Intratracheal budesonide mixed with surfactant to increase survival free of bronchopulmonary dysplasia in extremely preterm infants: statistical analysis plan for the international, multicenter, randomized PLUSS trial

To determine whether intratracheal budesonide mixed with surfactant increases survival free of bronchopulmonary dysplasia (BPD) at 36 weeks’ postmenstrual age (PMA) in extremely preterm infants born before 28 weeks’ gestation.

To determine whether intratracheal budesonide mixed with surfactant:

Extremely preterm infants (specifically 22–27 weeks’ completed gestation) admitted to a participating neonatal intensive care unit (NICU), who fulfill the entry criteria detailed below.

The randomization schedule is provided by the Clinical Epidemiology and Biostatistics Unit at the Murdoch Children’s Research Institute, Melbourne, Australia. Randomization with balanced variable block sizes is used, stratified by study center, gestational age (22–25 weeks’ vs. 26–27 weeks’ completed gestation), prior surfactant therapy, and mode of respiratory support at randomization (mechanical ventilation via an endotracheal tube vs. non-invasive respiratory support).

When eligibility of an infant is confirmed, and prospective consent obtained, the infant is assigned to either receive surfactant plus budesonide, or surfactant alone, using a web-based randomization system with an allocation ratio of 1:1. A checklist on the website is used to confirm eligibility prior to randomization. Multiple births where more than one infant is eligible are randomized individually. A sealed opaque envelope at the study site is identified by the unique study ID generated from the web-based server (https://​redcap.​mcri.​edu.​au) [2, 3]. The sealed envelope is opened by a dedicated intervention team who are not providing direct clinical care to the infant and will not be involved in any future outcome assessments. Inside the main envelope are a further two sealed envelopes for the first and second interventions respectively. Infants remain in their allocated group for repeat interventions (if applicable), with each envelope remaining sealed until the intervention team is ready to prepare the allocated medication.

Parents/caregivers, direct healthcare providers, outcome assessors, data analysts, and trial investigators are blinded to the randomization group. Our experience is that it is virtually impossible to distinguish between the control and intervention study drugs, although there theoretically may be a subtle difference in the appearance of the surfactant with budesonide admixed. Additionally, the volume of the study drug to be administered is 0.5 mL/kg greater in the active treatment arm. To maintain blinding, the study drugs will be prepared by an independent intervention team whose members are not directly involved in the clinical care of the infant, and not involved in data collection or outcome assessments for the study. Data on the dose and type of intervention, as well as other data required by hospital pharmacies, will be recorded by the intervention team on allocation cards and stored in a secure lockbox only accessible by hospital pharmacists. In addition, after preparation of the intervention, the contents of the syringe will be covered using a stick-on label to obscure the volume and appearance to the bedside clinical staff. The pharmacy departments of each participating center and the Clinical Epidemiology and Biostatistics Unit will be the only other personnel aware of the allocated study intervention; they also will not be involved in data collection or outcome assessments for the study. Pharmacies will maintain a logbook of allocated study drugs and doses.

Neither the PLUSS Trial Steering Committee nor site researchers will be aware of the allocated interventions and will not be permitted access to this information until trial completion.

The estimated incidence of the composite primary outcome of survival free of BPD is 50%, based on a review of data from the lead center (The Royal Women’s Hospital, Melbourne, Australia) and published studies enrolling extremely preterm infants. With a sample size of 1038 infants (519 in each group), the study has 90% power to detect a relative increase in survival free of BPD of 20% (an absolute increase of 10%), from the anticipated event rate of 50% in the control arm to 60% in the intervention (budesonide) arm, alpha error 0.05. To allow for up to 2% study withdrawals or losses to follow-up, PLUSS aimed to enroll a total of 1060 infants (530 in each arm).

Full details of the study procedures are presented in the published Study Protocol [1].

The intervention will be performed in participating tertiary NICUs with the medications prepared in the NICU (or potentially in the delivery room in some centers) after the birth weight of the infant has been confirmed. Following randomization, the first intervention will be administered as soon as possible. If the infant meets the same treatment criteria 6–12 h after the first intervention, a second (and final) intervention will be administered in the NICU.

In this pragmatic study, the following methods of intratracheal instillation will be permitted: standard bolus administration through an endotracheal tube that will remain in situ with ongoing mechanical ventilation, INSURE (INtubate, SURfactant, Extubate) technique via an endotracheal tube, or via a thin catheter in those infants receiving non-invasive respiratory support (including CPAP, NIPPV or nasal high-flow).

Dosing is as follows:

This statistical analysis plan (SAP) is designed to cover the analysis of:

This SAP does not cover planned analysis of:

Data will be exported from the study database to Stata (StataCorp 2019. Stata Statistical Software: Release 18.0 College Station, TX: StataCorp LLC) for analysis.

All data will be checked and cleaned by the trial data manager and trial statistician prior to analysis. Data analysis will not commence until the database is locked and the SAP has been submitted for publication.

The analysis population will comprise at least 1038 infants (as per the sample size calculation) randomized to either receive the intervention (budesonide mixed with surfactant) or control (surfactant alone).

AEs and SAEs are assessed from randomization until death, primary hospital discharge, or 52 weeks’ PMA, whichever occurs sooner. AEs are:

In addition to the two SAEs above, all deaths before hospital discharge are defined as an SAE and are reviewed by the independent Data Safety Monitoring Board (DSMB) for relationship to the trial intervention and classified as a “respiratory death” or not.

There will be no adjustment for multiplicity.

