KSE
Articles
- Page Path
- HOME > Epidemiol Health > Volume 11(1); 1989 > Article
- Special edition Statistical analysis of clinical trials
- Hae Hiang Song
-
Epidemiol Health 1989;11(1):18-26
DOI: https://doi.org/
- 6,562 Views
- 297 Download
- 0 Crossref
- 0 Scopus
Abstract
The aim of statistical analysis in a randomized clinical trial is the comparison of the benefit of treatment compared to control or other groups. If all other conditions except the treatment are the same for these groups and if the clinical trial is based on the randomized treatment assignment, the comparison is in general straightforward. Chi-square tests or t-tests can be applied appropriately. In a clinical trial the advantages of randomization are numerous and are mentioned in this paper. A few practical methods of randomization are also discussed. Not having used random assignment, the researchers could not easily certify to others the objectivity of the treatment results, no matter how unbiased the assignment may in fact have been. If in fact patients' baseline characteristics and medical conditions, influential in determining the course of disease, are different in two groups, the treatment evaluation should be taken into consideration the adjusted analysis for these covariates. In order to adjust for random allocation disparities the poststratification analysis and the adjustment methods can be adapted. One can also reduce bias and increase precision by modeling the influence of prognostic factors on treatment response. The analysis of variance or the analysis of covariance are excellent for this purpose. If follow-up intervals vary due to different entry of patients into the study, the distribution of survival times in the group should be analyzed. Survival analysis by the actuarial and product-moment methods are mentioned. The differences in the Survival rates in the two groups across time can be tested using Gehan’s test or log-rank test, also known as the Mantel-Haenszel test. The adoption of more complex analysis methods opens the way for far more extensive exploration of and adjustment for the effects of baseline risk factors and the Cox regression model is especially appealing for this purpose. The underlying assumption of the Cox model is discussed.
Cite
