A Bayesian survival model induced by hurdle zero-modified power series discrete frailty with dispersion: an application in lung cancer

Danilo Alvares; Joaqu\'in Mart\'inez-Minaya; Katy C. Molina; Vera D. Tomazella

REVIEW 2 major objections 5 minor 42 references

A Bayesian survival model induced by hurdle zero-modified power series discrete frailty with dispersion: an application in lung cancer

T0 review · 2 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read A Bayesian survival model driven by hurdle zero-modified power-series discrete frailty with a dispersion parameter can learn which patients are long-term survivors and classify each patient's zero-modification profile without knowing the…

desk verdict HZMGP frailty survival model is a reasonable incremental extension, but it ships with a sign error in the printed p.m.f. and an unspecified Lambert W branch; both are fixable. read the letter →

arxiv 2505.23568 v1 pith:FSLN23HB submitted 2025-05-29 stat.ME stat.AP

Katy C. Molina , Joaquín Martínez-Minaya , Danilo Alvares , Vera D. Tomazella This is my paper

classification stat.MEstat.AP MSC 62N0162F1562P10

keywords Bayesiansurvivalanalysisdiscretefrailtyzero-modifiedpowerserieshurdlereparameterizationcurefractionlong-termsurvivorslungcancerLambertWfunction

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 proposes a Bayesian survival model in which unobserved heterogeneity is a discrete frailty variable drawn from the hurdle zero-modified power-series (HZMPS) family, specifically the hurdle zero-modified generalized Poisson (HZMGP) distribution. The central claim is that the dispersion parameter lets the frailty distribution adapt to the data's zero structure, so one fit covers zero-inflated, zero-deflated, zero-truncated, and ordinary power-series frailty without knowing in advance which applies. If the claim is right, the model yields likelihood-based inference for cure fractions and long-term survivors while also giving a per-patient classification of risk profile. The authors support the proposal with simulations and a large lung cancer dataset in which the fitted model separates 65% poor-prognosis (zero-deflated), 32% favorable-prognosis (zero-inflated), and 3% standard-risk patients.

What carries the argument

The load-bearing object is the probability-generating function of the zero-truncated generalized-Poisson component, expressed through Lambert W, the function satisfying $W(x)e^{W(x)} = x$. Because the frailty's p.g.f. $G_V(s)$ appears directly in the marginal survival $S(t)=G_V(S_0(t))$, a closed form for $G_V$ turns the hierarchical model into a computable likelihood. The hurdle reparameterization $\omega = \rho(1 - \pi_{\mathrm{PS}}(0))$ makes the zero-structure parameter orthogonal to the mean, and the threshold $1 - e^{-\mu/(1+\mu\phi)}$ separates the zero-inflated, zero-deflated, and zero-truncated regimes used for classification.

What would settle it

For any posterior draw, compute $a(t) = -\frac{\mu \phi}{1+\mu\phi} S_0(t) e^{-\frac{\mu \phi}{1+\mu\phi}}$ for $t \in (0,\infty)$; if $a(t) < -1/e$ for some $t$, the Lambert W term has no real value and the likelihood that produced the draw is undefined, and checking $S(0)=1$ with $S(t)$ nonincreasing over posterior draws would provide the same verdict.

Watch

Extended reading notes

Core claim

The paper establishes on its own terms that the machinery of zero-modified power-series count distributions can be carried into survival analysis without fixing the zero-modification structure in advance. With frailty $V \sim \text{HZMGP}$, the marginal survival function is $S(t) = G_V(S_0(t))$, the frailty probability-generating function evaluated at the Weibull baseline survival; the p.g.f. of the zero-truncated generalized-Poisson component has a closed form in Lambert W, and the hurdle reparameterization makes the zero-structure parameter $\omega$ orthogonal to the mean $\mu$. The authors then place logit and log links on $\omega$ and $\mu$, assign weakly informative priors, sample the posterior with Hamiltonian Monte Carlo, and classify each individual by comparing the posterior draws of $\omega_i$ with the threshold $1 - e^{-\mu_i/(1+\mu_i \phi)}$. Applied to 30,900 lung cancer patients, the posterior classification assigns 65% to the zero-deflated class, 32% to the zero-inflated class, and 3% to the ordinary generalized-Poisson class, interpreted as poor-prognosis, long-term-survivor, and standard-risk profiles.

