The Choice of Two Baselines
- PDF / 4,596,969 Bytes
- 5 Pages / 612 x 792 pts (letter) Page_size
- 56 Downloads / 167 Views
David R. Bristol, PhD Monmouth junction, New jersey
The Choice of Two Baselines
Key Words Covariates; Baseline; Analysis of covariance
Correspondence Address David R. Bristol. PhD. 59 Cottonwood Court, Monmouth junction,
N J 08852 (e- mail address: drbris @msn. corn).
INTRODUCTION Randomized comparative clinical trials often include pretreatment assessments of efficacy variables, and these variables are usually included as covariates in the between-treatment comparisons. Such an analysis may be conducted using analysis of covariance. The use of this statistical technique is not a recent development. Cochran (1)described various aspects of the procedure and included several references. Although the procedure has been used for over 50 years, discussion of its use is ongoing. Recent discussion includes, but is not limited to, that of Tu et al. (2), Assmann et al. (3), Pocock et al. (4), and Berger (5). These references noted that inclusion of a covariate in general, and a pretreatment assessment in particular, in the analysis is important for several reasons. In general, inclusion of a covariate in the analysis will reduce the variability associated with the efficacy variable. The value of the variable of interest may depend on the pretreatment value. Furthermore, if the between-treatment comparison differs for different values of the pretreatment value (ie, a treatmen t-by-covariate interact ion), it is impor tan t to discover this relationship and to examine any plausible explanation for this relationship and any data that may contribute to this observation. Glueck and Muller (6) presented a detailed
A prerandomization assessment of an eficacy variable is often used as a covariate in the analysis to perform between-treatmentcomparisons. Such assessments may be made more than once. The variable to be used as the covariate must be chosen prior to any analyses. This choice may be the use of both assessmentsas covariates. The impact of the choice is examined here.
mathematical examination of the effect of the power when a baseline covariate is incorporated in linear models. Ford and Norrie (7) focused on the impact of the inclusion of covariates in Cox’s regression model, with some discussion of the normal linear model. Assmann et al. (3) suggested that, in general, unadjusted analyses that do not include the covariate should also be performed, but when the covariate is the baseline assessment of the efficacy variable, the adjusted analyses are more appropriate. Tu et al. (2),Pocock et al. (4),and others have noted that the model used for the analysis should be specified prior to any analysis. This should be done to validate the statistical analysis and to enhance the credibility of the results. This is particularly important when there are several covariates for possible inclusion in the model. An alternative approach, suggested by Pocock et al. (4) and others, is to specify the algorithm used to select the covariates included in the analysis. However, International Conference on Harmonisation (ICH) E9 (8) state
Data Loading...
