Pith. sign in

REVIEW 3 major objections 5 minor 59 references

Risk-adjusted disease networks from health trajectories yield four multimorbidity phenotypes with sharply different survival.

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

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

Risk-adjusted Word2Vec-plus-GGM comorbidity networks from UK Biobank trajectories recover clinically coherent cardiometabolic modules and four progression phenotypes with divergent survival.

T0 review reviewed 2026-07-11 challenge →

load-bearing objection Solid applied pipeline that actually ships a usable risk-adjusted cardiometabolic network and survival-separated phenotypes; the Δ-as-confounding step is heuristic but not fatal. the 3 major comments →

arxiv 2607.04702 v1 pith:ZO7SUTS7 submitted 2026-07-06 stat.AP

Enhancing comorbidity network inference with risk-enriched health trajectories embedding

Nicole Fontana , Alessia Mapelli , Emanuele Di Angelantonio , Francesca Ieva This is my paper
classification stat.AP
keywords comorbidity networkGaussian Graphical ModelsWord2Vec embeddingscardiometabolic diseaseslongitudinal trajectoriesconfounding adjustmentmultimorbidity phenotypesUK Biobank
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The reading

Comorbidity networks are usually built from snapshot co-occurrence counts. That approach ignores order of disease onset, confounds associations with shared risk factors, and produces dense graphs that mix direct and indirect links. This paper builds a pipeline that starts from full patient event sequences, embeds them so that temporal and semantic proximity become geometry, then measures how much each disease-pair similarity shrinks once risk factors are jointly embedded. Those shrinkages identify disease-specific confounder sets that are fed into a Lasso Gaussian graphical model, producing a sparse network of only direct conditional dependencies. Applied to UK Biobank cardiometabolic data, the network recovers four clinically coherent communities. Patient trajectories are then rewritten as transitions among those communities; clustering the transition fingerprints yields four progression phenotypes whose all-cause survival curves separate after age 60, with mortality ranging from roughly 12 % to 32 %. The claim is that this risk-enriched, trajectory-based route both cleans the network and supplies a usable stratification tool for multimorbidity care.

Core claim

Embedding complete disease-and-risk trajectories, quantifying confounding via the signed drop in cosine similarity after risk adjustment, and feeding the resulting disease-specific confounder sets into a Lasso Gaussian graphical model recovers a sparse cardiometabolic network whose four communities support patient transition representations that cluster into four progression phenotypes with significantly different long-term survival.

What carries the argument

Prior clinical knowledge constructed as disease-specific confounder sets C*_i: the union, for each disease, of risk factors that co-associate with it in the joint embedding space and that drive large negative similarity changes Δ_ij; these sets are included as covariates in each neighbourhood-selection Lasso regression so that the estimated edges are conditional on shared risk burden.

Load-bearing premise

The signed change in pairwise cosine similarity between disease-only and disease-plus-risk embeddings is treated as a quantitative measure of true confounding by shared risk factors; if that geometric change mainly reflects embedding artefacts rather than confounding, the adjusted edges and the downstream phenotypes are misspecified.

What would settle it

A controlled simulation or external cohort in which the true disease graph and true confounders are known: if the Δ-based confounder sets fail to recover the known confounders, or if the GGM edges recovered after adjustment still contain the known spurious links, the central confounding-identification claim fails.

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

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper proposes a pipeline for population-level comorbidity network inference from longitudinal UK Biobank trajectories. Individual event sequences (24 cardiometabolic diseases + 76 risk factors) are embedded via fine-tuned BioWordVec; signed changes in pairwise cosine similarity between disease-only and disease+risk embeddings (Δ, Eq. 3.4) are used to identify disease-specific confounder sets C*_i ("prior clinical knowledge"); these sets are inserted as covariates into neighbourhood-selection GGMs with Lasso (Eq. 3.7) to estimate a sparse risk-adjusted network. Topological analysis yields four communities aligned with cardiometabolic taxonomy; community-transition matrices are clustered into four progression phenotypes that show significantly different Kaplan–Meier all-cause survival (mortality 11.7%–32.1%). The authors report internal checks (Wilcoxon on Δ, Mann–Whitney linking |Δ| to SharedRF, τ sensitivity) and release code with simulated data.

