Studying Disease Occurrence and Drug Effects in Children: A global approach

Childhood diseases result from different causes and exhibit different characteristics. The occurrence of such diseases can be estimated from electronic healthcare records but the characteristics of both the diseases and the databases should be considered. Licensed drugs have limited or no direct evidence of safety and effectiveness in children, resulting in high levels of unlicensed and off-label drug use. To address this unfairness, evidence on safety and effectiveness should be generated using ‘real-world’ data in children. Methods need to be made ‘child-proof’. In this thesis, we studied and developed methods for assessing the occurrence of disease, and drug safety and effectiveness in children. Based on our findings, we made appropriate recommendations.

In the studies that comprise this thesis, we utilized different types/sources of data including the published literature, spontaneous reporting (suspected adverse drug reactions) data and electronic health records (data from general practitioners).

We found that disease occurrence can be estimated from electronic health care databases. Also, we demonstrated that both left censoring and the choice of episode duration impact the incidence and prevalence estimates of childhood diseases. For the common and recurrent diseases like acute otitis media (AOM), increasing the assumed length of an episode decreases the incidence but increases the prevalence estimates. For the chronic diseases, increasing the length of the run-in period prior to start of follow-up decreases the incidence rate with negligible effect on the prevalence. Of note, the length of the run-in period impacts both common and rare diseases. In order to test the performance of signal detection algorithms when applied to pediatric data specifically, we created a pediatric-specific reference set comprising 37 known (positive controls - PC) and 90 ‘unknown’ (negative controls - NC) DECs. We found that age-stratified analysis is important when performing pediatric-specific drug safety surveillance. Regarding pediatric pharmacoepidemiological safety studies, we discovered that such studies are few and the studies are almost exclusively conducted in high income countries and receive limited funding from pharmaceutical companies and other private sources. Designs differed a lot between drug and vaccine studies. Focus in these studies was only on a small fraction of registered drugs and mainly intermediate outcomes (signs/symptoms). Regarding the quality of the studies, cohort studies were most likely to be biased because historical time prior to study start was not considered for the identification of prevalent cases of the outcome of interest and because follow-up time after study start was not long enough to observe occurrence of the outcome(s). Case control studies were most likely to be of poor quality because of differences in non-response rate (regarding ascertainment of the exposure) between the cases and controls, and because the authors did not specifically report that the controls did not experience the outcome of interest. Regarding published pediatric comparative drug effectiveness studies, we found that studies were few, mainly conducted in North America and Europe and mainly based on public funding. Studies in newborns were rare. Regarding study design, the cohort design was the most commonly applied, probably also because it allows studying multiple outcomes. Regarding outcome, intermediate outcomes were often used. The investigated drugs did not always reflect routine drug utilization in paediatrics. Regarding the use of propensity scores to control confounding by indication in pediatric comparative drug effectiveness studies in asthmatic children; compared to ICS+LABA loose combination for treating asthma, ICS+LABA fixed protected against exacerbation. Calliper matching without replacement resulted in the largest adjustments of the crude hazard ratio (HR) especially in the 1 month prior to treatment start. Based on a survey of the current needs in pediatric pharmacoepidemiology, we found that different researchers used different databases to study the effects of different drug classes, perhaps reflecting the wide range of interests.

Based on our findings, we made several recommendations for the improvement of pediatric pharmacoepidemiology.