REVIEW 2 major objections 1 minor 15 references
The Ghosh-Lin and Fine-Gray models for a mix of administrative and random censoring
T0 review · 2 major / 1 minor · reviewed 2026-06-26 · grok-4.3
Pith's one-line read When administrative censoring times are known for all subjects, Ghosh-Lin and Fine-Gray models yield consistent estimates under mixed censoring by risk-set modification plus targeted inverse-probability weighting.
desk verdict The paper shows how to blend risk-set adjustment and IPCW for Ghosh-Lin and Fine-Gray under mixed censoring, provided administrative times are known for every subject. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
The hybrid adjustment that performs direct risk-set modification for known administrative censoring times while restricting inverse-probability-of-censoring weighting to the random-censoring component alone.
What would settle it
A simulation or real dataset in which the proposed estimators remain inconsistent after the administrative times are correctly inserted into the risk sets would falsify the claim.
Extended reading notes
Core claim
For the Ghosh-Lin model for recurrent events with terminal events and the Fine-Gray model for competing events, when both administrative and random censoring are present and administrative censoring times are known for all subjects, correct estimation is obtained by modifying the risk-set for administrative censoring and using IPCW adjustments only for random censoring, relying on a minimum of modeling assumptions.
Load-bearing premise
Administrative censoring times are known exactly for every subject.
Editorial extensions
If this is right
- Consistent parameter estimates become available for the Ghosh-Lin recurrent-event model under mixed censoring.
- Consistent parameter estimates become available for the Fine-Gray competing-risks model under the same mixed censoring.
- A full parametric model for the combined censoring distribution is no longer required.
- The approach applies directly to large registry data where administrative end-of-study dates are recorded for the entire cohort.
Reading between the lines
- The same separation of censoring types may apply to other marginal survival models that currently rely on full IPCW.
- In clinical-trial settings the method could be implemented by simply flagging the known administrative date for each participant and weighting only the observed random losses.
- Empirical checks could compare the hybrid estimator against naive full-IPCW estimators when the proportion of random censoring varies.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The manuscript addresses estimation in the Ghosh-Lin recurrent-events model (with terminal event) and the Fine-Gray competing-risks model when data are subject to a mixture of administrative censoring (known exactly for every subject) and random censoring. It proposes combining direct risk-set modification for the administrative component with inverse-probability-of-censoring weighting (IPCW) for the random component, claiming that this hybrid adjustment yields consistent estimators under minimal modeling assumptions.
Significance. If the proposed hybrid estimators are consistent, the work would supply a practical, low-assumption route for registry or trial data in which administrative follow-up times are recorded for the entire cohort while a subset of subjects are lost to random censoring. This would be directly useful in biostatistical practice and would extend the existing IPCW and risk-set literatures without requiring a joint model for the two censoring mechanisms.
major comments (2)
- [Abstract] Abstract and introduction: the central claim that the combined estimator is consistent rests on the assertion that administrative censoring times are known exactly for every subject, including those whose observed time is determined by random censoring. The manuscript does not specify how this information is obtained or recorded when random censoring occurs first, nor does it demonstrate that the risk-set modification remains valid under the data-generating process in which C_a,i is unobserved for subjects with C_r,i < C_a,i.
- [Methods] Estimating-equation section (presumably §3 or §4): the paper must show explicitly how the IPCW weights and the modified risk sets are combined inside the Ghosh-Lin and Fine-Gray estimating equations. Without the explicit form of the hybrid estimating function and a consistency argument (or at least a simulation study under the mixed-censoring regime), it is impossible to confirm that the two adjustments do not interfere.
minor comments (1)
- [Abstract] Notation for the two censoring times (C_a and C_r) should be introduced once and used consistently; the abstract uses “administrative censoring time” without a symbol.
Simulated Author's Rebuttal
We thank the referee for the constructive comments. We address each major comment below and will revise the manuscript accordingly to improve clarity.
read point-by-point responses
-
Referee: [Abstract] Abstract and introduction: the central claim that the combined estimator is consistent rests on the assertion that administrative censoring times are known exactly for every subject, including those whose observed time is determined by random censoring. The manuscript does not specify how this information is obtained or recorded when random censoring occurs first, nor does it demonstrate that the risk-set modification remains valid under the data-generating process in which C_a,i is unobserved for subjects with C_r,i < C_a,i.
Authors: In the registry and trial settings considered, the administrative censoring time C_a,i is the fixed study end date, which is known and recorded for every subject independently of the random censoring time. The observed data structure includes this known C_a,i for all i, with the observed time being the minimum of the event time, random censoring, and C_a,i. The risk-set modification therefore uses the known C_a,i values and remains valid under this data-generating process; we do not consider the case where C_a,i is unobserved. We will revise the abstract and introduction to explicitly describe this data structure and recording practice. revision: yes
-
Referee: [Methods] Estimating-equation section (presumably §3 or §4): the paper must show explicitly how the IPCW weights and the modified risk sets are combined inside the Ghosh-Lin and Fine-Gray estimating equations. Without the explicit form of the hybrid estimating function and a consistency argument (or at least a simulation study under the mixed-censoring regime), it is impossible to confirm that the two adjustments do not interfere.