Significance. If the pipeline is valid, it offers a concrete, reproducible alternative to cross-sectional co-occurrence networks that (i) uses full trajectories, (ii) attempts explicit risk-factor adjustment, and (iii) produces sparse conditional-dependence graphs rather than thresholded fully connected graphs. The downstream community-transition phenotypes with external survival separation are a useful demonstration of clinical utility. Strengths include open code, multi-source phenotype definitions, and transparent sensitivity analyses on τ. The work is of clear interest to statistical applications in multimorbidity and network medicine, provided the central confounding-identification step can be better justified.

major comments (3)
  1. [§3.3.2–3.3.5, Eq. (3.4), Eq. (3.7)] The load-bearing claim that Δ_ij = A^(D+R)_ij − A^(D)_ij (Eq. 3.4, §3.3.2–3.3.4) quantifies confounding by shared risk factors is only indirectly supported. The Mann–Whitney result (pairs in the 95th percentile of |Δ| share more risk factors, p = 2.57e-14) and τ robustness (Supp. Table S3) show association, not that Δ isolates classical confounding rather than embedding geometry, sequence-composition change, or multi-word MeSH averaging artefacts. Because both embedding spaces are learned from the same patients, redistribution of co-occurrence mass onto risk-factor tokens can produce negative Δ without corresponding to a causal confounder. This step defines the C*_i sets that enter the neighbourhood Lasso (Eq. 3.7); if it is misspecified, the "risk-adjusted" edges and the four survival phenotypes inherit the bias. A simulation with known ground-truth edges and known confounders (or an ab
  2. [§4, Figure 3, Table 1] No quantitative comparison is reported between the risk-adjusted GGM network and a disease-only GGM (or a standard co-occurrence / partial-correlation baseline). Table 1 and Figure 2 characterise similarity-matrix change, but the final sparse adjacency (Figure 3) is presented only in its adjusted form. Without edge-set overlap, precision/recall against a clinical reference, or a side-by-side topological comparison, it is impossible to assess how much the prior-knowledge step actually changes the estimated network or the downstream communities and survival clusters. This comparison is required to substantiate the claim that the pipeline "addresses" confounding.
  3. [§3.4] The multivariate Gaussian assumption for embedding vectors (treated as independent observations across dimensions) is stated and checked only via per-dimension Shapiro–Wilk tests (§3.4). Embeddings from skip-gram models are typically non-Gaussian and dimensions are not independent observations of a single multivariate law. The paper should either (i) report the actual Shapiro–Wilk outcomes and any remedial transforms, or (ii) justify neighbourhood selection under weaker assumptions (e.g., nonparanormal / copula GGMs). As written, the statistical foundation of the sparse network is under-specified.
minor comments (5)
  1. [§4] Typo in Results opening: "Staring from embeddings" should be "Starting from embeddings".
  2. [§2, Supplementary Material] Supplementary Tables S1–S2 are linked only as external Excel files; a short in-text summary of the 24 diseases and main risk-factor categories would help readers who cannot access the repository immediately.
  3. [§3.2] The stop criterion sim_preserved < 0.70 after 4 epochs (§3.2) is ad hoc; a brief sensitivity check on embedding dimension d or window c would strengthen reproducibility claims.
  4. [Figures 2–4] Figure 2 colour scale and caption are clear, but the adjacency matrix in Figure 3b would benefit from community-ordered rows/columns to make modular structure visible without referring to Figure 4.
  5. [§4, Figure 5] Kaplan–Meier analysis (§3.5.2, Figure 5) reports visual separation and mortality rates but no log-rank or Cox p-value in the main text; adding the test statistic would make the survival claim fully self-contained.

