SEARCH

SEARCH BY CITATION

TO THE EDITOR:

The recently published paper by French et al. has drawn some attention in the field (1). The authors claimed “The purpose of the present study was to evaluate an intervention to prevent weight gain among households (HHs) in the community. No significant intervention effects were observed for change in HH BMI-z score. Intervention HHs significantly reduced television viewing, snacks/sweets intake, and dollars per person spent eating out, and increased (adults only) physical activity and self-weighing frequency compared with control HHs” (1). However, we believe a disconnect exists between the design, analysis, and report of this study.

In a cluster randomized trial, whether an intervention is evaluated at the cluster level or the participant level has implications for the appropriate analysis of the outcome data. It is therefore important that the level at which outcomes is measured be explicit in the trial report. The authors stated “The primary outcome was change in HH mean BMI z-score”, and “The unit of randomization, intervention, and evaluation was the HH” (1). Despite that the claimed outcome was at the HH level, not at the individual level, the authors indicated “Data are presented individually to the calculation of the maximum likelihood (proc MIXED, SAS) but are modeled as correlated within HH” (1). When the individual level data are included, the coefficient associated with the intervention indicator will be used to evaluate the intervention effect, but this approach evaluated the individual level rather than the HH level. Therefore, the interpretation of the data presented in Table 2 should be at the individual level, rather than at the HH level. However, in the Results section, conclusions were drawn at the HH level for many study outcomes including BMI z-score. As a result, we believe many of the conclusions regarding the effect of the intervention at the HH level may not be appropriate. Alternately, if the HH BMI z-score is the primary outcome, an aggregated measure (e.g., an average within a HH) might have been used and a simple regression analysis rather than a linear mixed model would be more appropriate in that case.

Secondly, although the flow of clusters was reported in Table 1, we believe some critical pieces are missing and should have been reported in a cluster randomized trial—the flow of individuals through the trial from assignment to analysis is missing (item 13 of the CONSORT statement) (2). Thus, we do not know how many individuals are in control and intervention group (item 16) (2).

Furthermore, in Table 2, BMI z-score is 0.55 and 0.52 for baseline and follow-up intervention, respectively, but when it was broken down to adolescent and adult, it is unclear how can they become 0.71 and 0.71 in the baseline intervention, and 0.69 and 0.68 in the follow-up intervention. The overall BMI z-score should be a number between the corresponding results in adolescent and adult (a mathematical argument). However, the mentioned overall BMI z-scores are much smaller than the corresponding BMI z-scores in adolescent and adult.

References

  1. Top of page
  • 1
    French SA, Gerlach AF, Mitchell NR, Hannan PJ, Welsh EM. Household obesity prevention: Take action—A group-randomized trial. Obesity 2011; 10: 2082-2088.
  • 2
    Campbell MK, Elbourne DR, Altman DG. CONSORT statement: Extension to cluster randomized trials. BMJ 2004; 328: 702-704.