Spatial data analytics for assessing patterns of food safety in England and Wales

Approximately 2.4 million cases of foodborne illness occur each year in the United Kingdom, of which 60% are thought to be contracted outside the home. Despite a transformation in consumer-food environments in recent years, whereby 43% of the population are purchasing food from a restaurant or take-away at least once a week, local authority spending on food hygiene and food control has reduced. Many local authorities struggle to undertake the required number of food outlet inspections, leaving unsafe food practices and unsafe environments unchecked, increasing the risk to the consumer. To further compound the problem, routinely collected foodborne illness data are unsuitable for both the rapid detection of outbreaks and incidence calculation. They severely underestimate the true number of foodborne illness cases, due to problems of underreporting at both the General Practitioner and patient level. This thesis employs novel analytical techniques to explore several facets of food safety, including methods for monitoring foodborne illness via Consumer Generated Data, identifying neighbourhood determinants of non-compliant food outlets, machine learning approaches to predict food outlet compliance and assessing geographical variations in model metrics and predictions to inform local targeting of food outlet inspections. Emphasis is placed on the role of geography in improving food safety, a concept which is scarce in the literature, particularly the use of small area data to characterise the neighbourhoods of high-risk outlets and specific populations who may be more situationally vulnerable. This work culminates in the creation of a predictive model able to retrieve 85% of food serving businesses which are categorised as ‘non-compliant’, providing a method to effectively prioritise inspections. The geographical variation in this model is explored in the final piece of work where the model is applied to several case study areas to identify potential policy impact.