Minimizing inequalities in urinary tract infection care access and quality in São Caetano do Sul, Brazil: application of intelligent data linkage and machine learning decision-support and risk prediction.

• Grant number: 24/10166-8
• Support Opportunities: Regular Research Grants
• Start date: February 01, 2025
• End date: January 31, 2028
• Field of knowledge: Health Sciences - Medicine - Medical Clinics
• Principal Investigator: Silvia Figueiredo Costa
• Grantee: Silvia Figueiredo Costa
• Principal researcher abroad: Alison Helen Holmes
• Institution abroad: Imperial College London, England
• Host Institution: Faculdade de Medicina (FM). Universidade de São Paulo (USP). São Paulo , SP, Brazil
• Associated researchers: Anna Sara Shafferman Levin ; Bernard Hernandez Perez ; Erika Regina Manuli ; Ester Cerdeira Sabino ; Fabio Eudes Leal ; Fátima de Lourdes dos Santos Nunes Marques ; Marcio Katsumi Oikawa ; Nina Zhu

Abstract

Urinary tract infections (UTIs) stand out as one of the most prevalent bacterial infections occurring in all age groups, with an incidence of 50-60% in adult women, both in the community and in hospital settings. UTIs are a well-known cause of acute morbidity and mortality if treated inappropriately. Primary care is where most UTI cases present and are being treated. Improperly treated UTI can become persistent and recurrent, the bacteria can invade other parts of the body and cause complications such as bloodstream infections (BSI) and sepsis. Current diagnosis and follow-up for UTI management remain suboptimal. The integration of data from the Sistema Único de Saúde (SUS) including primary and secondary care is critical to understanding the epidemiology of UTI, the emergence and transmission of drug-resistant bacteria, and the outcomes of antibiotic treatment. Data linking and integration, supported by artificial intelligence (AI)/machine learning (ML) algorithms, are tools that help identify UTI cases, detect antimicrobial resistance, guide patient stratification and antibiotic prescribing, and predict treatment response. ML models can perform three main types of tasks: classification (e.g., Gaussian Naïve Bayes, decision tree, random forest, support vector machine), to draw conclusions to which category a new observation belongs, regression, to estimate the relationships between variables, and forecasting, to make predictions about the future. AM/AI models can be trained using clinical and non-clinical data to perform similar diagnostic tasks to not only identify UTIs but also recognize patients at higher risk of serious complications such as sepsis.This proposal also has a strong focus on addressing the social determinants of health and minimizing gaps in access to and quality of care, through the incorporation of mechanisms to identify particularly vulnerable patients, including those who are socially vulnerable, with low health or technology literacy, living in long-term care institutions or with multiple long-term conditions and polypharmacy.Algorithms for identifying community-acquired UTI cases will be validated and implemented based on clinical vocabularies incorporated into the UK and European electronic health record (EHR) systems to enable implementations in Brazil; translation of the list of antibiotic codes for the identification of prescriptions related to UTI in the British National Formulary for the Brazilian Pharmacopoeia. BF algorithms for case identification and risk stratification to estimate the prevalence of community-acquired UTI in different population groups in São Caetano do Sul - São Paulo, including those with social disadvantage; predicting subsequent hospital admission for urinary ICS and cross-validation using data from Brazil's national public health registries will also be validated.This proposal consists of deterministic linkage and analysis of 1) routinely collected data from primary care and hospitals, 2) open-sourced administrative/population census data in São Caetano do Sul, Brazil to improve case identification and care for urinary tract infection (UTI).The data that will be used in this proposal includes electronic health records (EHR) from primary and secondary care which are public funded (as part of the Brazil's universal healthcare system Sistema Único de Saúde (SUS) in municipality São Caetano do Sul. The primary care EHR is sourced from the primary care units in the region. The secondary care EHR is sourced from Albert Sabin Municipal Emergency Hospital and Maria Braido Municipal Hospital, two hospitals providing secondary and tertiary care. (AU)