Geographical variation of food discourse on Twitter and its correlation with the obesity rate in Argentina.

Abstract

Obesity is a significant health problem due to its increasing prevalence and impact on health. Social networks have proven to be a valid source for studying population health-related phenomena. This study aimed to evaluate the spatial distribution of food indicators constructed from food-related content posted on the social network Twitter and to compare it with the geographical variation at a provincial level of the prevalence of obesity in the adult population of Argentina.

An ecological study was conducted using the data from the 2018 National Risk Factor Survey (to calculate weighted obesity prevalence rates) and 6023548 geo-referenced tweets collected during 2021-2022. Food indicators (rate of tweets with food-related content, frequency of mention of food and food groups, and nutrient density index (NDI, the higher the value, the better the nutritional quality of the food mentioned) were constructed from the tweets for each province. Maps were produced and the correlation between food indicators and the prevalence of total obesity, by sex and age group, at the provincial level was estimated. In addition, the Moran Autocorrelation Index was calculated to detect spatial patterns of the variables studied.

The distribution of obesity prevalence, food tweet rate and NDI showed a non-random spatial distribution (p<0.05). The frequency of mention of some foods considered 'healthy' (cucumber, grapefruit, orange, mushroom, artichoke, tuna, beef, strawberry) was inversely correlated with the prevalence of obesity at the provincial level, while the mention of some foods considered 'unhealthy' (sweetbread, semi-hard cheese, chocolate, black pudding, hamburgers, candy) was positively correlated. In some cases, these results varied by gender and age group. Finally, higher food mentions in tweets were associated with better average NDI at the provincial level (p=0.04).

Twitter speeches could serve as a proxy indicator of dietary habits and their analysis would be useful for studying unfavourable health indicators at the population level.