Author keywords: Frailty models, Correlated survival times, Penalized likelihood, dementia, Aluminum

Lifetime Data Anal. 2003 June; 9(2): 139–53. Maximum penalized likelihood estimation in a gamma-frailty model Epidémiologie, santé publique et développement
INSERM : U330, Université Victor Segalen - Bordeaux II, 146, Rue Léo Saignat
33076 BORDEAUX,FR | ||||||||||||||||

Inference for Cox proportional hazards model (Cox, 1972) was developed under the assumption that the observations are statistically independent, at least conditionally upon covariates. However, this assumption may be violated. Thus in many epidemiological studies, failure times are clustered into groups such as families or geographical units: some unmeasured characteristics shared by the members of that cluster, such as genetic information or common environmental exposures could influence time to the studied event. In a different context, correlated data may come from recurrent events, i.e. events which occur several times within the same subject during the period of observation. In frailty models, dependence is produced by sharing an unobserved variable which is treated as a random effect, or frailty (Clayton, 1978; Hougaard, 1995; Petersen, Andersen and Gill, 1996). Semi-parametric inference for frailty models was introduced by Klein The aim of the present paper is to propose a method for semi-parametric inference in a stratified gamma frailty model: our focus is on the nonparametric estimation of the hazard function, and the approach is based on the penalized full likelihood (as opposed to the penalized partial likelihood). Parner (1998) proved the consistency of the nonparametric maximum likelihood estimator in the shared gamma-frailty model (and for the more general correlated gamma-frailty model). However, the usual nonparametric maximum likelihood estimation method leads to a discrete distribution and the hazard function cannot be derived from the estimated cumulative hazard. The article is structured as follows. In section 2, the shared frailty model is presented, and inference with Maximum Penalized Likelihood Estimation (MPnLE) is developed in section 3. In section 4 we describe simulation studies conducted to ascertain the properties of the proposed method and to compare it to a semi-parametric EM algorithm. We have also performed simulations to illustrate the estimation of the hazard function. In section 5 the method is applied to a study of the effect of aluminum on the risk of dementia in a large cohort (see Letenneur | ||||||||||||||||

We consider models in which the hazard function partly depends on an unobservable random variable thought to act multiplicatively on the hazard, so that a large value of the variable increases the hazard. We treat the case of right-censored and left-truncated data, and allow for stratum-specific baseline hazards. For the j (^{th}j = 1, …, n) individual of the _{ih}h stratum (^{th}h = 1, …, K) and the i group (^{th}i = 1, …, G), let T denote the survival times under study and let _{ihj}C be the corresponding right-censoring times. The observations are _{ihj}Y = min (_{ihj}T, _{ihj}C) and the censoring indicators _{ihj}δ = _{ihj}I_{{}_{Tihj}_{≤}_{Cihj}_{}}. The survival times may be left-truncated: only subjects with
are observed; we assume that the left-truncation times
are independent of the survival times T._{ihj}Our frailty model specifies that the hazard function conditional on the frailty is:
where λ t) is the baseline hazard function for stratum h; X = (_{ihj}X_{1} …,_{ihj},X)′ denotes the covariate vector for the _{pihj}j individual of stratum ^{th}h and group i, and β is the corresponding vector of regression parameters.Conditionally on the frailty T_{ih}_{1}, T_{ih}_{2}, …, T are assumed to be independent. It is assumed that the _{ihnih}Z’_{i}s are independently and identically distributed from a gamma distribution with mean 1 and unknown variance θ; the probability density function is thus:Large values of Z._{i}2.2 Inference in the shared gamma-frailty model The idea of the Cox partial likelihood does not carry over in a simple manner, since the integration over frailties induces a complicated form for this likelihood. Instead, we use the full likelihood. In the gamma-frailty model, a compact formula for the full likelihood can be obtained by integrating out the frailty Z from the joint likelihood (Klein _{i}et al., 1992; Nielsen et al., 1992). For treating left-truncation we must write a likelihood conditional on
