Pith. sign in

REVIEW 4 major objections 4 minor 49 references

Automated Hierarchical Graph Construction for Multi-source 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 This paper claims that a fully automated three-stage pipeline—neural optimal transport alignment, hyperbolic embedding, and recursive grouping—can both harmonize medical codes across institutions and construct clinically coherent…

desk verdict A sensible pipeline for a real EHR harmonization problem, but the recursive grouping criterion is unproven and the evaluation is partly circular; worth refereeing, but major revision needed. read the letter →

arxiv 2509.06576 v1 pith:KVGZ2VXO submitted 2025-09-08 stat.ML cs.LG

Yinjie Wang , Doudou Zhou , Yue Liu , Junwei Lu , Tianxi Cai This is my paper · ORCID
classification stat.MLcs.LG MSC 62H3068T0762P10
keywords electronichealthrecordsmedicalcodeharmonizationoptimaltransporthyperbolicembeddingshierarchicalclusteringlatenttreemodelsmulti-institutionalEHRdataVAlocallaboratorycodes
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

MASH (Multi-source Automated Structured Hierarchy) is a fully automated pipeline that tackles two obstacles to multi-institution electronic health record research at once: hospitals record the same clinical idea under different codes, and many of those codes have no usable hierarchy. It aligns co-occurrence and text-based embeddings across the U.S. Department of Veterans Affairs (VA) and Mass General Brigham (MGB) systems with neural optimal transport, then trains hyperbolic embeddings and recursively recovers a latent tree, inserting abstract concepts where needed. If the method works as claimed, researchers can pool records from systems with incompatible coding practices, and previously unstructured local codes—especially the VA's extensive laboratory codes—gain an interpretable taxonomy. The paper reports that MASH reconstructs known structures such as PheCode and RxNorm, integrates local lab codes with LOINC-mapped concepts, and outperforms baselines on embedding and hierarchy-quality metrics.

What carries the argument

The carrying mechanism is the recursive grouping step built on additivity of distances along tree paths. Given a node set $V$, the paper computes $D_{ij} = \max_{k\in V\setminus\{i,j\}} (d(z_i,z_k)-d(z_j,z_k)) - \min_{k\in V\setminus\{i,j\}} (d(z_i,z_k)-d(z_j,z_k))$, and treats $D_{ij}=0$ as the test for whether $i$ and $j$ belong to the same "bottom set"—nodes that are pairwise siblings or in a parent–child relation, with non-parent nodes being leaves. $k$-means on the $D$ matrix forms bottom sets, a latent node is inserted above each, distances are recomputed using the additivity relation, and the process iterates until the whole tree is built. The distances are Lorentz-model hyperbolic embeddings trained with an additivity loss, an information-preserving loss, and an InfoNCE contrastive loss, initialized from the optimal-transport-aligned Euclidean embeddings.

What would settle it

Construct a small known tree—say a root with two children A and B, where A has a child C and a grandchild D—and compute the hyperbolic or information distances. For the uncle C and nephew D, the criterion $D_{CD}$ will be zero for every node outside A's subtree, although C and D are not siblings or parent–child. Running Algorithm 2 on this tree should therefore place C and D in the same bottom set; if it does, the central recovery claim is contradicted.

Watch

Extended reading notes

Core claim

On its own terms, the paper's central claim is that code harmonization and hierarchy construction can be solved jointly by one three-stage procedure: an optimal transport map aligns institution-specific embedding spaces, hyperbolic-space training imposes tree-like geometry with an additivity loss plus contrastive supervision, and recursive grouping recovers the latent tree from the learned distances. This yields a common representational space for codes from different health systems and hierarchical graphs for diagnosis, medication, and laboratory codes. A specific claimed outcome is the first automated hierarchy for the VA's collection of local laboratory codes, which have no standardized ontology support.

Load-bearing premise

The entire hierarchy rests on the equivalence $D_{ij}=0$ if and only if codes $i$ and $j$ lie in the same bottom set; this equivalence fails for nodes at different depths, such as an uncle and nephew, because the distance difference to outside nodes is still constant even though the two nodes do not attach at the same level.