Circularity Check

2 steps flagged

Mild self-confirmation in the Δ/SharedRF confounding loop (same embedding geometry) plus non-load-bearing self-citation of the authors’ prior-informed GGM; survival separation of phenotypes remains an external, independent endpoint.

specific steps
  1. other [§3.3.2–3.3.4 (Eqs. 3.4, 3.6; confounding detection)]
    "We quantified similarity changes as: Δij = A(D+R)ij - A(D)ij. … a risk factor r ∈ R was considered associated with disease di when sim(edi, er) > τ. … SharedRFij = |Ri ∩ Rj| / min(|Ri|, |Rj|). Under our confounding framework, disease pairs with large changes in similarity are expected to share a greater burden of common risk factors. … Mann-Whitney U test … p-value = 2.57 × 10-14"

    Both Δ and the sets Ri that define SharedRF are computed from cosine geometry in the same (or closely related) embedding spaces learned on the identical patient sequences. The statistically significant association between large |Δ| and high SharedRF is therefore partly a geometric consequence of how co-occurrence mass is redistributed when risk-factor tokens are added, rather than an independent external test of classical confounding. The resulting C*_i “prior clinical knowledge” is thus derived from and validated inside the representation that later feeds the GGM.

  2. self citation load bearing [§3.4 Network estimation (neighbourhood selection)]
    "We employed a neighbourhood selection approach (Mapelli and others, 2026) to estimate Ω through D independent Lasso regression models. In this stage, the predictor set for each disease regression could be informed by the prior clinical knowledge derived in Section 3.3."

    The concrete estimation procedure that turns the embedding-derived confounders into a sparse network is justified by citation to Mapelli et al. (2026), whose author list overlaps with the present paper (Mapelli, Di Angelantonio, Ieva). While the underlying Meinshausen–Bühlmann framework is standard, the specific prior-informed variant is self-cited; the citation is not machine-checked or externally re-derived here and supplies the only reference for the exact regression setup used.

full rationale

The derivation chain is a data-driven pipeline (sequences → fine-tuned BioWordVec → cosine Δ → confounder sets C*_i → prior-informed neighbourhood Lasso GGM → modularity communities → patient transition matrices → k-means → KM survival). No step reduces a claimed prediction or first-principles result to its own inputs by algebraic identity or forced fit. Survival is never used in embedding, GGM, community detection or clustering, so the four phenotypes’ differential mortality is genuine external evidence. The only mild circularities are (1) the validation that high-|Δ| pairs share more risk factors, which lives inside the same cosine geometry that produces Δ, and (2) citation of the authors’ own 2026 prior-informed GGM paper for the neighbourhood-selection machinery (itself an extension of Meinshausen–Bühlmann). Neither is load-bearing for the survival claim, nor is any uniqueness theorem imported. Score 3 reflects these two non-central self-referential elements while recognising that the central empirical result is independently supported.

Axiom & Free-Parameter Ledger

8 free parameters · 6 axioms · 2 invented entities

The central claim rests on standard NLP/graphical-model machinery plus several domain and paper-specific modelling choices: that skip-gram geometry encodes pathophysiological relatedness, that Δ cosine similarity isolates confounding, that embedding coordinates behave as Gaussian observations for partial-correlation estimation, and that community transition counts are adequate patient fingerprints. Free parameters (window, dimension, fine-tune stop, τ, Lasso λ, k, event-count filter) are chosen by convention, median, CV, or elbow rather than theory. No new physical entities; the main invented constructs are the prior-clinical-knowledge confounder sets and the community-transition patient vectors.

