Supplemental Oxygen Strategies in children with BronchoPulmonary Dysplasia (the SOS BPD study): statistical analysis plan for a multicenter randomized controlled trial Authors: Rosemarie de Ridder, Maruschka P Merkus, Elianne J L E Vrijlandt, Irwin K M Reiss, Anton H van Kaam, Mariëlle W Pijnenburg, Wes Onland, SOS BPD study group Published: 2026 ABSTRACT AbstractBackgroundSupplemental oxygen remains the primary treatment for preterm infants with established bronchopulmonary dysplasia (BPD). Although randomized controlled trials have evaluated lower versus higher oxygen saturation targets in preterm infants before 36 weeks postmenstrual age, evidence beyond this age is lacking. The aim of the Supplemental Oxygen Strategies in children with BronchoPulmonary Dysplasia (SOS BPD) study is to evaluate the effect of a higher oxygen saturation lower limit of 95%, compared with a lower limit of 90%, on lung growth by proxy of weight standard deviation score (SDS), at 6 months corrected age in infants with established BPD. This document outlines the statistical analysis plan for the SOS-BPD study.Methods/designThe SOS BPD study is a multicenter, pragmatic, non-blinded, randomized trial in preterm infants with moderate or severe BPD. Eligible infants are randomized between 36 and 38 weeks postmenstrual age to a lower oxygen saturation limit of 95% or 90%, after parental informed consent. The assigned oxygen saturation lower limit is maintained until respiratory support can be discontinued. The primary outcome is weight SDS at 6 months corrected age. Secondary outcomes include anthropometrics measured at 6 and 12 months corrected age, rehospitalizations for respiratory reasons and respiratory complaints. This statistical analysis plan is written and submitted without knowledge of the data.Trial registrationOverview of Medical Research in the Netherlands (OMON), NL-OMON54694. Registered on 09-08-2018, URL: https://onderzoekmetmensen.nl/en/trial/54694.Keywords: Prematurity, Bronchopulmonary dysplasia, Neonatal intensive care unit, Supplemental oxygen, Oxygen saturation limit, Statistical analysis plan Background Bronchopulmonary dysplasia (BPD) is a major cause of morbidity in very or extremely preterm-born infants [1]. BPD is associated with long-term respiratory complications, such as recurrent respiratory hospitalizations, reduced lung function, and neurodevelopmental impairment [2, 3]. Supplemental oxygen is the most important treatment for infants with developing or established BPD. Multiple randomized controlled trials (RCTs) have evaluated lower versus higher oxygen saturation (SpO2) targets in preterm infants before 36 weeks postmenstrual age (PMA). These studies have shown that targeting a higher SpO2 range of 90–95% reduces the risk of death and necrotizing enterocolitis compared with a lower range of 85–89% [4–7]. No significant differences were observed in major disability including adverse neurodevelopmental outcomes, severe hearing loss, or blindness. In contrast, evidence guiding optimal SpO2 targets or SpO2 lower limit beyond 36 weeks PMA in infants with established BPD is lacking, resulting in wide variation in clinical practice. Several hypothetical reasons might prevent direct extrapolation of the results of the oxygen targeting RCTs before 36 weeks PMA to infants with established BPD approaching the term age. First, infants with established BPD are in a different stage of lung development, characterized by ongoing alveolarization, which may require different oxygen levels [8]. Second, compared with the very preterm infants included in the oxygen saturation target RCTs before 36 weeks PMA, infants with established BPD may be less susceptible to oxidative stress due to more matured antioxidant defenses [4, 9]. Finally, the pulmonary vasculature undergoes differentiation during the different stages of lung development, and structural and functional changes in the pulmonary vasculature associated with BPD may alter oxygen requirements [10]. Targeting a higher SpO2 lower limit may help prevent pulmonary vasoconstriction, potentially reducing pulmonary hypertension in infants with BPD [11]. Optimal oxygenation during early life is crucial for postnatal lung development. Hypoxemia may impair cell proliferation and angiogenesis, whereas hyperoxemia can induce oxidative stress and disrupt signaling pathways involved in secondary septation [12]. This underscores the importance of precise oxygen targeting to support lung growth while minimizing the risk of lung injury. Direct measurement of lung growth in infancy is limited by feasibility and the invasiveness of physiological assessments. Somatic growth, particularly weight and length/height gain, can serve as a practical, indirect marker of pulmonary development [13–15]. Adequate lung growth is closely intertwined with overall growth, as improved respiratory function reduces metabolic stress, enhances feeding tolerance, and supports better energy utilization [16]. Conversely, impaired pulmonary development may lead to growth faltering due to increased respiratory workload and limited nutritional reserves [17]. Given these considerations, the Supplemental Oxygen Strategies in children with BronchoPulmonary Dysplasia after the Neonatal Intensive Care Unit (SOS BPD) study is a multicenter RCT investigating the effects of a higher SpO2 lower limit (≥ 95%) compared with a lower SpO2 lower limit (≥ 90%) on body weight at 6 months corrected age (CA), as a surrogate marker for lung growth, in infants with moderate to severe BPD. The study protocol was published previously [18]. This paper