Editorial extensions

If this is right

  • Institutions can contribute to shared analyses without manually mapping every local code to a standard ontology; aggregated co-occurrence matrices and text descriptions are the required inputs.
  • VA local laboratory codes, previously lacking any curated hierarchy, become navigable and can anchor cohort definitions and clinical feature selection.
  • Established ontologies such as PheCode and RxNorm are reproduced, and the same machinery can extend them when new codes enter the system.
  • The pipeline tolerates partial overlap between institutions: only a set of shared codes is needed to fit the optimal transport map.

Reading between the lines

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

  • The paper leaves implicit that the design is naturally suited to federated learning settings: only summary SPPMI matrices and code descriptions need to leave each institution, never patient-level records.
  • The reported AUC and NMI gains are consistent with the claim, but the ablation shown means a reader should not attribute all gains to any single stage without further experiments.
  • A direct depth-aware audit on a known ontology—counting how often Algorithm 2 merges an uncle with a nephew—would test whether the $D_{ij}$ criterion's depth-sensitivity degrades the hierarchies in practice.
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 MASH, a three-stage framework for harmonizing multi-institution electronic health record codes and automatically constructing hierarchical graphs over them. Stage I aligns institution-specific SPPMI-SVD embeddings through neural optimal transport, Stage II maps the harmonized embeddings into hyperbolic space with additivity and contrastive losses, and Stage III recovers a latent tree by recursively identifying "bottom sets" via the criterion D_ij=0 and k-means. The authors apply MASH to VA and MGB data for PheCode diagnoses, RxNorm medications, and LOINC/local laboratory codes, reporting improvements over baselines in embedding-quality AUC, NMI/ARI against known ontologies, sibling precision/sensitivity, and GPT-4-based interpretability scores. The claimed headline outcomes are a common representational space enabling cross-system analyses and automated hierarchies for previously unstructured VA local laboratory codes.

Significance. If the technical claims were fully supported, MASH would address a real and timely barrier in multi-institutional EHR research: the heterogeneity of local coding systems and the absence of curated hierarchies for local codes. The paper has genuine strengths: it operates at substantial scale (12.6M VA and 2.5M MGB patients), integrates several complementary information sources, provides ablation comparisons, and includes a simulation study in the supplement. However, the central hierarchy-recovery criterion is not rigorously established, and the main empirical evaluations contain multiple sources of circularity. These issues directly affect the paper's two headline claims, so the significance is currently conditional rather than demonstrated.

major comments (4)
  1. [Supplementary Section B.3, Eq. (6), Algorithm 2] The claimed equivalence D_ij=0 iff i and j belong to the same bottom set is not proved. The converse direction in Supp B.3 only considers nodes k outside the subtree rooted at the nearest common ancestor p. For an uncle-nephew pair (i a child of p, j a grandchild of p through a sibling c of i), all such k yield a constant value of d(z_i,z_k)-d(z_j,z_k), so D_ij=0 even though i and j are neither siblings nor in a parent-child relationship. If the current node set V contains no node inside the subtrees that would expose the depth difference, the recursive grouping in Algorithm 2 will incorrectly merge such nodes into one bottom set and introduce a false latent parent. The proof must either handle k inside p's subtree or state an additional assumption (such as equal depth for all nodes in V) that is not satisfied in the EHR setting.
  2. [Sections 2.3 and 4.2.1, Table 3] The evaluation of known-code hierarchies is circular. PheCode, NDF-RT/RxNorm, and LOINC hierarchies are used as supervision in Stage II (Section 2.3) and then scored with NMI and ARI against those same ontologies in Table 3. For local laboratory codes, the reference tree in Section 4.2.1 is built from a manually curated LOINC-to-local mapping, and the same kind of LOINC-local links are used to create supervised training pairs in Section 2.3. This common-source overlap can inflate all reported alignment numbers. The authors should separate the supervision and evaluation mappings, for example by holding out a portion of the curated mapping that is never used in training.
  3. [Table 5 and Supplementary Section A.1] The semantic interpretability evaluation is self-referential. GPT-4o generates the latent-node annotations using the prompts in Table S3, and the same model is then asked to judge whether child annotations are sub-concepts of the parent and whether sibling annotations are distinct. This measures prompt consistency rather than independent clinical validity. An independent judge, a different model, or human review is needed before the near-perfect hierarchy and divergence scores in Table 5 can be interpreted as evidence of clinically meaningful structure.
  4. [Supplementary Section A.1, Section 3.3] The description of MASH as a fully automated hierarchy construction framework is not supported by the implementation details. Before Algorithm 2 runs, the user fixes 21 PheCode categories, 32 RxNorm categories, and the first 5 levels of the LOINC hierarchy, and the construction of latent nodes also reuses existing parent-child relations when labels are available. These manual choices should be explicitly scoped as user-specified inputs, and the paper should assess how sensitive the results in Tables 3 and 4 are to these choices.
minor comments (4)
  1. [Equation (1)] The notation for the marginal total is inconsistent: C^k(i,·) = Σ_i C^k(i,k) should presumably read C^k(i,·) = Σ_j C^k(i,j), since the sum should be over the second index.
  2. [Figure 3 and Table S4] There are small typos: the legend in Figure 3 uses "MESHI" instead of "MASH", and Table S4 says "consitent" instead of "consistent."
  3. [Data Availability] No code repository is provided. Given that the VA and MGB data cannot be shared, the supplemental simulation is helpful, but providing the exact code used for the pipeline would improve reproducibility.
  4. [Tables 3 and 4] The hierarchy evaluation reports only point estimates without confidence intervals or multiple-run variation. Since k-means initialization and the silhouette heuristic introduce randomness, the authors should report the variability of NMI/ARI and precision/sensitivity across seeds.