. This is obtained by dividing by the likelihood of
(for i = 1, …, G; h = 1, …, K; j = 1, …, n). Moreover, we can avoid using gamma functions in the expression of the log-likelihood, thus making it easier to compute. The full log-likelihood for left-truncated data can then be written:_{ih}
with λ _{0h}(.) are the cumulative baseline hazard functions and
is the number of observed events in the i group. If a parametric baseline hazard function is specified, then maximum likelihood estimates are available by directly maximizing (2) (see Costigan and Klein, 1993).^{th} | ||||||||||||||||

We introduce a semi-parametric approach to jointly estimate the parameters _{0}(t), β and θ as maximizing:
where h = 1, …, K) are positive smoothing parameters for each stratum. In practice, the range of the integral is restricted to the period when at least one subject is still at risk. This expression represents a trade-off between faithfulness to the data, as represented by l(.), and “smoothness” of the solution, as represented by the squared norm of the second derivative. For large k;, the term
will be forced toward zero and the curves λ̂_{h} (.) will approach linear functions of time. If _{oh}k is small, then the main contribution to _{h}pl (.) will be the log-likelihood l(λ_{0} (.), β, θ) and the curve estimate λ_{0} will track the data closely, but will be more irregular._{h}3.1 Spline-based approximations When the penalized likelihood is used to estimate nonparametric regression functions, it can be shown that the estimators are cubic splines with knots at every observed data point (see Silverman (1985)). In the present context of hazard function estimation, there is no such simplification. Exact computation of these estimators is not possible and the MPnLE of λ_{0}(t) must be approximated using splines. Splines are polynomial functions which are combined linearly to give
, where the M (.) are cubic M-splines, i.e. splines of order 4 (Ramsay, 1988). By direct integration and with the same vector of coefficients _{i}η = (_{h}η_{h}_{1}, …, η)′, it is also possible to obtain the cumulative intensity function:
where _{hm}I(.) are I-splines defined as
. In our approach, although there are different strata, we use the same basis of splines for each stratum, so only the coefficients _{i}η are different for the distinct strata._{hi}Note that the estimator is the MPnLE λ̂(.), while λ̃(.) is an approximation to the estimator λ̂(.). The approximation error can be made as small as desired by increasing the number of knots. 3.2 Variance of the estimates We drew our inspiration from Gray’s work (Gray, 1992; Gray, 1994). To simplify notations we will consider only the unstratified case. We make the assumption that λ_{0}(.) belongs to the space generated by m splines, so the model is specified by a parameter vector
where ζ belongs to a subspace χ⊂IR^{K+p}^{+1}. In the general case this is only an approximation. The penalized log-likelihood can then be written as:where G and on ζ but not on the data. P(ζ) may be a weighted sum of squared norms of second derivatives of the λ_{G}_{0}(.) as in (3), but the theory applies to other choices of penalization.We can show that the MPnLE ζ̂ asymptotically follows a multivariate normal distribution with mean
and variance-covariance matrix
where
and
, and ζ It follows from (4) that a necessary condition for consistency is or, under a condition of steady increase of information with G
. If
, this condition is equivalent to o(_{p}G). We conjecture that this is also a sufficient condition.Two estimators may be proposed for var(ζ̂). -
If the distribution of the observations belongs to the class of models considered, we have
and by substituting ζ̂ in (5) we obtain:
with and .
- A variance estimator deduced from a bayesian approach was proposed by O’ Sullivan (1988) for a similar problem. In this approach ζ is considered as a random variable. Up to a constant, the penalty term is regarded as the prior log-likelihood for ζ and the penalized log-likelihood as the posterior log-likelihood. After some manipulation we obtain a multivariate gaussian approximation distribution for ζ and this makes it possible to use simply as a variance estimator.
