41 2 Automated oxygen control in preterm infants, how does it work and what to expect workload when most adjustments are performed automatically, leaving more time for other patient-related care.29 There might be a trade-off in the rapidity of response of AOC systems. An algorithm making swift changes will result in quick resolution of hyperoxic overshoot, a commonly observed problem during recovery from hypoxia both with manual and automated control.39 However such an algorithm will tend to produce a lower mean SpO2 on the steeper part of the oxygen-dissociation curve, potentially leading to more instability in SpO2. 43 Smaller alterations would then have a bigger impact on SpO 2 resulting in a less stable oxygen saturation. Conversely, an algorithm making slow changes could prolong both hypoxic and hyperoxic episodes unnecessarily. A proxy for PaO2, oxygen saturation via pulse oximetry (i.e. SpO2), is used for continuous non-invasive monitoring of oxygenation, but is limited in accuracy.42, 44, 45 Unfortunately for now no adequate alternatives to pulse oximetry exist. The FiO 2SpO2 relationship shows substantial intra-subject variability in the change in SpO2 following an adjustment in FiO2. 22 Many factors will influence the SpO 2 response, including for example, the changes in the oxygen-dissociation relationship during transition from foetal to adult haemoglobin. This shift will be quite pronounced in preterm infants who receive transfusions of adult blood. A potential reason to apply lock-out periods or filtering out SpO2 values far out of range could be to prevent inappropriate FiO2 adjustments based on short-term fluctuations, not reflecting the actual oxygenation state. Rather than using these techniques and risking unnecessary delay, focus should lie on preventing erroneous values by developing new or improved methods of measuring oxygenation. Algorithms could benefit frommultiple input parameters to better assess oxygenation and overall condition of the premature infant, as pulse oximeters become more inaccurate in critically ill patients.42 Essential for this will be easy linking of bedside devices (e.g. patient monitors, ventilators, electronic patient records) from different manufacturers, an area in which improvements should be made. Modern techniques could be used to further improve AOC for preterm infants.46 For example, a prediction model involving artificial intelligence strategies such as deep learning could result in earlier mitigation or even prevention of hypoxia and hyperoxia. Another example is an algorithm evaluating its own performance. By continuous assessment of the effect of adjustments in FiO2 during operation, even better adaptivity could be achieved. Additionally, the risk for retinopathy of prematurity may be decreased if the algorithm would take gestation at birth and postnatal age into account and thus determine a vulnerability score for retinopathy and an appropriate
RkJQdWJsaXNoZXIy MTk4NDMw