Response to "Clustering effects in group-based behavioral weight loss trials".

We appreciate the thoughtful comments of Li and colleagues in response to our article entitled “Do weight loss and adherence cluster within behavioral treatment groups?”[1] and agree with the findings from their simulation model about the effects of clustering on statistical power. However, it is important to consider the clustering effect not only theoretically, but also empirically and to determine the magnitude of the clustering effect within specific types of groups, such as weight loss trials. We know of only two studies that have examined intra-class correlations coefficients (ICCs) in weight loss groups[1, 2] and both found little evidence of clustering. We hope that our article will encourage others to examine the extent to which there is a clustering of responses within their weight loss trials and to consider whether specific features of the intervention influence the ICCs. Although we did not analyze clustering in the control condition, we want to clarify that groups were held for both the intervention and the control condition in Look AHEAD, but with differing frequency and content. Moreover, intensive interventions with structured protocols, as used in Look AHEAD, are recommended for ensuring fidelity to treatment and optimizing weight loss outcomes.