Load-bearing premise

The model's likelihood is built from a probability-generating function that uses the Lambert W function, and the paper never specifies the real branch of W or the parameter restrictions that keep its argument in the real domain; if those restrictions fail, the marginal survival function is not guaranteed to be a valid survival function.

Editorial extensions

If this is right

A single fit of the model replaces model selection among zero-inflated, zero-deflated, zero-truncated, and ordinary power-series frailty specifications, because the posterior classifies each individual and yields population shares of each type.
In the lung cancer study the model assigns 65% of patients to the zero-deflated class (poor prognosis), 32% to the zero-inflated class (favorable prognosis and long-term survival), and 3% to the ordinary generalized-Poisson class.
The dispersion parameter $\phi$ is estimated with high precision in the application (posterior mean 0.140, 95% interval 0.134–0.147), indicating unmeasured heterogeneity beyond the observed covariates.
Simulation results with $n = 10{,}000$ and $n = 20{,}000$ show posterior means converging to the true values and coverage probabilities near the nominal level, while $n = 1{,}000$ leaves some parameters biased.
Individual-level classification at a threshold of $\alpha = 0.1$ stabilizes as sample size grows, so the zero-modification shares can be reported as population-level summaries.

Reading between the lines

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

Editorial extension: the same hurdle construction can be applied to other members of the power-series family, such as negative binomial or binomial, giving discrete frailty models that share the automatic zero-structure selection.
Editorial extension: the posterior classification rule could be used as a decision aid, sending zero-deflated patients to intensive follow-up and zero-inflated patients to de-escalated monitoring; the paper notes the clinical interpretation but does not propose a decision protocol.
Editorial extension: because the likelihood rests on Lambert W, a practical implementation should monitor the argument of W during sampling and report whether any posterior draw leaves the real domain; the paper does not provide this diagnostic.
Editorial extension: the roughly three-hour fit for 30,900 patients suggests approximate Bayesian computational methods could make the model practical for routine use, at the cost of checking whether the nonlinear W term is well approximated.

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.

Desk Editor's Note

read the letter

The HZMGP frailty model is a reasonable incremental extension of the ZMPS frailty framework, and the classification layer is a nice practical output. But the paper as written has an invalid p.m.f. in Table 1 and an unstated branch assumption for Lambert W. Both are fixable, but they need attention.

The genuinely new part is the hurdle reparameterization of the zero-modified power series family with a dispersion parameter, applied to discrete frailty in survival models. The authors correctly derive the marginal survival function as the p.g.f. of the frailty evaluated at the baseline survival, and the likelihood follows. The simulation study is a proper self-consistency check, and the lung cancer analysis uses a large real dataset. The code is on GitHub, which is good practice.

The soft spots. First, the p.m.f. in Table 1 is wrong as printed. The exponential term e^{μ/(1+μϕ)} multiplies the entire y>0 expression, but it should be e^{-μ/(1+μϕ)}. With the printed sign, the probabilities do not sum to 1: for μ=1, φ=1, P(Y=1) alone exceeds 1 before any renormalization. This is almost certainly a typo, because the p.g.f. in the same table is consistent with the negative sign, and the survival and hazard functions in Table 2 follow from that p.g.f. But as a published formula, it is not a valid distribution, and the paper needs to correct it.

Second, the Lambert W function appears throughout with a negative argument, and the paper never states which real branch is intended. The formulas are only valid on the principal branch W_0; using W_{-1} gives a negative hazard in Table 2. A careful reader can infer W_0 from context, but the omission is a rigor gap in a model whose entire likelihood is built on that function.

Third, the claimed flexibility is not tested against alternatives. The paper doesn't compare the HZMGP model with the simpler ZMPS model or other discrete frailty specifications, so we don't know whether the extra dispersion parameter is worth the computational cost (three hours on the real data). The classification into ZIGP, ZDGP, etc. is interesting, but it is a post-hoc summary of the posterior, not a validated model-selection tool.

Overall, the central derivation holds up on the principal branch, and the errors are technical, not conceptual. I would send this to a competent referee. The author response should fix the sign error and state the branch assumption explicitly. After that, it's a solid, usable contribution for people working on cure-rate survival models with discrete frailty.

