Hierarchical models in the analysis of trends in prevalence of congenital anomalies and risks associated with first trimester medications.
MetadataShow full item record
Abstract Background Early identification of risk factors, in particular first trimester teratogenic medications, for congenital anomalies (CAs) is essential. Despite similarities between different CAs and between different medications, current surveillance methods in Europe examine each CA and each medication separately. This thesis aims to investigate whether the use of hierarchical statistical methods combining information in groups can improve CA surveillance methods. Methods EUROCAT is a European network of population-based CA registries, with EUROmediCAT comprising those registries with additional information on medication use in pregnancy. Trends in CAs from 2003-2012 in 18 EUROCAT registries (n=81,147) were analysed using Poisson regression models considering each CA separately and using hierarchical models combining related subgroups. First trimester medication exposures from 1995-2011 in 13 EUROmediCAT registries (n=15,058) were analysed. Firstly, groupings of medications and/or CAs were considered when determining the statistical significance of each medication-CA combination, using False Discovery Rate (FDR) procedures to adjust for multiple testing. Secondly, Bayesian hierarchical models were applied to directly model the group effects. The Australian classification system for prescribing medicines in pregnancy was used to independently identify “high risk” medications. The number of “high risk” medications identified by the FDR methods and Bayesian models were compared. Results For analysis of trends, grouping EUROCAT CA subgroups using hierarchical models did not provide additional information over that obtained from independent analyses of each subgroup. The double FDR method grouping medications by ATC3 level codes performed better than other FDR methods. Use of Bayesian hierarchical models did not produce enough of an improvement to justify the increased effort of implementing such models.
- Theses