EUVIMEDAtrasti EUVIMED
Open access · CC BY 4.0

Smooth Penalized Marginal Intensity Models for Recurrent Event Data

Elsa Coz iD, Hadrien Charvat iD, Delphine Maucort-Boulch iD, Mathieu Fauvernier iD

Statistics in Medicine · 2026-11 · 45 (25-27)

DOI: 10.1002/sim.70828 ↗

At a glance

Journal article in Statistics in Medicine, published online on 2026-11-15. Written by Elsa Coz and 3 co-authors. Institution: Université de Lyon, Lyon, France. Major MEDLINE topics: Models, Statistical. 45 references. The full text is freely available (open access, CC BY 4.0). Funded by: EC, Horizon 2020 Framework Programme (EU Framework Programme for Research and Innovation H2020).

SantraukaSubjectsBibliografiniai duomenysCiteRelated studies

Abstract (original language)

The statistical analysis of a recurrent event process usually involves describing the cumulative number of events for a given subject. To this end, targeting the (conditional) intensity—that is, the instantaneous probability of an event occurring per unit of time, conditioned on the history of the process—can offer a detailed modeling of the event mechanisms and provide dynamic individual predictions. However, the intensity process may be very complex to specify and can lead to incorrect inference if it fails to capture all sources of correlation between events. When a more synthetic description of the process is desired, the marginal intensity, or rate function, is very useful and does not require any assumptions about the correlation structure between subjects' events. Unlike the conditional intensity, the marginal intensity is not sufficient to fully characterize the process, so inference is usually based on estimating equations rather than likelihood methods. The semiparametric marginal intensity model, also known as the Andersen‐Gill model with robust variance, is very popular but fails to adequately describe the process's dynamics over time or to capture non‐proportional effects of covariates. To address these limitations, we propose a flexible marginal log‐intensity model based on natural cubic splines, incorporating a quadratic penalization term to control overfitting. We conducted a novel simulation study to evaluate the models, using numerical methods to compute the true marginal intensity. We observed good coverage properties with the robust estimator without penalization, but undercoverage when penalization was applied. In such cases, bootstrap methods should be preferred. The proposed method was finally illustrated using real data.

Subjects

MeSH terms (MEDLINE)

Models, StatisticalHumansComputer SimulationRecurrenceLikelihood FunctionsData Interpretation, Statistical

Author keywords

marginal intensitypenalizationrecurrent eventssmoothingsplines

Authors and institutions

  • Université de Lyon, Lyon, France.
  • Université Lyon 1, Villeurbanne, France.
  • Hospices Civils de Lyon, Pôle Santé Publique, Service de Biostatistique et Bioinformatique, Lyon, France.
  • CNRS, UMR 5558, Laboratoire de Biométrie et Biologie Évolutive, Équipe Biostatistique-Santé, Villeurbanne, France.
  • Faculty of International Liberal Arts, Juntendo University, Tokyo, Japan.
  • Division of International Health Policy Research, Institute for Cancer Control, National Cancer Center, Tokyo, Japan.

ORCID profiles: Elsa Coz, Hadrien Charvat, Delphine Maucort-Boulch, Mathieu Fauvernier

Funding

  • EC, Horizon 2020 Framework Programme (EU Framework Programme for Research and Innovation H2020)

Cite

APA

Coz, E., Charvat, H., Maucort-Boulch, D., & Fauvernier, M. (2026). Smooth Penalized Marginal Intensity Models for Recurrent Event Data. Statistics in Medicine, 45(25-27). https://doi.org/10.1002/sim.70828

Vancouver

Coz E, Charvat H, Maucort-Boulch D, Fauvernier M. Smooth Penalized Marginal Intensity Models for Recurrent Event Data. Statistics in Medicine. 2026;45(25-27). doi:10.1002/sim.70828

BibTeX
@article{Coz2026,
  author = {Coz, Elsa and Charvat, Hadrien and Maucort-Boulch, Delphine and Fauvernier, Mathieu},
  title = {Smooth Penalized Marginal Intensity Models for Recurrent Event Data},
  journal = {Statistics in Medicine},
  year = {2026},
  volume = {45},
  number = {25-27},
  doi = {10.1002/sim.70828}
}
Start for free →