Referee Report

2 major / 5 minor

Summary. The paper proposes a Bayesian survival model in which the frailty term follows a hurdle zero-modified power series (HZMPS) distribution, specifically the hurdle zero-modified generalized Poisson (HZMGP) distribution. The marginal survival and hazard functions are obtained by evaluating the probability generating function of the frailty at the baseline survival function, following standard frailty-model theory. The authors specify a Weibull baseline, link the zero-modification probability and frailty mean to covariates, and perform Bayesian inference via Stan. They report a simulation study with two scenarios and three sample sizes, and an application to a large lung cancer registry data set. A novel feature is a posterior-based classification of each patient into zero-inflated, zero-deflated, zero-truncated, or ordinary generalized Poisson frailty specifications. The central claim is that the dispersion parameter and the hurdle structure make the model flexible enough to capture diverse heterogeneity patterns and to identify long-term survivors without prior knowledge of the zero-modification structure.

Significance. If the model were correctly specified, the paper would make a useful contribution to discrete-frailty cure-rate modeling: the p.g.f. route to the marginal survival function is theoretically sound, the classification of zero-modification structures has practical appeal, and the availability of R-Stan code plus a large real-data application are strengths. However, as printed, the defining distribution in Table 1 is not a valid probability mass function, and the Lambert W branch used in the p.g.f., survival, and hazard formulas is never stated. These issues affect every downstream quantity in the paper, including the likelihood, posterior, simulations, and patient classification. The central claim cannot be evaluated until those load-bearing formulas are corrected and the branch is specified.

major comments (2)

[Table 1] The p.m.f. printed for the HZMGP distribution is not a valid probability mass function. The zero-truncated component contains the factor e^{μ/(1+μφ)} where the exponent should be negative; the correct generalized-Poisson term has e^{-μ(1+φy)/(1+μφ)} rather than e^{μ(1-φy)/(1+μφ)}. As a concrete check, at μ=1 and φ=1 the printed zero-truncated probability for y=1 equals (1/2)/(1-e^{-1/2}) ≈ 1.27, which already exceeds 1. Thus the distribution cannot sum to 1, and the formulas in Table 2, Eq. (10), and all posterior summaries built on them are invalid as stated.
[Table 1 and Table 2] The paper never states which real branch of the Lambert W function is used, although the arguments in Tables 1 and 2 are negative (in (-1/e,0)). The formulas are internally consistent only on the principal branch W_0, for which W(x)∈(-1,0); on the W_{-1} branch the quantity W(-μφ/(1+μφ) S0(t) e^{-μφ/(1+μφ)}) + μφ/(1+μφ) can become negative, making the hazard in Table 2 negative. The authors must state the branch, justify that the argument stays in the real domain of that branch for all admissible parameter values and t, and then re-derive the survival and hazard expressions under that stated branch.

minor comments (5)

[Table 3 and Section 5] The text says that at n=1000 coverage is 'medium (~70%)' for a few parameters, but Table 3 reports CP values of 0.72, 0.74, 0.80, and 0.90 for γ, β(μ)_0, λ, and β(μ)_2 in Scenario I, and Scenario II also shows CP=0.73 for γ; this understates the extent of the small-sample estimation problem and should be described more precisely.
[Eq. (13)] The prior is written as N(0,102), which is ambiguous; it should be N(0,10^2) or the variance should be stated explicitly as 100.
[Eq. (9)] The logit link for ω_i is rendered incorrectly: it should read ω_i = exp(x_i^T β^{(ω)})/(1 + exp(x_i^T β^{(ω)})), with no stray 'p' in the denominator.
[Table 2 caption] The caption reads 'HZMPG model' but the model is called HZMGP elsewhere; the typo should be corrected.
[Eq. (1) and elsewhere] The indicator definition uses the Portuguese word 'se' instead of English 'if'; this should be fixed throughout for consistency.

