Preterm delivery (PTD) is the leading cause of neonatal mortality worldwide, yet its etiology remains largely unexplained. We propose that the genetic factors controlling this trait could act in a nonuniform manner during pregnancy, with each factor having a unique "window of sensitivity." We test this hypothesis by modeling the distribution of gestational ages (GAs) observed in maternal cousins from the Swedish Medical Birth Register (MBR) (n = 35,541 pairs). The models were built using a time-to-event framework, with simulated genetic factors that increase the hazard of birth either uniformly across the pregnancy (constant effect) or only in particular windows (varying effect). By including various combinations of these factors, we obtained four models that were then optimized and compared. Best fit to the clinical data was observed when most of the factors had time-variant effects, independently of the number of loci simulated. Finally, power simulations were performed to assess the ability to discover varying-effect loci by usual methods for genome-wide association testing. We believe that the tools and concepts presented here should prove useful for the design of future studies of PTD and provide new insights into the genetic architecture determining human GA.
Keywords: Cox regression; genetic simulations; nonproportional hazards; survival analysis.
Copyright © 2017 Juodakis et al.