These estimators for var(ζ̂) do not take into account the variability due to the choice of the smoothing parameters. 3.3 Confidence bands From
we deduce that var(λ̃_{0}(_{h}t)) = M′(t)[var(η̂)]_{h}M(t), so point-wise 95% confidence bands are of the formwhere t)) is the spline vector in t, and estimators
can be taken from one of the estimators of var(ζ̂). Wahba (1983), Silverman (1985), and O’Sullivan (1988) proposed such confidence bands based on Ĥ^{−1}(η̂) for conventional problems._{h}3.4 Choice of smoothing parameter To be practical, it is sometimes sufficient to choose the smoothing parameter heuristically, by plotting several curves and by choosing that which seems most realistic. We briefly present two other approaches to determine the smoothing parameters.In each stratum the smoothing parameters could be chosen by maximizing an approximate cross-validation score which was detailed by O’Sullivan (1988) for a Cox model:
where j,
is the information matrix
, and H(η) = I(η) + 2kΩ is minus the converged hessian of the penalized log-likelihood (Ĥ(η̂) = Î(η̂) + 2kΩ) and
. As in Gray (1992), if we interpret trace ([Î(η̂) + 2kΩ]^{−1} Î(η̂)) as an effective number of parameters or as the model degrees of freedom,
is equivalent to an AIC criterion (Akaike, 1974).Another approach introduces | ||||||||||||||||

We first present the computing algorithm, then two simulation studies: one to compare our results to those of Nielsen 4.1 The algorithm The estimated parameter ζ̂ was obtained by the robust Marquardt algorithm (Marquardt, 1963) which is a combination between a Newton-Raphson algorithm and a steepest descent algorithm. This algorithm is more stable than the Newton-Raphson algorithm but preserves its fast convergence property near the maximum. To be sure of having a positive function at all stages of the algorithm, we restricted all the spline coefficients η to be positive for all _{hi}i. This restriction does not have a major adverse effect on the approximation, while being very convenient numerically. We also imposed a positivity constraint for the variance parameter. When θ is very small, numerical problems may arise. When θ̂ ≤ 10^{−6}, we used an alternative formula of the log-likelihood (2) based on a third-order expansion in the expression of the log-likelihood. Thus, we replaced the two terms of the form 1/θ log(1 + θΔ) by Δ − θΔ^{2}/2 + θ^{2}Δ^{3} with Δ stands for
or
.4.2 First simulation study We conducted a simulation study to investigate the sampling properties of the MPnLE for θ. Our aim was to compare the statistical properties of the proposed method with the EM algorithm used by Nielsen et al. 1992. For the selected value of the variance parameter θ, we generated G independent pairs (t_{i}_{1}, t_{i}_{2}) of survival times with two strata of equal sample size and with only one subject (n = 1) in each stratum of each cluster. An example of this two-sample model with dependence between some of the individuals would be a study of a disease in a number (G) of families with _{ih}K = 2 corresponding to husbands (h = 1, n_{i}_{1}= 1) and wives (h = 2, n_{i}_{2} = 1). For each simulation run of M replicates (500 under H_{0} and 200 under H_{1}), the random variates were generated by the frailty model as follows:The survival times were not left-truncated (entry times equal to zero). All failure times were censored at fixed times ( We used cubic splines to approximate each hazard function. The number of knots was 8 for all simulations. For the first replicate of each simulation (i.e., for a given For the analyses of the uncensored life times, the results are summarized in Table 1. We observed a negative bias for The same general tendencies were observed when times were censored at The estimator
, which is theoretically less biased, was almost identical to Figure 1 illustrates the theoretical and estimated baseline hazard functions for a single simulation of 400 subjects and a variance parameter 4.3 Numerical illustration We also carried out a different simulation study to illustrate the quality of the estimation for a non-linear hazard function. A sample with similar features as the Paquid cohort (see section 5) was generated: there was 2,698 subjects and varying group sizes, between 13 and 232 subjects by group, (but without explanatory variables). The failure times were generated using a mixture of Weibull distributions (for stratum 1 : 0.35W(t; 20, 0.013) + 0.65W(t; 20, 0.0102); for stratum 2 : 0.35W(t; 15, 0.0107) + 0.65W(t; 15, 0.0107)). A shared frailty variable for each group of subjects was generated using a gamma distribution (with θ = 0.2). The data were left-truncated and right-censored at a fixed value, and the percentage of censoring was around 10%. The estimation of the hazard function for stratum 1 with three different numbers of knots (therefore, three different smoothing parameters) is illustrated by Figure 2. In each stratum we estimated a value for k using the automatic cross-validation method. Many authors have pointed out that the automatic cross validation method for estimating k often leads to an undersmoothed hazard function. In fact, we observed a slight oversmoothing. Figure (2) also shows that we obtained a good estimate of the theoretical hazard function. The more knots we used, the closer the approximation of MPnLE was to the true hazard function. Nevertheless, the benefit of using 15 knots compared with 8 knots was small. We also conducted a simulation run (for θ = 0.2) with this irregularly-shaped function to assess the inference for θ̂. The results obtained for 200 replicates were: mean(θ̂) = 0.232, empirical SE(θ̂) = 0.143, mean SE(θ̂) = 0.123. We again observed that the standard error estimators slightly under-estimated the empirical standard errors and θ̂ had a slight upward bias. | ||||||||||||||||

