Hylke Salverda



Financial support for the printing of this thesis was kindly provided by: Universitaire Bibliotheken Leiden Department of Paediatrics of the Leiden University Medical Center SLE Limited Cover by: Ridderprint, www.ridderprint.nl Printed by: Ridderprint, www.ridderprint.nl Layout by: Hylke Salverda ISBN: 978-94-6458-500-1 E-ISBN: 978-94-6458-501-8 The studies in this thesis were financially supported by an unrestricted research grant from SLE Limited; they had no role in study design nor in the collection, analysis, and interpretation of data, writing of the reports and decision to submit this manuscript or the papers for publication. Copyright © 2022 by Hylke H. Salverda All rights reserved. No part of this book may be reproduced, stored in a retrieval system of any system, or transmitted in any form or by any means, electronic, mechanical, photocopying, recording or otherwise, included a complete or partial transcription, without the prior written permission of the author.

Outcomes after automated oxygen control for preterm infants Proefschrift ter verkrijging van de graad van doctor aan de Universiteit Leiden, op gezag van rector magnificus prof.dr.ir. H. Bijl, volgens besluit van het college voor promoties te verdedigen op donderdag 3 november 2022 klokke 10:00 uur door Hylke Hendrik Salverda geboren te Alkmaar in 1988

Promotores: Prof.dr. A.B. te Pas Prof.dr. P.A. Dargaville University of Tasmania, Australia Leden promotiecommissie: Prof.dr. E. Lopriore Prof.dr. C.F. Poets Tübingen University Hospital, Germany Prof.dr. E. de Jonge Prof.dr. N.E. Schalij-Delfos

Voor mijn broer

What is now proved was once only imagined. – William Blake

Table of contents Preface 8 Part I General introduction Chapter 1 General introduction and outline of this thesis 13 Part II Currently available automated oxygen control algorithms Chapter 2 Automated oxygen control in preterm infants, how does it work and what to expect; a narrative review 27 Part III Effectivity of automated oxygen control algorithms on oxygenation of preterm infants in the NICU Chapter 3 Comparison of two devices for automated oxygen control in preterm infants – a randomised cross-over trial 51 Chapter 4 Comparison of two automated oxygen controllers in oxygen targeting in preterm infants during admission – an observational study 69 Chapter 5 Comparing descriptive statistics for retrospective studies from one-per-minute and one-per-second data 89 Part IV Clinical and long-term outcome after using automated oxygen controllers for preterm infants during NICU stay Chapter 6 The effect of automated oxygen control on clinical outcomes in preterm infants: a pre- and post-implementation cohort study 105 Chapter 7 Automated oxygen control for very preterm infants and neurodevelopmental outcome at two years – a retrospective cohort study 121 Chapter 8 Clinical outcomes of preterm infants while using automated controllers during standard care: comparison of cohorts with different automated titration strategies 137 Part V General discussion and summary Chapter 9 General discussion and future perspectives 153 Chapter 10 Summary 171 Chapter 11 Nederlandse samenvatting 179 Part VI Appendices Affiliations of co-authors 190 Abbreviations 192 List of publications 193 Curriculum Vitae 194 Dankwoord 195

8 Oxygen is crucial for the survival of all organisms. Without it, there would be no life. This much humankind has known since the first scientists discovered oxygen, one may have been as early as 1604. However, it was not until 1790 that oxygen was first mentioned for medical purposes and it would take another 110 years before it was first used in neonatal care. To get oxygen in our blood to enable the essential metabolic processes in our body, we use our lungs. The lungs of a preterm infant are not fully developed and they are thus not always capable of taking in enough oxygen. This prompts clinicians to give extra oxygen – a key treatment which has helped save many lives. Exactly how much oxygen clinicians can safely give is still unknown and can change within seconds for preterm infants. As with so many things in life, too little or too much can be harmful. A fine balance needs to be kept. In the first half of the 20th century, neonatal care saw little involvement from physicians. Few medical procedures were done to neonates. It was believed that handling during care led to cyanosis and apnoea and as a result, care was mostly limited to warming, feeding and isolation. Giving extra oxygen to preterm infants was also rare. The first mention of administering oxygen to preterm infants was by Budin in 1900, when he reported a beneficial effect of administering oxygen during cyanotic bouts. In the following years physicians noted that administering oxygen could also reduce irregular, also known as periodic, breathing. As a result, extended periods of oxygen administration were recommended and oxygen use for preterm infants became common practice. Oxygen hoods, funnels and even incubators were designed to administer oxygen, all with the aim of mixing in as little ambient air as possible. Preface

9 The first sign of the drawbacks of administering such a high fraction of oxygen in the air infants breathed appeared in 1940, when paediatrician Clifford noticed a new eye condition, later called retrolental fibroplasia. This new condition was meticulously studied by ophthalmologist Terry in the following years and another eleven years of research were needed to link this blindness-causing disease to administering oxygen. From then on, physicians began to realize that too much oxygen was harmful, and a more restrictive approach followed in the mid-1950s. Physicians lowered the oxygen content in supplied breathing air to 40%, and as a result the rate of retinopathy of prematurity - the contemporary name for retrolental fibroplasia - decreased. This change in practice was not based on evidence from research, but on clinical findings from individual paediatricians. Lowering the oxygen content also had a problem: both the rate of hyaline membrane disease, now known as respiratory distress syndrome, and the rate of cerebral palsy went up for preterm infants - keeping the balance between too much and too little was, and still is, difficult. Although a form of a pulse oximeter - a device to measure the oxygen saturation of the blood – was developed in 1935, titrating oxygen on the basis of the oxygen content in the blood only started in the 1960s, when blood gas monitoring became readily available. It would not be pulse oximetry as used today, as this was only developed by researcher Takuo Aoyagi in 1974. These early pulse oximeters were highly inaccurate when patients moved. This poses a particular problem for preterm infants, who cannot be instructed to lie still. Eventually, in 1995, Masimo developed Signal Extraction Technology, which is more resistant to motion, and this is now the basis for guiding how much oxygen to give preterm infants. Technology has become more and more sophisticated. In our unit, infants in need of respiratory support can receive breath volumes as low as 2 millilitres. The breath is given at the exact moment the baby attempts to breathe, and the amount of oxygen is automatically adapted to the infants’ need by measuring the oxygen saturation of the blood. By continuous automatic adjustment of the oxygen content we give, we are better than ever at keeping the fine balance between too much and too little. Despite these improvements many infants still suffer from the complications of prematurity every year. With this thesis, I hope to contribute to a better life for these infants by researching how the devices for automated oxygen control work, how well they work to balance the oxygen saturation in the blood and, most importantly, what the health outcomes are of the preterm infants treated with these devices.

