Machine learning as a tool to study the influence of chronodisruption in preterm births

Abstract

It is well known that there are some maternal and fetal issues that directly influence preterm births. However, all the variables provoking it are not completely determined. On the other hand, chronodisruption alters maternal circadian rhythms, with negative consequences for the maturation of the fetus. Thus, the objective of this work is to add other factors related to maternal chronodisruption factors and to check if all together can improve preterm birth prevention. The methodology followed to reach this objective is based on machine learning approach. The data are composed by a cohort of 380 births labelled as preterm or term births. Variables defining each individual are related to maternal habits, night exposure to light or sleep duration during gestation. In addition, maternal variables related to the gestation were obtained as well as fetal characteristics. Preliminary statistical tests confirm that cervix dilatation, fetus estimated weight and weight at birth were significantly lower ( $$p<0.05$$ ) in preterm group than in term group as expected. A deeper study based on machine learning highlights some interesting and non obvious relations between some factors related to night exposure to light and sleeping habits. In fact, the decision tree obtained as predictive model indicates that light coming in through the window or lightness level of the bedroom during the night are key features in predicting preterm delivery.