Statistical Analysis Plan for an Individual Patient Data Meta-analysis of Three, International Trials Comparing Protocolised With Usual Resuscitation in Patients Presenting to the Emergency Department With Severe Sepsis and Septic Shock

In 2001, Rivers and colleagues, published the results of a single-centre, proof-of-concept, randomised, controlled trial conducted in the USA comparing protocolised resuscitation, termed early goal-directed therapy (EGDT), with usual resuscitation in 263 patients presenting to the emergency department (ED) with severe sepsis or septic shock. The Rivers' trial demonstrated a significant 16% absolute risk reduction (ARR) in hospital mortality from 46.5% to 30.5%. With a view to informing the generalisability of these findings in their own health care settings, three independent, large, multicentre, randomised controlled trials testing EGDT were subsequently funded, one in the USA, one in Australasia and one in the UK:

USA - Protocolized Care for Early Septic Shock (ProCESS); Australasia - Australasian Resuscitation In Sepsis Evaluation (ARISE); and UK - Protocolised Management In Sepsis (ProMISe). Though independent trials, but with a view to performing a subsequent individual patient data meta-analysis (IPDMA), efforts were made to harmonise the three, contemporaneous trials on key areas of their design, for example, trial protocol, entry criteria, data and data collection, primary and secondary outcomes, etc.

As exist, and are published, for each of the individual trials, ProCESS, ARISE and ProMISe , the proposed IPDMA requires a pre-specified statistical analysis plan (SAP). The SAP set out below has been agreed between the three trial teams and prior to any knowledge of the results of any of the individual three trials.

It should be noted that, any proposed analyses added to the IPDMA SAP, post-knowledge of the results of ProCESS, ARISE and/or ProMISe, will be clearly indicated in any subsequent publication of the SAP and the results of this IPDMA.

Objectives

In patients presenting to the ED with severe sepsis and septic shock:

Primary objective

• to compare the effect of EGDT with usual resuscitation on 90-day all-cause mortality Secondary objectives
• to compare the effect of EGDT with usual resuscitation on 90-day all-cause mortality after adjustment for important baseline covariates
• to compare the effect of EGDT across countries
• to compare the effect of EGDT on secondary/intermediate outcomes
• to compare the effect of EGDT in pre-determined, clinically important subgroups

Data management Prior to pooling the data from the three trials, the clinical report forms for each trial will be compared and similarities/dissimilarities discussed across the trial teams to inform the final structure and specification of the IPDMA dataset. Similar variables will be double-checked for consistency across the trials (analysis of distribution, range and summary statistics) prior to being finally imported into the IPDMA database.

[Unlike ARISE and ProMISE - which are two-arm trials comparing EGDT with usual resuscitation, ProCESS is a three-arm trial with the additional arm evaluating protocolised usual resuscitation (termed protocolised standard care). Data from ProCESS for patients recruited and randomised to protocolised standard care (n=450) will be excluded from the analysis of the primary objective but retained for possible inclusion in the analyses of relevant secondary objectives.]

Analysis principles Primary analyses will be conducted on an intention-to-treat basis, with patients retained in their original, randomly assigned groups, and will be unadjusted for the effects of covariates.

Imputation of missing values will be considered and, if employed, the method fully described. For all analyses, the number of complete/missing observations will be reported.

Pre-determined, clinically important subgroup analyses will be conducted even if strong evidence of a treatment effect for the primary outcome is absent. All tests will be two-sided and a p-value of 0.05 will be used to indicate statistical significance. No formal adjustment will be made for multiple comparisons - however, with a large number of subgroup analyses planned, cautious interpretation will be employed.

Sample size calculation As indicated in the Introduction, the treatment effect in the Rivers' trial was a 16% ARR in hospital mortality (a 12.6% ARR in mortality at 60-days). Individually, ProCESS, ARISE and ProMISe have 80-90% power to detect a 6.5-8.0% ARR in mortality (hospital mortality censored at 60-days for ProCESS, 90-day mortality for ARISE and ProMISe), assuming a baseline mortality of 24-40% depending on the trial.