Part I General introduction

Chapter 1 General introduction and outline of this thesis

Supplemental oxygen for preterm infants is paramount against respiratory insufficiency related to preterm birth, but is a double-edged sword. On the one hand, supplemental oxygen helps prevent hypoxia which can lead tomorbidity andmortality. But, inappropriate administration of oxygen can lead to hyperoxia, which is related to morbidities such as bronchopulmonary dysplasia, retinopathy of prematurity and neurodevelopmental impairment. Oxygen must therefore be carefully titrated within a safe range, but this is a difficult task. Automated titration of the administered oxygen by a machine – an automated oxygen controller – can help reduce hypoxia and hyperoxia, and improve the time within this safe, or target, range. However, which commercially available automated oxygen controller is most effective is unknown, as is the effect of using an automated oxygen controller on clinical and long-term outcome. This thesis provides an overview of outcomes after using two different automated oxygen controllers employing different titration strategies.

15 1 General introduction and outline of this thesis Supplemental oxygen and effects on morbidity Supplemental oxygen is one of the most common therapies for preterm infants on respiratory support. Oxygen is fundamental for generating intracellular energy and is therefore essential to human life. A deficit of oxygen in the blood, or hypoxaemia, is common in preterm infants due to the immaturity of their lungs and respiratory centre, and can lead to oxygen shortage in the mitochondria which will eventually lead to cell death.1 Oxygen supplementation is therefore used to avert cell death and subsequent effects of hypoxaemia on the central nervous system, lungs, vasculature and other organs. As development of the respiratory system occurs until after term age,2 supplemental oxygen for preterm infants is often given for long periods of time. However, as with any drug, too much of it can lead to toxicity. The higher oxygen tension present in the extra-uterine atmosphere can harm a preterm infants’ organs, even without supplemental oxygen. The immaturity of a preterm infants’ anti-oxidant system renders them vulnerable to free radicals, also known as reactive oxygen and nitrogen intermediates, formed under the influence of excess oxygen in the blood. These free radicals cause membrane disruption and activate inflammatory pathways through lipid peroxidation.3 Indeed, infants developing a chronic lung condition called bronchopulmonary dysplasia were found to have elevated lipid peroxidation products.4 Causal evidence on the effect of hyperoxia is scarce for preterm infants, but plenty of evidence from animal experiments exists. Neonatal mice demonstrated that exposure of the lung to hyperoxia decreased the alveolarization, changed the vasculature of the lung, and increased lung fibrosis.5 Furthermore, intestinal histology of rat pups was markedly changed in pups continuously exposed to hyperoxia immediately after birth.6 Finally, hyperoxia has been shown to alter cerebral blood flow in mice, induce neuronal apoptosis and inflammation.7, 8 The first morbidity attributed to hyperoxaemia in preterm infants was retinopathy of prematurity (ROP), a neurovascular disease leading to blindness.9 The aetiology is not fully understood. Initially not enough retinal blood vessels are formed due to insufficient nutrition, insufficient growth factors, sepsis and a fluctuating oxygenation of the blood. The vasculogenesis of the retina is further reduced by iatrogenic hyperoxaemia inhibiting the formation of vascular endothelial growth factor. In this initial phase the retina is particularly susceptible to damage from intermittent hypoxia due to the scarce blood supply. After this phase uncontrolled blood vessel formation occurs under the influence of vascular endothelial growth factor eventually leading to retinal detachment.10 Given this aetiology, it is not surprising both intermittent

16 Chapter 1 hypoxia, hyperoxia and fluctuation in oxygenation all increase the rate of ROP in preterm infants.11-13 Oxygen and other improvements in neonatal care throughout the last century have led to increased survival, but have also increased the incidence of retinopathy of prematurity.14 Globally for the year 2010, it was estimated that ROP led to 20,000 blind infants, and left another 12,300 visually impaired.15 Bronchopulmonary dysplasia (BPD), a chronic disease of the lungs, is a major cause of respiratory illness in preterm infants leading to significant morbidity and mortality after discharge from the neonatal intensive care unit (NICU). The aetiology is multifactorial and involves disruption of the later phases of lung development and injury to the lung. Specifically for oxygen toxicity, high concentrations of free radicals are thought to cause chronic inflammation to the lung.16 This chronic inflammation in turn leads to changes in lung tissue: decreased alveolarization leading to less surface area for gas exchange; vascular remodelling leading to an increase in pulmonary resistance which in turn may lead to pulmonary hypertension;17 and changed lung elasticity.18 The incidence of BPD differs depending on the definition applied. When defined as supplemental oxygen requirement at 36 weeks postmenstrual age the overall incidence was estimated at 42% of infants born between 22-28 weeks gestational age, where a higher gestational age at birth reduced the change of having BPD.19 Titration of supplemental oxygen Mindful of the effects of supplemental oxygen on morbidity, its level must be carefully balanced within safe limits. Continuous monitoring using pulse oximetry (SpO2, the percentage of peripheral oxyhaemoglobin over total haemoglobin) is currently the most appropriate tool to guide the fraction of inspiratory oxygen (FiO2) delivered to the patient. In contrast to repeated arterial blood sampling, SpO2 is non-invasive and continuous. The ideal range for SpO2 in a given subject has been the subject of considerable debate,20 and remains unsolved to date. The accepted target in preterm infants has recently undergone refinement as a result of a series of randomised controlled trials comparing two SpO2 target ranges (SpO2 85-89% vs 91-95%). 21-23 These trials once again highlighted the impact of hypoxaemia and hyperoxaemia on preterm infants, with the lower target range associated with an increase in mortality and necrotizing enterocolitis, and higher target range with ROP. The need for supplemental oxygen is more common and prolonged in very preterm infants. These infants often present with respiratory instability and fluctuation in oxygenation. Whilst the need to target an SpO2 range is widely accepted, data from