describes the statistical analysis plan (SAP) and was written and submitted without knowledge of the data. Objective The primary aim of the SOS BPD study is to investigate whether targeting a higher SpO2 limit (lower limit of 95%) leads to superior growth of lung tissue, assessed indirectly by body weight standard deviation score (SDS) at 6 months CA (primary outcome) compared with targeting a lower SpO2 limit (lower limit of 90%) in infants with moderate or severe BPD diagnosed at 36 weeks PMA. Methods/design Design and setting The SOS-BPD study is a multicenter, pragmatic, non-blinded, two-group, parallel, superiority RCT in preterm infants diagnosed with moderate or severe BPD between 36 and 38 weeks PMA. The trial is conducted in 30 hospitals in the Netherlands, including nine of the ten Dutch neonatal intensive care units (NICUs) and 21 post-intensive care (post-IC)/high-care units. The Dutch healthcare system is referral based, meaning that clinically stable infants on non-invasive respiratory support are transferred from the NICU to a post-IC/high-care unit once they reach a minimum PMA of 30 weeks and a body weight of at least 1000 g. Study protocol development and conduct The SOS BPD study is registered in the Dutch Trial Register (NL7149/NTR7347; registered on 13 March 2019). This study is approved by the Medical Ethics Review Committee of the Erasmus University Medical Center in Rotterdam, The Netherlands (MEC-2018–1515). The local ethics committee of the participating hospitals approved the local feasibility of the study protocol. During the study, the ethics committee approved 3 substantial amendments. On February 1 st 2021, the acceptable time spent below the assigned SpO2 threshold was increased from 5 to 10% based on the feedback of the participating centers that nearly all infants in both randomization groups required intensification of respiratory support. Second, acquisition of SpO2 data from central patient monitors was approved for centers with the technical capability. For all other centers, a dedicated study pulse oximeter was used. Third, the study protocol was also amended to allow included participants in a participating center to be transferred to a non-participating hospital. In these cases, the local team was asked to adhere to the assigned SpO2 lower limit, but no SpO2 profiles were performed. In addition, on July 26th 2021, the Data Safety Monitoring Board (DSMB) charter was updated and retinopathy of prematurity (ROP) was considered a clinical significant safety outcome if the ROP progressed to need for treatment after randomization. The trial was conducted according to the principles of the Declaration of Helsinki, good clinical practice (GCP) guidelines and Dutch legislation regarding medical research involving human subjects [19, 20]. Written informed consent from parents or legal guardians of the participating infants was obtained prior to randomization of the participants. An independent DSMB monitored the progress and safety of the study, with a special focus on ROP. All participating sites were monitored by an independent Clinical Research Associate of the Neonatology Network Netherlands foundation (Amsterdam, the Netherlands). The trial will be reported according to the Consolidated Standards of Reporting Trials-Children (CONSORT-C) guidelines [21]. The inclusion and exclusion criteria are described in detail in the previously published study protocol [18]. In short, preterm infants born before 32 weeks of gestation and diagnosed with moderate or severe BPD were included between 36 and 38 weeks PMA. Moderate or severe BPD was defined using the 2001-National Institute of Child Health and Human Development (NIH) definition of Jobe and Bancalari [22]. Infants were categorized as moderate or severe BPD when needing supplemental oxygen and/or positive pressure ventilatory support at 36 weeks PMA. The oxygen reduction test was used according to Walsh [23]. Included infants were randomized to either a SpO2 lower limit of 95% or a SpO2 lower limit of 90%. For urgent medical reasons (e.g., pulmonary hypertension, ROP), healthcare providers were allowed to increase or decrease the assigned SpO2 lower limit. Randomization and data collection Eligible infants were randomly allocated in a 1:1 ratio to either a SpO2 lower limit of 95% or a SpO2 lower limit of 90%, stratified by 10 NICU hospital clusters and BPD severity (moderate or severe) [18]. Randomization was performed using randomly permuted blocks with varying block sizes (4, 6, or 8). The randomization was conducted using Castor Electronic Data Capture (Castor EDC), version 2025.2.1.0 (Castor, Ciwit BV, Amsterdam, the Netherlands) [24]. Randomization was performed by the treating physician or research nurse. The infants’ parents and treating physicians were not blinded to the treatment allocation. Data management was implemented according to GCP guidelines. All patient data were entered via Castor EDC, version 2025.2.1.0 (Castor, Ciwit BV, Amsterdam, the Netherlands), a GCP proof web-based database [24]. Collected baseline characteristics of the included patients are shown in Table 1. Table 1Baseline maternal and patients’ characteristics of randomised patientsSpO2 lower limit of 90% (n =)SpO2 lower limit of 95% (n =)Maternal characteristics Maternal age [years] Ethnicity Caucasian/European Mediterrean Afro-American Latin-American Other Unknown Pre-eclampsia Chorioamnionitis* Any antenatal corticosteroids Multiple birthsPatient characteristics Gestational age [weeks] Birth weight [grams] SGA** Male sex Duration of MV [days] Duration of NIPPV/nCPAP [days] Duration of nHF/nLF [days] Supplemental oxygen [days] Postnatal corticosteroids PDA needing medication BPD-PH NEC grade ≥2, Sepsis* IVH grade ≥2 ROP grade ≥2 ROP treatment Moderate BPD at 36 weeks PMA Severe BPD at 36 weeks PMA PMA at enrolment [weeks] Respiratory support at enrolment IMV NIPPV or nCPAP nHF ≥ 2 L/min nLF < 2 L/min Anthropometrics at enrolment Weight [grams] [SDS] Length [cm] [SDS] Head circumference [cm] [SDS]Data is presented as mean (± standard deviation) or median [interquartile ranges]. Categorical variables are presented as counts (percentage). *Chorioamnionitis and sepsis defined as reported by the local NICU or hospital. **SGA defined as birthweight <3th percentile of Dutch reference curve. SpO2 oxygen saturation, SGA small for gestational age. IMV invasivemechanical ventilation, NIPPV nasal intermittent positive pressure ventilation, nCPAP nasal continuous positive airway pressure, nHF nasal high flow, nLF nasal low flow, PDA patent ductus arteriosus, BPD-PH bronchopulmonary dysplasia associated pulmonary hypertension, NEC necrotizing enterocolitis, IVH intraventricular haemorrhage, ROP retinopathy of prematurity, BPD bronchopulmonary dysplasia, PMA postmenstrual age, SDS standard deviation score, cm centimeters Baseline maternal and patients’ characteristics of randomised patients Data is presented as mean (± standard deviation) or median [interquartile ranges]. Categorical variables are presented as counts (percentage). *Chorioamnionitis and sepsis defined as reported by the local NICU or hospital. **SGA defined as birthweight <3th percentile of Dutch reference curve. SpO2 oxygen saturation, SGA small for gestational age. IMV invasivemechanical ventilation, NIPPV nasal intermittent positive pressure ventilation, nCPAP nasal continuous positive airway pressure, nHF nasal high flow, nLF nasal low flow, PDA patent ductus arteriosus, BPD-PH bronchopulmonary dysplasia associated pulmonary hypertension, NEC necrotizing enterocolitis, IVH intraventricular haemorrhage, ROP retinopathy of prematurity, BPD bronchopulmonary dysplasia, PMA postmenstrual age, SDS standard deviation score, cm centimeters Intervention and study procedures Supplemental oxygen and/or respiratory support were weaned based on the assigned SpO2 lower limit. Adherence to the SpO2 limit was assessed by extracting SpO2 profiles twice weekly for 24 h from pulse oximeters, until respiratory support was stopped. If preterm infants were discharged home on supplemental oxygen, parents recorded a 24-h SpO2 profile once weekly. The study procedures are described previously [18]. If the time spent below the assigned SpO2 lower limit was ≤ 10% the research team would suggest to further wean the respiratory support of the infant; if the time spent below the assigned SpO2 was > 10%, increase of respiratory support was advised. If the SpO₂ profile did not indicate a need for either escalation or weaning, clinical management remained unchanged. Weaning of respiratory support was performed according to local protocols and the SpO2 profiles only served as a supportive tool. The allocated SpO2 lower limit was maintained from study entry until two weeks after discontinuation of the respiratory support. Primary outcome The primary outcome is body weight SDS at 6 months CA, as a surrogate marker for lung growth (Table 2) [13–15]. SDS values are calculated using Dutch reference standards from TNO (Netherlands Organisation for Applied Scientific Research – innovation for life) [25], and are corrected for age at measurement and sex. The measurement is conducted during the routine follow-up visit at 6 months CA. Table 2Outcome parameters and statistical analysisVariable/outcomeType of outcomeStatistical analysisPrimary Weight SDS at 6 months CALinear mixed model with random intercept for hospital cluster and fixed effects for treatment group, BPD severity stratification and SDS weight at randomization* plus sensitivity and sub-group analysesSecondary Time to supplemental oxygen independenceKaplan–Meier plot, log-rank test;Cox mixed effects model with random intercept for hospital cluster and fixed effects for treatment group and BPD severity stratification, where possible* Time to hospital discharge PMA at discontinuation of supplemental oxygenLinear mixed effects model with random intercept for hospital cluster and fixed effects for treatment group, BPD severity stratification and the baseline value of the respective outcome, where possible* PMA at discharge Need for home oxygenLogistic mixed effects model with random intercept for hospital cluster and fixed effects for treatment group and BPD severity stratification, where possible*If the number of (non)events is small, Fisher’s exact test will be performed BPD-PH at discharge New onset of ROP grade > 2 or ROP needing treatment Use of caffeine at discharge Use of diuretics at discharge Use of bronchodilators or inhalation corticosteroids at discharge Use of sildenafil or bosentan at dischargeSecondary at 6 and 12 months CA MortalityKaplan–Meier plot, log-rank testIf the number of events allow, a Cox mixed effects model with random intercept for hospital cluster and fixed effects for treatment group and BPD severity stratification* Anthropometrics (weight (12 months CA), length and head circumference SDS)Linear mixed effects model with random intercept for hospital cluster and fixed effects for treatment group, BPD severity stratification and the baseline value of the respective outcome, where possible* Need for home oxygenLogistic mixed effects model with random intercept for hospital cluster and fixed