Analysis of Safety Data of Drug Trials: An Update

In 2010, the 5th edition of the textbook, “Statistics Applied to Clinical Studies”, was published by Springer and since then has been widely distributed. The primary object of clinical trials of new drugs is to demonstrate efficacy rather than safety. However, a trial in humans which does not adequately address safety is unethical, while the assessment of safety variables is an important element of the trial.

An effective approach is to present summaries of the prevalence of adverse effects and their 95% confidence intervals. In order to estimate the probability that the differences between treatment and control group occurred merely by chance, a statistical test can be performed. In the past few years, this pretty crude method has been supplemented and sometimes, replaced with more sophisticated and better sensitive methodologies, based on machine learning clusters and networks, and multivariate analyses. As a result, it is time that an updated version of safety data analysis was published.

The issue of dependency also needs to be addressed. Adverse effects may be either dependent or independent of the main outcome. For example, an adverse effect of alpha blockers is dizziness and this occurs independently of the main outcome “alleviation of Raynaud ‘s phenomenon”. In contrast, the adverse effect “increased calorie intake” occurs with “increased exercise”, and this adverse effect is very dependent on the main outcome “weight loss”. Random heterogeneities, outliers, confounders, interaction factors are common in clinical trials, and all of them can be considered as kinds of adverse effects of the dependent type. Random regressions and analyses of variance, high dimensional clusterings, partial correlations, structural equations models, Bayesian methods are helpful for their analysis.

The current edition was written for non-mathematicians, particularly medical and health professionals and students. It provides examples of modern analytic methods so far largely unused in safety analysis. All of the 14 chapters have two core characteristics, First, they are intended for current usage, and they are particularly concerned with that usage. Second, they try and tell what readers need to know in order to understand and apply the methods. For that purpose, step by step analyses of both hypothesized and real data examples are provided.