Abstracts

Rationale: There are multiple ongoing anti-epileptic drug (AED) pregnancy registries. Many of these now capture substantial numbers of exposures (n>600) to several individual AEDs. While this increases power to detect increases in rare, specific defect types, the study multiplicity also increases the chance that signals for specific defects will occur by chance. A signal is defined as observing three or more of the same defect type in a cohort >=600 exposed individuals (as the probability of observing this according to a binomial distribution is less than 5%). A better understanding is needed of the likelihood of signal generation for a range of defect types. Methods: The example of the signal of increased risk of isolated oral clefts following in utero lamotrigine exposure was considered. A poisson distribution was used to calculate the probability of observing a signal for isolated oral clefts (defined as three or more clefts within a registry) given the registry denominators and the background risk for oral clefts appropriate to that registry. This method assumed independence of the different registries. The denominators for first trimester lamotrigine monotherapy exposure were available from six AED pregnancy registries (International Lamotrigine Registry, North American AED Registry, EURAP, UK Epilepsy and Pregnancy Register, Swedish Medical Birth Register and Danish Epilepsy and Pregnancy study). If the registry included an internal control group, this was used to calculate the expected background risk of oral clefts. Otherwise general population statistics from the country closest to that represented by the registry were used. Sensitivity analyses were also completed varying the background risks for oral clefts. Work will be extended to other defect types with varying background frequencies.Results: Allowing the background risk of isolated oral clefts to vary by registry, the probability of observing a signal for isolated oral clefts from one registry was 10%. If the lowest background risk (0.37/1000 from the Brigham and Women’s Surveillance program) was applied to all registries, the probability of observing a signal from one registry was 0.6%. However, if a higher background risk of 1.59/1000 from Sweden was applied to all registries, the probability of observing a signal from one registry was 24%.Conclusions: It is vital to understand the role of registry multiplicity and background risk assumptions in producing signals for specific defects types among mature registries. Further discussions are needed among regulators, investigators and prescribers to better develop a framework for the interpretation of these signals that will increase in frequency with the continued growth of the pregnancy registries.