Functions to implement group sequential clinical trial designs with early outcomes. Group sequential designs are one of the most widely used methodologies for adaptive design in randomized clinical trials. In such designs researchers collect data and undertake sequential analyses with the opportunity to either reject the null hypothesis, stop the study for futility or continue recruitment at an interim look, before reaching the planned sample size.
In situations where, for instance, outcomes are collected at long follow-up time-points, data at interim analyses are often available for not only the study primary (long-term) outcome time-point but also from early time-points for the same outcome (early outcomes); e.g. a primary outcome at 12 months and early outcomes at 3, 6 and 9 months. In settings where moderate to strong correlations exist between the sequence of such outcomes, information can be used from the early outcomes in addition to the final outcome at the interim analyses. The design, planning and sample size determination for such studies is more complex than for conventional group sequential designs and is generally achieved by simulating individual participant data for an assumed recruitment pattern as a means to determine information accrual during a proposed trial.
However, in practice, such simulations are complex and time-consuming to set-up and implement and provide a barrier to the use of group sequential designs. If we can assume approximate multivariate Normality for the distribution of the outcomes, and also make some assumptions about the expected correlation structure and recruitment patterns, then we can derive relatively simple analytic expressions for information accrual during a planned trial. Allowing a range of design options to be explored routinely without the burden of undertaking extensive simulation studies.
The two main functions (i) gsearlyModel and (ii)
gsearlyUser allow designs to be constructed based on a range of
typical clinical trial recruitment patterns and correlation models.
The function gsearlySimulate simulates multivariate Normal
datasets based on a previously fitted gsearly model.
Parsons NR, Basu J, Stallard N. Group sequential designs for pragmatic clinical
trials with early outcomes: methods and guidance for planning and
implementation. BMC Medical Research Methodology 2024; 24:42.
https://wrap.warwick.ac.uk/id/eprint/183449/
Parsons NR, Stallard N, Parsons H, Haque A, Underwood M, Mason J, Khan I,
Costa ML, Griffin DR, Griffin J, Beard DJ, Cook JA, Davies L, Hudson J,
Metcalfe A. Group sequential designs in pragmatic trials: feasibility and
assessment of utility using data from a number of recent surgical RCTs.
BMC Medical Research Methodology 2022; 22:256.
https://wrap.warwick.ac.uk/id/eprint/169801/
gsearlyModel, gsearlyUser,
gsearlySimulate