Obesity prevention in France

Yes, but how and why?

Madam

The report by Romon et al. describing a community-wide, multi-faceted intervention to prevent childhood overweight and obesity in France⁽Reference Romon, Lommez, Tafflet, Basdevant, Oppert, Bresson, Ducimetière, Charles and Borys¹⁾ asserts a substantial reduction over time in the prevalence of overweight and obesity in the intervention communities relative to two smaller comparison communities.

This novel approach to childhood obesity has been adopted by other European countries⁽²⁾ and will soon be implemented in South Australia⁽³⁾. It has been cited as a model for community-based solutions to obesity⁽Reference Katan⁴⁾. Reactions to this study have implications for public health worldwide.

We do not question the need for multi-level, multi-sector solutions to childhood obesity. We are, however, concerned about uncritical acceptance of the merits of this study, especially as the authors did not describe important elements of the intervention and its evaluation.

An intervention programme is a representation of a theoretical link between the known or perceived determinants of a problem, the problem itself, and predicted changes in the problem situation⁽Reference Cronbach, Ambron, Dornbusch, Hess, Hornik, Phillips, Walker and Weiner⁵⁾. The theoretical model underpinning this intervention study was not described. A theoretical model enables a systematic, evidence-based approach to planning and an explicit consideration of how change will be brought about and measured⁽Reference Koepsell, Wagner, Cheadle, Patrick, Martin, Diehr, Perrin, Kristal, Allan-Andrilla and Dey⁶⁾. In this intervention, while the aim was to reduce the prevalence of overweight and obesity in the community, it is unclear how this was to be achieved, or whether any socio-behavioural theory was employed to guide strategies for individual and community change processes.

Specification of a programme or ‘treatment’ theory is required to make explicit the processes and pathways by which a project is to achieve its objectives⁽Reference Koepsell, Wagner, Cheadle, Patrick, Martin, Diehr, Perrin, Kristal, Allan-Andrilla and Dey⁶⁾. In this study, the nature of the intervention components was not adequately described, nor the ‘dose(s)’, nor how community structures facilitated implementation, penetration and uptake of the intervention components.

Demonstrating beneficial change in potentially mediating variables such as diet and physical activity affecting overweight and obesity would have strengthened the conclusions that can be drawn from the study and thus indicate how the intervention worked. The authors note that data on diet and physical activity were collected in the Fleurbaix Laventie Ville Sante II study, but later state as a limitation that ‘the absence of measurement of mediating variables prevents us from defining which aspects of the interventions were actually effective’. This is confusing. Behavioural data on potential mediators were available from 1997 onwards and could conceivably have been included in the longitudinal analyses reported in Table 1.

Regarding evaluation, information was not given on the selection process for intervention and control communities, or of important characteristics of communities such as population size and baseline prevalence of overweight and obesity across age groups.

The authors improved the sensitivity of their statistical analyses by correctly accounting for multiple within-subject measurements. They did not however account for the intra-class correlation of individuals within communities nor of students within schools. Only the latter point was conceded as a limitation; the authors state that school-level clustering could not be dealt with as they did not have data on the schools attended by students. This seems strange, as the interventions and measurements were entirely school-based. The impact of clustering is important to consider as it acts to reduce statistical power and any basis for inference.

These and other challenges to evaluating the effectiveness of community trials are well documented. They also include analytic inefficiencies, unbalanced data, attrition from longitudinal cohorts, differential representation in successive cross-sectional samples (relevant to the social class ‘effects’ noted in this study), and ecological and individual-level bias⁽Reference Daniel and Green⁷⁾. Threats to validity, and alternative explanations for the results reported, were not adequately dealt with by the authors. It is not possible to assess whether factors other than the intervention could have yielded a changing prevalence of overweight and obesity between the intervention and control communities, or if the intervention is generalisable to different communities.

Given the potential significance of this study as a model for multi-level solutions to obesity among youth, and potential assumptions about the causal basis of its effects, we ask the authors to respond and also to publish a more detailed description to enable a transparent assessment of the intervention, basis for inference on its results and potential generalisability elsewhere.