Authors: We agree that the explicit combination of the two adjustments should be presented more clearly. The hybrid estimating equations are obtained by inserting the administratively modified risk-set indicators into the standard Ghosh-Lin and Fine-Gray estimating functions and then weighting the contributions by the IPCW terms for random censoring. We will add the precise mathematical form of these hybrid estimating functions, a brief consistency outline under independent censoring, and simulation results under the mixed-censoring regime to the methods section. revision: yes
Circularity Check
No significant circularity; estimators constructed from standard IPCW and risk-set rules
full rationale
The paper derives combined estimators for Ghosh-Lin and Fine-Gray models under mixed administrative and random censoring by direct modification of risk sets (when admin times are known for all) plus IPCW weighting for the random component. No step reduces a claimed prediction or uniqueness result to a parameter fitted from the target data itself, nor does any load-bearing premise rest on a self-citation chain that is unverified or self-referential. The central construction is an explicit combination of two existing adjustment techniques under an explicit observability assumption, remaining self-contained against external benchmarks.
Assumptions & free parameters
Cite this review
Pith. "Pith review of The Ghosh-Lin and Fine-Gray models for a mix of administrative and random censoring." pith.science (2026). https://pith.science/paper/JKOIR5KE
@misc{pith2026260619892,
author = {Pith},
title = {Pith review of: The Ghosh-Lin and Fine-Gray models for a mix of administrative and random censoring},
year = {2026},
howpublished = {\url{https://pith.science/paper/JKOIR5KE}},
note = {Machine review of arXiv:2606.19892}
}
read the original abstract
Recurrent events or competing risks regression models are often applied in the bio-medical setting and both can be considered as marginal models. In presence of right-censoring, such models need to be adjusted to give consistent estimators. When censoring is administrative, marginal regression models are particularly easy to estimate. However, when censoring is instead acting randomly, inverse probability of censoring weighting (IPCW) adjustments are typically considered to obtain parameter estimates. This technique relies on a censoring-weights adjustment via a correct censoring model, but for administrative censoring the adjustment is done correctly simply by modifying the risk-set. In practice for large central registries or some clinical trials, the administrative censoring time will be known for all subjects, but there will typically also be a proportion of subjects that are censored at random. In this work, we consider two frequently used regression approaches, the Ghosh-Lin model for recurrent events with terminal events and the Fine-Gray model for competing events. For these two settings, when both administrative and random censoring are present, we demonstrate how to obtain correct estimation by dealing with the combination of the two different types of censoring relying on a minimum of modeling assumptions.
Figures
Figures from the paper (2 more)
Reference graph
Works this paper leans on
-
[1]
R., Sargent, D
Alberts, S. R., Sargent, D. J., Nair, S., ... Grothey, A., and Goldberg, R. M. (2012). Effectofoxaliplatin, fluorouracil, andleucovorinwithorwithoutcetuximabonsurvival among patients with resected stage III colon cancer: a randomized trial.JAMA, 307(13):1383–1393
2012
-
[2]
Cook, R. J. and Lawless, J. (2007).The statistical analysis of recurrent events. Springer Science & Business Media. 15
2007
-
[3]
Cook, R. J. and Lawless, J. F. (1997). Marginal analysis of recurrent events and a terminating event.Statistics in medicine, 16(8):911–924
1997
-
[4]
Cook, R. J. and Lawless, J. F. (2018).Multistate models for the analysis of life history data. CRC Press
2018
-
[5]
Fine, J. P. and Gray, R. J. (1999). A proportional hazards model for the subdistribution of a competing risk.J. Amer. Statist. Assoc., 94:496–509
1999
-
[6]
and Lin, D
Ghosh, D. and Lin, D. (2000). Nonparametric analysis of recurrent events and death. Biometrics, 56(2):554–562
2000
-
[7]
and Lin, D
Ghosh, D. and Lin, D. (2002). Marginal regression models for recurrent and terminal events.Statistica Sinica, pages 663–688
2002
-
[8]
and Eguchi, S
Henmi, M. and Eguchi, S. (2004). A paradox concerning nuisance parameters and pro- jected estimating functions.Biometrika, 91(4):929–941
2004
Show all 15 references
-
[9]
Holst, K. K. and Scheike, T. (2026).mets: Analysis of Multivariate Event Times. R package version 1.3.9
2026
-
[10]
Thedanishnationalpatientregister
Lynge, E., Sandegaard, J.L., andRebolj, M.(2011). Thedanishnationalpatientregister. Scandinavian Journal of Public Health, 39(7 Suppl):30–33
2011
-
[11]
R., Scheike, T., and Springborg, J
Mirian, C., Ovesen, T., Jensen, L. R., Scheike, T., and Springborg, J. B. (2025). The burden of somatic comorbidities in patients surviving a traumatic brain injury.Acta neurochirurgica, 167:216
2025
-
[12]
R., Scheike, T., and Springborg, J
Mirian, C., Ovesen, T., Jensen, L. R., Scheike, T., and Springborg, J. B. (2026). Length of stay as a predictor of long-term mortality in patients surviving a traumatic brain injury: A nationwide tbi cohort study of 153 177 adults.The Journal of head trauma rehabilitation, 41:141–151
2026
-
[13]
M., Rotnitzky, A., and Zhao, L
Robins, J. M., Rotnitzky, A., and Zhao, L. P. (1994). Estimation of regression coefficients when some regressors are not always observed.Journal of the American Statistical Association, 89(427):846–866
1994
-
[14]
Scheike, T. H. (2025). Two-stage recurrent events random effects models.Lifetime data analysis, 32:6
2025
-
[15]
H., Martinussen, T., and Ozenne, B
Scheike, T. H., Martinussen, T., and Ozenne, B. (2023). Efficient estimation in the fine and gray model.Journal of the American Statistical Association, 118(544):2482–2490. 16 Appendix 6.1 Appendix A: Asymptotic Properties The basic structure of the proof follows closely the a...
2023
Reviewed June 26, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.