Minimising inequalities in care access and quality for patients with UTI in Brazil: application of intelligent data linkage and machine learning

Brazil reported the highest UTI incidence and UTI-associated mortality and morbidity in the world. Treating UTI has become increasingly difficult to due to antimicrobial resistance (AMR) in the causative pathogens, predominantly Gram-negative bacteria. Inappropriately treated UTI leads to recurrent cases and bacterial invasion to other body parts and drives AMR emergence and spread. Diagnosing, prescribing, and follow-up of UTI remain sub-optimal, with half of the UTI antibiotic prescriptions in primary care being inappropriate. Limited data is available to evaluate of UTI management as it requires tracking patients along care pathway to identify re-prescribing, (re)admissions to primary care and hospitals, and deaths and disabilities due to UTI complications and AMR. Artificial intelligent (AI)/machine-learning (ML) supported by data integration can improve care for UTI by identifying cases, detecting AMR, and guiding patient stratification and antibiotic prescribing.

In this proposal, University of São Paulo (USP) and Imperial College London (ICL) team will co-develop data linkage and case identification algorithms to better monitor UTI across primary and secondary care in Brazil. The linked data enables piloting and validation of three ML-based algorithms to perform risk stratification, guide antibiotic prescribing, and predict adverse events in hospitals. Routine electronic health records (EHR) and laboratory data from primary care units and hospitals in São Caetano do Sul, covering a population of 165,655 residents, will be deterministically linked. Using the linked data, we will develop tiered-case identification algorithms to identify cases and risk factors of community-acquired UTI, assess antibiotic prescribing appropriateness, and evaluate patient outcomes including urine-sourced BSI and other UTI complications. Three ML-based tools will be piloted, including a support Vector Machine (SVM) classifier to estimate the likelihood of UTI and...