Although the hypothesis of a link between aluminum and Alzheimer’s disease has been supported by several epidemiological studies (Martyn One of the characteristics of the Paquid cohort is the grouping of the individuals in geographical areas. Thus, subjects from the same group who may share the same environmental exposure are likely to be more similar than subjects from different groups. Our objective was to establish whether there was heterogeneity of the incidence rates of dementia between areas. If this were the case, our aim was to explain it by individual-specific or group-specific variables. We also wanted to correct the variance of the regression coefficients, especially for group-specific variables. For this analysis we used data on 2,698 subjects regularly followed-up for 8 years (1, 3, 5 and 8 years after the initial visit). Among these, 253 subjects were diagnosed with dementia during the 8-year follow-up. We chose to carry out a stratified analysis on gender and estimated a value for In a first analysis using a gamma-frailty model, we did not adjust for explanatory variables but age was chosen as the basic time scale and the model was stratified on gender. The variance of the random effects was estimated by We compared three approaches: a classical proportional hazards model using a Cox partial likelihood, a proportional hazard model with a penalized likelihood estimation for the hazard function (Joly, Commenges and Letenneur, 1998), and a shared gamma-frailty model using a penalized likelihood estimation. Table 3 shows the estimated regression coefficients and the standard errors. The different models and estimation methods led to very close estimates, which is not surprising in view of the low intra-group correlation. Although we did not find any intra-area dependency, we had to validate our previous results by using a model taking into account the potential correlation of the data. Furthermore, the advantage of the frailty model was to show that since the area variation ( The results confirmed a higher risk of dementia for subjects exposed to high concentrations of aluminum (≥ 0.1 Figure 3 shows the estimated hazard functions of dementia obtained with a shared frailty model. We distinguished women from men and subjects exposed to high levels of aluminum from those exposed to low levels. Since the basic time-scale is age, the hazard function is the age-specific incidence of dementia. It steadily increased with age, and again an increased risk of dementia for subjects exposed to high levels of aluminum was observed. Of course, we cannot exclude the possibilities that these results were obtained either by chance, or because of misspecification of the model. Furthermore we did not completely deal with the interval-censoring problem, which requires the evaluation of an integral in the estimation of the likelihood. There is probably little difference in considering interval-censoring or the middle of the interval (see Commenges et al. 1998) as long as we stay in the framework of survival models. A better approach would be an illness death model as in Joly | ||||||||||||||||

- Akaike H. A new look at the statistical model identification. IEEE Transactions on automatic Control. 1974;6:716–723.
- Alfrey AC, Legendre GR, Kaehny WD. The dialysis encephalopathy syndrome: possible aluminium intoxication. The New England Journal of Medicine. 1976;294:184–188.
- Buja A, Hastie T, Tibshirani R. Linear smoothers and additive models. The Annals of Statistics. 1989;17:453–555.
- Clayton D. A model for association in bivariate life tables and its application in epidemiological studies of familial tendency in chronic disease incidence. Biometrika. 1978;65:141–151.
- Commenges D, Andersen PK. Score test of homogeneity for survival data. Lifetime Data Analysis. 1995;1:145–56.
- Commenges D, Letenneur L, Joly P, Alioum A, Dartigues JF. Modelling age-specific risk: application to dementia. Statistics in medicine. 1998;17:1973–88.
- Costigan TM, Klein JP. Multivariate survival analysis based on frailty models. Advances in reliability; New-York: North Holland: 1993. pp. 43–58.
- Cox DR. Regression models and life tables (with discussion). Journal of the Royal Statistical Society B. 1972;34:187–220.
- Cox DR, Hinkley DV. Theoretical statistics. Chapman Hall; London: 1974.