Circularity Check

3 steps flagged · score 6.0 of 10

MASH's headline hierarchy results are partly in-sample: known-code hierarchies are supervised by the very PheCode/NDF-RT/LOINC structures used for NMI/ARI, the VA local-lab evaluation uses the same LOINC mappings that enter as training labels, and GPT-4 grades its own annotations; the core algorithm itself is not circular.

  1. fitted input called prediction [Section 2.3 (hierarchical supervision) and Section 4.2.1 (NMI/ARI evaluation)]
    "When available, curated hierarchies serve as valuable supervision signals and benchmarks. They encode parent–child relations that guide embeddings toward clinically meaningful structure. For diagnoses, hierarchical information is defined within the PheCode system. For medications, we relied on the VA classification system through NDF-RT (Nelson et al., 2011). For laboratory codes, we used the LOINC Multiaxial framework (McDonald et al., 2004). ..."

    The PheCode, RxNorm/NDF-RT, and LOINC parent–child relations are used as supervision for the hyperbolic embeddings (the additivity loss and contrastive labels), and then the reported NMI/ARI measures agreement with those same ontologies. The 'reconstruction' of known hierarchies is therefore partly a check of how well the model absorbed the supervised labels, not an independent recovery. The held-out split of labeled pairs mitigates the pair-level contrastive loss, but Supplementary A.1 states that 'the additivity loss is computed for PheCode, RxNorm, and all laboratory codes', so the full hierarchy structure used for the headline NMI/ARI is in the training objective.

  2. fitted input called prediction [Section 2.3 (LOINC-local lab labeled pairs) and Section 4.2.1 (local lab reference tree)]
    "To evaluate the hierarchies constructed for these codes, we leveraged available mappings from LOINC to local laboratory codes to impute hierarchical relations. ... These labeled pairs, covering PheCodes, RxNorm, laboratory codes, and LOINC–local lab links, were split evenly into training and testing sets for supervision and evaluation. ... Local laboratory codes do not have a predefined ontology. To evaluate their integration, we construct a reference tree by mapping 1897 VA local laboratory codes to LOINC or LP concepts using a manually curated dictionary."

    The evaluation reference tree for VA local laboratory codes is built from LOINC-to-local-lab mappings, and the same type of LOINC–local lab links are explicitly used as supervised labeled pairs for training. The paper does not state that the 1897-code reference tree is restricted to the held-out testing split; if any training links are among those 1897 mappings, the reported NMI/ARI and sibling precision/sensitivity for local labs are partly in-sample. This undercuts the headline claim that MASH 'generates the first automated hierarchies for unstructured local laboratory codes', because the evaluation standard is derived from the supervision source rather than an external gold standard.

