The food environment is considered a contributing factor to unhealthy diets. With out-of-home food consumption increasing globally, there is a growing interest in understanding the out-of-home food environment and its contribution to individual dietary behaviours. As well as the type and location of out-of-home food outlets (e.g., restaurants, takeaways, and cafés), the food choices available within these outlets also constitute an important dimension of the out-of-home food environment. However, our understanding of this dimension remains limited, in part due to a lack of data. This lack of data extends to both the healthiness of individual items sold (i.e., whether an item is "healthy"), and the overall healthiness of menus as a whole (i.e., whether an outlet is "healthy"). The aim of my thesis is to bridge this gap and advance the understanding of the out-of-home food environment by harnessing publicly available big data.

In this thesis, I used web scraping to harvest publicly available big data and established two databases: MenuTracker and TakeawayTracker. Using these two databases, I quantified the out-of-home consumer nutrition environment —a dimension of the food environment that encapsulates what consumers encounter within food outlets—at scale and in real time.

The first database, MenuTracker, captures nutrient composition data for foods served by large chain restaurants every quarter. It enables researchers, policymakers, and other stakeholders to understand the nutritional landscape of food served by large chain restaurants in the UK. In Chapter 2, I described the detailed methodology for developing this database, which involved automated data collection that saved 500 hours of manual work for each data collection wave. Using this database, in Chapter 3, I compared the nutritional landscape of restaurant foods in the United Kingdom (UK) vs the United States (US). The results suggest that UK menu items were healthier, as judged by their energy, fat, saturated fat, and sugar content, but 95% of all items were high in at least one of these measures in both countries. In Chapter 4, I further examined longitudinal trends in nutritional composition of out-of-home foods in the UK using MenuTracker from 2018 to 2020. While the sugar content of restaurant foods declined over time, levels of other nutrients remained unchanged. The unique finding for sugar may be attributed to the Soft Drinks Industry Levy (SDIL), which was implemented in the UK in 2018.

The second database, TakeawayTracker, captures menus of outlets on the UK's leading online delivery platform JustEat. In Chapter 5, I used this database and developed a novel deep learning model to characterise the menu healthiness of all out-of-home food outlets in the UK. The findings described in this chapter reveal that in more deprived areas in the UK, there are more out-of-home food outlets, and these tend to be less healthy. By linking these data with information on individuals, in Chapter 6, I investigated associations between exposure to the out-of-home food environment and both dietary behaviour and quality. Findings suggest that the availability of out-of-home food outlets, regardless of their menu healthiness, is the only aspect of the out-of-home food environment studied that is associated with dietary behaviour.

There are possible policy implications from the findings of my thesis. First, there are opportunities for policy regulation in the out-of-home food sector, as my findings show that out-of-home foods are largely unhealthy. Second, mandatory policies, especially fiscal policies, may prove to be more effective than voluntary approaches. Third, policy actions are needed to address the unhealthy out-of-home food environment, especially in more deprived areas. Lastly, policies designed to change physical access to outlets ("venue"), rather than just food offerings within outlets ("menu"), could be more effective in promoting healthy diets.

In summary, my thesis highlights the need for effective interventions in the out-of-home food environment, while showcasing the untapped potential of the combination of big data and epidemiological principles in dietary public health research.