17 1 General introduction and outline of this thesis cohort studies24-26 and randomised controlled trials27-29 point to the difficulty of SpO 2 targeting, with most studies reporting SpO2 values to be within the target range less than half of the time. Although bedside staff frequently adjusts FiO2 to maintain SpO2 within the target range prescribed by the clinician, their workload limits time availability and makes continuous tailoring of FiO2 to the infant’s needs difficult. This is further complicated by the neonatal oxygenation physiology being unpredictable and non-linear, with a long time delay between an adjustment in FiO2 and a stable SpO2. 30 Even in the presence of a dedicated respiratory therapist to titrate FiO 2, time within target range was only 66%.31 In premature infants with frequent fluctuations in oxygenation, clinical personnel usually respond to the occurrence of alarms in the pulse oximeter triggered by episodes of hypoxaemia with a manual increase in FiO2. When these episodes resolve and SpO2 returns to the desired range, FiO2 should be reset to the basal level. However, under routine clinical conditions, staff limitations can result in inconsistencies in response and timing. As a consequence, premature infants are often exposed to periods of insufficient oxygenation, unnecessary oxygen exposure and hyperoxaemia.32 Also, in these infants, the FiO 2 set by the bedside staff often exceeds the level required to maintain an acceptable range of SpO2. This is done in an attempt to reduce the frequency of the hypoxemic episodes. However, this is not always effective and can increase the exposure to supplemental oxygen and hyperoxaemia.33 Considering the effect of target range deviations and the difficulty of targeting SpO2, feedback-controlled adjustment of FiO2 by a machine –an automated oxygen controller (AOC)– is a logical improvement on current practice. In essence, SpO2 readings are continuously fed into a device holding a set of computational instructions (an algorithm), which then gives an output, an updated value for FiO2. The effectiveness of automated control of inspired oxygen and its effects on the fluctuation of oxygenation during the care of premature infants may result in improved neurodevelopmental outcomes.34 Randomised trials comparing automated FiO2 systems with manual titration for short periods demonstrated an increase in the proportion of time spent with SpO2 within target range varying between 8% and 24%.35-43 Automated FiO 2 control also decreased the required nursing time in preterm infants with frequent severe desaturations.36, 37, 44 Several automated oxygen control devices are commercially available and used in NICUs, but it is unknown whether these devices lead to different clinical or long-term outcome. Furthermore, it is unknown which of these controllers is most effective, as no comparisons have been made between the performance of different AOCs.

18 Chapter 1 Aim and outline of this thesis The general aim of this thesis was to evaluate the effects on outcome after automated oxygen control for preterm infants. This thesis aims to: describe currently available automated oxygen control algorithms and what to expect when they are used (Part II); compare effectiveness of automated oxygen control algorithms on oxygenation in the NICU (Part III); and investigate clinical and long-term outcome after using automated oxygen controllers (Part IV). This thesis comprises of observational studies and a randomised clinical trial. Part II consists of Chapter 1 in which an overview of approaches for algorithm design are described, after which the details on six commercially available oxygen control algorithms are set out. Per algorithm an outline follows on how the algorithm works, and what clinical effects were reported. In this narrative review we conclude that although all available controllers seem to improve time within target range and have a beneficial effect on the occurrence of hypoxia and hyperoxia, the most effective strategy is unknown, as available clinical studies were heterogenous. Part III reports on the effects of automated oxygen control on oxygenation. The NICU of the LUMC was the first to implement automated oxygen titration as standard of care. As of August 2015, all infants requiring respiratory support with supplemental oxygen received automated oxygen titration by the CLiO2 algorithm built into the AVEA ventilator (Vyaire, Yorba Linda, California, USA). In November 2018, all AVEA ventilators were replaced with SLE6000 ventilators (SLE Limited, South Croydon, UK), employing the OxyGenie algorithm for automated oxygen control. This led to a unique setting in which caregivers were competent to handle both ventilators allowing for a comparative, randomised, crossover trial. In Chapter 2 the results of this randomised crossover trial are presented. The effectiveness of an automated oxygen controller may vary depending on the postnatal age of the infant. The condition of the lungs and frequency of apnoea can change markedly during the course of admission. With the exception of one study, all studies report on achieved target range times while using automated oxygen control in an experimental setting and for a short period of time (maximum 24 hours). In Chapter 3 the achieved times within certain SpO2 ranges from birth to 32 weeks of postmenstrual age are compared when either using the CLiO2 or the OxyGenie controller as standard of care. Contemporary patient data management systems for intensive care may be limited to storing vital parameters once per minute, which is not sufficient to register all variation in vital parameters such as SpO2 and FiO2. In Chapter 4 we investigate whether one-per-minute data can be used to perform

19 1 General introduction and outline of this thesis retrospective comparisons of descriptive statistics. In Part IV we focus on clinical outcomes after using automated oxygen controllers as standard of care. Over 300 very preterm infants have received CLiO2 automated oxygen titration as standard of care. In Chapter 5 we compare the neonatal outcomes of these infants with infants born in the years before implementation of automated oxygen control. All preterm infants born under 30 weeks of gestation are invited to the outpatient clinic for standard follow-up at two years. In Chapter 6 we compare neurodevelopmental outcome of the same infants at a corrected age of two years. We conclude Part IV in Chapter 7 by comparing outcomes of infants treated with the CLiO2 algorithm with infants that received automated oxygen titration by the OxyGenie algorithm. Finally, in Part V of this thesis, the main findings of these studies are discussed and future perspectives are considered. The thesis is concluded with a summary of the studies, provided in English and Dutch.

