Early behavioral adherence predicts short and long-term weight loss in the POUNDS LOST study

It should be noted that the study did not emphasize changes in physical activity and it did not involve the use of medications or other medical approaches. Williamson and colleagues (2010) hypothesized that if the study had included these other tools or strategies, they would have identified dimensions of adherence related to these other behavioral features of weight management programs.

As reported by Williamson et al. (2010), computer tracking data from the first 6 months (but not from Months 7 to 24) were analyzed for two reasons: (1) the functionality of the computer tracking system was changed mid-way through the study (after all participants had at least 6 months in the study) and (2) utilization of the computer tracking system diminished over the course of the 24 month study, such that information from the computer tracking system during Months 7–24, is likely from a non-representative sample.

Comprehensive descriptions of the methods used to measure these outcome variables are provided in the paper authored by Sacks et al. (2009). The experimental design of the study called for measurement of percent body fat in a randomly selected sample (50% of the total sample, stratified by site and gender) due to budget constraints and guidance by tests of statistical power. Change scores were selected as the variables to be predicted by early adherence since the principal question concerned the prediction of relative improvement from baseline (by early adherence), not actual levels of outcomes at one or more assessment points in the study.

Comparison of the data shown in Table 2 to the data reported by Sacks et al. (2009 in Table 2 and Figs. 1, 2) shows that the mean changes in measures of adiposity and health for the subsample in this ancillary study may have been slightly greater than those reported for the full cohort (from either completer or intent-to-treat analyses). We did not test for statistical differences between the subsample and the full cohort, but it is logical based upon the results reported in this paper that a subsample that was more adherent (i.e., entered at least six computer tracking records) in the first 6 months of the intervention would have slightly greater improvement in outcomes related to adiposity and health.

As noted in the statistical methods section and as illustrated in Tables 2 and 3, data for body weight, waist circumference, and all health outcome variables were only collected at Months 6 and 24. Furthermore, percent body fat was only measured on 50% of the cohort. For these reasons, the canonical correlations were computed only at Months 6 and 24 and did not include percent body fat as a measure of changes in adiposity. It is noteworthy that in this smaller sample, changes in percent body fat were correlated with behavioral adherence at Month 6 (r = −0.45) and at Month 24 (r = −0.41). Also, the investigators calculated separate canonical correlations for participants in each of the four treatment arms of the POUNDS LOST study and found very similar results, i.e., two canonical correlations for both 6 and 24 month measurement periods, with behavioral adherence predicting adiposity outcomes, but not cardiovascular disease outcomes (R values at 6 months ranged from 0.51 to 0.58, P values < 0.001; R values at 24 months ranged from 0.41 to 0.43, P values < 0.07). With one exception, the canonical correlations for dietary adherence were not significant (R values ranged from 0.16 to 0.34, P values > 0.20). The only exception was in the Moderate Fat/High Protein treatment arm at 6 months; dietary adherence was associated with changes in insulin (R = 0.40, P < 0.01). It is important to note that statistical power for these within-treatment arm correlations was substantially lower than the primary findings reported for the entire study cohort. Nevertheless, the overall pattern of findings, especially pertaining to the relationship between early behavioral adherence and changes in adiposity is remarkably similar across all levels of analysis.