Pith. sign in

REVIEW 3 major objections 5 minor 42 references

This paper claims that a temporal machine-learning model turns longitudinal ALSFRS-R scores into individualized wheelchair-free survival curves, and that lower-limb decline (walking, stair-climbing) is the strongest driver of earlier wheelc

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 ↗

A transformer-based discrete-time survival model plus Cox regression predicts ALS domain decline and wheelchair timing from ALSFRS-R visits, but discrimination is near chance (mean C-index 0.549 on 28 test patients) and the wheelchair-survival model itself is never evaluated.

T0 review reviewed 2026-08-02 challenge →

load-bearing objection Novel ASTP architecture and honest held-out numbers, but the wheelchair-survival claim sits on an unevaluated Cox fit to 187/1,639 patients with no attrition analysis. the 3 major comments →

arxiv 2607.14190 v1 pith:SAIOVGTU submitted 2026-07-15 cs.LG cs.IRstat.ML

A Temporal Machine Learning-Based Time-to-Event Model for Predicting ALS Progression and Healthcare Utilization

Zongliang Yue , Qi Li , Terry Heiman-Patterson , Frank Bearoff , Zhaohui Qin , Huanmei Wu This is my paper
classification cs.LG cs.IRstat.ML
keywords Amyotrophic lateral sclerosisALSFRS-Rtime-to-event predictionsurvival analysisdigital twinwheelchair-free survivalfunctional declineCox proportional hazards
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

ALS progression is notoriously uneven, and predicting when a patient will need a wheelchair remains hard. This paper tries to establish two things: that a digital-twin-inspired time-to-event model can convert a patient's ALSFRS-R visit history into per-domain transition risks and personalized wheelchair-free survival curves; and that among all functional domains, lower-limb decline—especially walking and stair-climbing—is the strongest predictor of earlier assistive device use. A sympathetic reader would care because such dynamic, interpretable predictions are exactly what proactive care planning, clinical trial stratification, and precision medicine need. The authors test their claim on a harmonized cohort of ALS patients with longitudinal assessments and device logs, evaluating the model with C-indices, MAE, and Cox regression.

Core claim

The central discovery is that functional decline in ALS is not uniform: mobility loss dominates the timeline to wheelchair use. Using Cox proportional hazards on longitudinal ALSFRS-R data, the authors find that walking (ALSFRS8, coefficient −0.30) and stair climbing (ALSFRS9, coefficient −0.40) are the strongest predictors of earlier wheelchair access (p<0.005), far exceeding respiratory, bulbar, or age effects. They then embed this finding in a five-head discrete-time survival model (ASTP) that, from a single clinic visit, emits per-domain hazard curves over 24 30-day bins, computes survival probabilities, and extrapolates to a 5-year window with a constant-hazard tail. The result is an in

What carries the argument

The load-carrying machinery is the Attention-based Stage Transition Predictor (ASTP), a transformer-encoder model with cross-attention pooling that maps each patient's irregular visit history to five domain-specific hazard heads (bulbar, upper limb, axial, lower limb, respiratory) over 24 discrete 30-day time bins; per-domain survival is the product of one minus hazards, and the current score is fed into each head to prevent impossible transitions. It is paired with a regularized Cox proportional-hazards model (elastic net, λ=0.05, α=0.2) fitted to time-to-wheelchair-access, with the same ALSFRS-R items as predictors. The transformer handles irregular spacing via elapsed-time features rather

Load-bearing premise

Every wheelchair-timing result depends on the 187 patients (out of 1,639) who had assistive-device logs; if having a device record correlates with disease severity or care access, the survival curves are biased, and the paper never compares the included 187 with the excluded 1,452.

What would settle it

Re-estimate the Cox model after including all 1,639 linked patients, treating those without device logs as censored at their last follow-up, and compare baseline ALSFRS-R and follow-up durations between the 187 with logs and the 1,452 without; if the lower-limb coefficients (ALSFRS8 ≈ −0.30, ALSFRS9 ≈ −0.40) shrink substantially or flip, the central claim fails.