20 Chapter 1 References 1. Cohen PJ. Oxygen and intracellular metabolism. International anesthesiology clinics 1981;19(3):9-19. 2. Schittny JC. Development of the lung. Cell and tissue research 2017;367(3):427-44. 3. Mathias M, Chang J, Perez M, et al. Supplemental Oxygen in the Newborn: Historical Perspective and Current Trends. Antioxidants (Basel, Switzerland) 2021;10(12) 4. Ogihara T, Hirano K, Morinobu T, et al. Raised concentrations of aldehyde lipid peroxidation products in premature infants with chronic lung disease. Archives of disease in childhood Fetal and neonatal edition 1999;80(1):F21-5. 5. Warner BB, Stuart LA, Papes RA, et al. Functional and pathological effects of prolonged hyperoxia in neonatal mice. The American journal of physiology 1998;275(1):L110-7. 6. Giannone PJ, Bauer JA, Schanbacher BL, et al. Effects of hyperoxia on postnatal intestinal development. Biotechnic & histochemistry : official publication of the Biological Stain Commission 2007;82(1):17-22. 7. Liu Y, Jiang P, Du M, et al. Hyperoxia-induced immature brain injury through the TLR4 signaling pathway in newborn mice. Brain research 2015;1610:51-60. 8. Kennedy C, Grave GD, Sokoloff L. Alterations of local cerebral blood flow due to exposure of newborn puppies to 80-90 per cent oxygen. European neurology 1971;6(1):137-40. 9. Silverman WA. Retrolental Fibroplasia: A Modern Parable. New York, New York: Grune & Stratton, Inc. 1980. 10. Hellström A, Hård AL. Screening and novel therapies for retinopathy of prematurity - A review. Early human development 2019;138:104846. 11. Di Fiore JM, Bloom JN, Orge F, et al. A higher incidence of intermittent hypoxemic episodes is associated with severe retinopathy of prematurity. J Pediatr 2010;157(1):6973. 12. Martin RJ, Wang K, Koroglu O, et al. Intermittent hypoxic episodes in preterm infants: do they matter? Neonatology 2011;100(3):303-10. 13. Imanishi Y, Hirata K, Nozaki M, et al. Effect of fluctuation of oxygenation on the development of severe retinopathy of prematurity in extremely preterm infants. J Perinatol 2020;40(3):515-21. 14. Trzcionkowska K, Vehmeijer W, Kerkhoff FT, et al. Increase in treatment of retinopathy of prematurity in the Netherlands from 2010 to 2017. Acta ophthalmologica 2021;99(1):97103. 15. Blencowe H, Lawn JE, Vazquez T, et al. Preterm-associated visual impairment and estimates of retinopathy of prematurity at regional and global levels for 2010. Pediatric research 2013;74(1):35-49. 16. Alvira CM, Morty RE. Can We Understand the Pathobiology of Bronchopulmonary Dysplasia? The Journal of pediatrics 2017;190:27-37. 17. Mourani PM, Abman SH. Pulmonary vascular disease in bronchopulmonary dysplasia: pulmonary hypertension and beyond. Curr Opin Pediatr 2013;25(3):329-37. 18. Thibeault DW, Mabry SM, Ekekezie, II, et al. Lung elastic tissue maturation and perturbations during the evolution of chronic lung disease. Pediatrics 2000;106(6):14529. 19. Stoll BJ, Hansen NI, Bell EF, et al. Neonatal outcomes of extremely preterm infants from

21 1 General introduction and outline of this thesis the NICHD Neonatal Research Network. Pediatrics 2010;126(3):443-56. 20. Stenson BJ. Oxygen Saturation Targets for Extremely Preterm Infants after the NeOProM Trials. Neonatology 2016;109(4):352-8. 21. Al Hazzani F, Khadawardi E. Effects of Targeting Higher VS Lower Arterial Oxygen Saturations on Death or Disability in Extremely Preterm Infants: The Canadian Oxygen Trial. J Clin Neonatol 2013;2(2):70-2. 22. Tarnow-Mordi W, Stenson B, Kirby A, et al. BOOST-II Australia and United Kingdom Collaborative Groups. Outcomes of Two Trials of Oxygen-Saturation Targets in Preterm Infants. N Engl J Med 2016;374(8):749-60. 23. Carlo WA, Finer NN, Walsh MC, et al. SUPPORT Study Group of the Eunice Kennedy Shriver NICHD Neonatal Research Network. Target ranges of oxygen saturation in extremely preterm infants. N Engl J Med 2010;362(21):1959-69. 24. Hagadorn JI, Furey AM, Nghiem TH, et al. Achieved versus intended pulse oximeter saturation in infants born less than 28 weeks’ gestation: the AVIOx study. Pediatrics 2006;118(4):1574-82. 25. Laptook AR, Salhab W, Allen J, et al. Pulse oximetry in very low birth weight infants: can oxygen saturation be maintained in the desired range? J Perinatol 2006;26(6):337-41. 26. LimK, Wheeler KI, Gale TJ, et al. Oxygen saturation targeting in preterm infants receiving continuous positive airway pressure. The Journal of pediatrics 2014;164(4):730-36.e1. 27. Schmidt B, Whyte RK, Asztalos EV, et al. Effects of targeting higher vs lower arterial oxygen saturations on death or disability in extremely preterm infants: a randomized clinical trial. Jama 2013;309(20):2111-20. 28. Stenson BJ, Tarnow-Mordi WO, Darlow BA, et al. Boost II United Kingdom, Australia, New Zealand Collaborative Group. Oxygen saturation and outcomes in preterm infants. N Engl J Med 2013;368(22):2094-104. 29. Clarke A, Yeomans E, Elsayed K, et al. A randomised crossover trial of clinical algorithm for oxygen saturation targeting in preterm infants with frequent desaturation episodes. Neonatology 2015;107(2):130-6. 30. Sadeghi Fathabadi O, Gale TJ, LimK, et al. Characterisation of the Oxygenation Response to Inspired Oxygen Adjustments in Preterm Infants. Neonatology 2016;109(1):37-43. 31. Claure N, Gerhardt T, Everett R, et al. Closed-loop controlled inspired oxygen concentration for mechanically ventilated very low birth weight infants with frequent episodes of hypoxemia. Pediatrics 2001;107(5):1120-4. 32. van ZantenHA, TanRN, ThioM, et al. The risk for hyperoxaemia after apnoea, bradycardia and hypoxaemia in preterm infants. Arch Dis Child Fetal Neonatal Ed 2014;99(4):F26973. 33. van Zanten HA, Tan RNGB, van den Hoogen A, et al. Compliance in oxygen saturation targeting in preterm infants: a systematic review. 2015;174(12):1561-72. 34. Bancalari E, Claure N. Control of oxygenation during mechanical ventilation in the premature infant. Clinics in perinatology 2012;39(3):563-72. 35. Claure N, D’Ugard C, Bancalari E. Automated adjustment of inspired oxygen in preterm infants with frequent fluctuations in oxygenation: a pilot clinical trial. J Pediatr 2009;155(5):640-5 e1-2. 36. Plottier GK, Wheeler KI, Ali SK, et al. Clinical evaluation of a novel adaptive algorithm for automated control of oxygen therapy in preterm infants on non-invasive respiratory

