Simulation to Estimate Sample Size for the Strategic Timing of Anti-Retroviral Treatment (START) Trial

Rachel Isaksson

Plan B Thesis

Advisor: Dr. Jim Neaton

ABSTRACT

Purpose:

An important yet currently understudied question in the care of Human Immunodeficiency Virus (HIV)-infected individuals is when to start treatment. The Strategic Timing of Anti-Retroviral Treatment (START) Trial, an international randomized trial will address this question. They plan to enroll approximately 4,000 participants infected with HIV with CD4+ cell counts greater than 500 cells/mm3. They will be assigned 1:1 to immediate anti-retroviral treatment (ART) or to deferred ART, with the deferred arm waiting until their CD4+ cell count drops below 350 cells/mm3 to initiate treatment. The goal of this trial is to assess the long term effects of early treatment on morbidity and mortality. Because most of the information required for the sample size calculations for this trial is based on observational data, a simulation was used to evaluate the effect of various parameters, including the CD4+ cell count distribution of the participants at baseline, the CD4+ cell decline in the deferred arm before initiating ART, the CD4+ cell increase once starting ART, adherence and the event rates on sample size estimates.

Methods:

Each simulation replicated 1,000 experiments, each of which followed 10,000 patients over six years, generating their CD4+ cell counts at each visit. Two options for each of the five parameters of interest were considered and all 32 combinations were simulated. Annual and overall event rates in each arm and the hazard ratios were calculated. Sample size was calculated in each case using the results from the simulation.

Results:

The estimated hazard ratio of the immediate treatment to deferred treatment ranged from 0.40 to 0.52. The lowest and highest sample sizes calculated were 582 and 1076 (and up to 1423 when incorporating event adjudication). All parameters had some effect on the sample size, though beginning ART in the deferred group before the CD4+ cell count declined to less than 350 cells/mm3 and event rates assumptions had the greatest effect.

Conclusion:

The simulated data followed many of the expectations suggested by the observational and limited experimental data, yet ultimately led to smaller sample size estimates than anticipated. This simulation suggests that the sample size specified in the START protocol of 4,000 should be more than adequate to address the outcomes of interest. Limitations of the simulations include the use of CD4+ cell count as the only predictor of events (ignoring other important factors such as HIV viral load), the CD4+ cell count slopes are highly variable and hard to estimate, and the event rates used were from observational data and extrapolated to the higher CD4+ cell counts here.