The growth of evidence-based medicine has triggered an increased focus on the quality of clinical trials. However, careful scrutiny of the literature has revealed high rates of statistical errors in large numbers of scientific articles, even in the best journals.
Errors in statistical analysis of clinical trials are widespread, have occurred for some time and, perhaps surprisingly, concern basic and easily avoidable statistical concepts. The ten most common mistakes are:
- Concluding equivalence on the basis of a non-significant p-value
- Relying solely on p-values
- Using Pearson X² test on ordinal data
- Using Logrank test on ‘survival’ data displaying only a delay
- Removing non-compliant patients from the analysis
- Using paired t-tests back to baseline in treatment groups separately (rather than two sample t-tests on changes, to ‘compare’ the groups)
- Using anything other than the point of randomisation as the origin for a time to event measure when comparing treatments
- Using non-parametric methods unnecessarily
- Counting and comparing serious adverse events (rather than the number of patients suffering a serious adverse event)
- Using covariates which depend on changes during treatment.
Avoid these errors by asking your statisticians to review your manuscript before submission.