effects for treatment group and BPD severity stratification, where possible*. If the number of (non)events is small Fisher’s Exact test will be performed BPD-PH Use of diuretics Use of bronchodilators or inhalation corticosteroids Use of sildenafil or bosentan Used antibiotics for respiratory reason Used systemic steroids for respiratory reason Any respiratory symptom (wheezing, coughing for > 3 days, dyspnea, colds or exercise intolerance) Any respiratory hospital readmission since first discharge to home Any respiratory hospital ICU readmission since first discharge to homeSafety outcomes New onset of ROP grade > 2 or ROP needing treatmentSee above Serious adverse eventsDescriptive statisticsSDS standard deviation score, CA corrected age, PMA post menstrual age, BPD bronchopulmonary dysplasia, BPD-PH bronchopulmonary dysplasia-related pulmonary hypertension*Model assumptions will be assessed by goodness-of-fit testing, as appropriate. If required, continuous outcomes will be appropriately transformed. If a model faces convergence issues due to small numbers in some randomization strata, a simplified model will be used excludaing the respective randomization stratification factor(s) Outcome parameters and statistical analysis Logistic mixed effects model with random intercept for hospital cluster and fixed effects for treatment group and BPD severity stratification, where possible* If the number of (non)events is small, Fisher’s exact test will be performed Kaplan–Meier plot, log-rank test If the number of events allow, a Cox mixed effects model with random intercept for hospital cluster and fixed effects for treatment group and BPD severity stratification* SDS standard deviation score, CA corrected age, PMA post menstrual age, BPD bronchopulmonary dysplasia, BPD-PH bronchopulmonary dysplasia-related pulmonary hypertension *Model assumptions will be assessed by goodness-of-fit testing, as appropriate. If required, continuous outcomes will be appropriately transformed. If a model faces convergence issues due to small numbers in some randomization strata, a simplified model will be used excludaing the respective randomization stratification factor(s) Secondary outcomes Secondary outcomes include mortality (6 and 12 months CA), time to supplemental oxygen independence and PMA at discontinuation of supplemental oxygen, time to hospital discharge, need for home oxygen (at discharge, 6 and 12 months CA), BPD-related pulmonary hypertension (at discharge, 6 and 12 months CA), new-onset ROP grade > 2 or ROP treatment [26], use of respiratory medication (at discharge, 6 and 12 months CA), and use of sildenafil or bosentan (at discharge, 6 and 12 months CA) (Table 2). Other secondary outcomes at 6 and 12 months CA are weight (12 months), head circumference, and length SDS scores (6 and 12 months), used antibiotics or systemic steroids for respiratory reason during the previous 6 months, respiratory symptoms (defined as parent-reported wheezing, coughing for > 3 days, dyspnea, colds or exercise intolerance during the previous 6 months), and respiratory (ICU) readmissions. The quality of life (QOL) of caregivers (assessed with the CarerQol), unscheduled healthcare visits, and cost-effectiveness will be included in the health economic analysis and reported separately. Infant temperament (IBQ-r) was previously listed as a secondary outcome in the study protocol, however, due to the high burden of questionnaires to the parents, we decided to stop collecting this outcome after 25% of the patients were included. Additional secondary outcomes: predefined substudy Following center-specific assessments, additional secondary outcomes at 6 months CA will be collected in a subset of infants: lung structure on chest computed tomography, signs of pulmonary hypertension and/or right ventricular systolic function on echocardiography, and indices of overnight polysomnography. These assessments are part of routine follow-up in some centers and will be analyzed as a pre-defined substudy. The outcomes of these analyses will be reported in a separate paper. Analyses will be exploratory and hypothesis-generating due to limited sample size. Safety outcomes The safety outcomes were classified as follows (Table 2): New-onset ROP grade > 2 New-onset ROP grade > 2 Deterioration of ROP to needing treatment Deterioration of ROP to needing treatment Serious adverse events (SAEs). The definitions of the SAEs are described in the protocol [18]. Serious adverse events (SAEs). The definitions of the SAEs are described in the protocol [18]. Statistical methods specified in the study protocol Sample size calculation As described in the study protocol, no previous data are currently available on the effect of high versus low SpO2 lower limit thresholds on weight in infants with established BPD [18]. Therefore, we performed a simulation study with 4 scenarios to estimate the required sample size to obtain a statistically significant, clinically relevant difference in weight SDS at 6 months CA (primary outcome) between the two groups. A mixed effects model with a random intercept for hospital cluster was assumed to account for the correlation between patients treated in the same hospital cluster. The model structure reflects the design of the study, with clustering by NICU center (NICU with affiliated post-IC or high-care units). We assumed a total of 10 hospital clusters, each contributing approximately 16 (± 3) to 18 (± 3) patients. The mean weight at 6 months CA for infants randomized to the group with a SpO2 lower limit of 90% was assumed − 1.15 SDS (data from BPD cohort