Adjusting the Effect of Baseline Differences Between Groups in Trials with Which Have Two or More Groups

Ankarali H., Ankarali S.

TURKIYE KLINIKLERI TIP BILIMLERI DERGISI, vol.29, no.1, pp.91-98, 2009 (Peer-Reviewed Journal) identifier identifier identifier

  • Publication Type: Article / Article
  • Volume: 29 Issue: 1
  • Publication Date: 2009
  • Doi Number: 10.1016/j.scitotenv.2019.03.440
  • Journal Indexes: Science Citation Index Expanded, Scopus
  • Page Numbers: pp.91-98


Objective: In many clinical and experimental trials, researchers assess the effect of treatment by measuring the value of a continuous variable before and after the treatment. If there is an imbalance in baseline values between groups, some statistical comparisons may result with mistakes in estimation of the treatment effect. The aim of this study was to explain which statistical methods were more suitable in the estimation of the treatment effect when there was an imbalance for the baseline values between groups. Material and Methods: Different statistical methods, which are used in estimation of treatment effects, were briefly explained and were applied to a hypothetical data set, which had significant differences between groups according to baseline values of the related variable. In addition, a limited simulation study for several conditions was carried out to determine suitable statistical methods. Results: Baseline values were different between two groups and correlation was low between baseline and follow up values of related variable in each group for hypothetical data set. In this condition, comparison of simple differences between baseline and follow up values was the best method for the estimation of treatment effect. In the simulation study, the power of the test for simple differences was higher (85%) than the value in the analysis of covariance (40%) when correlations were low and sample sizes were small in each group. Moreover, the powers of these two tests were high and similar to each other, when sample sizes were moderate. When the correlation was high, the powers of both tests were high in both small and moderate sample sizes. Conclusion: The presence of a significant difference should be sought between groups according to baseline values of the related variable even though groups are randomly assigned. In addition, the degree of the correlation between baseline and follow up values should be taken into consideration. When significant differences exist between baseline values and the correlation is low, we suggest that the classical methods should be used to determine the significance of the effect; however, when the correlation is high, covariance analysis is a suitable method.