84 Chapter 4 The difference in algorithm design and responsiveness could explain our results. Exact and clear information on the workings of the CLiO2 algorithm is limited, but one can infer from the patent document that only in few cases FiO2 is titrated below the BaseFiO2 – a derivative of the average oxygen requirement. This could mean CLiO2 is slower to reduce the administered oxygen, possibly leading to more hyperoxia. Moreover, a longer averaging time employed by CLiO2 leads to a delay in pickup of deviations, which is further amplified by a timeout applied to prevent responding to erroneous SpO2 values. 19 These choices in algorithm design can lead to tardier algorithm responsiveness, and may therefore lead to less time within the oxygen saturation target range. Combined with the results of our cross-over study, the results show that choice of oxygen control device influences achieved time within oxygen saturation ranges in the preterm infant. Although choice of AOC may not be of great impact on achieved target range time over the entire course of admission, it could very well be that most morbidity finds its genesis in the periods of respiratory instability, during which a prominent difference is noted between the two AOC algorithms. A higher incidence of retinopathy of prematurity has been found to be associated with intermittent hypoxia1 as well as hyperoxia, both of which occur more often in periods of respiratory instability. Indeed, in a matched cohort study we reported a lower incidence of retinopathy of prematurity for infants under OxyGenie automated oxygen control.20 Although causality cannot be inferred from these results, all results of our recent studies are pointing in the same direction.