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

If this is right

  • Clinicians could enter a patient's current ALSFRS-R items at a single visit and receive a per-domain risk readout (transition within <6 mo, <1 yr, <2 yr, >2 yr) plus a wheelchair-free survival curve.
  • Monitoring walking and stair-climbing scores becomes a concrete, low-cost strategy for flagging patients who need early wheelchair planning.
  • The model's predictions are dynamic: as new visits accrue, the latent state updates and the survival curves shift, matching the digital-twin idea.
  • Trial designers could use predicted lower-limb transition time as a stratification variable or a surrogate endpoint in ALS studies.
  • The architecture is event-agnostic and could be retrained for other care milestones (e.g., feeding-tube placement) wherever longitudinal scales and event logs exist.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Because the wheelchair event is only observed for the 11% of patients with device logs, the Cox effect sizes are likely inflated if device-logged patients are sicker on average; a fuller follow-up could weaken the walking/stairs coefficients.
  • The constant-hazard tail used to extend the 24-bin (720-day) horizon to 5 years is an extrapolation the paper itself shades as out-of-distribution; in very slow progressors it may under- or over-estimate long-term survival probability.
  • The same framework could be pointed at other neurodegenerative diseases with functional rating scales (e.g., Parkinson's or multiple sclerosis) without architectural changes, an extension the authors hint at but do not test.
  • A natural test: in an independent ALS registry with complete device-dates, fit only ALSFRS8 and ALSFRS9 against time-to-wheelchair; if the discrimination (C-index) remains ~0.6 or better, the claim transfers.
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 time-to-event framework for ALS that combines correlation-based clustering of ALSFRS-R items, GAMM-based longitudinal modeling, an attention-based discrete-time survival model (ASTP) for per-domain stage transitions, and a Cox proportional hazards model for time to wheelchair use. The results are deployed in a Streamlit app that outputs personalized wheelchair-free survival curves. The main claims are that lower limb function (walking and stair climbing) is the strongest predictor of earlier wheelchair use and that the framework provides dynamically updated, individualized survival predictions.

Significance. If valid, the framework would offer a clinically interpretable tool for anticipating assistive-device needs in ALS, with potential utility in care planning and trial stratification. The paper has genuine strengths: a publicly available app and code repository, honest reporting of a held-out test set for the ASTP model, and explicit acknowledgment of several limitations. However, the central wheelchair-related claims rest on a small, unanalyzed subcohort and an unevaluated Cox model, which substantially undercuts the significance as currently presented.

major comments (3)
  1. [§3.1, Table 4, Fig. 7] The analytic cohort for the wheelchair time-to-event analysis collapses from 1,639 linked patients to 187 patients (1,409 cases) after merging assistive-device logs. The paper never compares these 187 patients with the 1,452 excluded patients, nor does it specify censoring or missingness mechanisms for the excluded group. If device-log availability correlates with severity, care access, or follow-up intensity, all Cox coefficients and the wheelchair-free survival curves in Fig. 7 are biased by selection. This directly threatens the abstract's claim that lower limb decline is the strongest predictor of earlier wheelchair use. Please provide an attrition analysis and a sensitivity analysis (e.g., inverse-probability weighting or multiple imputation) or explicitly reframe the claim to the device-log subpopulation.
  2. [§2.5, Table 4, Fig. 7] The Cox proportional hazards model that drives the wheelchair-free survival curves is never evaluated for discrimination or calibration on held-out data. The only quantitative evaluation in the paper is for the ASTP model (Table 2), and it shows C-indices of 0.513–0.598 on 28 held-out patients (194 landmarks), which is barely above chance, with no confidence intervals. The Figure 7 prediction of a median wheelchair-free survival of 75.5 months for a 'preserved baseline' patient is therefore unsupported. Please provide internal validation for the Cox model (e.g., bootstrap C-index, calibration plot, time-dependent AUC) and report confidence intervals for the ASTP C-indices; otherwise the headline 'dynamically predicts wheelchair-free survival' is an overclaim.
  3. [§2.4.3, Eq. (9)] The 5-year survival curves are produced by holding the hazard constant at the mean of the last three trained bins (Eq. 9). This is a strong distributional assumption with no stated justification or validation. While the manuscript does visually shade the extrapolated region, the abstract's claim that the model 'generates individualized survival curves' conflates in-distribution 720-day predictions with an unvalidated constant-hazard tail. Please either validate the extrapolation against long follow-up patients or explicitly state that the 5-year curves are scenario projections, not empirical predictions.
minor comments (5)
  1. [§1, last paragraph] Typo: 'we implement a this prototype' should read 'we implement this prototype' or 'we implement a prototype'.
  2. [Table 2] The Lower limb row is corrupted: '124 700.598169.4' should read '124 70 0.598 169.4' (three separate columns). Please fix the table formatting.
  3. [§3.4–§3.6] Section references in §3.4 point to '§2.4–§2.4.2' when describing evaluation; these are methods sections, not results sections. Please correct the cross-references.
  4. [§2.6] The app description says the Cox model has 'n=15 predictors: age at visit, follow-up since diagnosis, sex, and the 12 ALSFRS-R items,' but the abstract and §2.5 list individual ALSFRS items and the Cox model in Eq. (11) also includes ALSFRS1r–ALSFRS3r; please clarify whether the respiratory items are included in the deployed model and ensure the count matches.
  5. [References] Reference [27] is cited in the Conclusion as a disclaimer ('not a neural sparse-regression method'); this appears to be a response to a prior reviewer comment and reads oddly in a final manuscript. Please remove or integrate it more naturally.