Erasmus MC, Rotterdam, the Netherlands, data on file). A 0.5 higher SDS was considered clinically relevant. Therefore, the mean weight at 6 months CA for infants randomized to the group with a SpO2 lower limit of 95% was assumed − 0.65 SD. The variation in weight between individuals was set at 1.18 SDS. The SD variation of weight within NICU clusters was varied between 0.10 and 0.20 SDS across 4 simulation scenarios, to reflect potential differences in local practices or patient populations. All 4 simulated scenarios are reported in Table 3. Among these, the scenario with the largest cluster-level variance (0.20 SDS) resulted in a statistical power of 0.83, given a two-sided significance level of 0.05. This scenario was deemed the most conservative and was selected for the final sample size determination. Based on this scenario, the required sample size was 180 infants (90 per group). To account for an anticipated dropout rate of 10%, the target enrollment was set at 99 infants per arm, resulting in a total of 198 infants to be randomized. Sample size calculations were performed using Rstudio, version 1.2 (R Foundation for Statistical Computing, Vienna, Austria, packages knitr, MASS, nlme). Table 3Simulated scenarios for sample size calculationScenario 1Scenario 2Scenario 3Scenario 4Power0.720.820.730.83Variation of weight (hospital)0.100.100.200.20Total patients160180160180 Simulated scenarios for sample size calculation Original proposed analyses The originally proposed analyses are described in the published study protocol [18]. The effect of a higher SpO2 lower limit ≥ 95% on weight SDS at 6 months CA, compared with a lower SpO2 lower limit ≥ 90%, will be presented as the estimated mean difference (MD) with corresponding 95% confidence intervals (CI), and evaluated using a linear mixed effects model. This model will include a random intercept to account for the correlation between patients born in the same hospital cluster and a fixed effect for BPD severity and weight SDS at randomization, since these variables are considered relevant for the primary outcome. All analyses are based on the intention to treat (ITT) principle. A p-value of < 0.05 is considered significant. In this paper, we present our final and more detailed statistical analysis plan. Interim analyses and safety reporting A formal DSMB was installed to protect the safety of the patients, and to assist and advise the trial's Steering Committee in ensuring its safety, validity, and credibility. The DSMB included an independent pharmacist and researcher, neonatologist, and biostatistician. All members had no competing interests and were not involved in the trial. The DSMB charter (version 3, 26-07−2021) described the composition, responsibilities, tasks and working procedures (available on request). Safety interim analyses were performed after approximately 25%, 50%, and 75% of the patients reached the end of the follow-up at 12 months CA. Prior to the planned interim analyses, the DSMB received a report from the trial coordinator including patient recruitment status, drop-outs, serious adverse events (SAEs), and the predefined safety endpoint of new-onset ROP grade > 2 or deteriorating of ROP to needing treatment after randomization. In principle, the trial was not to be stopped early for a beneficial effect on the primary outcome. Also, futility was not assessed. Consequently, no alpha spending adjustment was applied for the interim analyses. Early stopping for safety was considered if an absolute difference of at least 20% in new or deteriorating ROP was detected between the groups, and if this difference reached a two-sided p-value of < 0.05 in unblinded analyses. Statistical analysis plan Overall principles The first phase of data analysis will commence once all in-hospital and 6 months’ CA follow-up data for the last enrolled patient have been collected, the database has been cleaned and locked, and the present SAP has been submitted for publication. The second phase of data analysis will include all outcomes up to 12 months CA and will be performed once the 12 months’ CA follow-up data for the last enrolled patient have been collected, and the database has been cleaned and fully locked. All analyses will be conducted by RR, in consultation with MM and a biostatistician. Analyses will be performed according to the ITT principle. Given the possibility of intervention cross-over and protocol deviations, an additional analysis of the primary outcome will be conducted in the per protocol (PP) population to evaluate robustness of the findings. Descriptive statistics will be used to summarize baseline characteristics and outcome parameters using the mean and standard deviation or the median and interquartile range for continuous, normally and non-normally distributed outcomes respectively. Categorical outcomes will be summarized using counts and percentages. Treatment effect estimates will be expressed in absolute and relative effect sizes, as appropriate. For all treatment effect estimates, 95% CIs will be reported, and p-values < 0.05 will be considered statistically significant. Secondary analyses, although pre-specified, will be considered exploratory, so no formal adjustments for multiplicity of statistical testing will be made. Additional ad hoc analyses motivated by new information stemming from this clinical trial or other outside research may also be performed. Any deviations from this plan will be clearly documented and justified in the study report. All statistical analyses will be performed in RStudio, version 4.4.3 (R Foundation for Statistical