The PROSPECT trial was conducted between July 2015 and March 2019. We enrolled a total of 2653 patients across 44 sites (288 patients in the pilot and vanguard phases, [6, 7] and 2365 patients in the main phase of the trial, which is the focus of this report). There were 42 centers in Canada, 1 in the United States, and 1 in Saudi Arabia. In PROSPECT, we randomized invasively mechanically ventilated patients to receive the study product (1 × 1010 colony-forming units of Lactobacillus rhamnosus GG) or an identical enteral placebo for up to 60 days, until discharge from ICU [8]. We found that L. rhamnosus GG did not reduce the risk of ventilator-associated pneumonia, Clostridioides difficile or other infections, or any other clinically important endpoints [5]. We summarize the trial timeline in Fig. 1.

PROSPECT Timeline and Milestones. Timeline of the PROSPECT trial, including major milestones, beginning with pilot trial registration in June 2013 and completing with final manuscript published in July 2020. NOL = no objection letter; RCT = randomized controlled trial

In this retrospective analysis of the PROSPECT multicenter RCT, we included data sources from prospectively collected case report forms (CRF) of site-level data captured at site initiation, the methods center’s internal tracking documents, and self-reported screening logs submitted by site research coordinators to the methods center. For each participating site, we extracted the following information: level of care (tertiary care vs. non-tertiary care), number of ICU beds at trial initiation, prior trial engagement (new randomized trial collaboration with coordinating group vs. previous collaboration), participation in the PROSPECT pilot or vanguard trial (yes/no), date of study initiation (Health Canada approval for each Canadian site, and opening in randomize.net for the Saudi Arabian site), date of first patient randomization, and years of ICU research experience of the site’s lead research coordinator (≥ 10 years vs. < 10 years) at trial initiation. Additional metrics included the number of months in which the site was open, total screening months, and total patient enrolment (per month and overall).

All investigators caring for critically ill adults affiliated with the Canadian Critical Care Trials Group (CCCTG) have the opportunity to participate in trials associated with this consortium, including PROSPECT. Certain sites which previously collaborated with the methods center participated in this trial, as well as additional new sites. Sites engaged serially rather than concurrently, such that the trial was launched with three open sites. On-boarding of additional sites occurred over the course of the trial up a final total of 44 sites. During the trial, three independent centers merged into one hospital institution. For the purpose of this analysis, we focused on early site-level factors possibly predicting enrolment and include the initial three centers but excluded data from the merged center. We excluded two sites that participated in the pilot study but not in the main trial. Therefore, this analysis includes 41 sites and each site was audited or monitored once during the trial [9].

We defined open sites as those with research ethics approval and fully executed contracts. Recruiting sites were defined as those actively screening for patients to enroll which followed an official site initiation visit. A screening month was defined as a month in which at least one patient was enrolled or in which a research coordinator screened ICU patients for trial eligibility during at least 2 weeks.

The primary outcome measure for this study was average monthly enrolment rate per site over the duration of the trial, accounting for the number of ICU beds. This approach avoids the confounding effect of number of ICU admissions and serial site engagement during the trial. As all else being equal, a larger ICU with a greater number of beds and a site with more months to recruit patients could have more enrolment opportunities. We excluded patients enrolled in either the pilot or vanguard phase and included only enrolments from the main trial.

To explore the association between monthly enrolment rate accounting for number of ICU beds, site characteristics, and trial metrics, we were limited to three independent variables to avoid model overfitting [10]. These variables were selected a priori based on the following criteria: (1) hypothesized influence on trial enrolment patterns based on factors identified in the literature, (2) collected prior to the site’s first randomization, and (3) availability in the PROSPECT trial database or internal documents.

We used time (in days) from the site’s regulatory approval to the first randomization to represent the time it takes to begin enrolling patients. In this trial, for Canadian sites, Health Canada’s acknowledgement of receipt of the clinical trial site information was the final regulatory step before sites could enroll patients. For one site in Saudi Arabia, regulatory approval was not required as probiotics in this jurisdiction are not considered a drug. Thus, we used the official opening date in the randomization software, which, comparably, was the last step before recruitment could be initiated after ethics approval.

We divided total RCT time into quartiles and determined within which quartile each site enrolled their first patient (1st quartile of the trial vs. 2nd, 3rd, or 4th quartile of the overall trial recruitment).

Lead research coordinator years of ICU research experience (≥ 10 vs. < 10 years) at the time of Health Canada approval is a binary variable which we used to explore whether more experienced research coordinators supporting trial initiation influenced the average monthly recruitment rate.

We report descriptive statistics of trial and site characteristics. For continuous variables, we report means and standard deviations (SD) if normally distributed and medians and interquartile ranges (IQR) if not.

We performed multivariable negative binomial regression to estimate the association between predictor variables (time to first randomization, quartile of site engagement, research coordinator years of experience) and the primary outcome (site enrolment rate). We report negative binomial regression results as incident rate ratios (IRR) and 95% confidence intervals (CI), with associated p-values. All p-values are reported to three decimal places. We set the criterion for statistical significance at alpha = 0.05. All analyses were performed using the Statistical Analysis System version 9.4 (SAS Institute Inc., Cary, North Carolina, USA).

The PROSPECT RCT was approved by each hospital’s research ethics board using an a priori consent model wherein written informed consent was obtained from each patient or their substitute decision maker prior to enrolment. PROSPECT was conducted in accordance with the principles of Good Clinical Practice following the Tri-Council Guidelines [11]. We did not require ethical approval to carry out this post-hoc analysis since data included only site-level variables with no patient-level data or identifiers.