Circularity Check

0 steps flagged · score 2.0 of 10

No substantive circularity: the frailty survival model is a direct p.g.f. construction, and the cited HZMPS background work is not load-bearing.

full rationale

The paper's central derivation chain is: choose a discrete frailty distribution (HZMGP), compute the marginal survival function as S(t) = G_V(S0(t)) via the frailty p.g.f. (Eq. 8), then form the likelihood, posterior, and classification rule. This is a standard and self-contained application of probability generating function theory, not an assumption of the target result. The HZMPS/HZMGP family is taken from prior work by Conceicao et al. (2017) and Molina et al. (2021), which includes coauthors of the present paper, so there is self-citation; however, those citations supply background distributional definitions and do not force the survival-model conclusion. No uniqueness theorem, fitted parameter renamed as a prediction, or ansatz smuggled via citation is used. The simulation study generates data from the same model and recovers parameters, which is a self-consistency check rather than external validation, but it is not circular in the sense of fitting a subset and predicting the same quantity. The Section 4.4 classification rule is a decision rule based on the parameter regions of the HZMPS family shown in Figure 1; it is a definitional application, not a circular derivation. I note two correctness concerns that are not circularity: Table 1's printed p.m.f. appears to have a sign error in the exponential factor (the exponent should be negative for the distribution to sum to one), and the paper does not state which real branch of the Lambert W function is used; both are correctness risks, not circular-reasoning risks. Overall, the derivation is independent of its conclusions and scores low on circularity.

Assumptions & free parameters 3 free parameters · 5 assumptions · 2 invented entities

The central model rests on a standard frailty identity, a Weibull baseline, the external p.g.f. formula for HZMGP, an unstated Lambert W branch, and an unexamined identifiability assumption. No new physical or mathematical constants are introduced; all parameters are estimated from data.

free parameters (3)

Threshold alpha for classification = 0.1
Chosen by hand in decision rule (16) to assign patients to zero-modification groups; the reported proportions of ZIGP, GP, ZDGP, ZTGP depend on this value.
Dispersion parameter phi = 0.140 (posterior mean, lung cancer data)
Dispersion in the HZMGP frailty distribution; estimated from the survival data and used to interpret unmeasured heterogeneity.
Weibull baseline parameters lambda, gamma = lambda = 0.187, gamma = 1.266 (posterior means, lung cancer data)
Scale and shape of the baseline hazard, estimated jointly with the frailty parameters.

assumptions (5)

standard math The marginal survival function is S(t) = G_V(S0(t)) for a discrete frailty V under proportional hazards.
Equation (8); standard frailty identity from Vaupel et al. (1979) and Cox and Oakes (1984), used to derive all model functions.
domain assumption The baseline hazard is Weibull with scale lambda and shape gamma.
Section 3; a modeling choice, not derived from data or theory.
domain assumption The p.g.f. of the zero-truncated generalized Poisson distribution, involving the Lambert W function, is correct as reported.
Table 1; taken from Ambagaspitiya and Balakrishnan (1994), without proof in this paper.
ad hoc to paper Lambert W is evaluated on the principal branch so that W(x) is in (-1,0) for the negative arguments in Tables 1 and 2.
The paper never states the branch; the hazard formula in Table 2 is positive only on this branch, so the model is incomplete without this assumption.
domain assumption The model parameters are identifiable from right-censored survival data.
Section 4 assumes a well-behaved posterior; no identifiability proof is given, and the simulation shows high posterior SDs at n=1000.

invented entities (2)

Discrete frailty latent variable V with HZMGP distribution
purpose: Captures unobserved heterogeneity; zero frailty corresponds to long-term survivors.
A modeling construct; no independent observable is proposed to validate the distribution beyond the fitted model.
Individual zero-modification classification (ZIGP, GP, ZDGP, ZTGP)
purpose: Stratifies patients into risk categories for clinical interpretation.
Labels are produced by the decision rule (16) with hand-chosen alpha; they are not compared with clinical outcomes or external biomarkers.

how reviews work

0 comments

Cite this review

Pith. "Pith review of A Bayesian survival model induced by hurdle zero-modified power series discrete frailty with dispersion: an application in lung cancer." pith.science (2026). https://pith.science/paper/FSLN23HB

@misc{pith2026250523568,
  author       = {Pith},
  title        = {Pith review of: A Bayesian survival model induced by hurdle zero-modified power series discrete frailty with dispersion: an application in lung cancer},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/FSLN23HB}},
  note         = {Machine review of arXiv:2505.23568}
}

read the original abstract

Frailty survival models are widely used to capture unobserved heterogeneity among individuals in clinical and epidemiological research. This paper introduces a Bayesian survival model that features discrete frailty induced by the hurdle zero-modified power series (HZMPS) distribution. A key characteristic of HZMPS is the inclusion of a dispersion parameter, enhancing flexibility in capturing diverse heterogeneity patterns. Furthermore, this frailty specification allows the model to distinguish individuals with higher susceptibility to the event of interest from those potentially cured or no longer at risk. We employ a Bayesian framework for parameter estimation, enabling the incorporation of prior information and robust inference, even with limited data. A simulation study is performed to explore the limits of the model. Our proposal is also applied to a lung cancer study, in which patient variability plays a crucial role in disease progression and treatment response. The findings of this study highlight the importance of more flexible frailty models in survival data analysis and emphasize the potential of the Bayesian approach to modeling heterogeneity in biomedical studies.

