Pith. sign in

REVIEW 4 major objections 6 minor 24 references

Cluster-weighted modeling of lifetime hierarchical data for profiling COVID-19 heart failure patients

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

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

Pith's one-line read The paper proposes a multilevel cluster-weighted survival model with cluster-specific shared frailties and shows it identifies three COVID-19 heart-failure patient profiles with distinct survival and hospital effects.

desk verdict A sound methodological extension of cluster-weighted models to survival data, with a real application whose clinical reading rests on an independence assumption the paper itself flags. read the letter →

arxiv 2507.12230 v1 pith:54V5NZKX submitted 2025-07-16 stat.AP

Luca Caldera , Andrea Cappozzo , Chiara Masci , Marco Forlani , Barbara Antonelli , Olivia Leoni , Anna Maria Paganoni , Francesca Ieva This is my paper
classification stat.AP MSC 62H3062N0162P10
keywords cluster-weightedmodelsfrailtysurvivalright-censoreddatalatentpatientprofileshealthcarehierarchyCOVID-19heartfailureEMalgorithmmodel-basedclustering
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

This paper develops a statistical model that finds hidden subgroups of patients while simultaneously estimating how hospitals and respiratory complications affect survival within each subgroup. It extends cluster-weighted models, which describe a population as a mixture of local regressions, to right-censored time-to-event outcomes nested in hospitals. In the Lombardy COVID-19 heart-failure cohort, the model selects three patient profiles with different survival curves and different hospital effects. If the method holds up, hospital comparisons and personalized risk assessment for this population can be adjusted for patient mix instead of treating all patients alike.

What carries the argument

The central object is a cluster-weighted mixture whose survival term is a parametric shared-frailty model: within each latent cluster $g$, the hazard for patient $i$ in hospital $j$ is $h_0(y;\gamma_g)m_{jg}\exp\{x_{ij}^T\beta_g\}$, with hospital-specific frailty $m_{jg}$ following a Gamma law with cluster-specific variance $\theta_g$. The frailty is integrated out analytically using derivatives of its Laplace transform, so the cluster log-likelihood has a closed classification form. Estimation alternates soft posterior weights with hard assignment (CEM) or stochastic assignment (SEM), which lets each cluster's survival parameters be fit independently with standard frailty-survival routines.

What would settle it

Fit a variant that lets the binary respiratory covariates correlate within clusters (for example, through an Ising or log-linear dependence term) on the same Lombardy data and compare cluster allocations, $PNA$ and $RF$ coefficients, and hospital frailty rankings to the paper's Table 5 and Figure 4; disagreement would indicate the independence assumption is load-bearing. In simulation, generate data with correlated binary covariates inside clusters and measure the drop in adjusted Rand index from the paper's reported 0.882.

Watch

Extended reading notes

Core claim

The paper claims that a cluster-weighted model with a shared-frailty survival component can decouple two sources of survival heterogeneity: latent patient subpopulations and hospital-level effects. The method factorizes the joint distribution of covariates and survival time into cluster-specific covariate densities and a parametric frailty survival density, integrating out hospital frailties through the Laplace transform of their distribution. Applied to 3086 heart-failure patients hospitalized with COVID-19 in 32 Lombardy hospitals, the model selects three latent clusters with a lognormal baseline hazard. Cluster 1 is small, high-comorbidity, and has the worst survival with negligible hospital effects; Cluster 2 consists entirely of COPD patients with significant pneumonia risk and pronounced hospital heterogeneity; Cluster 3 is the healthiest, largest group, with a borderline significant pneumonia effect and several hospitals showing protective or harmful frailties. The clinically central finding is that hospital influence matters most for relatively healthier patients and least for the most severely comorbid ones.

Load-bearing premise

Within each latent cluster, the categorical covariates are treated as independent, so correlations among COPD, bronchitis, pneumonia, and respiratory failure are ignored; the discovered clusters are defined almost entirely by COPD status, so a failure of this independence could change which patients land in which profile and what survival effects the model reports.

