Saturday, 18 July 2020

Poor research doesn't help us understand the relationship between fast food availability and health

I've done quite a bit of research on the effects of alcohol outlets on neighbourhoods, focusing on the relationship between the number of outlets and an area and measures of harm, such as violence. The challenge in that sort of research is establishing causality. In particular, alcohol outlets tend to locate in poorer neighbourhoods, and those neighbourhoods tend to have more social problems. Also, people who want to drink more might want to live closer to places where they can easily access alcohol, and they might also be the kind of people who cause (or are victims of) social problems. Because of those reasons (and others), there is a lot of confounding of how large the relationship between alcohol outlets and harm actually is.

A similar problem is faced in the literature that attempts to link fast food outlets to health impacts at the neighbourhood level. So, even more than in the alcohol outlets literature, there isn't a consistent set of statistical results linking fast food outlets with obesity, for instance. However, it might be possible to deal with at least some of the issues by looking at the effects on children, because at least children aren't going to move to be closer to fast food outlets (probably!). However, that still doesn't solve all of the issues.

One example is this 2019 article by Matthew Pearce (NHS Gloucestershire Clinical Commissioning Group), Isabelle Bray, and Michael Horswell (both University of the West of England), published in the Journal of Public Health (open access). They look at accessibility to fast food outlets and obesity among children in South Gloucestershire, England, over the period from 2006 to 2013. Children were surveyed at the start and the end of the period. Essentially, Pearce et al. look at the relationship between obesity, or rather weight gain (proxied by whether each child gained 50 percentile points in a standardised weight distribution over the period between the two surveys) and access to fast food outlets (measured as a weighted score based on how many outlets are near where the child lives, and how close they are).

To me, this paper initially had a lot of promise. However, it then finds quite conflicting results that are difficult to reconcile. First:
The Spearman’s rank correlation coefficient between change in weight between Reception and Year 6 and accessibility score found no evidence of an association (r = −0.07, P = 0.768).
In essence, there was no correlation between fast food outlets and their measure of obesity. However, they then break their fast food accessibility measure into three groups, and find that:
The group of children (tertile 3) most exposed to fast-food outlets were more likely to gain significant weight (>50 percentile points) than those least exposed (odds ratio = 1.89, 95% confidence interval: 1.0–3.6, P = 0.04).
In other words, children in the top one-third (tertile) of the distribution of fast food accessibility experienced more weight gain than those in the bottom one-third. This is of course a very different finding than their initial result of no correlation.

The problem here stems from the choices the researchers have made in their analysis, and that they don't make those choices transparent to the reader of the article. Why do they use tertiles in the analysis, and not quartiles or quintiles, or indeed using the actual distribution of accessibility scores which is a continuous variable? As a journal reviewer and editorial board member, one of my pet hates is when researchers take a continuous variable and turn it into a categorical (or ordinal) variable instead. At the very least, I'd want some additional analyses to demonstrate that these results don't simply arise as a result of the researchers' choices about how they specify this variable in their analysis.

The lack of transparency is good reason to discount these results, and should be a good lesson to other researchers on what not to do if you want your research to be credible. This paper may have generated some buzz at the time of its release (including here in New Zealand), but it does nothing to help us actually understand the relationship between fast food availability and health.

No comments:

Post a Comment