Figures

Figures reproduced from arXiv: 2505.23568 by the authors.

**Figure 2.** Classification of zero-modification specification for the first four patients in the lung cancer data. [PITH_FULL_IMAGE:figures/full_fig_p011_2.png] view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

42 extracted references · 42 canonical work pages

[1]

Modelling survival data in medical research

Collett D. Modelling survival data in medical research . 3rd ed. Boca Raton, FL: Chapman and Hall/CRC, 2015

work page 2015
[2]

Statistical methods for reliability data

Meeker WQ, Escobar LA and Pascual FG. Statistical methods for reliability data . 2nd ed. Hoboken, NJ: John Wiley & Sons, 2022

work page 2022
[3]

Actuarial mathematics for life contingent risks

Dickson DCM, Hardy MR and Waters HR. Actuarial mathematics for life contingent risks . 3rd ed. Cambridge: Cambridge University Press, 2020

work page 2020
[4]

Machine learning for survival analysis: A survey

Wang P, Li Y and Reddy CK. Machine learning for survival analysis: A survey. ACM Computing Surveys (CSUR) 2019; 51(6): 1–36

work page 2019
[5]

World Health Organization. Cancer. Fact sheet, 2024. URL https://www.who.int/news-room/ fact-sheets/detail/cancer

work page 2024
[6]

Cancer statistics, 2024

Siegel RL, Miller KD, Wagle NS et al. Cancer statistics, 2024. CA: A Cancer Journal for Clinicians 2024; 74(1): 12–49

work page 2024
[7]

Global cancer observatory: Cancer today

Ferlay J, Ervik M, Lam F et al. Global cancer observatory: Cancer today. Lyon, France: International Agency for Research on Cancer 2024; 2018-2020

work page 2024
[8]

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

Vaupel JW, Manton KG and Stallard E. The impact of heterogeneity in individual frailty on the dynamics of mortality. Demography 1979; 16(3): 439–454

work page 1979

Show all 42 references

[9]

Analysis of survival data

Cox DR and Oakes D. Analysis of survival data . 1st ed. Boca Raton, FL: Chapman and Hall/CRC, 1984. 11 A Bayesian survival model induced by HZMPS discrete frailty with dispersion

1984
[10]

Frailty models in survival analysis

Wienke A. Frailty models in survival analysis . 1st ed. Boca Raton, FL: Chapman and Hall/CRC, 2010

2010
[11]

Proportional hazards models with discrete frailty

Caroni C, Crowder M and Kimber A. Proportional hazards models with discrete frailty. Lifetime Data Analysis 2010; 16(3): 374–384

2010
[12]

Survival functions for the frailty models based on the discrete compound Poisson process

Ata N and Özel G. Survival functions for the frailty models based on the discrete compound Poisson process. Journal of Statistical Computation and Simulation 2013; 83(11): 2105–2116

2013
[13]

Bayesian cure rate models induced by frailty in survival analysis

Souza D, Cancho VG, Rodrigues J et al. Bayesian cure rate models induced by frailty in survival analysis. Statistical Methods in Medical Research 2017; 26(5): 2011–2028

2017
[14]

A new long-term survival model with dispersion induced by discrete frailty

Cancho VG, Macera MA, Suzuki AK et al. A new long-term survival model with dispersion induced by discrete frailty. Lifetime Data Analysis 2020; 26: 221–244

2020
[15]

Survival models induced by zero-modified power series discrete frailty: Application with a melanoma data set

Molina KC, Calsavara VF, Tomazella VD et al. Survival models induced by zero-modified power series discrete frailty: Application with a melanoma data set. Statistical Methods in Medical Research 2021; 30(8): 1874–1889

2021
[16]

