Page 194 - Prevention and Treatment of Incisional Hernia- New Techniques and Materials

Chapter 9
received the treatment or not(18). According to the International Conference on Harmonisation (ICH) guidelines of Good Clinical Practice (GCP) a statistical test decision of a study should be conservative. This is the rationale to use the ITT population for superiority studies and the PP population for equivalence studies. For non-inferiority trials the correspondence between ITT and PP should be used or a hybrid population.
In some speci c clinical situations, an equivalence or non-inferiority design is preferred. An equivalence study investigates whether a new treatment is not worse than the control. The analysis will be performed on the Per Protocol Population (PP), i.e. the patients who adhered strictly to the protocol and actually received the intervention called for by the protocol. These di erent types of analysis aid investigators in determining if a new treatment or device is better or as good as, but cheaper than what is now available. Like most clinical studies, the use of a biomedical statistician at both the study design and study analysis stage is recommended.
The sample size
When designing a clinical trial it is important to estimate the number of patients needed to answer the research question. Performing a clinical trial is time- consuming and expensive. It is also ethically mandatory to keep the number of patients that allow for valid study results as small as possible. Therefore it is important to estimate the number of patients that should be included in the study at the onset to answer the clinical question and the scienti c hypothesis the study is exploring. If the sample size is too small the study might not be able to reject the H0. In other words the study sample is too small to show a di erence in the primary outcome, although in reality there is a di erence (false negative; type II error). On the other hand if the sample size is too large, scares resources will be a spent unnecessarily. To calculate the sample size needed, there has to be agreement on several elements. First, the hypothesis type has to be clear: superiority, equivalence or non-inferiority. The expected mean value of the primary outcome parameter in the two groups and the di erence in outcome considered clinically important has to be estimated, based on preliminary ndings or results from similar studies in the literature. The signi cance level, i.e. the α or Type I error we accept (usually 5%) and the statistical power (usually 80% = 1- β, where β denotes the Type II error level) have to be de ned. These assumptions will provide the number of patients
192