Editorial extensions

If this is right

  • If the claim is right, hospital-level comparisons for COVID-19 heart-failure patients should be made cluster-by-cluster: a hospital can look protective for one profile and neutral or harmful for another.
  • The significant pneumonia coefficients in Clusters 2 and 3 imply pneumonia raises the death hazard by about 23.5% and 16.7% within those profiles, while respiratory failure shows no significant cluster-specific effect.
  • The near-zero frailty variance in Cluster 1 means that for the sickest, highest-comorbidity patients, the treating hospital matters little; for healthier profiles, hospital choice can matter a lot.
  • The BIC selection procedure choosing $G=3$ with a lognormal baseline indicates that a single survival curve for the whole cohort would mask the profile structure.
  • The simulation study shows the estimation procedure recovers latent clusters with an average adjusted Rand index of 0.882, supporting the method's ability to identify the true grouping in a controlled setting.

Reading between the lines

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

  • A natural extension, explicitly listed by the paper as future work, is replacing the independent-multinomial covariate model with an Ising-type dependence term; this could be tested on the same data and should be checked for whether the COPD-driven clusters persist.
  • The method's cluster-specific frailty decoupling suggests a hospital profiling dashboard: report each hospital's frailty within each patient profile, which administrators could use to target care pathways.
  • Because the clusters are so strongly driven by COPD, the 'healthiest' third cluster may actually encode unmeasured respiratory severity; a replication with spirometry or oxygen-saturation data would test whether Cluster 3's survival is truly explained by monitoring prioritization.
  • The same machinery should transfer to other nested survival settings, such as patients in intensive-care units or treatment centers, wherever latent subgroups and facility effects both matter.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 6 minor

Summary. The manuscript introduces a cluster-weighted model for right-censored survival data with a hierarchical structure, where each latent cluster contains a parametric shared-frailty survival model for the response and independent Gaussian/multinomial models for continuous and categorical covariates. Estimation is performed with CEM and SEM algorithms based on a classification log-likelihood, and model selection uses BIC. The method is applied to 3086 Lombardy heart-failure patients hospitalized with COVID-19, yielding three latent patient clusters with cluster-specific PNA/RF effects and hospital frailties. A simulation study with G=3 clusters and a Weibull-gamma frailty data-generating process is used to demonstrate parameter recovery.

Significance. If the empirical findings withstand scrutiny, the paper makes a useful methodological contribution by extending cluster-weighted models to time-to-event outcomes while allowing group-level frailties to differ by latent cluster; the likelihood derivation in Appendix A is careful, and the R implementation is openly available. The simulation confirms that the estimation algorithms can recover parameters when the model is correctly specified. However, the real-data conclusions rest on a strong independence assumption for categorical covariates and on modeling PNA/RF as fixed covariates, and the simulation does not probe misspecification; these issues currently limit the strength of the clinical claims.