The primary analysis will be based on an intention to treat population accounting for all infants randomized. If necessary, multiple imputation methods will be used for missing data; however, it is expected there will be very few instances in which the primary outcome cannot be determined. The reason for this is that infant death is always recorded and BPD assessment at 36 weeks will likely occur for nearly all infants when they are still in the hospital and their outcome can be recorded directly into the trial database. Therefore, we do not anticipate needing to address missing data for the primary or secondary outcomes in this study.

All infants at participating hospitals were screened for eligibility for the trial. The CONSORT flow diagram (Fig. 1) will be used to report enrolment, randomization, intervention allocation, follow-up, and analysis groups.

Baseline characteristics will be summarized by group as shown in Example Table 1.

The primary outcome is survival without physiological BPD at 36 weeks’ PMA.

Physiological BPD will be assessed between 36⁺⁰ and 36⁺⁶ weeks’ PMA, and infants will be defined as having BPD if any of the following criteria are met:

For infants receiving oxygen by nasal prongs at < 2L/min, the effective FiO₂will be determined using the Benaron-Benitz formula [4].

Any infants who are discharged home prior to 36 + 0 weeks’ PMA without any respiratory support or supplemental oxygen will be classified as “no BPD.”

The Trial Steering Committee has approved an algorithm for determining a diagnosis of BPD in cases where the BPD assessment is incorrectly or inadequately performed (e.g., too early, too late, or incomplete data). This algorithm uses data collected at exactly 36⁺⁰ weeks’ PMA and is assessed blinded to the treatment group. When this algorithm is required, infants will be defined as having BPD if any of the following criteria are met at 36⁺⁰ weeks’ PMA:

If a diagnosis of BPD is unable to be determined by this method, the Trial Steering Committee will assign a BPD status based on all available clinical evidence, blinded to the treatment group.

We will provide a table in the Supplementary Appendix of the number of infants in whom the BPD algorithm was used to determine BPD status and why the TSC was required to make a decision on the diagnosis of BPD, the number who required the effective FiO₂ to be calculated to determine BPD, and the number who were discharged home prior to 36 + 0 weeks’ PMA without any respiratory support or supplemental oxygen (see Table S1).

The binary primary outcome will be reported as the number and percentage of infants who were alive without BPD at 36 weeks’ PMA for each randomization group. To estimate the difference in risk (with 95% CI), between the treatment and control arms we will use binary regression (a generalized linear multivariable model with an identity link function and binomial distribution). In the regression, the primary outcome will be the dependent variable, group allocation (intervention/control) is the predictor variable, and the stratification factors used in the randomization are covariates. Standard errors will be adjusted to take into account the clustering of multiple births.

The three randomization strata that will be reported by the treatment group in the primary outcome table and will be covariates in the adjusted binary regression model are:

Along with the primary outcome, the individual components of the primary outcome—death or physiological BPD at 36 weeks’ PMA, will be reported by a randomization group (n (%)). Again, binary regression as defined above will be used to estimate the risk difference between the two treatment groups for survival and physiological BPD at 36 weeks’ PMA (Example Table 2).

Additional information about deaths will appear in the Supplemental Appendix by treatment group, including a summary of the age at death, causes of death (categorized), what proportion of deaths were classed as “respiratory” by the DSMB, and the modes of death (see Table S2).

The addendum to the International Council for Harmonisation of Technical Requirements for Pharmaceuticals for Human Use (ICH) E9 (R1) guidelines [6] promotes the use of the estimand framework. An estimand consists of five attributes: population, treatment, variable of interest, i.e., outcome, summary measure, and possible intercurrent events (which is a post randomization event that can occur and preclude or affect the interpretation of the variable of interest, e.g., discontinuation of treatment). The primary outcome for the PLUSS trial is presented in the estimand framework below.

Primary outcome for the PLUSS trial presented in the estimand framework (table layout based on the proposed estimand framework reporting proposed by Kang et al. [7]).

If an imbalance in demographics from Table 1 is detected, a sensitivity analysis adjusting for the relevant demographics will be conducted for the primary outcome and its components.

A sensitivity analysis will be run for the primary outcome, excluding those infants who had their BPD status (yes/no) determined by using the BPD algorithm, and reported if an important effect on the primary outcome is seen.

We plan to include a Kaplan–Meier curve of survival up to hospital discharge or 52 weeks’ PMA (whichever comes first). If requested by journal editors or reviewers, we may perform other analyses that are not mentioned in this protocol. These will be performed consistently with the principles of this analysis plan, as far as possible.

The secondary outcomes will be split into two tables “in hospital secondary outcomes” and “adverse events.” Binary secondary outcomes will be presented as the number and percentage of infants, by randomization group. The risk difference (with 95% CI) between the treatment and control arms will be estimated by binary regression (a generalized linear multivariable model with an identity link function and binomial distribution).

Continuous secondary outcomes will be graphed and visually inspected to determine if the distribution of values is skewed. If there is no indication that the distribution is skewed the mean and standard deviation (SD) will be reported for the variable, by randomization group. Linear regression will be used to estimate the difference of means (with 95% CI) between the treatment and control groups. For continuous variables where the data are judged to be skewed the median and interquartile range (IQR) will be reported, by randomization group. Quantile regression will be used to estimate the difference of medians (with 95% CI) between the treatment and control arms for these variables.

We will report the number of interventions received by both groups in the text.

There will be no adjustment for stratification variables for the secondary outcomes. Standard errors will be adjusted to take into account the clustering of multiple births. The analysis will be performed on the intention-to-treat population. There will be no adjustment for multiplicity. The results of these analyses will be presented in Example Tables 4 and 5.

Key respiratory parameters in the first 14 days after randomization (e.g., receiving mechanical ventilation via an endotracheal tube, FiO₂) may be presented graphically. These figures may be included in the main paper or Supplementary Appendix.