Hylke Salverda

130 Chapter 7 respiratory support may play only a minor role in the eventual neurodevelopmental outcome, meaning only very large randomised studies are able to demonstrate a statistically significant difference. Thirdly, neonatal care is a rapidly developing field with frequent changes to standard of care. Some of these unmeasured factors may influence the results in either direction. Finally, some of the adverse outcomes are relatively rare. If the effect of automated oxygen control is modest, a large clinical trial would be needed to observe an effect. Currently, the FiO2-C trial randomises between automated oxygen control or manual titration during the entire NICU stay, and will investigate the effect on clinical and neurodevelopmental outcome at 24 months of corrected age.29 The study is projected to run until December 2022. A change in target range may influence the time spent in (mild) hypoxia. In our case one would expect that 76% (223/293) of infants in the pre-implementation group born before November 2014 spent more time in the 85% - 90% range, as the lower limit was changed from 85% to 90%. The achieved proportion of time in the 85% - 90% range based on 1 minute-values of the pre- and post-implementation data show no difference while infants received oxygen (pre-AOC 10.9 [8.6 – 13.5]%, post-AOC 10.4 [7.7 – 12.7]%, p=0.09), and a 1.8% difference when considering the entire period of respiratory support (pre-AOC 5.5 [1.7 – 9.8]%, post-AOC 3.7 [1.6 – 7.6]%, p=0.002; unpublished data). Van Zanten et al. reported before that the change of lower limit led to a reduction in achieved time within the 80%-90% range in our unit, but time spent in hypoxia (SpO2 < 80%) was not different 31. One of the inherent limitations of a retrospective design is the rate of missing data (loss to follow-up in this study: pre-AOC 6.9%, post-AOC 10.6%), which is unfortunately frequently high in follow-up research. The majority of missing children were transferred to another university NICU in the neonatal phase and had subsequent follow-up there, therefore we expect them to be missing at random and not related to neurodevelopmental outcome. However, children lost to follow-up may be under treatment in a special care facility and therefore not missing at random. Parents may be less inclined to present their child for follow-up when they already receive regular tests in such a facility. To limit biased results due to missing such children, we requested data for all children tested elsewhere. A strength of the study is that we have a relatively large cohort in which we had few exclusion criteria, meaning the results are generalisable to other NICUs in a similar setting. Furthermore children are tested by trained professionals as part of a standardized national followup programme, improving the repeatability and reliability of the assessment of neurodevelopmental outcome. Finally, most data was collected prospectively during standard follow-up, minimising recall bias.