major comments (4)
  1. [Section 3.1, Eq. (1); Section 4.2, Table 5; Appendix B; Discussion] The categorical covariate distribution is a product of independent multinomials. In the fitted solution, clusters 2 and 3 are fully separated on COPD (pi_COPD = (0,1) and (1,0)), and Appendix B shows that the ICD-9 lists for COPD and BRH overlap (491.*). Under the product assumption, any within-cluster dependence between COPD and BRH cannot be represented, so the posterior allocations in Eq. (8) and the cluster-specific survival and frailty estimates are potentially distorted. The authors note the Ising-model extension as future work, but the three profiles are presented as empirical findings. A sensitivity analysis with a model that permits within-cluster dependence between the binary covariates (for example, an Ising or log-linear interaction term) is needed to establish that the profiles and the cluster-specific beta_g, (eta_g, nu_g), and theta_g estimates are not artifacts of this assumption.
  2. [Section 4.1 and Abstract] PNA and RF are described as complications that arise during COVID-19 infection and are included as time-fixed explanatory variables in the survival regression. If these conditions are diagnosed after hospital admission, conditioning on them in the hazard model introduces the standard time-dependent-covariate or immortal-time bias, which can distort the regression coefficients and the baseline survival estimates. The paper should either justify that PNA/RF status is known at the start of follow-up or model these variables as time-dependent covariates; at a minimum, the timing of these diagnoses relative to admission needs to be documented.
  3. [Table 5 and Section 4.2] For cluster 3, beta_PNA = 0.167 with p = 0.08, which is not significant at the conventional alpha = 0.05 level, yet the text states that the effect of PNA is significant in this cluster and Figure 3 displays the PNA hazard curve. This overstates the evidence and should be corrected in the text and figures; the cluster-3 PNA conclusion should be tempered accordingly.
  4. [Section 5] The simulation data-generating process is the same model that is fitted, including independent categorical covariates, a Weibull baseline, and a gamma frailty, so the simulation verifies computational self-consistency rather than robustness to misspecification. Because the real-data analysis is the main deliverable, at least one misspecification scenario, such as correlated binary covariates within clusters or a different baseline hazard, should be added, and a goodness-of-fit assessment for the chosen lognormal/gamma model should be reported. This would strengthen the claim that the method effectively uncovers latent profiles in practice.
minor comments (6)
  1. [Table 2] The BRH row reports 17.70% present and 83.30% not present, which sum to 101.00%; the not-present percentage is likely intended to be 82.30%.
  2. [Appendix C] The delta-method variance formula for the survival function omits the covariance between eta_g and nu_g; if these estimates are correlated, the confidence bands in Figure 2 may be miscalibrated.
  3. [Section 5.1] The statement that the model achieves effective recovery in 93 of 100 runs needs an explicit threshold for what counts as effective; the ARI and misclassification averages are reported but the criterion for a successful run is not defined.
  4. [Eq. (7)] The bracket notation in the frailty term, specifically log[(-1)^{d_jg} L^{(d_jg)}(...)], is difficult to parse; reformatting this expression would improve readability.
  5. [Section 4.2 and Figure 4a] Cluster 1 contains only 151 patients spread over 29 hospitals, so the average number of patients per hospital is fewer than six; the paper should comment on the stability of the cluster-1 hospital frailty estimates.
  6. [Section 3.1] The phrase 'finite space Omega' is misleading because the data space is a continuous/discrete product space rather than a finite set; this wording should be revised.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the paper fits a proposed multilevel cluster-weighted survival model and reports the fitted clusters, covariate effects, and frailties; no central claim reduces by construction to the model's inputs or to a load-bearing self-citation chain.

full rationale

The paper's central deliverable is an estimated mixture model for right-censored hierarchical survival data, not a derivation of a known result from first principles. The joint density in Eq. (1) is a model specification, and Eq. (7) is derived honestly from that specification and from the frailty likelihood of Eq. (5); Appendix A supplies the algebra. The application then fits this model to the Lombardy data, selects G=3 by BIC, and describes the resulting cluster-specific estimates; these are estimates, not predictions forced by construction. The simulation study (Section 5) draws data from a DGP that matches the fitted model family and checks parameter recovery, which is a self-consistency exercise rather than independent validation, but that is a limitation of evidence strength, not circularity. The Discussion explicitly flags the independence assumption for binary covariates and the single-profile assumption as limitations and lists Ising-model extensions as future work; these are model-misspecification and generalizability concerns, not cases where an output is equivalent to an input by definition. The near-deterministic separation of clusters 2 and 3 on COPD is an estimated empirical pattern (Table 5), and while it illustrates sensitivity to the covariate model, it does not constitute a circular step. Self-citations to Caldera et al. (2025) appear for a modified comorbidity score, for standard M-step formulas, and as a pointer for future Ising-model work; none of these citations carries a load-bearing uniqueness claim or supplies the paper's central identification result. No fitted parameter is renamed as a prediction, no uniqueness theorem is imported from the authors' prior work, and no ansatz is smuggled in via citation. Accordingly, there are no specific circular steps to list, and the appropriate score is 0.

