sontag, D. (2022). Large language models are few-shot clinical information extractors. arXiv preprint arXiv:2205.12689 . Akter, S., Liu, Z., Simoes, E. J. and Rao, P. (2025). Using m...

    arXiv 2019

  2. [2]

    , He, F.-F

    Zhang, C.-Y. , He, F.-F. , Su, H. , Zhang, C. and Meng, X.-F. (2020). Association between chronic kidney disease and alzheimer’s disease: an update. Metabolic Brain Disease 35 883–894. Zhang, X. S. , Tang, F., Dodge, H. H. , Zhou, J. and Wang, F. (2019). Metapred: Meta- learning for clinical risk prediction with limited patient electronic health records. ...

    work page arXiv 2020

  3. [3]

    Beam, A. L. , Kompa, B., Schmaltz, A., Fried, I., Weber, G., Palmer, N., Shi, X., Cai, T. and Kohane, I. S. (2020). Clinical concept embeddings learned from massive sources of multimodal medical data. In Pacific Symposium on Biocomputing. Pacific Symposium on Bio- computing, vol

    work page 2020

  4. [4]

    , Zhou, J

    Li, L. , Zhou, J. , Gao, Z. , Hua, W. , Fan, L., Yu, H. , Hagen, L. , Zhang, Y. , Assimes, T. L., Hemphill, L. et al. (2024a). A scoping review of using large language models (llms) to investigate electronic health records (ehrs). arXiv preprint arXiv:2405.03066 . Li, Q., Yang, X., Xu, J., Guo, Y., He, X., Hu, H., Lyu, T., Marra, D., Miller, A., Smith, G....

    arXiv 2024

  5. [25]

    24 Bonnet, A., Herrera, M. M. and Sangnier, M. (2021). Maximum likelihood estimation for hawkes processes with self-excitation or inhibition. Statistics & Probability Letters 179 109214. Cai, B. , Zhang, J. and Guan, Y. (2024). Latent network structure learning from high- dimensional multivariate point processes. Journal of the American Statistical Associ...

    work page Pith review arXiv 2021

  6. [76]

    30 Vapnik, V. (2013). The nature of statistical learning theory . Springer science & business media. Venkatesh, S., Wang, L., Morris, M., Moro, M., Srivastava, R., Han, Y., Patira, R., Berman, S., Lopez, O., Visweswaran, S. et al.(2025). Leveraging electronic health records to examine differential clinical outcomes in people with alzheimer’s disease. medR...

    work page Pith review arXiv 2013

  7. [86]

    K., Delen, D

    Reddy, B. K., Delen, D. and Agrawal, R. K. (2019). Predicting and explaining inflammation in crohn’s disease patients using predictive analytics methods and electronic medical record data. Health Informatics Journal 25 1201–1218. 29 Rockenschaub, P., Nguyen, V., Aldridge, R. W., Acosta, D., Garc´ıa-G´omez, J. M. and S´aez, C. (2020). Data-driven discovery...

    work page 2019

  8. [151]

    , Guo, Z

    Hou, J. , Guo, Z. and Cai, T. (2023). Surrogate assisted semi-supervised inference for high dimensional risk prediction. Journal of Machine Learning Research 24 1–58. Hu, M., Shi, X. and Song, P. X.-K. (2024). Collaborative inference for treatment effect with distributed data-sharing management in multicenter studies. Statistics in Medicine 43 2263–

    work page 2023

Show all 20 references
  1. [194]

    and Yu, H

    Yang, Z., Mitra, A., Liu, W., Berlowitz, D. and Yu, H. (2023). Transformehr: transformer- based encoder-decoder generative model to enhance prediction of disease outcomes using elec- tronic health records. Nature Communications 14

    2023

  2. [198]

    Gronsbell, J. L. and Cai, T. (2018). Semi-supervised approaches to efficient evaluation of model prediction performance. Journal of the Royal Statistical Society Series B: Statistical Methodology 80 579–594. 26 Guo, X. , Wei, W. , Liu, M. , Cai, T. , Wu, C. and Wang, J. (2023)...