Circularity Check

0 steps flagged

No significant circularity: held-out evaluations and explicit extrapolations keep the derivation self-contained.

full rationale

The derivation chain is self-contained. The ASTP model is trained to predict a clearly defined event (first future >=1-point ALSFRS-R domain drop; Sec. 2.4.1) and is evaluated on a held-out 28-patient test set (Table 2, C-index 0.513-0.598), so its reported discrimination is not forced by construction. The Cox wheelchair model (Eq. 11-13) uses device-log event times separate from the ALSFRS-R covariates, and the lower-limb coefficients are estimated rather than imposed. The two places where predicted quantities are generated from fitted values — the 5-year constant-hazard tail (Eq. 9) and the time-to-floor calculation in Sec. 3.5 ('by scaling the predicted slope by the current score') — are explicitly labeled as extrapolation/post-processing, not as independent empirical predictions. The paper contains no load-bearing self-citations: the prior ALSNHC landmark and semi-competing-risk models cited as refs. [5] and [7] are by different author groups, and the digital-twin refs. [24]-[31] are external. The Discussion's 'observational data... selection bias' limitation is a data-quality/attrition concern, not a definitional reduction. No step exhibits a prediction that reduces by construction to its inputs, so circularity is not present.

Axiom & Free-Parameter Ledger

8 free parameters · 7 axioms · 0 invented entities

The paper's central claims rest on a chain of modeling choices: a Gaussian treatment of ordinal ALSFRS-R items, the 1-point-drop event definition, the equivalence of device codes 4/6 to wheelchair access, and, most fragile, the representativeness of the 187-patient device-log subset. The pipeline also embeds several hand-set hyperparameters (lambda=0.05, alpha=0.2, kappa=0.70, K=24, Delta=30) and an explicit constant-hazard extrapolation (Eq. 9) that converts fitted bins into 5-year predictions. No new physical entities are introduced; the digital twin is a deployment label.

