This does a bunch of simulations to help understand the optimal design of phase 2 trials of SARS-CoV-2 antivirals. The main idea is that if viral dyanmics are predictive of treatment effect (clinical benefit) then using rate of clearance is much better than time to clearance as the primary trial endpoint.
The model and data are taken from https://github.com/gradlab/CtTrajectories. The model has been slightly re-parameterised so that clearance slopes are inferred rather than clearance times. This helps make the posterior predictive behave better (otherwise the clearance slope is dependent on peak viral load).
The RMarkdown script Comparing_time_vs_rate.Rmd does a series of power calculations for different effect sizes and sample sizes, comparing the use of time-to-clearance and rate-of-clearance. This produces Figures 1-4 in the paper.
The RMarkdown script Adaptive_randomisation.Rmd simulates an adaptive randomisation trials (the randomisation ratios of intervention arms are made proportional to the probability of superiority compared to control) under two scenarios:
- Scenario 1: there are 4 intervention arms, 1 control arm. One of the 4 intervention arms increases viral clearance by 10% relative to control.
- Scenario 1: there are 4 intervention arms, 1 control arm. None of the 4 intervention arms increase viral clearance relative to control.
Trials can stop early for success (>99% probability that the viral clearance is increased by at least 1% for at least 1 drug); or for futility (<10% probability that the viral clearance is increased by 1% for all drugs). This RMarkdown produces Figure 5.