An exploration of potential risk factors for gastroschisis using decision tree learning

Julie M Petersen; Jaimie L Gradus; Martha M Werler; Samantha E Parker

doi:10.1016/j.annepidem.2024.12.004

An exploration of potential risk factors for gastroschisis using decision tree learning

Ann Epidemiol. 2024 Dec 8:101:19-26. doi: 10.1016/j.annepidem.2024.12.004. Online ahead of print.

Authors

Julie M Petersen¹, Jaimie L Gradus², Martha M Werler³, Samantha E Parker⁴

Affiliations

¹ Department of Epidemiology, Boston University School of Public Health, 715 Albany Street, Boston, MA 02118, USA; Department of Epidemiology, College of Public Health, University of Nebraska Medical Center, 984395 Nebraska Medical Center, Omaha, NE 68198-4395, USA. Electronic address: julpetersen@unmc.edu.
² Department of Epidemiology, Boston University School of Public Health, 715 Albany Street, Boston, MA 02118, USA. Electronic address: jgradus@bu.edu.
³ Department of Epidemiology, Boston University School of Public Health, 715 Albany Street, Boston, MA 02118, USA. Electronic address: werler@bu.edu.
⁴ Department of Epidemiology, Boston University School of Public Health, 715 Albany Street, Boston, MA 02118, USA. Electronic address: separker@bu.edu.

PMID: 39657869
DOI: 10.1016/j.annepidem.2024.12.004

Abstract

Purpose: Despite a wealth of research, the etiology of the abdominal wall defect gastroschisis remains largely unknown. The strongest known risk factor is young maternal age. Our objective was to conduct a hypothesis-generating analysis regarding gastroschisis etiology using random forests.

Methods: Data were from the Slone Birth Defects Study (case-control, United States and Canada, 1998-2015). Cases were gastroschisis-affected pregnancies (n = 273); controls were live-born infants, frequency-matched by center (n = 2591). Potential risk factor data were ascertained via standardized interviews. We calculated adjusted odds ratios (aOR) and 95 % confidence intervals (CIs) using targeted maximum likelihood estimation.

Results: The strongest associations were observed with young maternal age (aOR 3.4, 95 % CI 2.9, 4.0) and prepregnancy body-mass-index < 30 kg/m² (aOR 3.3, 95 % CI 2.4, 4.5). More moderate increased odds were observed for parents not in a relationship, non-Black maternal race, young paternal age, marijuana use, cigarette smoking, alcohol intake, lower parity, oral contraceptive use, nonsteroidal anti-inflammatory drug use, daily fast food/processed foods intake, lower poly- or monounsaturated fat, higher total fat, and lower parental education.

Conclusions: Our research provides support for established risk factors and suggested novel factors (e.g., certain aspects of diet), which warrant further investigation.

Keywords: Congenital abnormalities; Decision trees; Gastroschisis; Machine learning; Random forest; Risk factors.