Risk factors associated with preterm birth: identification, prediction and evaluation in the BRISA cohort

Abstract

Problem: Preterm birth is the leading cause of death and can result in significant long-term loss of physical and psychological capacity among survivors.
Background: An estimated 15 million babies are born preterm every year. Prediction models based on machine learning methods have reported promising results.
Aims: To identify risk factors associated with preterm birth and to develop and validate a prediction model for this outcome in a Brazilian birth cohort.
Methods: Cross-sectional study of all births that occurred in Ribeirão Preto-SP and of one in three births that occurred in São Luís-MA, Brazil, in 2010. Questionnaires were applied to obtain pregnancy-related data. Multivariate adaptive regression splines were used to determine the independent variables. Preterm birth, defined as birth before 37 weeks gestational age, was the dependent variable. A random forest model was developed and its performance was evaluated by ROC analysis.
Findings: The preterm birth rates were 12.7% (RP) and 14.1% (SL). The prediction and validation accuracies of the RF-based model were 91.3% and 85.5% respectively. The model can be applied starting in the third month of gestation and is more effective in identifying preterm infants with GA<31 weeks and 6 days (AUC=0.98).
Discussion: It was possible to build a prediction model based on easily accessible low-cost data, without the need for complementary tests, providing results similar to those of other studies.
Conclusions: Previous preterm birth and prenatal care were determinants. The use of an application for individualized patient monitoring an early stage can have positive effects on the quality of life of mother and child.