1 more flagged steps
  1. self definitional [Section 3.3 (GPT-4 annotation) and Section 4.2.1 (GPT-4 semantic scores)]
    "Finally, to improve interpretability, we use GPT-4o (Hurst et al., 2024), with prompt given in Supplementary Section A.1, to annotate each latent node with a concise human-readable summary synthesized from its descendants. ... Beyond structure, we assess interpretability using GPT-4 annotations... Each internal node in the constructed tree is labeled using our annotation module. Two scores are defined: 1) Hierarchy score: For each internal node, GPT-4 assigns a score of 1 if all child nodes represent valid sub-concepts of the parent annotation and 0 otherwise..."

    The same GPT-4o model generates the latent-node summaries and then evaluates whether those summaries are valid, distinct sub-concepts. A model instructed to produce 'distinctly different' cluster summaries will tend to pass its own divergence test by construction, so the hierarchy and divergence scores measure self-consistency rather than external clinical or semantic validity. Table 5's near-perfect interpretability scores are therefore not independent evidence that the constructed hierarchies are clinically meaningful.

full rationale

The algorithmic derivation chain itself—neural optimal transport alignment, hyperbolic embedding, and recursive grouping—does not reduce to its inputs by construction. I found no load-bearing self-citation chain, no imported uniqueness theorem, and no ansatz smuggled in via citation: the method's core mechanism is independent of its evaluation. The score of 6 reflects three evaluation circularities in the paper's headline claims. First, the reconstructed PheCode, RxNorm, and LOINC hierarchies are supervised by those same ontologies through the additivity loss and contrastive labels, and then scored with NMI/ARI against those same ontologies; this is at least partially an in-sample fit. Second, the VA local laboratory hierarchy, a central contribution, is evaluated against a reference tree built from LOINC-to-local-lab mappings, while LOINC–local lab links are also used as supervised training labels, with no statement that the 1897-code reference tree uses only held-out mappings. Third, GPT-4o both annotates latent nodes and grades its own annotations for hierarchy and divergence scores, making the near-perfect semantic scores a self-consistency check rather than an external validation. The Supplementary B.3 converse proof that D_ij = 0 implies membership in a common bottom set is mathematically incomplete—the argument only treats k outside the nearest common ancestor's subtree—but this is a correctness gap, not a circularity, so it does not raise the circularity score by itself. Similarly, the Supplementary A.4 simulation is generated from CODER embeddings and label-guided updates, sharing the method's own priors, but it is presented as a replication aid rather than a first-principles prediction. Overall, the central algorithm is not circular by construction, but several of the load-bearing evaluation results are partially forced by their own supervision, giving a partial-circularity score of 6.

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

The central claim rests on a supervised pipeline where ontologies and manual categories are inputs, not on a self-contained derivation. The additivity assumption is only approximately enforced, the local lab reference mapping likely enters both training and evaluation, and GPT-4o is used as annotator and judge. These factors raise the circularity burden and limit the independence of the validation.

free parameters (5)
  • OT hyperparameters (eta, omega, M, N1, N2, learning rate) = eta=1e-5, omega=1e-4, M=2, N1=1000, N2=50, lr=1e-3
    Set in Algorithm 1 without sensitivity analysis; they control how strongly the transport plan and mapping are regularized.
  • Hyperbolic loss weights and training schedule = w_a=0.1, w_c=0.1, w_e=1, lr=0.01, 1000 epochs, 100 warm-up
    Chosen by hand in Supp A.1; they balance additivity, information preservation, and contrastive supervision, directly shaping the hierarchy.
  • Manual top-level category partitions = 21 PheCode classes; 32 RxNorm VA drug classes; LOINC first 5 levels
    Supplied by the authors before recursive grouping; these categories determine the upper levels of the output graph and are not learned.
  • Rare-code frequency thresholds = VA at least 5000; MGB at least 1000
    Codes below these frequencies are excluded; this changes the code set and the resulting graph and embedding evaluation.
  • Number of clusters in recursive k-means = chosen by silhouette
    Bottom-set cardinality is selected by silhouette scores; no validation of the resulting graph sensitivity.
assumptions (5)
  • standard math Information distances in a latent tree are additive along paths (Supp B.1, Eq. S3).
    Classical latent-tree result used to justify the additivity loss and the D_ij criterion.
  • ad hoc to paper Trained hyperbolic embeddings satisfy the parent-child additivity relation in Eq. (5) closely enough for D_ij to identify bottom sets.
    The additivity loss only penalizes violations; the proof in Supp B.3 assumes exact additivity, which real embeddings do not have.
  • domain assumption SPPMI co-occurrence statistics and CODER text embeddings together contain enough signal to infer clinically valid hierarchies, including for VA local lab codes.
    For local lab codes there is no gold hierarchy; the evaluation relies on a manually curated LOINC mapping that may also be used as supervision.
  • domain assumption GPT-4o provides reliable semantic labels and relevance scores.
    Used as ground truth for feature selection and for hierarchy and divergence scores, while also generating the annotations being scored.
  • domain assumption Codes shared between VA and MGB are semantically equivalent anchors for OT alignment.
    The alignment is trained on shared code embeddings; if shared codes are not truly equivalent across institutions, the harmonized space inherits the error.
invented entities (1)
  • Latent parent nodes in constructed hierarchy graphs
    purpose: Unobserved abstract concepts introduced to connect leaf medical codes in a tree
    These nodes are outputs of the algorithm, not independently measured clinical concepts; they are validated only by GPT-4 annotations and ontology overlap.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Automated Hierarchical Graph Construction for Multi-source Electronic Health Records." pith.science (2026). https://pith.science/paper/KVGZ2VXO