22 Chapter 1 support. Arch Dis Child Fetal Neonatal Ed 2017;102(1):F37-F43. 37. Lal M, Tin W, Sinha S. Automated control of inspired oxygen in ventilated preterm infants: crossover physiological study. Acta Paediatr 2015;104(11):1084-9. 38. van Kaam AH, Hummler HD, Wilinska M, et al. Automated versus Manual Oxygen Control with Different Saturation Targets and Modes of Respiratory Support in Preterm Infants. J Pediatr 2015;167(3):545-50 e1-2. 39. Urschitz MS, Horn W, Seyfang A, et al. Automatic control of the inspired oxygen fraction in preterm infants: a randomized crossover trial. Am J Respir Crit Care Med 2004;170(10):1095-100. 40. Hallenberger A, Poets CF, Horn W, et al. Closed-loop automatic oxygen control (CLAC) in preterm infants: a randomized controlled trial. Pediatrics 2014;133(2):e379-85. 41. Waitz M, Schmid MB, Fuchs H, et al. Effects of automated adjustment of the inspired oxygen on fluctuations of arterial and regional cerebral tissue oxygenation in preterm infants with frequent desaturations. J Pediatr 2015;166(2):240-4 e1. 42. Claure N, Bancalari E, D’Ugard C, et al. Multicenter crossover study of automated control of inspired oxygen in ventilated preterm infants. Pediatrics 2011;127(1):e76-83. 43. Zapata J, Gomez JJ, Araque Campo R, et al. A randomised controlled trial of an automated oxygen delivery algorithm for preterm neonates receiving supplemental oxygen without mechanical ventilation. Acta Paediatr 2014;103(9):928-33. 44. Van Zanten HA, Kuypers K, Stenson BJ, et al. The effect of implementing an automated oxygen control on oxygen saturation in preterm infants. Arch Dis Child Fetal Neonatal Ed 2017;102(5):F395-F99.

Part II Currently available automated oxygen control algorithms

What is known about this topic • All contemporary automated oxygen control algorithms increase time spent within oxygen saturation target range. • The automated oxygen control algorithms are all different in design and function. What this study adds • The basal oxygen requirement, the amount and magnitude of interventions by an automated oxygen controller can possibly be used as an indicator of clinical deterioration. • Studies involving automated oxygen controllers are heterogenous and cannot be compared. • A head-to-head comparison of algorithms is required to understand how to best utilise this technology.

Chapter 2 H.H. Salverda, S.J.E. Cramer, R.S.G.M. Witlox, P.A. Dargaville, A.B. te Pas Archives of Disease in Childhood - Fetal and Neonatal Edition Mar 2021, 106 (2) 215-221; DOI: 10.1136/ archdischild-2020-318918 Automated oxygen control in preterm infants, how does it work and what to expect; a narrative review

Abstract Background: Automated oxygen control systems are finding their way into contemporary ventilators for preterm infants, each with its own algorithm, strategy and effect. Objective: To provide guidance to clinicians seeking to comprehend automated oxygen control and possibly introduce this technology in their practice. Method: A narrative review of the commercially available devices using different algorithms incorporating rule-based, proportional-integral-derivative and adaptive concepts are described and explained. An overview of how they work and, if available, the clinical effect is given. Results: All algorithms have shown a beneficial effect on the proportion of time that oxygen saturation is within target range, and a decrease in hyperoxia and severe hypoxia. Automated oxygen control may also reduce the workload for bedside staff. There is concern that such devices could mask clinical deterioration, however this has not been reported to date. Conclusions: So far, trials involving different algorithms are heterogenous in design and no head-to-head comparisons have been made, making it difficult to differentiate which algorithm is most effective and what clinicians can expect from algorithms under certain conditions. Keywords: Hypoxemia; hyperoxia; closed-loop; algorithm; neonate

29 2 Automated oxygen control in preterm infants, how does it work and what to expect Introduction Preterm infants often receive respiratory support, including supplemental oxygen therapy for a prolonged period of time during their first hospitalisation. Provision of supplemental oxygen must aim to keep oxygen saturation within a normoxic range, thereby minimising occurrence of hypoxia and hyperoxia, both of which are associated with organ injury.1-6 But titrating oxygen therapy manually within the narrow therapeutic range is a difficult task to perform.7, 8 Over the last four decades, a number of algorithms have been developed to facilitate automatic titration of oxygen for preterm infants. All the contemporary algorithms use oxygen saturation as measured by pulse oximetry (SpO2) as their input, but each has a different design in processing the input and computing an adjustment in the fraction of inspired oxygen (FiO2). An important prerequisite to successfully applying automated oxygen control (AOC) in clinical practice will be for clinicians to understand how the control algorithms operate and what effect differences in design have.9-21 Six oxygen control algorithms are currently embedded in commercially available neonatal ventilators. This narrative review will consider the different approaches to algorithm design, explain how the contemporary algorithms operate and, based on available data, discuss their clinical effects in preterm infants. For this, we performed a search on PubMed, Embase, Web of Science, Cochrane and Emcare for (pre)clinical studies and reviews comparing AOC with manual titration of oxygen in preterm infants receiving respiratory support and oxygen therapy. A cross-check for search completeness was made using the reference list of primary resources, aiming to identify any studies not located in the initial search. Additionally, patent documents and device operating manuals were studied where available.

