Artificial Intelligence in Allergy and Immunology: Comparing Risk Prediction Models to Help Screen Inborn Errors of Immunity
Marina Mayumi Vendrame Takao, Luiz Sérgio Fernandes de Carvalho, Paula Garcia Pereira Silva, Maisa Moraes Pereira, Ana Carolina Viana, Marcos Tadeu Nolasco da Silva, Adriana Gut Lopes Riccetto
Abstract
BACKGROUND: Inborn errors of immunity (IEI) are underdiagnosed disorders, leading to increased morbimortality and expenses for healthcare system. OBJECTIVES: The study aimed to develop and compare risk prediction model to measure the individual chance of a confirmed diagnosis of IEI in children at risk for this disorder. METHOD: Clinical and laboratory data of 128 individuals were used to derive machine learning (ML) and logistic regression risk prediction models, to measure the individual chance of a confirmed diagnosis of IEI in children with suspected disorder, according to previous general pediatrician/clinician judgement. Their performances were compared. RESULTS: Statistically significant variables were mainly leucopenia, neutropenia, lymphopenia, and low levels of immunoglobulins A/G/M. ML models performed better. CONCLUSION: The enhanced predictive power provided by ML models could be a resource to track IEI, providing better healthcare outcomes.