free parameters (8)
  • Word2Vec embedding dimension d
    Fixed at 200 following BioWordVec; controls capacity of disease/risk representations used for all downstream similarity and GGM steps.
  • Skip-gram context window c
    Fixed at 5; defines which co-occurring events shape embeddings and thus all association estimates.
  • Semantic retention stop threshold sim_preserved
    Fine-tuning stopped when cosine similarity to pretrained embeddings fell below 0.70 (4 epochs); ad hoc balance between adaptation and biomedical semantics.
  • Risk-factor association threshold τ
    Main analysis uses median pairwise risk-factor similarity (τ≈0.73) to define R_i; sensitivity 0.6–0.8 reported but primary confounder sets depend on this choice.
  • Highly-affected pair percentile
    95th percentile of |Δ_ij| used to validate SharedRF enrichment; defines which pairs count as confounding evidence.
  • Lasso penalty λ
    Selected by 10-fold CV minimising MSE in each neighbourhood regression; directly determines edge sparsity of the reported network.
  • k-means cluster count
    Chosen via elbow on within-cluster SS; yields the four progression phenotypes whose survival is the clinical payoff claim.
  • Minimum main-disease events filter
    Participants with fewer than three main disease events excluded (502,297 → 113,973); shapes the trajectory distribution and network.
axioms (6)
  • domain assumption Skip-gram Word2Vec on tokenised clinical sequences yields embeddings whose cosine geometry reflects semantic and temporal disease relatedness.
    Section 3.2; foundation for all similarity matrices and confounder detection.
  • ad hoc to paper Shared risk-factor confounding inflates disease-only cosine similarity; jointly embedding risk factors redistributes that signal so Δ_ij measures confounding removal.
    Section 3.3.2 conceptual framework; not derived from causal identification theory.
  • domain assumption Embedding vectors (or dimensions treated as observations) are multivariate Gaussian so zeros in the precision matrix encode conditional independence.
    Section 3.4 citing Zhelezniak et al.; Shapiro–Wilk mentioned but not fully reported.
  • standard math Neighbourhood selection with OR-rule symmetrisation recovers the undirected conditional dependence graph of diseases after conditioning on C*_i.
    Meinshausen–Bühlmann style Lasso neighbourhood selection (Section 3.4).
  • domain assumption Multi-source UK Biobank phenotype definitions (ICD-10, OPCS-4, primary care, labs) and MeSH mapping correctly label the 24 diseases and 76 risk factors.
    Section 2; expert consultation claimed, definitions in Supp Tables S1–S2.
  • domain assumption Fast-greedy modularity communities and K×K community transition counts are meaningful summaries of pathophysiological domains and patient progression.
    Section 3.5; used for both taxonomy alignment and survival phenotypes.
invented entities (2)
  • Prior clinical knowledge confounder sets C*_i no independent evidence
    purpose: Disease-specific unions of pairwise shared risk factors used as covariates in GGM neighbourhood regressions to adjust for confounding.
    Constructed entirely inside the pipeline from embedding similarities and Δ ranking; not an external clinical ontology.
  • Community-transition patient fingerprint (flattened normalised T_p) no independent evidence
    purpose: Individual-level representation of progression dynamics across detected modules for clustering and survival stratification.
    Derived from the estimated network’s communities; prognostic value shown only within this study’s KM analysis.

reviewed 2026-07-11 · how reviews work

0 comments
Cite this review

Pith. "Pith review of Enhancing comorbidity network inference with risk-enriched health trajectories embedding." pith.science (2026). https://pith.science/paper/ZO7SUTS7

