#epidemiology: Ecological analysis of fast food tweets in relation to Behavioral Risk Factor Surveillance System data

Recent public health research has described the use of Internet search engines and social media for tracking infectious disease.  Influenza, in particular, has attracted the attention of several researchers, who have developed statistical models of prevalence based upon the rate of geo-located tweets mentioning key words likely to be related to infection—“flu”, “H1N1”, “Tamiflu”, etc.  Such research has yielded encouraging results with regard to the utility of Twitter data for tracking health outcomes, but has not yet included analysis of outcomes in relation to exposures.  Furthermore, there are very few examples of researchers using Twitter to track chronic diseases or their related health behaviors. 

To explore these possibilities, we calculated the rate of geo-located tweets mentioning well-known fast food chains, for each state in the U.S., as a measure of the acceptability and popularity of fast food.  Using scatter plots and correlation coefficients, we then compared the rates of fast food tweets to the average number of fruit and vegetable servings consumed per day by respondents to the Behavioral Risk Factor Surveillance System (BRFSS).  We also conducted a parallel intra-state analysis using 41 regions of Kentucky developed specifically for aggregating BRFSS data.  Lastly, we repeated these analyses, but substituted rates of overweight and obesity for average fruit and vegetable intake.

The results of this study demonstrate that residents of U.S. states and regions of Kentucky with higher rates of fast food tweets consumed fewer fruit and vegetable servings per day on average (r=-0.52, p<0.0001; rho=-0.59, p=0.0001; respectively).   We observed weaker but still statistically significant trends in the analysis of overweight and obesity.  These results suggest analysis of Twitter data might be useful for tracking health behaviors related to chronic disease.