30 Chapter 2 Algorithm design Different approaches have been used for designing AOC algorithms. Generally, a combination of three methods is used: rule-based, proportional-integral-derivative (PID) and adaptive. Central to each of these approaches is the derivation of an SpO2 error, the deviation from the desired value, i.e. the positive or negative difference between the current measured SpO2 and the middle or upper/lower limit of the target range (TR). Rule-based A rule-based algorithm uses a set of rules to decide on an FiO2 adjustment much like a decision tree. The rules are often derived from expert knowledge (for example: “in case of mild hyperoxia, lower FiO2 by 0.02”; Figure 1). Incorporating clinical experience makes this type of algorithm quick to develop and intuitive for clinicians. By combining a large set of rules an attempt is made to cover all possible scenarios. However, there is great heterogeneity in response to a change in FiO2 22, 23, making it virtually impossible to have an exhaustive set of rules for each circumstance. Oxygen saturation trend 100 90 80 70 Time SpO No change FiO + 0.02 FiO – 0.02 No change Start Enter SpO value Hypoxia FiO + 0.02 FiO – 0.02 Hyperoxia Normoxia Oxygenation status FiO alteration Figure 1: A simple rule-based algorithm and its possible effect on the oxygen saturation trend.

31 2 Automated oxygen control in preterm infants, how does it work and what to expect Proportional-integral-derivative Proportional-integral-derivative (PID) control is a longstanding mathematical approach widely used in industry (e.g. automotive cruise-control, thermostats). The proportional term reflects the current error; the integral term reflects the sum of previous errors; and the derivative term accounts for the direction the SpO2 error is heading (Figure 2). The computation includes a unique coefficient for each of the P, I and D terms, thus balancing their relative influence. The sum of the P, I and D terms – which may be opposite in sign – determines the magnitude of the change in FiO2. Although PID control is more abstract and its function more difficult to understand, it makes use of the breadth of available oxygenation information. However, the choice of PID coefficients is vital, as inappropriate coefficients could lead to major fluctuations or oscillations in oxygen saturation and result in oxygen control that is worse than manual titration.21 Oxygen saturation trend 100 90 80 70 Time SpO Proportional error - 0.08 0.005 - 0.01 sum baseFiO PID Terms new FiO Integral Derivative slope Figure 2: The three components of the PID algorithm hidden in the oxygen saturation trend.

32 Chapter 2 Adaptive Adaptive control is an element of an algorithmwhereby the behaviour of the controller changes while it is in use. The aim is to tailor the controller to the oxygenation system of the baby by including patient-specific parameters as input to the algorithm. The most apparent example is to change the magnitude of adjustments in response to a change in the severity of lung disease (estimated by baseline oxygen requirements). The basal oxygen requirement (baseFiO2) of a preterm infant almost always changes over time, as will the response of SpO2 to an FiO2 adjustment. Adding baseFiO2 to the algorithm will lead to an adaptation to the degree of lung disease. An example of the behaviour of such an algorithm is given in Figure 3. Figure 3: The possible effect and responses of an adaptive algorithm in two cases with a different baseFiO2 Oxygen saturation trend, baseFiO = 0.25 100 90 80 70 Time SpO Oxygen saturation trend, baseFiO = 0.40 100 90 80 70 Time SpO FiO – 0.03 FiO – 0.06 FiO + 0.04 FiO + 0.08 Do nothing Do nothing

33 2 Automated oxygen control in preterm infants, how does it work and what to expect Commercially available algorithms Details of six commercially available oxygen control algorithms are set out below, with a precis of the known function of the algorithm followed by a section on clinical effect where data are available. Closed-Loop Automatic oxygen Control17 How it works The Closed-Loop Automatic oxygen Control (CLAC) is a rule-based algorithm commercially available in the Leoni ventilator (Löwenstein medical, RheinlandPfalz, Germany). The algorithm deduces two parameters from the SpO2 signal once per second: the state and trend parameters. The state parameter is calculated by taking SpO2 values of the last three minutes, filtering out values far out-of-range, and forming a so-called ‘spread’ from the remaining information: a regression line combined with an adapted standard error. Using the middle of this spread the algorithm labels the state ‘substantially above’, ‘above’, ‘normal range’, ‘below’, or ‘substantially below’ target range, in response to which FiO2 will be adjusted in the range of -0.02 to +0.05 (-0.02, -0.01, +0.01, +0.02 and +0.05). The trend parameter – the slope of SpO2 trend in the last 60 seconds – can postpone an adjustment, dependent on an increasing, stable or decreasing trend. This to account for when the SpO2 is outside the TR but normalising. 17, 24, 25 After making an adjustment the algorithm pauses for 180 seconds to allow the baby to reach a new steady oxygenation state. Recently, the alternative to pause for 30 seconds was added.26 The system can pause for safety (reasons are: adapted standard error above a cut-off value; missing or invalid oximetry input; acute hypoxia (SpO2 < 80%) for more than 4 seconds) during which bedside staff is alerted and can intervene if necessary. Clinical effect Three randomised crossover trials reported the use of the CLAC algorithm (Table 1). While infants spent more time within TR during AOC (90.5% vs 81.7%, P = 0.01) during the first study, the short study span probably led to the high proportion of time in TR in both groups. The reduction in manual adjustments (89%) is however striking and could be beneficial in reducing workload for bedside caregivers.17 In a subsequent multicentre study27 infants on (non)invasive ventilation were studied