Computing, Vienna, Austria) and IBM SPSS Statistics for Windows, Version 25.0 (IBM Corp., Armonk, NY, USA). The results of the analyses of the primary and secondary outcomes up to 12 months CA will be published together in a peer-reviewed journal. Withdrawal As previously described in the study protocol [18], parents or caregivers could withdraw their child from study participation at any time and for any reason, without consequences. All reasons for withdrawal were documented in the study records. However, all patients continued to be followed as part of the national routine clinical neonatology follow-up program for preterm infants, regardless of study withdrawal. Parental consent was obtained to use follow-up data after discontinuation at parents’ request. Handling of missing data If data are missing, every effort will be made to retrieve data. Both NICUs and post-IC/high-care hospitals will be contacted, as many (transferred) infants were included in the study in a post-IC/high-care unit. Because all preterm infants routinely visit the neonatal follow-up clinic, we expect minimal missing data for the primary outcome. All cases lost to follow-up will be documented, including the patient characteristics, in-hospital outcomes and the reason for loss to follow-up. In the unlikely event that more than 10% of the primary outcome data are missing, missing data will be imputed. Multiple imputed datasets will be generated from an imputation model that includes the variables from the primary analysis model and ancillary variables that are thought to explain the missingness or predict weight SDS at 6 months CA. Definition of analysis sets Intention-to-treat population The intention-to-treat (ITT) population includes all randomized patients, regardless of protocol deviations. Infants who withdrew from the study and did not provide consent for use of data will be excluded. All patients will be analyzed according to the treatment group they were assigned to. Per-protocol population The per-protocol (PP) population includes all patients who received treatment according to randomized group assignment and the study protocol, thereby excluding infants with major protocol deviations (see below and Table 4). Table 4Definition of population analysis setsAnalysis populationRandomization armIntention-to-treatIncluding all protocol deviationsPer-protocol• Excluding deviations from the eligibility criteria• Excluding infants with withdrawal of study intervention• Excluding non-adherence to the assigned SpO2 lower limit (e.g., intervention crossover based on medical reasons or other causes, a median SpO2 of < 90% or < 95%, among all saturation profiles made during the intervention period)SpO2 oxygen saturation Definition of population analysis sets • Excluding deviations from the eligibility criteria • Excluding infants with withdrawal of study intervention • Excluding non-adherence to the assigned SpO2 lower limit (e.g., intervention crossover based on medical reasons or other causes, a median SpO2 of < 90% or < 95%, among all saturation profiles made during the intervention period) SpO2 oxygen saturation Statistical analyses Patient flow The flow of participants is displayed in the CONSORT flow diagram (Fig. 1). Reasons why infants are not eligible or why infants are excluded will be summarized. We will collect patient characteristics from eligible infants that are not included in this study. These include gestational age, birth weight, gender, and BPD severity. Consolidated Standards of Reporting Trials (CONSORT) 2010 flow diagram. SpO2, oxygen saturation Protocol deviations Major protocol deviations are defined as deviations from the eligibility criteria, non-compliance with the assigned SpO2 lower limit, a median SpO2 < 90% or < 95% across all SpO2 profiles per patient (depending on allocation), or premature withdrawal from the study intervention (Table 4). All major protocol deviations will be line-listed according to treatment allocation. Baseline characteristics We will present maternal and patient characteristics of the randomized infants. The following maternal characteristics will be presented: maternal age; ethnicity; chorioamnionitis; pre-eclampsia; any antenatal corticosteroids; and multiple births (Table 1). The following patient characteristics at birth will be presented: gestational age; birth weight; small for gestational age (SGA) defined as birth weight < 3rd percentile of Dutch birth weight reference curve [27]; and sex. Next, we will present the medication use and comorbidities during NICU stay: duration of respiratory support (invasive mechanical ventilation (IMV), nasal intermittent positive airway pressure ventilation (NIPPV) and nasal continuous positive airway pressure (nCPAP), nasal high flow (nHF) and nasal low flow (nLF)), and supplemental oxygen; postnatal corticosteroids; persistent ductus arteriosus (PDA) needing medication; BPD-associated pulmonary hypertension; necrotizing enterocolitis (NEC) defined as Bell stage 2 or higher [28]; sepsis as reported by the NICU or regional hospital; intraventricular hemorrhage (IVH) stage 2 or higher [29]; and ROP grade 2 or higher and treatment [30]. Finally, we will present the characteristics at enrollment of the study: BPD severity; PMA; respiratory support (IMV, NIPPV or nCPAP, nHF or nLF), and anthropometrics. All characteristics and comorbidities will be reported as summary statistics according to allocation treatment of the trial. The characteristics of the SpO2 profiles will be presented in the supplementary material (e.g., number of SpO2 profiles, average SpO2 during all