Data donation of individual shopping data to help predict the occurrence of disease: A pilot study linking individual loyalty card and health survey data to investigate COVID-19

Introduction & BackgroundPrevious studies have found shopping data could increase the predictive accuracy of disease surveillance systems and illuminate behavioural responses in self-management symptoms disease. Yet, accessing individual sales datasets for linkage to health is challenging, recruitment appropriate sample sizes medical research has been limited.
 Objectives ApproachObjectivesCollect link investigate COVID-19. Assess feasibility scaling-up this method, use collected using loyalty card machine learning (ML) models disease.
 MethodsBased on recommendations public’s preferences donation a new protocol was designed collecting, linking analysing data. Participants were requested Tesco Clubcard website portability function share their complete an online survey. An exploratory analysis conducted linked dataset. recruited (18/01/2022 04/02/2022) with target 200.
 Relevance Digital FootprintsThe collection transactional research.
 Results197 participants shared survey contained 893,414 transactions 65,310 uniquely named items purchased from 2015 2022. Average per participant 4,653 (SD 5256) average timeframe recorded five years 6 months 30 days 836 days). A total 6,993 medication accounting 1% sales, 81% (159/197) bought medications 44 (STD 68) individual. Most (196/197) status survey, 94% (81/86) those names. reported donating do good (79%, 155/197), help NHS (77%, 152/197), be socially responsible (74%, 144/197) because secure anonymised (78%, 153/197).
 Conclusions ImplicationsUsing which enables convenient sharing transparent safeguards, public willing both into To apply robust ML analysis, particularly explore self-medication at level, must significantly scaled collect enough individuals high regular frequency, or techniques developed address sparseness key purchasing events related health. The study suggests readiness research, but investment needed large-scale AI application.