[GSb15] The effect of missing visits on GEE, a simulation studyAtelier, Poster ou Démonstration dans une Conférence Internationale : Miss Data 2015, June 2015, pp.xx, Rennes, France,
Mots clés: Longitudinal data, repeated correlated data, correlation, missing data, simulations, Generalized Estimating Equations.
Résumé: Clinical research is often interested in longitudinal follow-up over several visits. All scheduled visits are not carried out and it is not unusual to have a different number of visits by patient. The Generalized Estimating Equations can handle con- tinuous or discrete autocorrelated response. The method allows a different number of visits by patients. The GEE are robust to missing completely at random data, but when the last visits are fewer, the estimator may be biased. We propose a simula- tion study to investigate the impact of missing visits on the estimators of the model parameters under different missing data pattern. Different types of responses are studied with an exchangeable or autoregressive of order one structure. The number of subjects affected by the missing data and the number of visits removed vary in order to assess the impact of the missing data. Our simulations show that the estimators obtained by GEE are resistant to a certain rate of missing data. The resultsare homogeneous regardless to the imposed missing data structure.
Commentaires: 18-19 juin 2015