@misc{pith2026260704702,
  author       = {Pith},
  title        = {Pith review of: Enhancing comorbidity network inference with risk-enriched health trajectories embedding},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ZO7SUTS7}},
  note         = {Machine review of arXiv:2607.04702}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Multimorbidity poses a growing challenge for individual health, reducing quality of life and increasing treatment burden, resulting in a multiplicative impact on healthcare system management and fragmented care trajectories. Comorbidity networks could provide crucial insight into characterising multimorbidity and disease relationships. However, existing approaches to comorbidity network construction face critical limitations: they overlook temporal information by relying on cross-sectional statistics, produce biased association estimates by ignoring confounding due to shared risk factors, and fail to distinguish between direct and indirect disease associations, thereby yielding fully connected networks. To address these limitations, we develop a methodological framework for population-level disease network inference that uses individual health trajectories to learn disease associations, capturing semantic similarity and temporal co-occurrence. Sparse network estimation is achieved via Gaussian Graphical Models with Lasso regularisation, informed by prior clinical knowledge on shared risk factors derived from a dedicated confounding evaluation step. Applied to UK Biobank data comprising 24 cardiometabolic diseases and 76 risk factors, the resulting network revealed clinically meaningful disease patterns. Topological analysis identifies key pathological hubs, reveals potential actionable targets for multimorbidity management, and identifies four distinct disease communities that align with the established cardiometabolic taxonomy. Building on this community structure, we derive community-based patient representations that capture disease progression dynamics. Clustering these representations reveals four progression phenotypes with significantly different long-term survival trajectories, highlighting the potential of the framework for risk stratification and personalised care.

Figures

Figures reproduced from arXiv: 2607.04702 by Alessia Mapelli, Emanuele Di Angelantonio, Francesca Ieva, Nicole Fontana.

Figure 1
Figure 1. Figure 1: Schematic representation of the methodological framework for risk-adjusted comorbidity [PITH_FULL_IMAGE:figures/full_fig_p008_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Signed changes in disease similarity after risk factor adjustment (∆). Blue indicates de [PITH_FULL_IMAGE:figures/full_fig_p017_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Estimated cardiometabolic risk-adjusted disease network (a), and corresponding adjacency [PITH_FULL_IMAGE:figures/full_fig_p019_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Community structure of the risk-adjusted cardiometabolic disease network. Node colours [PITH_FULL_IMAGE:figures/full_fig_p021_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Kaplan-Meier survival curves stratified by transition cluster. [PITH_FULL_IMAGE:figures/full_fig_p022_5.png] view at source ↗

discussion (0)

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

Reference graph

Works this paper leans on

59 extracted references · 3 canonical work pages

  1. [1]

    2026 , eprint=

    Prior-informed conditional Gaussian graphical models: an application to protein interaction network reconstruction , author=. 2026 , eprint=

    2026

  2. [2]

    BioData Mining , volume =

    Network analysis of longitudinal electronic health records using linear mixed models , author =. BioData Mining , volume =. 2026 , doi =

    2026

  3. [3]

    and Pearson, Jonathan and Rafferty, James and Schofield, Daniel , title =

    Burke, Jamie and Akbari, Ashley and Bailey, Rowena and Fasusi, Kevin and Lyons, Ronan A. and Pearson, Jonathan and Rafferty, James and Schofield, Daniel , title =. 2023 , doi =. https://www.medrxiv.org/content/early/2023/09/05/2023.08.31.23294903.full.pdf , journal =

    2023

  4. [4]

    The Lancet , volume=

    Epidemiology of multimorbidity and implications for health care, research, and medical education: a cross-sectional study , author=. The Lancet , volume=. 2012 , publisher=

    2012

  5. [5]

    Journal of the American College of Cardiology , volume=

    Global burden of cardiovascular diseases and risk factors, 1990--2019: update from the GBD 2019 study , author=. Journal of the American College of Cardiology , volume=. 2020 , publisher=

    1990

  6. [6]

    PLoS Computational Biology , volume=

    A dynamic network approach for the study of human phenotypes , author=. PLoS Computational Biology , volume=. 2009 , publisher=

    2009

  7. [7]

    Proceedings of the National Academy of Sciences , volume=

    The human disease network , author=. Proceedings of the National Academy of Sciences , volume=. 2007 , publisher=

    2007

  8. [8]

    Nature Communications , volume=

    Temporal disease trajectories condensed from population-wide registry data covering 6.2 million patients , author=. Nature Communications , volume=. 2014 , publisher=

    2014

  9. [9]

    British Journal of General Practice , year=

    Impact of multimorbidity on healthcare costs and utilisation: a systematic review of the UK literature , author=. British Journal of General Practice , year=