@misc{pith2026250906576,
  author       = {Pith},
  title        = {Pith review of: Automated Hierarchical Graph Construction for Multi-source Electronic Health Records},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/KVGZ2VXO}},
  note         = {Machine review of arXiv:2509.06576}
}
read the original abstract

Electronic Health Records (EHRs), comprising diverse clinical data such as diagnoses, medications, and laboratory results, hold great promise for translational research. EHR-derived data have advanced disease prevention, improved clinical trial recruitment, and generated real-world evidence. Synthesizing EHRs across institutions enables large-scale, generalizable studies that capture rare diseases and population diversity, but remains hindered by the heterogeneity of medical codes, institution-specific terminologies, and the absence of standardized data structures. These barriers limit the interpretability, comparability, and scalability of EHR-based analyses, underscoring the need for robust methods to harmonize and extract meaningful insights from distributed, heterogeneous data. To address this, we propose MASH (Multi-source Automated Structured Hierarchy), a fully automated framework that aligns medical codes across institutions using neural optimal transport and constructs hierarchical graphs with learned hyperbolic embeddings. During training, MASH integrates information from pre-trained language models, co-occurrence patterns, textual descriptions, and supervised labels to capture semantic and hierarchical relationships among medical concepts more effectively. Applied to real-world EHR data, including diagnosis, medication, and laboratory codes, MASH produces interpretable hierarchical graphs that facilitate the navigation and understanding of heterogeneous clinical data. Notably, it generates the first automated hierarchies for unstructured local laboratory codes, establishing foundational references for downstream applications.

Figures

Figures reproduced from arXiv: 2509.06576 by the authors.

Figure 1
Figure 1. Pipeline of MASH in three stages: (I) Embeddings from co-occurrence and [PITH_FULL_IMAGE:figures/full_fig_p009_1.png] view at source ↗
Figure 2
Figure 2. Recursive grouping algorithm example. Dashed lines and grey nodes represent [PITH_FULL_IMAGE:figures/full_fig_p014_2.png] view at source ↗
Figure 3
Figure 3. Feature selection performance. Average correlations: SDNE = 0 [PITH_FULL_IMAGE:figures/full_fig_p017_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Partial views of constructed hierarchy graphs for (a) diagnosis codes (PheCode), [PITH_FULL_IMAGE:figures/full_fig_p021_4.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

49 extracted references · 39 canonical work pages

  1. [1]

    Armijo, L. (1966). Minimization of functions having lipschitz continuous first partial derivatives. Pacific Journal of mathematics\/ 16\/ (1), 1--3

    work page 1966

  2. [2]

    Beam, A. L., B. Kompa, A. Schmaltz, I. Fried, G. Weber, N. Palmer, X. Shi, T. Cai, and I. S. Kohane (2020). Clinical concept embeddings learned from massive sources of multimodal medical data. In Pacific Symposium on Biocomputing 2020 , pp.\ 295--306. World Scientific

    work page 2020

  3. [3]

