Choosing marginal or random-effects models for longitudinal binary responses: application to self-reported disability among older persons

BackgroundLongitudinal studies with binary repeated outcomes are now widespread in epidemiology. The statistical analysis of these studies presents difficulties and standard methods are inadequate.MethodsWe consider strategies for modelling binary repeated responses and focus on two specific issues: the choice between marginal and random-effects models, and the choice of the time point origin. These issues are addressed using the example of self-reported disability in older women assessed annually for 6 years. The indicator of disability "needing help to go outdoors or home-confined" is used.ResultsIn view of the observed associations between the responses for consecutive years, the baseline response was considered as a covariate. We compared the marginal and random-effects models first when only the influence of time and age is analysed and second when individual risk factors are studied in an aetiological perspective. There were substantial differences between the parameter estimates. They were due to differences between specific concepts related to the two models and the large between-individual heterogeneity revealed by the analysis.ConclusionsA random-effects model appears to be most suitable for the analysis of self-reported disability in older women.

[1]  N. Breslow,et al.  Approximate inference in generalized linear mixed models , 1993 .

[2]  G Molenberghs,et al.  Analysis of incomplete public health data. , 1999, Revue d'epidemiologie et de sante publique.

[3]  J. Robins,et al.  Analysis of semiparametric regression models for repeated outcomes in the presence of missing data , 1995 .

[4]  J. Kalbfleisch,et al.  A Comparison of Cluster-Specific and Population-Averaged Approaches for Analyzing Correlated Binary Data , 1991 .

[5]  L. Beckett,et al.  Analysis of change in self-reported physical function among older persons in four population studies. , 1996, American journal of epidemiology.

[6]  Noreen Goldman,et al.  An assessment of estimation procedures for multilevel models with binary responses , 1995 .

[7]  G. Breart,et al.  Fall-related factors and risk of hip fracture: the EPIDOS prospective study , 1996, The Lancet.

[8]  Nicole A. Lazar,et al.  Statistical Analysis With Missing Data , 2003, Technometrics.

[9]  P S Albert,et al.  Longitudinal data analysis (repeated measures) in clinical trials. , 1999, Statistics in medicine.

[10]  M. Lindstrom,et al.  A survey of methods for analyzing clustered binary response data , 1996 .

[11]  P. Diggle.  Analysis of Longitudinal Data , 1995 .

[12]  P. Albert,et al.  Models for longitudinal data: a generalized estimating equation approach. , 1988, Biometrics.

[13]  D. Bates,et al.  Approximations to the Log-Likelihood Function in the Nonlinear Mixed-Effects Model , 1995 .

[14]  R. Wolfinger,et al.  Generalized linear mixed models a pseudo-likelihood approach , 1993 .

[15]  R C Littell,et al.  Mixed Models: Modelling Covariance Structure in the Analysis of Repeated Measures Data , 2005 .

[16]  Scott L. Zeger,et al.  Generalized linear models with random e ects: a Gibbs sampling approach , 1991 .

[17]  D. Spiegelhalter,et al.  Factors associated with functional decline and improvement in a very elderly community-dwelling population. , 1999, American journal of epidemiology.

[18]  J K Lindsey,et al.  On the appropriateness of marginal models for repeated measurements in clinical trials. , 1998, Statistics in medicine.

[19]  S. Zeger,et al.  Longitudinal data analysis using generalized linear models , 1986 .

[20]  N. Breslow,et al.  Bias correction in generalised linear mixed models with a single component of dispersion , 1995 .