The combined recruitment into ProCESS, ARISE and ProMISE is 4210 patients with 3760 patients randomised either to receive EGDT or usual resuscitation. Based on a control event rate ranging from 25%-35%, an 80% power and a two-sided p-value of 0.05, this IPDMA will be able to detect an ARR in 90-day mortality ranging from 4-5% (with no allowance for heterogeneity of treatment effect or clustering of outcomes across the three trials).

For subgroups, again based on a control event rate ranging from 25%-35%, an 80% power and a two-sided p-value of 0.05, this IPDMA will be able to detect an interaction effect (odds ratio) of around 1.5 for a subgroup representing half of the total sample and an interaction effect of around 1.6 for a subgroup representing one quarter of the total sample. If a treatment-subgroup interaction is not significant, then it will be interpreted that the treatment effect for that subgroup is best informed by the overall treatment effect in the IPDMA.

Analysis plan The IPDMA will be performed using one stage, multi-level (patients nested in sites nested in trials), mixed modelling. Heterogeneity between trials will be determined by fitting a fixed interaction term between treatment and trial, while overall treatment effect will be reported with trial treated as a fixed effect and site treated as a random effect. A secondary analysis will adjust for important baseline covariates, including: age; sex; APACHE II score; SBP<90 mm Hg; and use of invasive mechanical ventilation.

Primary outcome 90 day all-cause mortality - logistic, mixed modelling, with terms for trial and site, reported as odds ratios with 95% confidence intervals (CI) Secondary/intermediate outcomes Hospital (censored at 60 days) and 28-day mortality - binomial, mixed modelling reported as odds ratios with 95% CI Survival analysis - Appropriate survival analysis techniques, e.g. Cox proportional hazards regression reported as Hazards Ratio with 95% CI if proportionality assumption holds Duration of stay in ED, ICU and hospital - assessed for normality, appropriate transformation reported as ratios of geometric means with 95% CI, accounting for impact of survivorship Receipt of and duration of mechanical ventilation, vasopressor support and renal replacement therapy - binomial, mixed modelling reported as odds ratios with 95% CI Where relevant, any assumptions underlying analyses will be detailed and reported. All results will be reported in tabular form and displayed using forest plots with 95% CI. All analyses will be performed using SAS version 9.3 (SAS Institute Inc., Cary, NC, USA). A two-sided p-value of 0.05 will be considered to be statistically significant.

Subgroup analyses To determine if the relationship between treatment and the primary outcome differs between pre-determined, clinically important subgroups, fixed interaction terms between treatment and subgroup will be reported. To further ascertain if the treatment-subgroup interaction varied between trials, a three-way fixed interaction between trial, treatment and subgroup will also be reported. Furthermore, where subgroups are to be defined by a continuous variable, they will be analysed both as categorical and continuous data with the linearity of the continuous relationship explored. Additional sensitivity analyses will be conducted adjusting for important baseline covariates.

Pre-determined, clinically important, pre-randomisation subgroups of interest will relate to site, patient and care delivery factors.

Site factors:

• Country
• Type of hospital
• Annual admissions
• Annual ED presentations
• Number ICU beds
• Ratio of ICU to hospital beds
• Annual ICU admissions
• Specialist staffing in ICU
• EGDT delivery model

Patient factors:

• Age
• Sex
• Race/ethnicity
• Obesity
• APACHE II score
• MEDS score
• SOFA score
• Source of infection
• Infectious aetiology
• Presentation - refractory hypotension
• Presentation - hypoperfusion
• Receipt of vasopressors
• Receipt of invasive ventilation

Care delivery factors:

• Interval between ED presentation and first administration of antimicrobials
• Interval between ED presentation and starting intervention
• Time of admission (day/night and weekend/weekday)
• Volume of fluid

Causal mediation analysis Exploratory analyses on post-randomisation variables, to further understand the delivery and therapeutic effect of EGDT, will be conducted. Potential mediators of the causal effect of EGDT on outcome will be considered and the associations between intervention and mediator and between intervention and outcome, adjusted for mediator, will be estimated.

Prior to commencing these IPDMA analyses, a detailed, final SAP will be published.