Bibian van der Voorn

196 CHAPTER 12 also discuss the possibility of selecting only ‘healthy’ very preterm newborns, e.g., excluding infants born small for gestational age (SGA) or who have developed bronchopulmonary dysplasia. However, this results in a selection of infants with lower risks of adverse outcomes, making it difficult to study effects of (standard) care in an unbiased way. Accordingly, in our studies we aimed to be as explorative as possible, and therefore applied no strict selection criteria within very preterm cohorts. The impactofperformanceanddetectionbiases inobservational studieshaspreviously been questioned by Vandenbroucke 9 . He advocated that unintended effects, such as those resulting from standard care, can be studied reliably within an observational study design, on the premise that study outcomes were not predictable at the time of allocation, and therefore were unrelated to intended effects. We therefore believe that the effects that were studied in chapter 7, 9 and 10 , were studied in an unbiased way, i.e., long-term effects of antenatal glucocorticoid treatment in infants with a genetic predisposition to altered glucocorticoid sensitivity, and sex differences in HPA axis (re)activity. Another challenge of observational studies is the sample size, which is ‘offered by nature’. Therefore, it could have been too small or be substantially decreased due to attrition at follow-up, making it difficult to detect effects in clinically relevant subgroups, such as male infants ( Chapter 2, 3, 5, 7 and 8 ) or very preterm newborns that were born SGA ( Chapter 7 and 8 ). Nonetheless, post-hoc analyses revealed valuable working hypotheses that can be used in future studies, designed to study such groups of subjects. In addition, The International Committee of Medical Journal Editors (ICMJE) recently advocated for responsible data sharing, while they believe it is ethically right to use data as thorough as possible 10 . The advantages of combining data from several cohorts are: enlargement of power, and a reduction of biases due to the heterogeneity of subjects and allocated care. In this context, it should again be emphasized that data are collected and presented in standardized ways (see for a more extensive argumentation chapter 10 ). Moreover, the selection of subjects and classification of cohorts also needs more consistency. In previous trials both infants who were born very preterm and/or with a very low birth weight have been included ( chapter 7 and 8 ). However, our research group has recently quantified the impact of such classification criteria on later-life outcomes 11 . To conclude, while the observational study designs that were used cannot rule out any form of bias, with our approach we tried to minimize it. When having the aim to explore unclarified mechanisms by use of such data it is important to present and