Assumptions & free parameters 8 free parameters · 8 assumptions · 0 invented entities

The model relies on standard survival and mixture assumptions, plus several choices specific to this paper (Gamma frailty with mean 1, independent categorical covariates, BIC for model selection). No new physical or conceptual entities are introduced; the 'latent clusters' are statistical constructs.

free parameters (8)
  • Number of clusters G = 3
    Selected by BIC among G=1..5; central to the profiling claim.
  • Baseline hazard distribution = Lognormal
    Selected by BIC among Exponential, Weibull, Gompertz, Lognormal.
  • Frailty distribution = Gamma(1, theta_g)
    Assumed form; theta_g estimated but mean fixed to 1 for identifiability.
  • Observation window = 90 days
    Chosen by the authors; all survival estimates and censoring depend on this.
  • Hospital minimum size = 50 patients
    Exclusion threshold chosen to ensure reliable frailty estimates; affects sample composition.
  • Exclusion rule for transfers and repeated infections
    Data filtering choice that defines the cohort; potential selection bias.
  • Convergence threshold (CEM) = 1e-5
    Relative difference in objective function; chosen by hand.
  • Initialization method = k-prototypes
    Clustering on the five random covariates; deterministic but arbitrary in distance metric.
assumptions (8)
  • domain assumption The joint distribution factorizes as in Eq. (1): within each cluster, covariates and survival are conditionally independent given the latent cluster.
    This is the CWM factorization; it assumes the cluster explains the dependence between covariates and survival.
  • domain assumption Censoring time C is independent of survival time T.
    Standard survival analysis assumption, stated in Section 3.1 but not verified.
  • domain assumption Frailty terms m_jg are independent Gamma(1, theta_g) random variables across hospitals and clusters.
    Gamma frailty with mean 1 is a standard choice; independence of hospital frailties across clusters is assumed.
  • domain assumption The baseline hazard is parametric (Lognormal in the selected model), and its form is correctly specified.
    The validity of cluster-specific survival estimates depends on the parametric baseline being adequate.
  • ad hoc to paper Categorical covariates are independent within each cluster.
    The paper models V as a product of independent multinomials (Section 3.1), an explicit simplification acknowledged as a limitation; dependencies among COPD and BRH are not captured.
  • standard math The Laplace transform and its derivatives can be computed for the Gamma frailty (Appendix A).
    Used to derive the closed-form likelihood in Eq. (7).
  • standard math The delta method is valid for the survival function confidence intervals.
    Used in Supplement C; requires large-sample normality of parameter estimates.
  • domain assumption BIC with the specified penalty is an appropriate model selection criterion.
    Used to choose G and baseline distribution; BIC consistency for mixture models is not guaranteed in this setting.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Cluster-weighted modeling of lifetime hierarchical data for profiling COVID-19 heart failure patients." pith.science (2026). https://pith.science/paper/54V5NZKX

@misc{pith2026250712230,
  author       = {Pith},
  title        = {Pith review of: Cluster-weighted modeling of lifetime hierarchical data for profiling COVID-19 heart failure patients},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/54V5NZKX}},
  note         = {Machine review of arXiv:2507.12230}
}
read the original abstract

This study investigates the heterogeneity in survival times among COVID-19 patients with Heart Failure (HF) hospitalized in the Lombardy region of Italy during the pandemic. To address this, we propose a novel mixture model for right-censored lifetime data that incorporates random effects and allows for local distributions of the explanatory variables. Our approach identifies latent clusters of patients while estimating component-specific covariate effects on survival, taking into account the hierarchical structure induced by the healthcare facility. Specifically, a shared frailty term, unique to each cluster, captures hospital-level variability enabling a twofold decoupling of survival heterogeneity across both clusters and hierarchies. Two EM-based algorithms, namely a Classification EM (CEM) and a Stochastic EM (SEM), are proposed for parameter estimation. The devised methodology effectively uncovers latent patient profiles, evaluates within-cluster hospital effects, and quantifies the impact of respiratory conditions on survival. Our findings provide new information on the complex interplay between the impacts of HF, COVID-19, and healthcare facilities on public health, highlighting the importance of personalized and context-sensitive clinical strategies.