34 Chapter 2 Table 1: Studies on automated oxygen control Oxygen control algorithm Author (year) Study duration SpO2 TR Gestational age in weeks Age at entry study n CLACslow Urschitz (2004)17 3x90 min 87%-96% 25.5 (24-33) 20.5 days (4 - 78) 12 Hallenberger (2014)27 2x24 hrs 90%-95% 80%-92% 83%-93% 85%-94% 26.4 (23.0-35.3) 29.9 weeks (26.0 – 35.6) 34 CLACfast Schwarz (2019)26 3x8 hrs 90%-95% 85%-93% 85%-96% 26.4 (24.0-32.7) 24 days ± 10 19 CLiO2 Claure (2001)18 2x2 hrs 88%-96% 25 ±1.6 26 days ± 11 14 Claure (2009)28 2x4 hrs 88%-95% 24.9 ±1.4 33 days ± 15 16 Claure (2011)29 2x24 hrs 87%-93% 25 (24-27) 27 days [17-36] 32 Waitz (2015)30 2x24 hrs 88%-96% 25 (23-28) 34 days [19-75] 15 Van Kaam (2015) 31 2x24 hrs 91%-95% (1) 89%-93% (2) 26 (25-28) 18 days [10-29] 80 Lal (2015)32 2x12 hrs 90%-95% 25 (24-27) 16 days [9-27] 27 Van Zanten (2017)33 3-28 days 90%-95% 27+6 (26+3-28+4) 27+3 (26+0-28+2) n/a 42 PRICO ‡ n/a n/a n/a n/a n/a n/a SPO2C Gajdos (2018)34 2x24 hrs 88%-96% 25.3 (23-26) 31.5 days [12 – 62] 12 IntellO2 Reynolds (2018)35 2x24 hrs 93% (A) 90%-95% (M) 26 (24-27) 29 days [18-53] 30 VDL1.0 Plottier (2017)36 3x4 hrs 91%-95% (A) 90%-94% (M) 27.5 [26-30] 8 days [1.8-34] 20 Mean ±SD or median[IQR] or (range). TR, target range; n/a, not applicable/available; RMC, routine manual control; NCPAP, nasal continuous airway pressure; MV, mechanical ventilation. P-values are two sides unless indicated otherwise. *one sided superiority test; †one sided non-inferiority test. ‡For PRICO no clinical data are available to date.

35 2 Automated oxygen control in preterm infants, how does it work and what to expect Number on mode of support Control group Proportion of time spent in TR (%) Control AOC p-value 12 nCPAP 1) RMC 2) Fully dedicated operator 1) 82 (39-100) 2) 91(41-99) 91 (59 - 99) 0.01 23 nCPAP 11 MV RMC 61 ±15 72 ±14 <0.001 18 nCPAP 1 MV 1) RMC 2) CLAC slow 1) 58 ±11 2) 65 ±11 68 ±11 1) 0.0001 * 2) 0.0005 † 14 MV Fully dedicated operator 66 ±14 75 ±13 <0.05 16 MV RMC 42 ±9 58 ±10 <0.001 32 MV RMC 32 ±13 40 ±14 <0.001 14 nCPAP/NIPPV 1 MV RMC 69 ±8.2 76 ±9.2 <0.01 50 nCPAP 30 MV RMC (1) 58 ±16 (2) 54 ±16 (1) 62 ±17 (2) 62 ±17 <0.05 <0.001 27 MV RMC 60 (49-73) 73 (59-83) 0.031 nCPAP or MV RMC 48 (42-56) 62 (49-72) <0.01 n/a n/a n/a n/a n/a 10 nCPAP/NIPPV 2 MV RMC 69 ± 7.7 78 ± 7.1 0.0012 30 HFNC RMC 49 (40-57) 80 (70-87) <0.0001 7 HFNC 13 nCPAP RMC 56 (48-63) 81 (76-90) <0.0001

36 Chapter 2 for 2x24 hours, and a similar difference in time in TR was seen (71.2% vs. 61.4%, P < 0.001). Different TRs (range 80-95%) were used between participating centres which make results potentially more generalisable but also harder to interpret, as a different target range will influence controller performance. Finally, the last study showed that the 30-second pause was superior to manual control and non-inferior to the 180-second pause, with similar increases in TR time.26 CLAC is not designed to treat acute SpO2 deterioration; the algorithm will raise an alarm and cease operation if low SpO2 values occur. Also, suspending action after an adjustment might be undesirable, as an algorithm could also use SpO2 feedback to try to resolve hypoxia more swiftly and diminish subsequent overshoot. Closed-loop Inspired Oxygen Control (CLiO2) 18 How it works The first algorithm for AOC to be embedded in a neonatal ventilator was CLiO2 available in the AVEA ventilator (Vyaire Medical, Mettawa, United States). This algorithm, a hybrid of rule-based and proportional-derivative control with an adaptive element, runs through a large set of instructions each second.37 The algorithm starts with SpO2 validation, and determination of the status of oxygenation (normoxic, hyperoxic or hypoxic). After a change in oxygenation status for ≥ 3 seconds (to filter out short fluctuations), an initial FiO2 adjustment is made, proportional to the magnitude of the error. For the CLiO2 algorithm, the SpO2 error is calculated in relation to the upper and lower limits of the TR in hyperoxia and hypoxia, respectively, rather than the mid-point. The timing and magnitude of further FiO2 adjustments are then determined in relation to the SpO2 error (via a proportional term), the SpO2 trend (via a derivative term) and the baseFiO2 (via an adaptive component). The FiO2 increments will be amplified if SpO2 is deviating further from the TR limits, and with progressively smaller increments as SpO2 approaches the TR. With the CLiO2 algorithm, the adjustments in FiO2 rarely lead to an FiO2 below baseFiO2; only long-lasting periods of mild hyperoxia will result in an FiO2 below the base value. BaseFiO2 is updated periodically after at least 5 minutes, using the last 300 FiO2 values that meet specific conditions. The average of these values is limited to ±10% of the current baseFiO2 and then again averaged with the current baseFiO2. 37