    2018

  3. [259]

    Si, Y. , Du, J. , Li, Z. , Jiang, X. , Miller, T. , Wang, F., Zheng, W. J. and Roberts, K. (2021). Deep representation learning of patient data from electronic health records (ehr): A systematic review. Journal of Biomedical Informatics 115 103671. Stanciu, G. D., Ababei, D. C...

    2021

  4. [563]

    A., Corbin, C

    Steinberg, E., Jung, K., Fries, J. A., Corbin, C. K., Pfohl, S. R. and Shah, N. H. (2021). Language models are an effective representation learning technique for electronic health record data. Journal of Biomedical Informatics 113 103637. Sun, J., Liao, K. P. and Cai, T. (2024...

    2021

  5. [1107]

    and Liu, M

    Zhan, K., Xiong, X., Guo, Z., Cai, T. and Liu, M. (2024). Transfer learning targeting mixed population: A distributional robust perspective. arXiv preprint arXiv:2407.20073

    2024 arXiv

  6. [1622]

    G., Xiong, X., Hong, C., Weber, G

    Wang, X., Zhang, H. G., Xiong, X., Hong, C., Weber, G. M., Brat, G. A., Bonzel, C.-L., Luo, Y., Duan, R., Palmer, N. P. et al. (2022). Survmaximin: robust federated approach to transporting survival risk prediction models. Journal of Biomedical Informatics 134 104176. Wang, Y....

    2022 arXiv

  7. [2202]

    , Peissig, P

    Bao, Y., Kuang, Z. , Peissig, P. , Page, D. and Willett, R. (2017). Hawkes process mod- eling of adverse drug reactions with longitudinal observational data. In Machine Learning for Healthcare Conference. PMLR. Bartholomew, D. J., Knott, M. and Moustaki, I. (2011). Latent vari...

    2017

  8. [2279]

    Jiang, S., Gai, X., Treggiari, M. M. , Stead, W. W. , Zhao, Y., Page, C. D. and Zhang, A. R. (2024). Soft phenotyping for sepsis via ehr time-aware soft clustering.Journal of Biomedical Informatics 152 104615. Knight, P. , Jobe, N. I. and Duan, R. (2025). Fast and robust invar...

    2024 arXiv

  9. [4439]

    and Cai, T

    Xiong, X., Guo, Z. and Cai, T. (2023b). Distributionally robust transfer learning. arXiv preprint arXiv:2309.06534 . Xu, Z., Chou, J., Zhang, X. S. , Luo, Y., Isakova, T., Adekkanattu, P., Ancker, J. S. , Jiang, G., Kiefer, R. C., Pacheco, J. A. et al. (2020). Identifying sub-...

    2023 arXiv

  10. [5968]

    and Tighe, P

    Giordano, C., Brennan, M., Mohamed, B., Rashidi, P., Modave, F. and Tighe, P. (2021). Accessing artificial intelligence for clinical decision-making. Frontiers in Digital Health 3 645232. Goldstein, B. A. , Navar, A. M., Pencina, M. J. and Ioannidis, J. P. (2016). Opportuni- t...

    2021

  11. [7155]

    and Cai, T

    Lu, J., Yin, J. and Cai, T. (2023). Knowledge graph embedding with electronic health records data via latent graphical block model. arXiv preprint arXiv:2305.19997 . 28 Luo, Y. , Liu, Z. and Liu, Q. (2022). Deep stable representation learning on electronic health records. In 2...

    2023 arXiv

  12. [7857]

    , Feng, Z

    Yu, J. , Feng, Z. , Lu, J. , Cai, T. and Zhou, D. (2025). Time-aware attention for enhanced electronic health records modeling. arXiv preprint arXiv:2507.14847 . Yuan, W., Beaulieu-Jones, B. K. , Yu, K.-H., Lipnick, S. L. , Palmer, N., Loscalzo, J., Cai, T. and Kohane, I. S. (...

    2025 arXiv

Pith tools

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