Abstract
We have proposed 24-mo between-treatment difference (active-placebo) in mean percent change in total hip bone mineral density (%THBMD) be used to evaluate whether a drug will likely reduce fracture risk, such that a mean %THBMD change greater than the surrogate threshold effect (STE) would indicate fracture benefit. However, this approach does not consider trial size. Here, we investigate using the lower limit of the 95% confidence interval (LCL) of %THBMD as an alternative to the mean to account for the impact that trial size has on estimator uncertainty. We compared the performance of these two measures (mean and LCL of %THBMD) in indicating fracture risk reduction relative to the STE by simulating trials of various sizes (100, 250, 500, 750, and 1000) based on data re-sampling from existing large trials with THBMD at 24 mo; this included 11 studies with radiographic vertebral fracture, three studies with hip fracture, and five studies with all clinical fractures. We re-sampled the THBMD data from each study 1000 times with equal numbers in treatment groups to estimate the reliability of these measures in being consistent with the observed fracture risk reduction due to treatment. Concordance between the %THBMD-STE comparisons and observed fracture risk reduction generally converged at a sample size of 500 (250 per treatment group). For vertebral fracture using mean %THBMD, 9 of the 11 studies had ≥90% of trials consistent with the observed fracture risk reduction if the sample size exceeded 500, which decreased to 7 of the 11 studies using the LCL. For both hip and all clinical fracture, all included studies had ≥90% of trials consistent with the observed fracture risk reduction if the sample size exceeded 500, regardless of using mean or LCL. Overall, the %THBMD-STE comparisons were generally consistent with the original studies' fracture risk reduction observed.