Bayesian survival model induced by frailty for lifetime with long-term survivors

Cancho VG, Barriga GD, Cordeiro GM et al. Bayesian survival model induced by frailty for lifetime with long-term survivors. Statistica Neerlandica 2021; 75(3): 299–323

2021
[17]

A survival model for lifetime with long-term survivors and unobserved heterogeneity

Espirito Santo APJ, Cancho VG, Louzada F et al. A survival model for lifetime with long-term survivors and unobserved heterogeneity. Brazilian Journal of Probability and Statistics 2022; 36(4): 692–703

2022
[18]

Estimating cure rates from survival data: An alternative to two- component mixture models

Tsodikov AD, Ibrahim JG and Yakovlev AY . Estimating cure rates from survival data: An alternative to two- component mixture models. Journal of the American Statistical Association 2003; 98(464): 1063–1078

2003
[19]

Stan: A probabilistic programming language

Carpenter B, Gelman A, Hoffman MD et al. Stan: A probabilistic programming language. Journal of Statistical Software 2017; 76: 1–32

2017
[20]

The analysis of rates using Poisson regression models

Frome EL. The analysis of rates using Poisson regression models. Biometrics 1983; 39(3): 665–674

1983
[21]

A test for the Poisson distribution

Brown LD and Zhao LH. A test for the Poisson distribution. Sankhy¯a: The Indian Journal of Statistics, Series A 2002; 64(3): 611–625

2002
[22]

Understanding Poisson regression

Hayat MJ and Higgins M. Understanding Poisson regression. Journal of Nursing Education 2014; 53(4): 207–215

2014
[23]

A review of multivariate distributions for count data derived from the Poisson distribution

Inouye DI, Yang E, Allen GI et al. A review of multivariate distributions for count data derived from the Poisson distribution. Wiley Interdisciplinary Reviews: Computational Statistics 2017; 9(3): e1398

2017
[24]

Analysis of overdispersed count data by Poisson model

Kurnia A, Sadik K et al. Analysis of overdispersed count data by Poisson model. European Journal of Molecular & Clinical Medicine 2021; 7(10): 1400–1409

2021
[25]

Estimation in zero-inflated generalized Poisson distribution

Kamalja KK and Wagh YS. Estimation in zero-inflated generalized Poisson distribution. Journal of Data Science 2018; 16(1): 183–206

2018
[26]

Zero-modified power series distribution and its hurdle distribution version

Conceição KS, Louzada F, Andrade M et al. Zero-modified power series distribution and its hurdle distribution version. Journal of Statistical Computation and Simulation 2017; 87(9): 1842–1862

2017
[27]

Biparametric zero-modified power series distributions: Bayesian analysis under a reference prior approach

Conceição KS, Tomazella V , Andrade MG et al. Biparametric zero-modified power series distributions: Bayesian analysis under a reference prior approach. Communications in Statistics-Theory and Methods 2017; 46(21): 10518–10536

2017
[28]

On the compound generalized Poisson distributions

Ambagaspitiya RS and Balakrishnan N. On the compound generalized Poisson distributions. ASTIN Bulletin: the Journal of the IAA 1994; 24(2): 255–263

1994
[29]

On the Lambert W function

Corless RM, Gonnet GH, Hare DE et al. On the Lambert W function. Advances in Computational Mathematics 1996; 5(1): 329–359

1996
[30]

Life table methods for heterogeneous populations: Distributions describing the heterogeneity

Hougaard P. Life table methods for heterogeneous populations: Distributions describing the heterogeneity. Biometrika 1984; 71(1): 75–83

1984
[31]

Nonparametric hazard versus nonparametric frailty distribution in modelling recurrence of breast cancer

Santos DM, Davies RB and Francis B. Nonparametric hazard versus nonparametric frailty distribution in modelling recurrence of breast cancer. Journal of Statistical Planning and Inference 1995; 47(1-2): 111–127

1995
[32]

Frailty models power variance function with cure fraction and latent risk factors negative binomial

Calsavara V , Rodrigues A, Tomazella V et al. Frailty models power variance function with cure fraction and latent risk factors negative binomial. Communications in Statistics-Theory and Methods 2017; 46: 9763–9776

2017
[33]

Bayesian data analysis

Gelman A, Carlin JB, Stern HS et al. Bayesian data analysis. 3rd ed. Boca Raton, FL: Chapman and Hall/CRC, 2013

2013
[34]