    Bennett, C. C. (2012). Utilizing rxnorm to support practical computing applications: capturing medication history in live electronic health records. Journal of Biomedical Informatics\/ 45\/ (4), 634--641

    work page 2012

  4. [4]

    Bodenreider, O. (2004). The unified medical language system (umls): integrating biomedical terminology. Nucleic Acids Research\/ 32\/ (suppl\_1), D267--D270

    work page 2004

  5. [5]

    Chen, J., Y. Chen, Z. Wang, H. Zhang, and X. Peng (2024). Spectral embedding fusion for incomplete multiview clustering. IEEE Transactions on Image Processing\/ 33 , 4116--4130

    work page 2024

  6. [6]

    Choi, M. J., V. Y. Tan, A. Anandkumar, and A. S. Willsky (2011). Learning latent tree graphical models. The Journal of Machine Learning Research\/ 12 , 1771--1812

    work page 2011

  7. [7]

    Mauksch, C

    Chunchu, K., L. Mauksch, C. Charles, V. Ross, and J. Pauwels (2012). A patient centered care plan in the ehr: improving collaboration and engagement. Families, Systems, & Health\/ 30\/ (3), 199

    work page 2012

  8. [8]

    Cowell, R. G., A. P. Dawid, S. L. Lauritzen, and D. J. Spiegelhalter (1999). Probabilistic Networks and Expert Systems . Statistics for Engineering and Information Science. New York: Springer-Verlag

    work page 1999

Show all 49 references
  1. [9]

    Erd \"o s, P. L., M. A. Steel, L. Sz \'e kely, and T. J. Warnow (1999). A few logs suffice to build (almost) all trees: Part ii. Theoretical Computer Science\/ 221\/ (1-2), 77--118

    1999

  2. [10]

    Unger, A

    Federico, P., J. Unger, A. Amor-Amor \'o s, L. Sacchi, D. Klimov, and S. Miksch (2015). Gnaeus: Utilizing clinical guidelines for knowledge-assisted visualisation of ehr cohorts. In EuroVA@ EuroVis , pp.\ 79--83

    2015

  3. [11]

    Goldberg, Y. and O. Levy (2014). word2vec explained: deriving mikolov et al.'s negative-sampling word-embedding method. arXiv preprint arXiv:1402.3722\/

    2014 arXiv

  4. [12]

    Gu, Y., R. Tinn, H. Cheng, M. Lucas, N. Usuyama, X. Liu, T. Naumann, J. Gao, and H. Poon (2021). Domain-specific language model pretraining for biomedical natural language processing. ACM Transactions on Computing for Healthcare (HEALTH)\/ 3\/ (1), 1--23

    2021

  5. [13]

    He, Y., J. Chen, D. Antonyrajah, and I. Horrocks (2022). Bertmap: a bert-based ontology alignment system. In Proceedings of the AAAI Conference on Artificial Intelligence , Volume 36, pp.\ 5684--5691

    2022

  6. [14]

    Hong, C., E. Rush, M. Liu, D. Zhou, J. Sun, A. Sonabend, V. M. Castro, P. Schubert, V. A. Panickan, T. Cai, et al. (2021). Clinical knowledge extraction via sparse embedding regression (keser) with multi-center large scale electronic health record data. npj Digital Medicine\/ ...

    2021

  7. [15]

    Hou, C., F. Nie, D. Yi, and D. Tao (2014). Discriminative embedded clustering: A framework for grouping high-dimensional data. IEEE Transactions on Neural Networks and Learning Systems\/ 26\/ (6), 1287--1299

    2014

  8. [16]

    Hoyos-Idrobo, A. (2020). Aligning hyperbolic representations: an optimal transport-based approach. arXiv preprint arXiv:2012.01089\/

    2020 arXiv

  9. [17]

    Hripcsak, G. and D. J. Albers (2013). Next-generation phenotyping of electronic health records. Journal of the American Medical Informatics Association\/ 20\/ (1), 117--121

    2013

  10. [19]

    Ruder, R

    Kementchedjhieva, Y., S. Ruder, R. Cotterell, and A. S gaard (2018, October). Generalizing P rocrustes analysis for better bilingual dictionary induction. In Proceedings of the 22nd Conference on Computational Natural Language Learning , Brussels, Belgium, pp.\ 211--220. Assoc...

