Assessing the Treatment Effect Based on Percentage Change from Baseline in a Controlled Trial - A Simulation Study
Sarfaraz Sayyed M.Sc; Ashwini Mathur Ph.D; Asha Kamath Ph.D
Abstract
Background: Several possibilities exist for analyzing continuous endpoints in randomized clinical trials. These include regressing post-treatment response, absolute change from baseline, or percent change from baseline on factors (gender, region, etc.) with/without baseline response as a covariate. If response variable follows a Gaussian distribution, the percent change from baseline will be the ratio of two correlated Gaussian distributions. The assumption that percent change from baseline follows a Gaussian distribution may be incorrect and biased. Additionally, missing data could complicate the behavior of the percent change variable. It is also shown by Vickers (Vickers, 2001) that percentage changes from baseline are statistically inefficient when analyzed traditionally.
Methods: We propose an alternative solution using the Delta method to get estimates under different missing data imputation techniques and investigate the distribution for percent change from baseline for all values in numerator and denominator except zero.
Results: Delta method estimates on simulated data were compared with traditional point estimates with confidence intervals.
Conclusions: The Proposed method provides results that are better, and this study would be useful to researchers in choosing methods for analysis and decision-making when the endpoint of interest is the ratio of correlated Gaussian distribution, and the data has missing responses.
Full Text: PDF DOI: 10.15640/ijhs.v11n2a7
Today
All Days
Online