- Gill RD. Discussion of the paper by D.Clayton and J.Cuzick. Journal of the Royal Statistical Society A. 1985;148:108–109.
- Gray RJ. Flexible methods for analyzing survival data using splines, with applications to breast cancer prognosis. Journal of the American Statistical Association. 1992;87:942–951.
- Gray RJ. Spline-based tests in survival analysis. Biometrics. 1994;50:640–652.
- Hastie TJ, Tibshirani RJ. Varying-coefficient models (with discussion). Journal of the Royal Statistical Association, series B. 1993;55:757–796.
- Hougaard P. Frailty models for survival data. Lifetime data analysis. 1995:255–273.
- Jacqmin-Gadda H, Commenges D, Letenneur L, Barberger-Gateau P, Dartigues JF. Components of Drinking Water and risk of Cognitive Impairment in the elderly. American Journal of Epidemiology. 1994;139:48–57.
- Joly P, Commenges D, Letenneur L. A penalized likelihood approach for arbitrarily censored and truncated data: application to age-specific incidence of dementia. Biometrics. 1998;54:185–194.
- Joly P, Commenges D, Helmer C, Letenneur L. A penalized likelihood appproach for an illness-death model with interval-censored data: application to age-specific incidence of dementia. Biostatistics. 2002 in press.
- Klein JP, Moeschberger ML, Li YH, Wang ST. Survival analysis: State of the art. Kluwer Academic; Boston, Massachusetts: 1992. Estimating random effects in the Framingham heart study; pp. 99–120.
- Lee EW, Wei LJ, Amato DA. Cox-type regression analysis for large numbers of small groups of correlated failure time observations. In: Klein JP, God PK. , editors. Survival Analysis: State of the art. 1992. pp. 237–247.
- Letenneur L, Commenges D, Dartigues JF, Barberger-Gateau P. Incidence of dementia and Alzheimer’s disease in elderly community residents of south-western France. International Journal of Epidemiology. 1994;23:1256–1261.
- Liang KY, Self SG, Chang YC. Modelling marginal hazards in multivariate failure time data. Journal of the Royal Statistical Society B. 1993;55:441–453.
- Marquardt D. An algorithm for least-squares estimation of nonlinear parameters. SIAM Journal of Applied Mathematics. 1963:431–441.
- Martyn CN, Barker DJP, Osmond C, Harris EC, Edwardson JA, Lacey RF. Geographical relation between Alzheimer’s disease and Aluminium in drinking water. Lancet. 1989;1:59–62.
- Nielsen GG, Gill RD, Andersen PK, Sorensen TIA. A counting process approach to maximum likelihood estimation in frailty models. Scandinavian Journal of Statistics. 1992;19:25–43.
- O’Sullivan F. Fast computation of fully automated log-density and log-hazard estimators. SIAM Journal of Science and Statistical Computation. 1988;9:363–379.
- Parner E. Asymptotic theory for the correlated gamma-frailty model. The Annals of Statistics. 1998;26:183–214.
- Petersen JH, Andersen PK, Gill RD. Variance components models for survival data. Statistica Neerlandica. 1996;50:193–211.
- Ramsay JO. Monotone regression splines in action. Statistical Science. 1988;3:425–461.
- Rondeau V, Commenges D, Jacqmin-Gadda H, Dartigues JF. Relationship between aluminum concentrations in drinking water and Alzheimer’s disease: an 8-year follow-up study. American Journal of Epidemiology. 2000;152:59–66.
- Serfling RJ. Approximation theorems of mathematical statistics. Wiley; New-York: 1980.
- Silverman BW. Some aspects of the spline smoothing approach to non-parametric regression curve fitting. Journal of the Royal Statistical Society B. 1985;47:1–52.
- Therneau T, Grambsch P. Modeling survival data: extending the Cox model. Springer-Verlag; New York: 2000.
- Wahba G. Bayesian “confidence intervals” for the cross-validated smoothing spline. Journal of the Royal Statistical Society B. 1983;45:133–150.
- Wei LJ, Lin DY, Weissfeld L. Regression analysis of multivariate incomplete failure time data by modeling marginal distributions. Journal of the American Statistical Association. 1989;84:1065–1073.
| ||||||||||||||||