free parameters (8)
  • elastic_net_lambda = 0.05
    Cox penalty strength (Section 2.5), hand-set with no tuning curve.
  • elastic_net_alpha = 0.2
    L1/L2 mixing ratio (Section 2.5), hand-set.
  • survival_threshold_kappa = 0.70
    Default transition-time threshold in Eq. 10; user-adjustable but drives all reported transition times.
  • time_bins_K = 24 x 30 days
    Horizon discretization (Section 2.4.1); fixes the 720-day in-distribution horizon.
  • correlation_threshold = 0.7
    Pearson threshold that justifies upper-limb/axial/lower-limb subdivision (Section 3.2); value chosen by inspection of the dendrogram.
  • gamm_spline_df = unstated
    Degrees of freedom for the natural cubic spline in Eq. 1 are never reported.
  • clustering_k_or_cutoff = unspecified
    Ward clustering either fixes k or applies a distance cutoff (Section 2.2); neither value is reported.
  • constant_hazard_tail = mean of final 3 bins
    Eq. 9 extrapolation rule; the 5-year predictions are pinned to the fitted tail.
axioms (7)
  • domain assumption ALSFRS-R ordinal items (0-4) are valid numeric outcomes for Pearson correlation, Gaussian GAMMs, and Cox regression.
    Pearson correlations and Gaussian mixed models are applied to 5-point ordinal items without a stated ordinal-treatment justification (Sections 2.2, 2.3, Eq. 1).
  • domain assumption A first drop of at least 1 point on a domain subscale is a clinically meaningful transition event.
    Event definition for the ASTP target (Section 2.4.1); no clinical anchoring is provided.
  • domain assumption Assistive-device codes 4 and 6 correspond to wheelchair access, and missing device records mean no event.
    Device-event definition in Section 2.1; no validation of coding accuracy.
  • domain assumption The 187 patients with device logs represent the 1,639-patient linked cohort for survival modeling.
    Section 3.1 cohort funnel; no comparison of included vs excluded patients; this is the paper's most load-bearing data assumption.
  • domain assumption Proportional hazards and within-subject exchangeability (after clustering) hold for the Cox model.
    Section 2.5, Eq. 11; the paper itself concedes the PH assumption may not fully capture complex, time-varying effects (Section 4).
  • ad hoc to paper Future hazard equals the mean of the last three fitted hazard bins.
    5-year extrapolation rule, Eq. 9; it converts the 720-day forecast into the headline 5-year view with no uncertainty accounting.
  • domain assumption Missing diagnosis or follow-up dates are ignorable; row deletion is safe.
    Section 2.1 excludes subjects with missing dates without a missingness analysis.

reviewed 2026-08-02 · how reviews work

0 comments
Cite this review

Pith. "Pith review of A Temporal Machine Learning-Based Time-to-Event Model for Predicting ALS Progression and Healthcare Utilization." pith.science (2026). https://pith.science/paper/SAIOVGTU

@misc{pith2026260714190,
  author       = {Pith},
  title        = {Pith review of: A Temporal Machine Learning-Based Time-to-Event Model for Predicting ALS Progression and Healthcare Utilization},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/SAIOVGTU}},
  note         = {Machine review of arXiv:2607.14190}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Amyotrophic lateral sclerosis (ALS) is a progressive and heterogeneous neurodegenerative disease in which predicting clinically meaningful milestones, such as assistive device use, remains challenging. We developed a time-to-event, digital-twin-inspired framework that integrates longitudinal ALS Functional Rating Scale-Revised (ALSFRS-R) trajectories with survival modeling to support individualized prediction of functional decline and assistive device utilization. We constructed a harmonized longitudinal dataset by integrating diagnosis records, ALSFRS-R assessments, activities of daily living, and demographic information, followed by preprocessing to ensure data quality, temporal alignment, and cohort consistency. Correlation-based clustering identified coherent functional domains spanning bulbar, upper limb, axial, lower limb, and respiratory systems. Generalized additive mixed models characterized nonlinear, domain-specific functional decline across all domains. In addition, a temporal machine learning model was developed to predict longitudinal functional decline and capture stage-dependent disease progression. Cox proportional hazards modeling further identified lower limb function, particularly walking and stair climbing, as the strongest predictors of earlier wheelchair access. Building on these results, we implemented a digital twin-inspired temporal machine learning-based time-to-event (TTE) model that generates individualized survival curves and dynamically predicts wheelchair-free survival. This framework provides a scalable, interpretable, and clinically actionable approach for linking ALS progression with personalized decision support, with applications in proactive care planning, clinical trial stratification, and precision medicine.

Figures

Figures reproduced from arXiv: 2607.14190 by Frank Bearoff, Huanmei Wu, Qi Li, Terry Heiman-Patterson, Zhaohui Qin, Zongliang Yue.