    2018

  11. [20]

    Papadopoulos, M

    Krioukov, D., F. Papadopoulos, M. Kitsak, A. Vahdat, and M. Bogun \'a (2010). Hyperbolic geometry of complex networks. Physical Review E---Statistical, Nonlinear, and Soft Matter Physics\/ 82\/ (3), 036106

    2010

  12. [21]

    Kush, R. D., E. Helton, F. W. Rockhold, and C. D. Hardison (2008). Electronic health records, medical research, and the tower of babel. New England Journal of Medicine\/ 358\/ (16), 1738--1740

    2008

  13. [22]

    Li, J., Y. Wu, J. Zhao, and K. Lu (2016). Low-rank discriminant embedding for multiview learning. IEEE Transactions on Cybernetics\/ 47\/ (11), 3516--3529

    2016

  14. [23]

    Lipton, Z. (2015). Learning to diagnose with lstm recurrent neural networks. arXiv preprint arXiv:1511.03677\/

    2015 arXiv

  15. [24]

    McDonald, C., S. Huff, J. Suico, and K. Mercer (2004). Logical observation identifiers names and codes (loinc ) users' guide. Indianapolis: Regenstrief Institute\/

    2004

  16. [25]

    McDonald, C. J., S. M. Huff, J. G. Suico, G. Hill, D. Leavelle, R. Aller, A. Forrey, K. Mercer, G. DeMoor, J. Hook, et al. (2003). Loinc, a universal standard for identifying laboratory observations: a 5-year update. Clinical Chemistry\/ 49\/ (4), 624--633

    2003

  17. [26]

    Sutskever, K

    Mikolov, T., I. Sutskever, K. Chen, G. S. Corrado, and J. Dean (2013). Distributed representations of words and phrases and their compositionality. Advances in Neural Information Processing Systems\/ 26

    2013

  18. [27]

    Nelson, S. J., K. Zeng, J. Kilbourne, T. Powell, and R. Moore (2011). Normalized names for clinical drugs: Rxnorm at 6 years. Journal of the American Medical Informatics Association\/ 18\/ (4), 441--448

    2011

  19. [28]

    Nickel, M. and D. Kiela (2017). Poincar \'e embeddings for learning hierarchical representations. Advances in Neural Information Processing Systems\/ 30

    2017

  20. [29]

    Murphy, V

    Nickel, M., K. Murphy, V. Tresp, and E. Gabrilovich (2015). A review of relational machine learning for knowledge graphs. Proceedings of the IEEE\/ 104\/ (1), 11--33

    2015

  21. [30]

    Oord, A. v. d., Y. Li, and O. Vinyals (2018). Representation learning with contrastive predictive coding. arXiv preprint arXiv:1807.03748\/

    2018 arXiv

  22. [31]

    Organization, W. H. (2004). International Statistical Classification of Diseases and related health problems: Alphabetical index , Volume 3. World Health Organization

    2004

  23. [32]

    Organization, W. H. et al. (1978). International classification of diseases:[9th] ninth revision, basic tabulation list with alphabetic index. In International Classification of Diseases:[9th] Ninth Revision, Basic Tabulation List with Alphabetic Index

    1978

  24. [33]

    Pathak, J. and C. G. Chute (2010). Analyzing categorical information in two publicly available drug terminologies: Rxnorm and ndf-rt. Journal of the American Medical Informatics Association\/ 17\/ (4), 432--439

    2010

  25. [34]

    Courty, R

    Perrot, M., N. Courty, R. Flamary, and A. Habrard (2016). Mapping estimation for discrete optimal transport. Advances in Neural Information Processing Systems\/ 29

    2016

  26. [35]

    Cuturi, et al

    Peyr \'e , G., M. Cuturi, et al. (2019). Computational optimal transport: With applications to data science. Foundations and Trends in Machine Learning\/ 11\/ (5-6), 355--607

    2019

  27. [36]

    Phecode map 1.2 with icd-9 and icd-10 codes

    PheWAS Catalog (2021). Phecode map 1.2 with icd-9 and icd-10 codes. https://phewascatalog.org/phecodes. Accessed: 2025-04-17

    2021

  28. [37]