profiles, percentage of time spent below the randomized SpO2 lower limit, and time spent above SpO2 limit of 95%). Primary outcome Primary analysis Crude estimate with 95%CI of the mean difference in the primary outcome (weight SDS at 6 months CA) will be calculated between infants with the SpO2 lower limit of 95%, and those with lower SpO2 limit of 90%. The primary analysis of the effect of the SpO2 lower limit of 95% as compared with 90% on weight SDS at 6 months CA will be performed using a linear mixed effects model with random intercept for hospital cluster, and fixed effects for treatment group, randomization stratification factor BPD severity, and weight SDS at randomization (Table 2). Model assumptions will be assessed by goodness-of-fit testing, as appropriate. If required, appropriate transformation of the outcome will be applied. If the model faces convergence issues due to small numbers in some randomization strata, a simplified model will be used excluding the respective randomization stratification factor(s). Subgroup analyses We plan to perform pre-specified subgroup analyses for the primary outcome by examining treatment × subgroup interaction effects in the primary analysis model, if possible. We plan to perform five subgroup analyses, each examining one distinct subgroup: BPD severity (moderate or severe BPD); gestational age (less than or greater or equal to 27 weeks); small for gestational age (yes/no); multiple births (yes vs. no); and sex (male vs. female). Treatment effect estimates and associated 95% CIs will be reported in each subgroup, as well as a p-value for the test of the interaction term, if possible. Subgroup analyses, although pre-specified, should be interpreted as hypothesis-generating, and no adjustments for multiple comparisons will be made. The results of subgroup analyses will be interpreted taking into account accepted criteria for credible subgroup effects [31]. Sensitivity analyses To assess the robustness of the primary analysis of the primary outcome, the following sensitivity analyses are planned. First, to study the potential confounding effect of important biologically plausible risk factors for growth, parental smoking [32] and dietary intake [33] will be added as fixed effects to the primary analysis model, if possible. Second, a per-protocol analysis will be conducted, excluding infants with major protocol deviations (Table 4). This will allow evaluation of the potential impact of major protocol deviations on the estimated treatment effect. Third, analysis restricted to infants with at least 4 weeks of study data will be performed, in line with the post hoc analyses from a pilot RCT on oxygen saturation targeting in infants with BPD [34]. This study reported a higher incidence of intermittent hypoxic events in the lower SpO2 target group compared with the higher SpO2 target group in infants with at least 4 weeks of study data from randomization. Secondary outcomes The secondary outcomes are listed in Table 2. Secondary outcomes will be analyzed using mixed effects linear, logistic, and Cox regression models, as appropriate. These models will include a random intercept for hospital cluster and fixed effects for treatment group and randomization stratification factor BPD severity, and, if applicable, the baseline value of the respective outcome, where possible. If the number of (non)events is small, Fisher’s exact test will be performed for binary outcomes. Model assumptions will be assessed by goodness-of-fit testing, as appropriate. If required, continuous outcomes will be appropriately transformed. If a model faces convergence issues due to small numbers in some randomization strata, a simplified model will be used excluding the respective randomization stratification factor(s). Time to event outcomes will be presented as Kaplan–Meier plots and survival distributions of both study treatment groups will be compared with the log-rank test. Time to event will be calculated as the time between randomization and the event of interest, end of follow-up (censoring event), or in case of lost to follow-up, date of last contact (censoring event), whichever occurs first. Infants who died before reaching independence of SpO2 or discharge home will be considered to have failed the specific event (censoring event), and they will be assigned the longest observation time for the respective analysis. SAEs will be reported using descriptive statistics. Differences between the study protocol and statistical analysis plan In addition to the analyses specified in the published protocol [18], sensitivity and subgroup analyses were added for the primary outcome. We further specified the secondary outcomes and clarified the analyses. Trial status Thirty participating centers were initiated between January 2020 and January 2024. Four centers did not include patients. Due to COVID-19, recruitment was temporarily stopped between January 2020 and September 2020. Recruitment was completed at May 1 st 2025. The last follow-up at 12 months CA is expected around July 2026. Database lock and final analysis will be performed in two stages: the database will be partially locked once data collection, data monitoring, and data cleaning up to the 6 months CA visit of all included patients has been completed. Subsequently the first statistical analyses of these outcomes up to 6 months CA will be performed. Following completion, monitoring, and cleaning of the 12 months CA visits of all included patients, the database will be fully locked. Next, the statistical analysis of all outcomes at 12 months CA will be performed. All