  10. [10]

    BMC health services research , volume=

    Economic burden of multimorbidity among older adults: impact on healthcare and societal costs , author=. BMC health services research , volume=. 2016 , publisher=

    2016

  11. [11]

    BMC medicine , volume=

    Costs of multimorbidity: a systematic review and meta-analyses , author=. BMC medicine , volume=. 2022 , publisher=

    2022

  12. [12]

    A scoping review , author=

    Impact of multi-morbidity on quality of healthcare and its implications for health policy, research and clinical practice. A scoping review , author=. European Journal of General Practice , volume=. 2015 , publisher=

    2015

  13. [13]

    European Heart Journal Open , volume=

    Impact of multimorbidity patterns on outcomes and treatment in patients with coronary artery disease , author=. European Heart Journal Open , volume=. 2024 , publisher=

    2024

  14. [14]

    Journal of Internal Medicine , volume=

    Evidence supporting the best clinical management of patients with multimorbidity and polypharmacy: a systematic guideline review and expert consensus (vol 285, pg 272, 2019) , author=. Journal of Internal Medicine , volume=

    2019

  15. [15]

    BMJ open , volume=

    Functional decline and associated factors in patients with multimorbidity at 8 months of follow-up in primary care: the functionality in pluripathological patients (FUNCIPLUR) longitudinal descriptive study , author=. BMJ open , volume=. 2018 , publisher=

    2018

  16. [16]

    Social networks , volume=

    Centrality in social networks conceptual clarification , author=. Social networks , volume=. 1978 , publisher=

    1978

  17. [17]

    International Conference on Learning Representations , year=

    Efficient Estimation of Word Representations in Vector Space , author=. International Conference on Learning Representations , year=

  18. [18]

    Scientific data , volume=

    BioWordVec, improving biomedical word embeddings with subword information and MeSH , author=. Scientific data , volume=. 2019 , publisher=

    2019

  19. [19]

    arXiv preprint arXiv:1905.07790 , year=

    Correlation coefficients and semantic textual similarity , author=. arXiv preprint arXiv:1905.07790 , year=

    Pith/arXiv arXiv 1905

  20. [20]

    Annals of Statistics , volume=

    High-dimensional graphs and variable selection with the Lasso , author=. Annals of Statistics , volume=. 2006 , month=. doi:10.1214/009053606000000281 , url=

    doi:10.1214/009053606000000281 2006

  21. [21]

    New England Journal of Medicine , volume=

    Acute kidney injury and chronic kidney disease as interconnected syndromes , author=. New England Journal of Medicine , volume=. 2014 , publisher=

    2014

  22. [22]

    Minerva Cardioangiologica , volume =

    Tricuspid valve regurgitation: epidemiology and pathophysiology , author =. Minerva Cardioangiologica , volume =. 2018 , doi =

    2018

  23. [23]

    Secondary tricuspid regurgitation: Do we understand what we would like to treat? , journal =

    Anne Guérin and Julien Dreyfus and Thierry. Secondary tricuspid regurgitation: Do we understand what we would like to treat? , journal =. 2019 , note =. doi:https://doi.org/10.1016/j.acvd.2019.04.010 , url =

    work page doi:10.1016/j.acvd.2019.04.010 2019

  24. [24]

    European Heart Journal , volume=

    Staging classification of aortic stenosis based on the extent of cardiac damage , author=. European Heart Journal , volume=. 2017 , publisher=

    2017

  25. [25]

    Circulation , volume=

    Atrial fibrillation begets heart failure and vice versa: temporal associations and differences in preserved versus reduced ejection fraction , author=. Circulation , volume=. 2016 , publisher=

    2016

  26. [26]

    World Journal of Cardiology , volume =

    Atrial fibrillation and coronary artery disease: An integrative review focusing on therapeutic implications of this relationship , author =. World Journal of Cardiology , volume =. 2023 , doi =