    Rajkomar, A., E. Oren, K. Chen, A. M. Dai, N. Hajaj, M. Hardt, P. J. Liu, X. Liu, J. Marcus, M. Sun, et al. (2018). Scalable and accurate deep learning with electronic health records. npj Digital Medicine\/ 1\/ (1), 1--10

    2018

  29. [38]

    Schwarz, G. (1978). Estimating the dimension of a model. The Annals of Statistics\/ , 461--464

    1978

  30. [39]

    Tseng, P. (2001). Convergence of a block coordinate descent method for nondifferentiable minimization. Journal of Optimization Theory and Applications\/ 109 , 475--494

    2001

  31. [40]

    Cui, and W

    Wang, D., P. Cui, and W. Zhu (2016). Structural deep network embedding. In Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining , pp.\ 1225--1234

    2016

  32. [41]

    Wang, X., X. Han, W. Huang, D. Dong, and M. R. Scott (2019). Multi-similarity loss with general pair weighting for deep metric learning. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pp.\ 5022--5030

    2019

  33. [42]

    Gifford, X

    Wu, P., A. Gifford, X. Meng, X. Li, H. Campbell, T. Varley, J. Zhao, R. Carroll, L. Bastarache, J. C. Denny, et al. (2019). Mapping icd-10 and icd-10-cm codes to phecodes: workflow development and initial evaluation. JMIR Medical Informatics\/ 7\/ (4), e14325

    2019

  34. [43]

    Mao, and Y

    Yao, L., C. Mao, and Y. Luo (2019). Kg-bert: Bert for knowledge graph completion. arXiv preprint arXiv:1909.03193\/

    2019 arXiv

  35. [44]

    Yuan, Z., Z. Zhao, H. Sun, J. Li, F. Wang, and S. Yu (2022). Coder: Knowledge-infused cross-lingual medical term embedding for term normalization. Journal of Biomedical Informatics\/ 126 , 103983

    2022

  36. [45]

    Yuan, and S

    Zeng, S., Z. Yuan, and S. Yu (2022). Automatic biomedical term clustering by learning fine-grained term representations. arXiv preprint arXiv:2204.00391\/

    2022 arXiv

  37. [46]

    Zhou, D., Z. Gan, X. Shi, A. Patwari, E. Rush, C.-L. Bonzel, V. A. Panickan, C. Hong, Y.-L. Ho, T. Cai, et al. (2022). Multiview incomplete knowledge graph integration with application to cross-institutional ehr data harmonization. Journal of Biomedical Informatics\/ 133 , 104147

    2022

  38. [47]

    Zhou, D., H. Tong, L. Wang, S. Liu, X. Xiong, Z. Gan, R. Griffier, B. Hejblum, Y.-C. Liu, C. Hong, et al. (2025). Representation learning to advance multi-institutional studies with electronic health record data. arXiv preprint arXiv:2502.08547\/

    2025 arXiv

  39. [48]

    @esa (Ref

    \@ifxundefined[1] #1\@undefined \@firstoftwo \@secondoftwo \@ifnum[1] #1 \@firstoftwo \@secondoftwo \@ifx[1] #1 \@firstoftwo \@secondoftwo [2] @ #1 \@temptokena #2 #1 @ \@temptokena \@ifclassloaded agu2001 natbib The agu2001 class already includes natbib coding, so you should ...

  40. [49]

    \@lbibitem[] @bibitem@first@sw\@secondoftwo \@lbibitem[#1]#2 \@extra@b@citeb \@ifundefined br@#2\@extra@b@citeb \@namedef br@#2 \@nameuse br@#2\@extra@b@citeb \@ifundefined b@#2\@extra@b@citeb @num @parse #2 @tmp #1 NAT@b@open@#2 NAT@b@shut@#2 \@ifnum @merge>\@ne @bibitem@firs...

  41. [50]

    @open @close @open @close and [1] URL: #1 \@ifundefined chapter * \@mkboth \@ifxundefined @sectionbib * \@mkboth * \@mkboth\@gobbletwo \@ifclassloaded amsart * \@ifclassloaded amsbook * \@ifxundefined @heading @heading NAT@ctr thebibliography [1] @ \@biblabel @NAT@ctr \@bibset...

    2019 arXiv

Pith tools

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