analyses of the primary and secondary outcomes will be performed after this statistical analysis plan is submitted for publication in a peer-reviewed journal and the database is (partially) locked. Acknowledgements We would like to thank the members of the DSMB of the SOS BPD study: Prof. dr. S Bambang Oetomo, paediatrician/neonatologist; Prof. dr. A H Maitland van der Zee, pharmacist/researcher; Prof. dr. ir. R Eijkemans, biostatistician; and Dr. D Cianci, biostatistician. In addition, we would like to thank Dr. J A Bogaards, biostatistician, for his advice for this statistical analysis plan. SOS-BPD study group members: S Balink MD, Department of Paediatrics, Amphia Hospital, P H Dijk MD, PhD, Department of Neonatology, UMCG, A A Kroon MD, PhD, Department of Neonatology, Erasmus MC, A J Sprij MD, Department of Paediatrics, Juliana Children’s Hospital, D H Nuytemans, Department of Neonatology, Amsterdam UMC, L E Wirsching-Smit, Department of Neonatology, Amsterdam UMC, A Hulsmann MD, Department of Neonatology, Amphia Hospital, J C A Klerk MD, Department of Neonatology, Maasstad Hospital, M A G van Scherpenzeel-de Vries MD, Department of Neonatology, Frisius Medical Centre, I A M Schiering MD, PhD, Spaarne Gasthuis, Department of Neonatology, G J Blok MD, PhD, Department of Neonatology, Northwest Hospital Group, R Geurtzen MD, PhD, Department of Neonatology, Radboud University Medical Centre, H D Buiter MD, Department of Neonatology, Martini Hospital, C E Counsilman MD, Department of Neonatology, Flevo hospital, C A Dalen Meurs MD, Department of Neonatology, Meander Medical Centre, A C M Dassel MD, Department of Neonatology, Deventer hospital, A M de Grauw MD, PhD, Department of Neonatology, Juliana Children’s Hospital, M E N van den Heuvel MD, Department of Neonatology, OLVG, J L A M van Hillegersberg MD, Department of Neonatology, Antonius Hospital, J C R van Hoften MD, Department of Paediatrics, Elisabeth Tweesteden Hospital, C H ten Hove MD, PhD, Department of Neonatology, Rijnstate Hospital, M de Jong MD, PhD, Department of Neonatology, Albert Schweitzer Hospital, A Kamerbeek MD, Department of Neonatology, Franciscus Hospital, A A M W van Kempen MD, PhD, Department of Neonatology, OLVG, J S von Lindern MD, Department of Neonatology, Groene Hart Hospital, R M J Moonen MD, PhD, Department of Neonatology, Zuyderland Hospital, E E M Mulder PhD, Department of Neonatology, Isala Medical Centre, H J Niemarkt MD, PhD, Department of Neonatology, Máxima MC, L G M van Rooij MD, PhD, Department of Neonatology, Medical Spectrum Twente, R N G B Tan MD, Department of Neonatology, Leiden University Medical Centre, K Van Mechelen MD, Department of Neonatology, Maastricht University Medical Centre. Consortia On behalf of the SOS BPD study group. Abbreviations Bronchopulmonary dysplasia Corrected age Confidence interval Consolidated Standards of Reporting Trials Data and safety monitoring board Good clinical practice Intensive care unit Intention to treat Invasive mechanical ventilation Intraventricular hemorrhage Nasal continuous airway pressure Necrotising enterocolitis Nasal high flow Neonatal intensive care unit National Institute of Child Health and Human Development Nasal intermittent positive airway pressure ventilation Nasal low flow Persistent ductus arteriosus Postmenstrual age Post-intensive care Per protocol Randomized controlled trial Retinopathy of prematurity Serious adverse event Statistical analysis plan Standard deviation score Small for gestational age Supplemental Oxygen Strategies in children with BronchoPulmonary dysplasia study Oxygen saturation Quality of life Authors' contributions RR, WO, MM, and MP were involved in the conception and design of the statistical analysis plan and drafted the manuscript. EV, AvK, AK, IR, PD, SB, and AS are member of the SOS-BPD steering committee. All SOS-BPD study group members mentioned in the acknowledgements are involved in the SOS-BPD study as local investigator at the participating centers. All authors read, edited, and approved the final manuscript. Funding This trial was supported by the Lung Foundation Netherlands under grant number 4.1.17.162 and the Netherlands Organisation for Health Research and Development (ZonMW) – Efficiency Studies Programme under grant number 843 002 927. Data availability No datasets were generated or analysed during the current study. Declarations Ethics approval and consent to participate This study was approved by the Medical Ethics Review Committee of the Erasmus University Medical Center, Rotterdam, The Netherlands (MEC-2018–1515), on 13 March 2019. Written consent to participate in the study was obtained from all parents or legal guardians of the participating infants prior to randomization. Consent for publication Not applicable. Competing interest The authors declare no competing interests. Footnotes Publisher's Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Contributor Information Wes Onland, Email: w.onland@amsterdamumc.nl. SOS BPD study group: S Balink, P H Dijk, A A Kroon, A J Sprij, D H Nuytemans, L E Wirsching-Smit, A Hulsmann, J C A Klerk, M A G van Scherpenzeel-de Vries, I A M Schiering, G J Blok, R Geurtzen, H D Buiter, C E Counsilman, C A Dalen Meurs, A C M Dassel, A M de Grauw, M E N van den Heuvel, J L A M van Hillegersberg, J C R van Hoften, C H ten Hove, M de Jong, A Kamerbeek, A A M W van Kempen, J S von Lindern, R M J Moonen, E E M Mulder, H J Niemarkt, L G M van Rooij, R N G B Tan, and K Van Mechelen References Associated Data Data Availability Statement No datasets were generated or analysed during the current study.