    2023

  27. [27]

    Physical Review E , volume=

    Finding community structure in very large networks , author=. Physical Review E , volume=. 2004 , publisher=

    2004

  28. [28]

    and Khan, Muhammad S

    Usman, Muhammad S. and Khan, Muhammad S. and Butler, John , title =. Chronic Kidney Disease and Type 2 Diabetes , publisher =. 2021 , month =

    2021

  29. [29]

    and Calderón-Larrañaga, A

    Prados-Torres, A. and Calderón-Larrañaga, A. and Hancco-Saavedra, J. and Poblador-Plou, B. and van den Akker, M. , title =. Journal of Clinical Epidemiology , year =. doi:10.1016/j.jclinepi.2013.09.021 , pmid =

    work page doi:10.1016/j.jclinepi.2013.09.021 2013

  30. [30]

    and Wang, M

    Yan, J. and Wang, M. and Cao, Y. , title =. BMC Public Health , year =. doi:10.1186/s12889-022-14124-6 , url =

    work page doi:10.1186/s12889-022-14124-6

  31. [31]

    Nature Reviews Genetics , year =

    Barabási, Albert-László and Gulbahce, Natali and Loscalzo, Joseph , title =. Nature Reviews Genetics , year =. doi:10.1038/nrg2918 , note =

    doi:10.1038/nrg2918

  32. [32]

    del Valle, E. P. G. and Lagunes García, G. and Prieto Santamaría, L. and others , title =. Journal of Biomedical Informatics , year =

  33. [33]

    Representation learning for natural language processing , pages=

    Representation learning and NLP , author=. Representation learning for natural language processing , pages=. 2023 , publisher=

    2023

  34. [34]

    Artificial Intelligence in Medicine , volume=

    Comparing neural language models for medical concept representation and patient trajectory prediction , author=. Artificial Intelligence in Medicine , volume=. 2025 , publisher=

    2025

  35. [35]

    Information , volume=

    Recurrent neural networks: A comprehensive review of architectures, variants, and applications , author=. Information , volume=. 2024 , publisher=

    2024

  36. [36]

    BMC medical informatics and decision making , volume=

    Transformer models in biomedicine , author=. BMC medical informatics and decision making , volume=. 2024 , publisher=

    2024

  37. [37]

    Nature Communications , volume=

    Disease trajectory browser for exploring temporal, population-wide disease progression patterns in 7.2 million Danish patients , author=. Nature Communications , volume=. 2020 , publisher=

    2020

  38. [38]

    Scientific Reports , volume =

    Network-based analysis of diagnosis progression patterns using claims data , author =. Scientific Reports , volume =. 2017 , doi =

    2017

  39. [39]

    Boyd and Diego Garlaschelli and Peter J

    Naoki Masuda and Zachary M. Boyd and Diego Garlaschelli and Peter J. Mucha , keywords =. Introduction to correlation networks: Interdisciplinary approaches beyond thresholding , journal =. 2025 , note =. doi:https://doi.org/10.1016/j.physrep.2025.06.002 , url =

    doi:10.1016/j.physrep.2025.06.002 2025

  40. [40]

    Nature Machine Intelligence , volume=

    Causal inference and counterfactual prediction in machine learning for actionable healthcare , author=. Nature Machine Intelligence , volume=. 2020 , publisher=

    2020

  41. [41]

    IEEE Transactions on Knowledge and Data Engineering , volume=

    Patient flow prediction via discriminative learning of mutually-correcting processes , author=. IEEE Transactions on Knowledge and Data Engineering , volume=. 2017 , publisher=

    2017

  42. [42]

    2024 , issue_date =

    Gong, Chang and Zhang, Chuzhe and Yao, Di and Bi, Jingping and Li, Wenbin and Xu, YongJun , title =. 2024 , issue_date =. doi:10.1145/3705297 , journal =

    doi:10.1145/3705297 2024

  43. [43]

