A machine learning approach to investigate potential risk factors for gastroschisis in California

Abstract

Background To generate new leads about risk factors for gastroschisis, a birth defect that has been increasing in prevalence over time, we performed an untargeted data mining statistical approach. Methods Using data exclusively from the California Center of the National Birth Defects Prevention Study, we compared 286 cases of gastroschisis and 1,263 non‐malformed, live‐born controls. All infants had delivery dates between October 1997 and December 2011 and were stratified by maternal age at birth (<20 and ≥ 20 years). Cases and controls were compared by maternal responses to 183 questions (219 variables) using random forest, a data mining procedure. Variables deemed important by random forest were included in logistic regression models to estimate odds ratios and 95% confidence intervals. Results Among women younger than 20, of variables deemed important, there were higher odds observed for higher consumption of chocolate, low intake of iron, acetaminophen use, and urinary tract infections during the beginning of pregnancy. After adjustment, the higher odds remained for low iron intake and a urinary tract infection in the first month of pregnancy. Among women aged 20 or older, of variables deemed important, higher odds were observed for US‐born women of Hispanic ethnicity and for parental substance abuse. There were lower odds observed for obese women, women who ate any cereal the month before pregnancy, and those with higher parity. Conclusions We did not discover many previously unreported associations, despite our novel approach to generate new hypotheses. However, our results do add evidence to some previously proposed risk factors.

Artificial Intelligence in Pregnancy: A Scoping Review

“…35.7% (n = 5) of studies reported using imbalanced data sets [40], [44], [79]- [81]. K-means algorithm (unsupervised learning, clustering technique) was reported to be used in [79] for reducing the population of the majority class, and over-under sampling techniques have been used in [40].…”

Three machine learning algorithms and their utility in exploring risk factors associated with primary cesarean section in low‐risk women: A methods paper

“…Only one study reported finding no additional associations with the outcome of interest beyond what was already known (K. A. Weber et al, 2019). In comparison to traditional regression models, the CART and random forest algorithms offer a number of potential advantages.…”

“…CART is only as good as the variables provided to it; asking CART to choose important variables can lead to humorous results (K. A. Weber et al, 2019). While CART might optimize each split, the overall model may not be stable.…”

Genetic Susceptibility to Drug Teratogenicity: A Systematic Literature Review

“…Investigation of complex teratogenic processes, likely involving non-linear and non-additive interactions of multiple genetic and environmental factors, are further supported by the development of novel methods within bioinformatics (Weber et al, 2019 ). Analyzing and deciphering such higher-order interactions prompts the application of machine learning models, which can detect patterns and predict drug teratogenicity and child outcomes.…”