Bayesian survival analysis with INLA

Alvares D, Van Niekerk J, Krainski ET et al. Bayesian survival analysis with INLA. Statistics in Medicine 2024; 43(20): 3975–4010

2024
[35]

MCMC using Hamiltonian dynamics

Neal RM. MCMC using Hamiltonian dynamics. Handbook of Markov Chain Monte Carlo 2011; 2(11): 1–51. 12 A Bayesian survival model induced by HZMPS discrete frailty with dispersion

2011
[36]

The No-U-Turn sampler: Adaptively setting path lengths in Hamiltonian Monte Carlo

Hoffman MD and Gelman A. The No-U-Turn sampler: Adaptively setting path lengths in Hamiltonian Monte Carlo. Journal of Machine Learning Research 2014; 15(1): 1593–1623

2014
[37]

Câncer de pulmão, 2023

Instituto Nacional de Câncer (INCA). Câncer de pulmão, 2023. URL https://www.inca.gov.br/ tipos-de-cancer/cancer-de-pulmao

2023
[38]

Nonproportional hazards model with a frailty term for modeling subgroups with evidence of long-term survivors: Application to a lung cancer dataset

Gazon AB, Milani EA, Mota AL et al. Nonproportional hazards model with a frailty term for modeling subgroups with evidence of long-term survivors: Application to a lung cancer dataset. Biometrical Journal 2022; 64(1): 105–130

2022
[39]

Comprehensive molecular profiling of lung adenocarcinoma

Cancer Genome Atlas Research Network. Comprehensive molecular profiling of lung adenocarcinoma. Nature 2014; 511: 543–550

2014
[40]

Multi-omics characterization and machine learning of lung adenocarcinoma molecular subtypes to guide precise chemotherapy and immunotherapy

Zhang Y , Wang Y and Qian H. Multi-omics characterization and machine learning of lung adenocarcinoma molecular subtypes to guide precise chemotherapy and immunotherapy. Frontiers in Immunology 2024; 15: 1497300

2024
[41]

Multi-omics analyses reveal biological and clinical insights in recurrent stage I non-small cell lung cancer

Wang C, Li J, Chen J et al. Multi-omics analyses reveal biological and clinical insights in recurrent stage I non-small cell lung cancer. Nature Communications 2025; 16(1): 1–19

2025
[42]

Approximate Bayesian inference for latent Gaussian models by using integrated nested Laplace approximations

Rue H, Martino S and Chopin N. Approximate Bayesian inference for latent Gaussian models by using integrated nested Laplace approximations. Journal of the Royal Statistical Society: Series B (Statistical Methodology) 2009; 71(2): 319–392. 13

2009

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.

[1] [1]

Modelling survival data in medical research

Collett D. Modelling survival data in medical research . 3rd ed. Boca Raton, FL: Chapman and Hall/CRC, 2015

work page 2015

[2] [2]

Statistical methods for reliability data

Meeker WQ, Escobar LA and Pascual FG. Statistical methods for reliability data . 2nd ed. Hoboken, NJ: John Wiley & Sons, 2022

work page 2022

[3] [3]

Actuarial mathematics for life contingent risks

Dickson DCM, Hardy MR and Waters HR. Actuarial mathematics for life contingent risks . 3rd ed. Cambridge: Cambridge University Press, 2020

work page 2020

[4] [4]

Machine learning for survival analysis: A survey

Wang P, Li Y and Reddy CK. Machine learning for survival analysis: A survey. ACM Computing Surveys (CSUR) 2019; 51(6): 1–36

work page 2019

[5] [5]

World Health Organization. Cancer. Fact sheet, 2024. URL https://www.who.int/news-room/ fact-sheets/detail/cancer

work page 2024

[6] [6]

Cancer statistics, 2024

Siegel RL, Miller KD, Wagle NS et al. Cancer statistics, 2024. CA: A Cancer Journal for Clinicians 2024; 74(1): 12–49

work page 2024

[7] [7]

Global cancer observatory: Cancer today

Ferlay J, Ervik M, Lam F et al. Global cancer observatory: Cancer today. Lyon, France: International Agency for Research on Cancer 2024; 2018-2020

work page 2024

[8] [8]

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

Vaupel JW, Manton KG and Stallard E. The impact of heterogeneity in individual frailty on the dynamics of mortality. Demography 1979; 16(3): 439–454

work page 1979