A recent study in the British Journal of Nutrition reportedly showed that regular consumption of chocolate could reduce the risk of developing type 2 diabetes. Naturally, I wondered if that was really what the study showed.
Looking at the study, there were a few things that stood out to me. The research was done using a food frequency questionnaire, a notoriously inaccurate measure of diet. Besides the fact that this measure is often inaccurate, is the fact that we couldn’t tell if it distinguished between types of chocolate consumed. While the authors made much of the potential link between polyphenols in chocolate and reduced risk of T2 diabetes, we don’t know if the study actually looked at types of chocolate that were rich in polyphenols. By the article, we can’t tell if they made any distinction between dark chocolate, milk chocolate, white chocolate, chocolate bars, chocolate cake, chocolate ice cream, and so on. Without accounting for different types of chocolate (many of which contain negligible quantities of polyphenols) there’s no way to attribute the reduced risk of T2 diabetes to the consumption of polyphenol-rich chocolate.
Perhaps more importantly though, there’s no way we can draw any conclusions regarding causation. This wasn’t a longitudinal study so we don’t know if people who have T2 diabetes are avoiding eating chocolate (quite plausible) or if there’s some other reason why people who eat chocolate are less likely to have T2 diabetes than people who don’t.
I also wondered about the true significance of the results. For that I consulted with my math expert, Scott. His take was that the sample size wasn’t very large and that it was limited to Luxembourg. This makes it difficult to generalize the results to populations outside of Luxembourg, for example, North America, as there could be other differences between Canadians and Americans and Luxembourgians (is that the right term?) that would make it impossible to apply the findings to our population.
He also said:
Although they followed proper testing and analysis, I’d be concerned about variables that they did not include in this study, such as location and what might be in their environment or particular diet (food items not mentioned) that may distinguish this sample from say a sample in North America. I am also wary anytime the analysis includes a questionnaire or feedback rather than pure conclusions based on observed tests and results. As you well know from interviewing people at stats can, there are more than admitted “fake” stats and responses… Yes, I do see a correlation between the two, I would require further testing to be conclusive on the hypothesis.
I followed up this analysis by asking him if he thought the standard deviations were of concern. To my untrained eye, I thought that it was possible that the range for each result was large enough that there might, in actuality, be no real difference between each group. Scott said:
I would support that claim, you would want the SD to be much closer to the mean than those results. I suspect the SD would fluctuate with any other sample size tested under those conditions.
And there you have it. While it’s possible that there’s a reduced risk of having diabetes to chocolate consuming Luxembourgians, there’s more research to be done before anything definitive, especially for other populations, can be concluded.