The use of supermarket loyalty card transaction records as a source of population dietary information

Recognising the need to intervene on the food system to reduce obesity and associated non-communicable disease, governments have begun to apply legislative sanctions for the sale and advertising of unhealthy foods. Suitable metrics are therefore critical to monitor population dietary behaviours, supporting the design and evaluation of policy interventions. National dietary surveys reveal trends, but their coverage is limited, leaving hard-to-reach low-income and minority ethnic groups largely underrepresented by population dietary statistics. Supermarket loyalty card transactions offer simple, consistent, scalable dietary metrics to complement existing approaches. In partnership with a large UK retailer, this thesis explores the utility of supermarket loyalty card transaction data as a novel tool for population dietary monitoring. This begins with presenting results from a systematic literature review, which identifies research gaps addressed in the remainder of the thesis. Using transactions for around 50,000 loyalty card customers, in Leeds, I present a unique spatial exploration of dietary purchases at the neighbourhood level. Spatial clustering observed in fruit and vegetable purchasing patterns supports the association between deprivation and poor dietary quality, as well as identifying areas which oppose this trend. This demonstrates the capacity of transactions data to contribute to hypothesis generation in ecological research and the targeting of local dietary policy strategies. Through conduct of the STRIDE study, this thesis for the first time quantifies agreement with an online food frequency questionnaire for 686 participants, adding to knowledge of the validity of transactions as a nutrient-level dietary metric. Purchases demonstrate good agreement with intake for energy adjusted metrics, making them a good marker of individual-level dietary composition and diet quality. Yet, the evidence for agreement with absolute measures is less clear, with variation by household size and loyalty. Identification and characterisation of appropriate customer samples will be key for the generalisability of absolute purchase metrics as a population-level dietary proxy.