Statistical methods

Analyses will be intention-to-treat (ITT), consisting of all patients who have consented and have been randomised to a treatment arm.

For the primary outcome measure (composite of non-fatal myocardial infarction, non-fatal stroke and cardiovascular death [excluding confirmed intracranial haemorrhage and other fatal cardiovascular haemorrhage]), deaths from other causes (including fatal bleeding) will be treated as competing events. Patients who do not experience a major vascular event will be censored at the date of last follow-up.

As non-fatal major bleeding and anticoagulation are events, which, in the intervention arm, may lead to aspirin cessation, sensitivity analyses of the primary outcome measure (for the ITT population) will also include:

Censoring patients who experience non-fatal major bleeding (adjudicated), clinically relevant non-major bleeding, or anticoagulation at the date of the event (whichever occurs first)

Censoring only patients who experience non-fatal major bleeding (adjudicated) at the date of the event

For the secondary outcomes of time to fatal/non-fatal major haemorrhage (both intracranial and extracranial), the following competing risk models will be used to assess impact of assumptions over competing risk and censoring:

Deaths from other causes (excluding fatal bleeding) will be treated as competing events. Patients who experience a major vascular event will be censored at the date of the event. Patients who do not experience either a major vascular event or fatal/non-fatal major event will be censored at the date of last follow-up

Major vascular events and deaths from other causes (excluding fatal bleeding) will be treated as competing events. Patients who do not experience a fatal/non-fatal major vascular event will be censored at the date of last follow-up

Major vascular events and deaths from other causes (excluding fatal bleeding) will be treated as competing events. Patients who experience anticoagulation or clinically relevant non-major bleeding will be censored at the date of the event (whichever occurs first). Patients who do not experience either anticoagulation, clinically relevant non-major bleeding, or a fatal/non-fatal major vascular event will be censored at the date of last follow-up

Deaths from other causes (excluding fatal bleeding) will be treated as competing events. Patients who experience anticoagulation, clinically relevant non-major bleeding, or a major vascular event will be censored at the date of the event (whichever occurs first). Patients who do not experience either anticoagulation, clinically relevant non-major bleeding, a major vascular event or a fatal/non-fatal major event will be censored at the date of last follow-up

Time to event data will be described using Kaplan-Meier curves (or Cumulative Incidence curves for time to event outcomes involving competing risks). Analyses of time to event outcomes will be performed using a Cox proportional hazards model (or Fine and Gray’s adaptation of the Cox proportional hazards model for the subdistribution of a competing risk [87], i.e. a Competing Risk regression model for time to event outcomes involving competing risks), both unadjusted and adjusted for stratification factors: age, diabetes and CKD severity. The proportional hazards assumption will be assessed graphically with a log-log plot and a Schoenfeld test based on scaled Schoenfeld residuals.

The adjusted competing risk regression model for time to first major vascular event, with deaths from other causes (including fatal bleeding) treated as competing events, and patients who do not experience a major vascular event censored at the date of last follow-up, will form the primary endpoint analysis model.

Negative binomial regression will be used to analyse unplanned hospitalisations, both unadjusted and adjusted for stratification factors: age, diabetes and CKD severity.

For other secondary and tertiary endpoints, we will compare proportions for categorical data and means/medians for continuous data using Pearson’s χ² test and T test/Mann-Whitney U test, respectively.

The amount of missing data and reasons for the incompleteness will be explored and presented overall, i.e. not by group. If the amount of missing data is deemed too high and if appropriate (i.e. assuming the missing data is either missing at random or missing completely at random and censoring assumed to be non-informative), multiple imputation will be performed accordingly, for which all covariates included in the multivariable model, together with the censoring/event indicator and the cumulative baseline hazard will be included in the multiple imputation model.

All statistical analyses will be carried out using Stata v15 or higher, or SAS v9.4 or higher.