A comparison of nonparametric and parametric methods to adjust for baseline measures

Martin O Carlsson; Kelly H Zou; Ching-Ray Yu; Kezhen Liu; Franklin W Sun

doi:10.1016/j.cct.2014.01.002

A comparison of nonparametric and parametric methods to adjust for baseline measures

Contemp Clin Trials. 2014 Mar;37(2):225-33. doi: 10.1016/j.cct.2014.01.002. Epub 2014 Jan 21.

Authors

Martin O Carlsson¹, Kelly H Zou², Ching-Ray Yu³, Kezhen Liu⁴, Franklin W Sun⁵

Affiliations

¹ Pfizer Inc., New York, NY, USA; Department of Statistics, Rutgers, The State University of New Jersey, New Brunswick, NJ, USA. Electronic address: Martin.Carlsson@pfizer.com.
² Pfizer Inc., New York, NY, USA. Electronic address: Kelly.Zou@pfizer.com.
³ Pfizer Inc., New York, NY, USA. Electronic address: Ching-Ray.Yu@pfizer.com.
⁴ Department of Statistics, Rutgers, The State University of New Jersey, New Brunswick, NJ, USA. Electronic address: LiuKeZhen@gmail.com.
⁵ Pfizer Inc., New York, NY, USA. Electronic address: Franklin.Sun@pfizer.com.

PMID: 24462567
DOI: 10.1016/j.cct.2014.01.002

Abstract

When analyzing the randomized controlled trial, we may employ various statistical methods to adjust for baseline measures. Depending on the method chosen to adjust for baseline measures, inferential results can vary. We investigate the Type 1 error and statistical power of tests comparing treatment outcomes based on parametric and nonparametic methods. We also explore the increasing levels of correlation between baseline and changes from the baseline, with or without underlying normality. These methods are illustrated and compared via simulations.

Keywords: Analysis of covariance; Covariate imbalance; Percent change from baseline; Randomized controlled trials; Robust regression.

MeSH terms

Bias*
Biometry
Humans
Models, Statistical*
Randomized Controlled Trials as Topic / methods*
Research Design*
Statistics, Nonparametric