Background: Despite the growing interest in the relation between adiposity in children and different lifestyle clusters, few studies used a longitudinal design to examine a large range of behaviors in various contexts, in particular eating-and sleep-related routines, and few studies have examined these factors in young children. The objectives of this study were to identify clusters of boys and girls based on diet, sleep and activity-related behaviors and their family environment at 2 and 5 years of age, and to assess whether the clusters identified varied across maternal education levels and were associated with body fat at age 5. Methods: At 2 and 5 years, respectively, 1436 and 1195 parents from the EDEN mother-child cohort completed a questionnaire including behavioral data. A latent class analysis aimed to uncover gender-specific behavioral clusters. Body fat percentage was estimated by anthropometric and bioelectrical impedance measurements. Association between cluster membership and body fat was assessed with mutivariable linear regression models. Results: At 2 years, two clusters emerged that were essentially characterized by opposite eating habits. At 5 years, TV exposure was the most distinguishing feature, but the numbers and types of clusters differed by gender. An association between cluster membership and body fat was found only in girls at 5 years of age, with girls in the cluster defined by very high TV exposure and unfavorable mealtime habits (despite high outdoor playing and walking time) having the highest body fat. Girls whose mother had low educational attainment were more likely to be in this high-risk cluster. Girls who were on a cluster evolution path corresponding to the highest TV viewing time and the least favorable mealtime habits from 2 to 5 years of age had higher body fat at 5 years. Conclusions: Efforts to decrease TV time and improve mealtime routines may hold promise for preventing overweight in young children, especially girls growing up in disadvantaged families. These preventive efforts should start as early in life as possible, ideally before the age of two, and should be sustained over the preschool years.

Variable selection in cluster analysis is important yet challenging. It can be achieved by regularization methods, which realize a trade-off between the clustering accuracy and the number of selected variables by using a lasso-type penalty. However, the calibration of the penalty term can suffer from criticisms. Model selection methods are an efficient alternative, yet they require a difficult optimization of an information criterion which involves combinatorial problems. First, most of these optimization algorithms are based on a suboptimal procedure (e.g. stepwise method). Second, the algorithms are often greedy because they need multiple calls of EM algorithms. Here we propose to use a new information criterion based on the integrated complete-data likelihood. It does not require the maximum likelihood estimate and its maximization appears to be simple and computationally efficient. The original contribution of our approach is to perform the model selection without requiring any parameter estimation. Then, parameter inference is needed only for the unique selected model. This approach is used for the variable selection of a Gaussian mixture model with conditional independence assumption. The numerical experiments on simulated and benchmark datasets show that the proposed method often outperforms two classical approaches for variable selection. The proposed approach is implemented in the R package VarSelLCM available on CRAN.