Figures

Figures reproduced from arXiv: 2507.12230 by the authors.

Figure 1
Figure 1. Scatterplots of patients in the Age–MCS feature space colored according to the three estimated clusters; [PITH_FULL_IMAGE:figures/full_fig_p011_1.png] view at source ↗
Figure 2
Figure 2. Estimated survival curves for the three clusters. The orange curve represents the baseline survival curve for [PITH_FULL_IMAGE:figures/full_fig_p013_2.png] view at source ↗
Figure 3
Figure 3. Estimated baseline hazard in the three clusters and effect of the significant covariates (black curve). [PITH_FULL_IMAGE:figures/full_fig_p014_3.png] view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Estimated frailties and their 95% confidence intervals in the three clusters. Hospitals that significantly increase [PITH_FULL_IMAGE:figures/full_fig_p015_4.png]
Figure 5
Figure 5. Figure 5: Comparison of BIC distributions for G ∈ {1, 2, 3, 4}. Each boxplot represents the BIC distribution for a given G across the 100 repetitions of the simulated experiment. 0.4 0.5 0.6 0.7 0.8 0.9 ARI ARI Value Adjusted Rand Index (ARI) 0.05 0.15 0.25 0.35 0.45 Misclassifi…
Figure 6
Figure 6. Figure 6: ARI and percentage of missclassified observations across the 100 simulations. [PITH_FULL_IMAGE:figures/full_fig_p016_6.png]
Figure 7
Figure 7. Figure 7: Comparison of the distributions of estimated parameters across the R=100 repetitions with the true values of [PITH_FULL_IMAGE:figures/full_fig_p017_7.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

24 extracted references · 16 canonical work pages

  1. [7]

    doi: 10.1007/ s00180-016-0661-7

    ISSN 0943-4062. doi: 10.1007/ s00180-016-0661-7. URL http://link.springer.com/10.1007/s00180-016-0661-7 . Charles Bouveyron, Gilles Celeux, T Brendan Murphy, and Adrian E Raftery. Model-Based Clustering and Clas- sification for Data Science , volume

    work page doi:10.1007/s00180-016-0661-7

  2. [8]

    doi: 10.1017/9781108644181

    ISBN 9781108644181. doi: 10.1017/9781108644181. URL https://www.cambridge.org/core/product/identifier/9781108644181/ type/book. Peter Bryant and John A Williamson. Asymptotic Behaviour of Classification Maximum Likelihood Estimates. Biometrika, 65(2):273, aug

    arXiv

  3. [10]

    URL http://www.jstor.org/stable/2984875https://onlinelibrary

    1111/j.2517-6161.1977.tb01600.x. URL http://www.jstor.org/stable/2984875https://onlinelibrary. wiley.com/doi/10.1111/j.2517-6161.1977.tb01600.x. Rong-Hui Du, Li-Rong Liang, Cheng-Qing Yang, Wen Wang, Tan-Ze Cao, Ming Li, Guang-Yun Guo, Juan Du, Chun-Lan Zheng, Qi Zhu, et al. Predictors of mortality for patients with covid-19 pneumonia caused by sars-cov-2...

    arXiv 1977

  4. [12]

