Inspired oxygen control for preterm infants

Inspired oxygen control for preterm infants is performed to maintain oxygen saturation (`SpO_2`) in the blood in a target range. Exposure to insufficient or excess levels of fraction of inspired oxygen (`FiO_2`) may increase the risk of mortality and morbidity in these patients. Despite this sensitivity, manual control is still the common practice largely due to immaturity of the existing control algorithms in dealing with the challenging task. Accordingly, this thesis aimed to initially identify the shortcomings of the existing automated controllers and then to provide indications for addressing these shortcomings. Through comprehensive analysis of the literature, the main design challenges of automated controllers were identified as oxygenation variability, technologic insufficiencies of infant monitoring and safety considerations. The thesis then largely focused on addressing the variability issue. Oxygenation variability means that a given `FiO_2` adjustment may lead to a different `SpO_2` response on different occasions. A first order transfer function characterised by a delay, a time-constant and a gain was used to model the `FiO_2-SpO_2` relationship following `FiO_2` adjustments in a large dataset from preterm infants receiving supplemental oxygen. The model was found representative for 37% of the adjustments from which an image of the parameter variations was obtained. The model was more representative for `FiO_2` increments than decrements and predictability was low in the collective set of model parameters. The study was followed by a thorough characterisation of the oxygenation response which most notably indicated intra and inter-patient variability as well as influence of severity of lung dysfunction on the gain of the oxygenation system. These findings support the superiority of adaptive control algorithms over robust and rule-based approaches. Moreover, these results provide a quantitative basis for developing adaptive algorithms and point to the severity of lung dysfunction reflected in baseline `FiO_2` as a viable basis for the adaptation. Frequent fluctuations of `SpO_2` being another challenging aspect of variability was then investigated. Apnoea, loss of circuit pressure and motion artefact concomitant with falls in `SpO_2` (e.g. hypoxia) were of interest. The high frequency of these factors as well as relevance of respiratory pauses with the characteristics of the following hypoxic events indicated potential benefits of incorporating respiratory rate in automated control methods. Finally, the issue of oximetry signal dropouts was studied and the results indicated that pre-emptive increments to `FiO_2` when `SpO_2` is missing during automated control may not be necessary. Parts of the outcomes of this thesis were used in development of a neonatal oxygen control algorithm for which a patent application is in progress. In a nutshell, the main contributions of this thesis to the research area include 1) Identification of the main challenges in automated control of `FiO_2` for preterm infants, indications for overcoming the challenges, 2) Providing a quantitative image of the characteristics of oxygenation system in preterm infants with a representation suitable for developing automated control algorithms, 3) Identifying the severity of lung dysfunction as a predictor of oxygenation response variability, 4) Revealing the frequency and relevance of factors such as apnoea and motion artefact concomitant to hypoxic events which can complicate automated `FiO_2` control, 5) Obtaining information concerning the `SpO_2` changes before and after episodes of signal dropout which assists in decision-making of a controller during these periods and 6) Providing information which acted as a basis for developing a control algorithm with commercialisation prospects.