Figure 2
Figure 2. Figure 2: Correlation structure and domain organization of [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 1
Figure 1. Figure 1: Workflow for cross-linked dataset construction and [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 3
Figure 3. Figure 3: ASTP model output for a representative held-out [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Predicted 5-year domain survival curves for three [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Best-case cohort-wide decline-to-zero projections, one representative patient per visit-count group: Short, Medium, [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Decline-projection accuracy by visit-count group [PITH_FULL_IMAGE:figures/full_fig_p008_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Digital twin-inspired temporal machine learning [PITH_FULL_IMAGE:figures/full_fig_p008_7.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

42 extracted references

  1. [1]

    Amyotrophic lateral sclerosis,

    E. L. Feldmanet al., “Amyotrophic lateral sclerosis, ”The Lancet, vol. 400, no. 10360, pp. 1363–1380, 2022

    2022

  2. [2]

    Amyotrophic lateral sclerosis,

    M. C. Kiernanet al., “Amyotrophic lateral sclerosis, ”The Lancet, vol. 377, no. 9769, pp. 942–955, 2011

    2011

  3. [3]

    Amyotrophic lateral sclerosis,

    M. A. van Eset al., “Amyotrophic lateral sclerosis, ”Lancet, vol. 390, no. 10107, pp. 2084–2098, Nov. 2017

    2084

  4. [4]

    Diagnostic elements in amyotrophic lateral sclerosis: A case report,

    I. R. Marcu, O. C. Rogoveanu, R. Padureanu, V. Padureanu, and D. Dop, “Diagnostic elements in amyotrophic lateral sclerosis: A case report, ”Biomed Rep., vol. 21, no. 4, p. 141, Oct. 2024

    2024

  5. [5]

    Time-to-event prediction in als using a landmark modeling approach, using the als natural history consortium dataset,

    D. Schnecket al., “Time-to-event prediction in als using a landmark modeling approach, using the als natural history consortium dataset, ”Amyotroph Lateral Scler Frontotemporal Degener, vol. 26, no. 5–6, pp. 417–425, Aug. 2025

    2025

  6. [6]

    The alsfrs-r: A revised als functional rating scale that incorporates assessments of respiratory function,

    J. M. Cedarbaumet al., “The alsfrs-r: A revised als functional rating scale that incorporates assessments of respiratory function, ”J. Neurol. Sci., vol. 169, no. 1–2, pp. 13–21, 1999

    1999

  7. [7]

    Risk prediction for als using semi-competing risk models with applications to the als natural history consortium dataset,

    A. Arguedaset al., “Risk prediction for als using semi-competing risk models with applications to the als natural history consortium dataset, ”Amyotroph Lateral Scler Frontotemporal Degener, vol. 26, no. 5–6, pp. 550–557, Aug. 2025

    2025

  8. [8]

    The semi-competing risk problem revisited,

    R. L. Prentice and A. K. Aragaki, “The semi-competing risk problem revisited, ” Lifetime Data Anal., vol. 32, no. 2, Apr. 2026

    2026

  9. [9]

    Development and evaluation of a clinical staging system for amyotrophic lateral sclerosis,

    A. Chioet al., “Development and evaluation of a clinical staging system for amyotrophic lateral sclerosis, ”J. Neurol. Neurosurg. Psychiatry, vol. 86, no. 1, pp. 38–44, Jan. 2015

    2015

  10. [10]

    Practice parameter update: the care of the patient with amy- otrophic lateral sclerosis,

    R. G. Milleret al., “Practice parameter update: the care of the patient with amy- otrophic lateral sclerosis, ”Neurology, vol. 73, no. 15, pp. 1218–1226, Oct. 2009

    2009

  11. [11]

    Effects of non-invasive ventilation on survival and quality of life in patients with amyotrophic lateral sclerosis: a randomised controlled trial,

    S. C. Bourkeet al., “Effects of non-invasive ventilation on survival and quality of life in patients with amyotrophic lateral sclerosis: a randomised controlled trial, ” Lancet Neurol., vol. 5, no. 2, pp. 140–147, Feb. 2006

    2006

  12. [12]

    Mapping of critical events in disease progression through binary classification: Application to amyotrophic lateral sclerosis,

    Ö. O. Dalgıçet al., “Mapping of critical events in disease progression through binary classification: Application to amyotrophic lateral sclerosis, ”J. Biomed. Inform., vol. 123, p. 103895, 2021

    2021

  13. [13]

    Power wheelchair users with als: Impact of an alerting system on complications with prolonged sitting and power feature utilization,