    22 A PREPRINT - S EPTEMBER 25, 2025 A. P. Dempster, N. M. Laird, and D. B. Rubin. Maximum likelihood from incomplete data via the em algorithm. Journal of the Royal Statistical Society: Series B (Methodological) , 39:1–22, 9

    work page 2025

  5. [21]

    doi: 10.18637/jss.v051.i11

    ISSN 1548-7660. doi: 10.18637/jss.v051.i11. URL http://www.jstatsoft.org/v51/i11/. Søren Feodor Nielsen. The Stochastic EM Algorithm: Estimation and Asymptotic Results.Bernoulli, 6(3):457, jun

    work page doi:10.18637/jss.v051.i11

  6. [23]

    doi: 10.1214/aos/1176344136

    ISSN 0090-5364. doi: 10.1214/aos/1176344136. URL http://projecteuclid.org/euclid.aos/1176344136. Angelica Tiotiu, Herberto Chong Neto, Andras Bikov, Krzysztof Kowal, Paschalis Steiropoulos, Marina Labor, Ivan Cherrez-Ojeda, Hector Badellino, Alexander Emelyanov, Rocio Garcia, et al. Impact of the covid-19 pandemic on the management of chronic noninfectiou...

    arXiv

  7. [1978]

    doi: 10.2307/2335205

    ISSN 00063444. doi: 10.2307/2335205. URL https://www.jstor.org/ stable/2335205?origin=crossref. Luca Caldera, Chiara Masci, Andrea Cappozzo, Marco Forlani, Barbara Antonelli, Olivia Leoni, and Francesca Ieva. Uncovering mortality patterns and hospital effects in COVID-19 heart failure patients: a novel multilevel logistic cluster-weighted modeling approac...

    work page arXiv

  8. [1985]

    doi: 10.1007/BF01908075

    ISSN 0176-4268. doi: 10.1007/BF01908075. URL http://link.springer.com/10.1007/BF01908075. Salvatore Ingrassia, Simona C. Minotti, and Giorgio Vittadini. Local statistical modeling via a cluster-weighted approach with elliptical distributions. Journal of Classification , 29:363–401, 10

    doi:10.1007/bf01908075

Show all 24 references
  1. [1992]

    doi: 10.1016/0167-9473(92)90042-E

    ISSN 01679473. doi: 10.1016/0167-9473(92)90042-E. URL https://www.sciencedirect.com/science/article/pii/016794739290042E. Didier Chauveau. A stochastic em algorithm for mixtures with censored data. Journal of statistical planning and inference, 46(1):1–25,

  2. [1997]

    doi: 10.1111/j.1749-6632.1997.tb51651.x

    ISSN 0077-8923. doi: 10.1111/j.1749-6632.1997.tb51651.x. URL http://doi.wiley.com/ 10.1111/j.1749-6632.1997.tb51651.x. Davide Giustivi, Francesco Bottazzini, and Mirko Belliato. Respiratory monitoring at bedside in covid-19 patients. Journal of clinical medicine, 10(21):4943,

    1997

  3. [2000]

    doi: 10.2307/3318671

    ISSN 13507265. doi: 10.2307/3318671. URL https://www.jstor.org/stable/3318671?origin=crossref. R Core Team. R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing, Vienna, Austria,

  4. [2007]

    doi: 10.1016/j.csda.2006.03.009

    ISSN 01679473. doi: 10.1016/j.csda.2006.03.009. Ernest A Adeghate, Nabil Eid, and Jaipaul Singh. Mechanisms of covid-19-induced heart failure: a short review. Heart failure reviews, 26:363–369,

    2006 doi

  5. [2008]

    doi: 10.1214/07-AOS515

    ISSN 0090-5364. doi: 10.1214/07-AOS515. URL http://projecteuclid.org/euclid.aos/1211819566. Neil Gershenfeld. Nonlinear inference and cluster-weighted modeling. Annals of the New York Academy of Sciences, 808:18–24, 1

  6. [2012]

    doi: 10.1007/s00357-012-9114-3

    ISSN 0176-4268. doi: 10.1007/s00357-012-9114-3. URL http://link.springer.com/10.1007/s00357-012-9114-3 . Salvatore Ingrassia, Simona C. Minotti, and Antonio Punzo. Model-based clustering via linear cluster-weighted models. Computational Statistics and Data Analysis , 71:159–182, 3

    doi

  7. [2014]

    doi: 10.1016/j.csda.2013

    ISSN 01679473. doi: 10.1016/j.csda.2013. 02.012. URL http://dx.doi.org/10.1016/j.csda.2013.02.012https://linkinghub.elsevier.com/ retrieve/pii/S016794731300056X. Salvatore Ingrassia, Antonio Punzo, Giorgio Vittadini, and Simona C. Minotti. The generalized linear mixed cluster-...

