Newborn Cystic Fibrosis Diagnosis Made Accurate and Efficient with Machine Learning to Reduce False Positives in IRT-Trypsinogen Immunoreactive Screening Program

Abstract

This paper presents a methodology for developing a predictive model using random forests to identify true positive cases of cystic fibrosis in neonatal screening, aiming to improve early detection and care for patients. The current heel prick test used in Brazilian neonatal screening has a high incidence of false positives, leading to unnecessary anxiety and medical interventions for patients and their families. Our methodology, developed using synthetic data and varied model parameters, showed promising results in improving the sensitivity of the model for identifying true positive cases. Our methodology prioritizes sensitivity in the model and uses general indices from the neonatal screening laboratory APAE in São Luís do Maranhão. In conclusion, our approach offers a promising avenue for developing a predictive model for cystic fibrosis in neonatal screening and highlights the potential benefits of improving early detection and care for patients.