    A. L. Ward, C. Nooijen, and J. Bernstein, “Power wheelchair users with als: Impact of an alerting system on complications with prolonged sitting and power feature utilization, ”Assistive Technology, pp. 1–10, 2025

    2025

  14. [14]

    Risk prediction for als using semi-competing risk models,

    A. Arguedaset al., “Risk prediction for als using semi-competing risk models, ” Amyotroph Lateral Scler Frontotemporal Degener, vol. 26, no. 5–6, pp. 550–557, 2025

    2025

  15. [15]

    Tutorial in biostatistics: competing risks and multi-state models,

    H. Putter, M. Fiocco, and R. B. Geskus, “Tutorial in biostatistics: competing risks and multi-state models, ”Stat. Med., vol. 26, no. 11, pp. 2389–2430, 2007

    2007

  16. [16]

    Deep learning predictions of survival based on mri in amyotrophic lateral sclerosis,

    H. K. van der Burghet al., “Deep learning predictions of survival based on mri in amyotrophic lateral sclerosis, ”NeuroImage: Clinical, vol. 13, pp. 361–369, 2017

    2017

  17. [17]

    A machine-learning based objective measure for als disease severity,

    F. G. Vieiraet al., “A machine-learning based objective measure for als disease severity, ”npj Digit. Med., vol. 5, no. 1, p. 45, 2022

    2022

  18. [18]

    Deep learning methods to predict amyotrophic lateral sclerosis disease progression,

    C. Pancottiet al., “Deep learning methods to predict amyotrophic lateral sclerosis disease progression, ”Sci. Rep., vol. 12, no. 1, p. 13738, 2022

    2022

  19. [19]

    Identifying and predicting amyotrophic lateral sclerosis clinical subgroups: a population-based machine-learning study,

    F. Faghriet al., “Identifying and predicting amyotrophic lateral sclerosis clinical subgroups: a population-based machine-learning study, ”Lancet Digit. Health, vol. 4, no. 5, pp. e359–e369, 2022

    2022

  20. [20]

    Rizopoulos,Joint Models for Longitudinal and Time-to-Event Data: With Appli- cations in R

    D. Rizopoulos,Joint Models for Longitudinal and Time-to-Event Data: With Appli- cations in R. Boca Raton, FL: CRC Press, 2012

    2012

  21. [21]

    Why most published research findings are false,

    J. P. A. Ioannidis, “Why most published research findings are false, ”PLoS Medicine, vol. 2, no. 8, p. e124, 2005

    2005

  22. [22]

    Transparent reporting of a multivariable prediction model for individual prognosis or diagno- sis (tripod): The tripod statement,

    G. S. Collins, J. B. Reitsma, D. G. Altman, and K. G. M. Moons, “Transparent reporting of a multivariable prediction model for individual prognosis or diagno- sis (tripod): The tripod statement, ”Journal of British Surgery, vol. 102, no. 3, pp. 148–158, 2015

    2015

  23. [23]

    E. W. Steyerberg,Clinical Prediction Models: A Practical Approach to Development, Validation, and Updating. New York, NY: Springer, 2009

    2009

  24. [24]

    Digital twin in industry: State-of-the- art,

    F. Tao, H. Zhang, A. Liu, and A. Y. C. Nee, “Digital twin in industry: State-of-the- art, ”IEEE Trans. Ind. Inform., vol. 15, no. 4, pp. 2405–2415, 2018

    2018

  25. [25]

    Digital twins to personalize medicine,

    B. Björnssonet al., “Digital twins to personalize medicine, ”Genome Med., vol. 12, no. 1, p. 4, 2019

    2019

  26. [26]

    The ‘digital twin’ to enable the vision of precision cardiol- ogy,

    J. Corral-Aceroet al., “The ‘digital twin’ to enable the vision of precision cardiol- ogy, ”Eur. Heart J., vol. 41, no. 48, pp. 4556–4564, 2020

    2020

  27. [27]

    Data-driven discovery of digital twins in biomedical research,

    C. Métayer, A. Ballesta, and J. Martinelli, “Data-driven discovery of digital twins in biomedical research, ”Briefings in Bioinformatics, vol. 27, no. 1, p. bbaf722, 2026

    2026

  28. [28]