    2013 doi

  8. [2015]

    doi: 10.1007/s00357-015-9175-1

    ISSN 0176-4268. doi: 10.1007/s00357-015-9175-1. URL http://link.springer.com/10.1007/s00357-015-9175-1 . Anil K Jain and Richard C Dubes. Algorithms for clustering data. Prentice-Hall, Inc.,

    doi

  9. [2016]

    doi: 10.1007/s40300-016-0098-3

    ISSN 0026-1424. doi: 10.1007/s40300-016-0098-3. URL http://link.springer.com/10.1007/s40300-016-0098-3 . Paolo Berta, Salvatore Ingrassia, Giorgio Vittadini, and Daniele Spinelli. Latent heterogeneity in COVID-19 hospitali- sations: a cluster-weighted approach to analyse morta...

    doi

  10. [2017]

    doi: 10.1111/insr.12214

    ISSN 03067734. doi: 10.1111/insr.12214. URL http://doi.wiley.com/10.1111/ insr.12214. Feras Bader, Yosef Manla, Bassam Atallah, and Randall C Starling. Heart failure and covid-19. Heart failure reviews, 26(1):1–10,

    doi

  11. [2018]

    doi: 10.18637/jss.v086.i04

    ISSN 1548-7660. doi: 10.18637/jss.v086.i04. URL http://www.jstatsoft.org/v86/i04/. Marco Munda, Federico Rotolo, and Catherine Legrand. parfm : Parametric frailty models in r. Journal of Statistical Software, 51,

    doi

  12. [2019]

    doi: 10.1002/sam.11421

    ISSN 1932-1864. doi: 10.1002/sam.11421. URL https://onlinelibrary.wiley.com/doi/10.1002/sam.11421. Paolo Berta, Salvatore Ingrassia, Antonio Punzo, and Giorgio Vittadini. Multilevel cluster-weighted models for the evaluation of hospitals. METRON, 74:275–292, 12

    1932 doi

  13. [2020]

    doi: 10.1177/0962280220921889

    ISSN 14770334. doi: 10.1177/0962280220921889. Alida Benfante and Nicola Scichilone. Prioritizing care for severe asthma during sars-cov-2 pandemic. Pulmonology, 27(3):189–190,

    doi

  14. [2021]

    The impact of heterogeneity in individual frailty on the dynamics of mortality

    23 A PREPRINT - S EPTEMBER 25, 2025 James W Vaupel, Kenneth G Manton, and Eric Stallard. The impact of heterogeneity in individual frailty on the dynamics of mortality. Demography, 16(3):439–454,

    2025

  15. [2024]

    doi: 10.1111/anzs.12407

    ISSN 1369-1473. doi: 10.1111/anzs.12407. URL https://onlinelibrary.wiley.com/doi/ 10.1111/anzs.12407. Laurent Bordes and Didier Chauveau. Stochastic em algorithms for parametric and semiparametric mixture models for right-censored lifetime data. Computational Statistics, 31:15...

    doi

  16. [2025]

    doi: 10.1093/biomtc/ujaf046

    ISSN 0006-341X. doi: 10.1093/biomtc/ujaf046. URL https://academic.oup.com/biometrics/article/doi/10.1093/biomtc/ujaf046/8121043. Gilles Celeux. The sem algorithm: a probabilistic teacher algorithm derived from the em algorithm for the mixture problem. Computational statistics ...

    doi

Pith tools

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