    Biotechnology Reports , pages=

    Integrating multi-omics data: Methods and applications in human complex diseases , author=. Biotechnology Reports , pages=. 2025 , publisher=

    2025

  44. [44]

    Ndumele and Janani Rangaswami and Sheryl L

    Chiadi E. Ndumele and Janani Rangaswami and Sheryl L. Chow and Ian J. Neeland and Katherine R. Tuttle and Sadiya S. Khan and Josef Coresh and Roy O. Mathew and Carissa M. Baker-Smith and Mercedes R. Carnethon and Jean-Pierre Despres and Jennifer E. Ho and Joshua J. Joseph and Walter N. Kernan and Amit Khera and Mikhail N. Kosiborod and Carolyn L. Lekavich...

    2023

  45. [45]

    Cardiovascular diseases (CVDs) , year =

  46. [46]

    Annals of Statistics , volume=

    On the definition of a confounder , author=. Annals of Statistics , volume=. 2013 , publisher=

    2013

  47. [47]

    2009 , publisher=

    Causality: Models, Reasoning, and Inference , author=. 2009 , publisher=

    2009

  48. [48]

    PLoS Medicine , year =

    Sudlow, Cathie and Gallacher, John and Allen, Naomi and Beral, Valerie and Burton, Paul and Danesh, John and Downey, Paul and Elliott, Paul and Green, Jane and Landray, Martin and Liu, Bette and Matthews, Paul and Ong, Geraldine and Pell, Jill and Silman, Alan and Young, Alan and Sprosen, Tim and Peakman, Tim and Collins, Rory , title =. PLoS Medicine , y...

    doi:10.1371/journal.pmed.1001779

  49. [49]

    2020 international conference on decision aid sciences and application (DASA) , pages=

    Research paper classification based on Word2vec and community discovery , author=. 2020 international conference on decision aid sciences and application (DASA) , pages=. 2020 , organization=

    2020

  50. [50]

    The Lancet , volume=

    Cardiovascular remodelling in coronary artery disease and heart failure , author=. The Lancet , volume=. 2014 , publisher=

    2014

  51. [51]

    2012 , month = may, howpublished =

    2012

  52. [52]

    2016 , month = sep, howpublished =

    2016

  53. [53]

    and Huepenbecker, S

    Zhao, B. and Huepenbecker, S. and Zhu, G. and Rajan, S. S. and Fujimoto, K. and Luo, X. , title =. Frontiers in Big Data , volume =. 2023 , doi =

    2023

  54. [54]

    Scientific Data , volume=

    Comorbidity networks from population-wide health data: aggregated data of 8.9 M hospital patients (1997--2014) , author=. Scientific Data , volume=. 2025 , publisher=

    1997

  55. [55]

    arXiv preprint arXiv:2506.22136 , year=

    Characterization Of Diseases In Temporal Comorbidity Networks , author=. arXiv preprint arXiv:2506.22136 , year=

    Pith/arXiv arXiv

  56. [56]

    arXiv preprint arXiv:2008.05377 , year=

    Network reinforcement driven drug repurposing for COVID-19 by exploiting disease-gene-drug associations , author=. arXiv preprint arXiv:2008.05377 , year=

    Pith/arXiv arXiv 2008

  57. [57]

    Cancers , volume=

    Healthcare Costs by Comorbidity Patterns in Lung Cancer Patients , author=. Cancers , volume=. 2025 , publisher=

    2025

  58. [58]

    Archives of internal medicine , volume=

    Prevalence, expenditures, and complications of multiple chronic conditions in the elderly , author=. Archives of internal medicine , volume=. 2002 , publisher=

    2002

  59. [59]

    Health Information Science and Systems , volume=

    Comorbidity progression analysis: patient stratification and comorbidity prediction using temporal comorbidity network , author=. Health Information Science and Systems , volume=. 2024 , publisher=

    2024

This paper was first reviewed by grok-4.5 on July 11, 2026.