gms | German Medical Science

Drug safety in pregnancy is typically evaluated by observational studies and adequate handling of the complex timing of events is crucial. Motivated by a recent cohort study of the German Embryotox Pharmacovigilance Institute in Berlin [1], my master thesis aims to evaluate the effect of fluoroquinolones exposure on the risk of adverse pregnancy outcomes like spontaneous abortion.

Statistical methodology for analyzing such dynamic event data is accompanied by diverse challenges: first, gestational age is the natural timescale of pregnancy, but women enter the cohort several weeks after conception. This phenomenon is known as left-truncation. Second live birth may be precluded by the competing events ‘elective termination’ or ‘spontaneous abortion’. Third antibiotics are temporary treatments with perhaps multiple usage during pregnancy. The methodological novelty of my thesis is to generalize landmarking and super modeling to competing risks and a left-truncated data situation. The idea is to condition on the time-varying exposure history at fixed landmark time points for prediction by adequately accounting for delayed study entry.

As a proof of concept, the thesis performs an extensive simulation study and compares the results to well-established procedures evaluating exposure as a time-dependent covariate or incorporating exposure in a multistate models. The methods are also applied to the motivating study examples.

Results indicate that landmarking can handle simultaneous left-truncation and competing risks quite well. More specifically, the estimates of the prediction probabilities perform, regarding bias and variance, only slightly worse than the data-generating multi-state models. Especially the supermodels can compensate small numbers around time zero by stacking the dataset from all landmark times together. Considering the coefficients, the true values are attenuated quite much as compared to the true values, especially for big differenced in the never before exposed, the currently exposed and the previously exposed and for greater values of the transition hazards between these states. Left-truncation there, can even lessen the attenuation by censoring observations at early landmark times.

In conclusion, the proposed concept is not restricted to pregnancy outcome studies, but is also important in other epidemiological fields. The findings are highly relevant for counseling health care providers and their patients with respect to risk and safety of medication during pregnancy.

The authors declare that they have no competing interests.

The authors declare that a positive ethics committee vote has been obtained.