CCBY Open access
Research

Associations between exposure to takeaway food outlets, takeaway food consumption, and body weight in Cambridgeshire, UK: population based, cross sectional study

BMJ 2014; 348 doi: https://doi.org/10.1136/bmj.g1464 (Published 13 March 2014) Cite this as: BMJ 2014;348:g1464

Re: Associations between exposure to takeaway food outlets, takeaway food consumption, and body weight in Cambridgeshire, UK: population based, cross sectional study

After reading the interesting article by Burgoine et al. I was at first irritated by the lack of a table to compare the characteristics (as shown in Table 1) of participants grouped according to quarters of take-away environment. Further, I missed a simple presentation of outcome variables (mean take-away consumption, mean BMI, percentage overweight and obese) grouped according to these same quarters. Usually one would expect such tables in order to assess the comparability of the groups with respect to possible confounders and for a direct, unadjusted comparison of outcomes, respectively.

Then I discovered this information in Web table 3 of the online appendix. Here, we see systematic differences between quarters with respect to education, smoking and car ownership. I think the authors should have presented these tables and drawn attention to these differences in the main printed article, even if the multiple linear regression models adjusted for the covariables concerned.

What surprised me even more in Web table 3 was the fact that mean take-away consumption was slightly inversely correlated with combined take-away availability, varying between 36.3 g/day in Q1 and 34.2 g/day in Q4. This contrasts completely with the results of the multivariate analysis (Fig. 1) in which a significant positive correlation between take-away availability and consumption was obtained. Moreover, In Web table 3 mean BMI is almost constant in all quarters of take-away availability, contrasting with the significant positive correlation between take-away availability and BMI derived from the multiple linear model (Fig. 2). While I accept that the multivariate analysis adjusting for potential confounders is the analysis of choice for such an observational study, the complete lack of agreement with the simple univariate analysis is worrying and should be presented and discussed.

A hint on the possible explanation for these inconsistencies is given under ‘sensitivity analyses’. ‘In models that omitted supermarket exposure as a covariate, the associations between combined take-away food outlet exposure, consumption of take-away food and body mass index were attenuated towards the null…’. These sensitivity results are given in Web figures 5 and 6. The expression ‘attenuated towards the null’ is an understatement: no association remains at all, in agreement with the simple univariate comparison. Thus, supermarket exposure and take-away exposure seem to interact with each other in their relationship to take-away consumption and BMI. Possible, the results obtained reflect not just take-away outlets but the type of environment as a whole, for instance, rural versus urban. Information on supermarket exposure and urban/rural residence (possibly also urban/rural work-place) per quarter should have been included in Web table 3.
In summary, the incomplete presentation of results in the printed article tends to obscure inconsistencies which shed doubt on the conclusions drawn.

Competing interests: No competing interests

11 June 2014
Jeremy Franklin
statistician
University of Cologne
Kerpener Str. 62, 50924 Koeln, Germany