Shared Frailty Model for Clustered Survival Data
Also known as: shared frailty model, random effects survival model, Frailty Modeli (Paylaşılan Kırılganlık)
The shared frailty model, introduced by Vaupel, Manton, and Stallard in 1979, extends standard survival regression by incorporating a random effect — the 'frailty' — that captures unobserved heterogeneity among subjects or clusters. When survival outcomes are measured on individuals who share a common environment (patients in the same hospital, members of the same family, animals in the same litter), a frailty term accounts for the within-cluster dependence that ordinary Cox regression ignores.
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
When to use it
Use the frailty model when you have clustered or hierarchical time-to-event data — for example multi-centre clinical trials, family studies, litter experiments, or repeated measurements on the same individual. The method is appropriate when you suspect unobserved cluster-level factors that inflate within-cluster correlation. A minimum of roughly 50 subjects is needed, with at least three events per cluster for reliable estimation of the frailty variance. The frailty distribution (Gamma or Log-Normal) must be specified; the Gamma is the default because it is computationally convenient and covers most practical cases. If the proportional hazards assumption is violated at the conditional level, the model can give misleading results and a stratified Cox model or time-dependent coefficients should be considered instead.
Strengths & limitations
- Accounts for within-cluster dependence without requiring an explicit model for every source of heterogeneity.
- Provides unbiased hazard ratio estimates in the presence of unmeasured confounders shared within clusters.
- The variance parameter θ is directly interpretable as the degree of unobserved heterogeneity.
- Naturally extends Cox regression, so all familiar tools (partial likelihood, Breslow estimator, Schoenfeld residuals) remain applicable at the conditional level.
- Requires specifying the frailty distribution; misspecification can bias estimates.
- The frailty variance θ cannot be reliably estimated if clusters are very small (fewer than three events per cluster).
- Marginal and conditional hazard ratios differ in interpretation; this distinction is often overlooked in applied reports.
- Computational cost is higher than standard Cox regression, particularly for Log-Normal frailty.
Frequently asked
What is a frailty term?
A frailty is an unobserved random variable, one per cluster, that multiplies the hazard of every individual in that cluster. It captures the collective influence of all unmeasured cluster-level risk factors. Clusters with a frailty greater than one have higher hazard (more fragile); clusters with a frailty less than one have lower hazard (more robust). The distribution of frailties across clusters is what the model estimates.
How does the frailty model differ from Cox regression?
Standard Cox regression treats all subjects as independent. The frailty model introduces a cluster-level random effect that induces correlation among subjects who share a cluster. If you apply Cox regression to clustered data, the regression coefficient estimates may be unbiased but the standard errors will be too small, leading to inflated type I error. The frailty model corrects this by explicitly modelling the between-cluster variance.
What does the frailty variance θ mean?
The variance parameter θ quantifies the degree of unobserved heterogeneity among clusters. When θ is close to zero the clusters are essentially identical in their unobserved risk, and the model reduces to ordinary Cox regression. A large θ indicates substantial between-cluster variation that must be accounted for. A confidence interval for θ that excludes zero is evidence that the frailty model is preferable to the standard Cox model.
Should I report marginal or conditional hazard ratios?
Both quantities are valid, but they answer different questions. The conditional hazard ratio describes the effect of a covariate within a cluster, holding the frailty fixed — a cluster-specific interpretation. The marginal hazard ratio averages over all clusters and is closer to a population-average statement. Shared frailty models directly estimate the conditional quantity; the marginal ratio must be derived separately. Report whichever is relevant to your research question, and state clearly which you are reporting.
Sources
- Vaupel, J.W., Manton, K.G. & Stallard, E. (1979). The Impact of Heterogeneity in Individual Frailty on the Dynamics of Mortality. Demography, 16(3), 439–454. DOI: 10.2307/2061224 ↗
- Hougaard, P. (2000). Analysis of Multivariate Survival Data. Springer. DOI: 10.1007/978-1-4612-1304-8 ↗
How to cite this page
ScholarGate. (2026, June 1). Shared Frailty Model for Clustered Survival Data. ScholarGate. https://scholargate.app/en/survival/frailty-model
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- Joint Model for Longitudinal and Survival DataSurvival↔ compare
- Kaplan-MeierSurvival↔ compare
- Recurrent Event ModelSurvival↔ compare