    Digital twins in health care: ethical implications of an emerging engineering paradigm,

    K. Bruynseels, F. Santoni de Sio, and J. Van den Hoven, “Digital twins in health care: ethical implications of an emerging engineering paradigm, ”Front. Genet., vol. 9, p. 31, 2018

    2018

  29. [29]

    Digital twin in healthcare: Recent updates and challenges,

    T. Sun, X. He, and Z. Li, “Digital twin in healthcare: Recent updates and challenges, ” Digit. Health, vol. 9, p. 20552076221149651, 2023

    2023

  30. [30]

    High-performance medicine: the convergence of human and artificial intelligence,

    E. J. Topol, “High-performance medicine: the convergence of human and artificial intelligence, ”Nat. Med., vol. 25, no. 1, pp. 44–56, 2019

    2019

  31. [31]

    Personalized medicine: time for one-person trials,

    N. J. Schork, “Personalized medicine: time for one-person trials, ”Nature, vol. 520, no. 7549, pp. 609–611, 2015

    2015

  32. [32]

    Generalized additive models: An introduction with r,

    V. Gómez-Rubio, “Generalized additive models: An introduction with r, ”Journal of Statistical Software, vol. 86, pp. 1–5, 2018

    2018

  33. [33]

    Fast stable restricted maximum likelihood and marginal likelihood estimation of semiparametric generalized linear models,

    S. N. Wood, “Fast stable restricted maximum likelihood and marginal likelihood estimation of semiparametric generalized linear models, ”Journal of the Royal Statistical Society: Series B (Statistical Methodology), vol. 73, no. 1, pp. 3–36, 2011

    2011

  34. [34]

    Discrete-time methods for the analysis of event histories,

    P. D. Allison, “Discrete-time methods for the analysis of event histories, ”Socio- logical Methodology, vol. 13, pp. 61–98, 1982

    1982

  35. [35]

    Time-to-event prediction with neural networks and cox regression,

    H. Kvamme, Ø. Borgan, and I. Scheel, “Time-to-event prediction with neural networks and cox regression, ”Journal of Machine Learning Research, vol. 20, no. 129, pp. 1–30, 2019

    2019

  36. [36]

    Deephit: A deep learning approach to survival analysis with competing risks,

    C. Lee, W. Zame, J. Yoon, and M. van der Schaar, “Deephit: A deep learning approach to survival analysis with competing risks, ” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 32, no. 1, 2018

    2018

  37. [37]

    Attention is all you need,

    A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, Ł. Kaiser, and I. Polosukhin, “Attention is all you need, ” inAdvances in Neural Information Processing Systems, vol. 30, 2017

    2017

  38. [38]

    On layer normalization in the transformer architecture,

    R. Xiong, Y. Yang, D. He, K. Zheng, S. Zheng, C. Xing, H. Zhang, Y. Lan, L. Wang, and T. Liu, “On layer normalization in the transformer architecture, ” inInterna- tional Conference on Machine Learning. PMLR, 2020, pp. 10 524–10 533

    2020

  39. [39]

    Regression models and life-tables,

    D. R. Cox, “Regression models and life-tables, ”Journal of the Royal Statistical Society: Series B (Methodological), vol. 34, no. 2, pp. 187–202, 1972

    1972

  40. [40]

    Regularization and variable selection via the elastic net,

    H. Zou and T. Hastie, “Regularization and variable selection via the elastic net, ” Journal of the Royal Statistical Society: Series B (Statistical Methodology), vol. 67, no. 2, pp. 301–320, 2005

    2005

  41. [41]

    Artificial intelligence and black-box medical decisions: Accuracy versus explainability,

    A. J. London, “Artificial intelligence and black-box medical decisions: Accuracy versus explainability, ”Hastings Center Report, vol. 49, no. 1, pp. 15–21, 2019

    2019

  42. [42]

    Deep extended hazard models for survival analysis,

    Q. Zhong, J. W. Mueller, and J.-L. Wang, “Deep extended hazard models for survival analysis, ” inAdvances in Neural Information Processing Systems, vol. 34, 2021, pp. 15 111–15 124

    2021

This paper was first reviewed by deepseek-v4-flash on August 2, 2026.