37 2 Automated oxygen control in preterm infants, how does it work and what to expect Clinical effect CLiO2 is the most researched algorithm in clinical setting. 18, 28-33 All seven studies reported a significant increase in time spent within TR, including a reduction of28-30, 32, 33 or equal18, 31 time spent above it (Table 1). Three studies demonstrated a small but significant increase of mild hypoxia.28, 29, 33 But, in the largest multicentre trial31 the gain in TR time was mainly attributed to less hypoxia, rather than hyperoxia. There was heterogeneity in sample size, study duration and modes of respiratory support in the studies. The TR upper and lower limits also varied, this being relevant for the calculation of SpO2 error, thus influencing algorithm function. The CLiO2 algorithm has thus been seen to be superior to manual oxygen titration in a range of situations, including when used for longer periods of time as shown by van Zanten et al.33 In a recent meta-analysis38 on AOC, the effect on SpO 2 targeting was somewhat diminished when only including the CLiO2 studies compared to when all studies were included (8.9% vs 12.8% increase in time within TR). In our experience33, the CLiO 2 algorithm performs well in both stable and unstable preterm infants. However, there appears to be room for improvement, especially regarding the rapidity of FiO2 reduction in the event of clinical improvement, for example after surfactant therapy. The algorithm needs substantial time to adjust baseFiO2 downwards in this circumstance, which poses a problem as most FiO2 adjustments are limited to be ≥ baseFiO2. In addition, in unstable babies with frequent hypoxia the baseFiO2 does not reflect the basal oxygen requirement. AOC with the CLiO2 algorithm has been implemented in the NICU in Leiden University Medical Centre since mid-2015. A practical solution for the baseFiO2 problem has been to reset the algorithm when needed, which adopts the clinician-set FiO2 as baseFiO2. IntellO2 35 How it works The IntellO2 algorithm is a recent PID algorithm implemented in the Vapotherm Precision Flow device for delivering nasal high flow.35 Details on the algorithm are limited. It is reported to be a modified version of an earlier algorithm described by Bhutani.9 As with other algorithms, SpO 2 is measured by a built-in pulse-oximeter. If the SpO2 is lost for two minutes or is degraded by more than 50% in the last four minutes, the algorithm reverts to the back-up FiO2 value (the highest of a clinician-set back-up value or the median of the last three FiO2 values).

38 Chapter 2 Clinical effect Reynolds et al. reported an increase of 31% more time within TR (80% during automated, 49% during manual; P < 0.0001) in 30 preterm infants (Table 1). This was accompanied by less time in hypoxia and hyperoxia, albeit with more hyperoxic episodes in the automated arm. These so-called overshoot episodes are common39, and perhaps impossible to prevent. Predictive Intelligent Control of Oxygenation19 How it works PRICO (Predictive Intelligent Control of Oxygenation) is a rule-based algorithm originally designed for the delivery room. It is now available for NICU use in the Fabian ventilator (Acutronic, Hirzel, Switzerland). The algorithm uses the current SpO2, its trend, and a prediction of what SpO2 will become to make step-wise adjustments. Comparable to CLAC, each adjustment is followed by a pause of at least 30 seconds. The set FiO2 is limited to an adjustment range which is specified by the caregiver. If the limits of this range are met, PRICO will alarm and pause until the issue is resolved. Information on how exactly the algorithm operates is limited. Whilst in TR stepwise adjustments are limited to ±1%, whereas outside the TR adjustments vary from ±1-10%. Large, swift changes are recognised using the SpO2 trend and are used to fine tune the magnitude of the FiO2 adjustments. A prediction based on this trend is used to limit possible under/overshoots. Before an adjustment safety checks are performed (reliability of the connection, assessment of the correctness of parameters).19, 40 There is currently not enough data available to give a complete appraisal of this algorithm. Clinical effect So far, for PRICO there is no clinical data available to date, feasibility has only been tested on preterm lambs while using volume guarantee ventilation.19 The time spent within target range was significantly higher with AOC (93.2% vs 84.0%, P < 0.05). A 30 second lockout could potentially delay appropriate intervention against hypoxia and hyperoxia. SPO2C34 How it works The SPO2C algorithmwas developed in Ulm, Germany and is commercially available

39 2 Automated oxygen control in preterm infants, how does it work and what to expect in the Sophie ventilator (Stephan GmbH, Gackenbach, Germany). Information on the workings of the algorithm is limited.34 The control loop iterates every two seconds, conveniently using the SpO2 measured by an existing bedside monitor. Separate PID controllers update the set FiO2 and the baseFiO2. The primary controller adjusting set FiO2 unconventionally uses variable PID coefficients which depend on the range in which SpO2 falls and the speed and direction of the change. In an attempt to account for the shape of the oxygendissociation curve, the proportional term is exponentially weighted. The second PID loop updates the baseFiO2 every 5 minutes by comparing recent values for set FiO2 to the previous baseFiO2. Clinical effect Although the data was only recently published34 the algorithm was clinically tested in 2014 in a randomised crossover trial (Table 1). The algorithm proved superior to manual titration in twelve preterm infants on non-invasive respiratory support (manual titration: 68.5% within TR; SPO2C controller: 77.8%; p<0.001). VDL 1.121 How it works The VDL 1.1 algorithm, an adaptive PID algorithm, is available as the Oxygenie option on the SLE6000 ventilator (SLE, Croydon, United Kingdom). Like other algorithms it uses the baseline oxygen requirement, known as reference FiO2, which is calculated every 30 minutes using the preceding 60 minutes of data. Combining this value and the P, I and D terms the algorithm calculates a new value for the set FiO2 each second, rounded to 0.5%. Measurements with low signal IQ are labelled as missing, in which case the last set value for FiO2 is used. The algorithm does not use a lock-out period. To account for known pathophysiological idiosyncrasies, as well as the limitations of pulse oximeters, each of the PID terms is modified in different ways. The proportional term is adapted to the degree of lung dysfunction by multiplying with 0.5-1, corresponding to a reference FiO2 in the range of 21%-40%. Furthermore, the error in the proportional term is attenuated whilst SpO2 is in TR to minimise adjustments during good control.21, 36, 41 Finally, to correct for the increasing imprecision of pulse oximetry at SpO2 values below 80%42, the negative error is limited to 13%. During protracted hypoxia, the integrand (sum of past errors) will rapidly increase in