Bradford is home to one of the country’s leading hospital-at-home services: the Ambulatory Care Experience or ACE. Children suffering with health problems, that may traditionally have been admitted to a hospital ward, can instead receive treatment at home. The familiar household setting reduces anxiety for child and parent and frees up much needed space and resources in hospitals.
The service is highly successful, with 85% of children who are accepted onto the service avoiding the need for hospital treatment. Recent work has investigated whether a machine learning approach can refine the acceptance criteria to further reduce the proportion of children admitted to hospital by identifying key factors that predict successful treatment at home.
However, this work found that the ability of machine learning algorithms to predict treatment success did not outperform that of the experienced ACE admissions team. It was theorised that this lack of predictive performance may have been inevitable, given that machine learning algorithms were trained using the same measurements and variables available to the admissions team.
This project seeks to extend that work by attempting to find variables beyond those usually found in a clinical dataset, which may lend additional context to the standard physiological and demographic variables. These can then be used either as part of another machine learning algorithm, or to refine and update the ACE admission criteria. Candidate variables include local levels of air pollution, socio-economic indices of deprivation, and data derived from primary care records.