Statistics play a crucial role in the planning, implementation and evaluation of clinical investigations of medical devices, in order to be able to demonstrate the safety and effectiveness of the products. To meet this importance, statistics are prescribed and anchored in national laws and regulations as well as in harmonised standards as an essential part of a clinical investigation.

Normative principles

In accordance with the relevant standards (DIN EN ISO 14155: Clinical investigations of medical devices for human subjects), the statistics of a clinical investigation must meet the requirements of scientific methodology. For this reason, corresponding sections were included in DIN ISO 14155. The federal higher authorities recommend that the following points be taken into account when planning the statistics:

1. Primary and secondary hypotheses to be accepted or rejected
   As a rule, the alternative hypothesis should describe the compliment of the null hypothesis.
   A distinction must be made here between an exploratory (early phase) and a confirmatory study. An exploratory study may (but does not have to) dispense with a statistical test and thus with a null hypothesis.
   In a confirmatory study, it is important to keep the number of primary hypotheses to a minimum. The "success criterion" of the study is usually derived from the primary hypotheses, e.g. the effect is considered proven if the primary hypothesis(es) is/are rejected.
   When formulating the null hypotheses, it is important to define them precisely, stating the corresponding distribution parameters (e.g. population means). The formulation must be unambiguous, as a subsequent selection of the exact null hypothesis would increase the true significance level and thus the validity of the statistical test would be lost.
2. Information on the statistical design, procedure and analytical methods
   In a confirmatory study, the statistical tests or analytical procedures to be applied must be specified exactly and unambiguously, since a subsequent selection would again increase the true significance level.
   The populations of the evaluation: ITT (Intention-To-Treat population or the full analysis set), PP (per protocol: study population excluding protocol violators) and possibly others, e.g. a safety data set.
   The population on which the primary analysis is based should be described (in a confirmatory study, this is the ITT population) and the populations should be defined precisely. For the PP population, the possible protocol violations must be listed precisely.
3. Sample size planning
   Sample size planning: information on power, expected or desired treatment effect, assumed variability, drop-out rate
   The significance level should be given as the primary parameter.
   What sample size is required to reject the primary null hypothesis with a given probability, assuming a certain treatment effect and variability of the data? More complicated statistical models require further information.
4. The significance level and the power of the clinical investigation
   The significance level is usually set at 5%. In general, the significance level refers to a two-sided test or hypothesis (e.g. mean values equal versus mean values different). One-sided tests or hypotheses (e.g. mean value1 </= mean value2 vs. mean value1 > mean value2) are the exceptions. In most cases, the two-sided test at the 5% level corresponds to the corresponding one-sided test at the 2.5% level.
5. Drop-out rates
   Information on the expected drop-out rates. In some cases, it may also be useful to specify the drop-out incidence per unit of time.
6. Pass/fail criteria to be applied
   Criteria that lead to the rejection of the primary null hypotheses. Several null hypotheses may require a more precise formulation.
7. Information on planned interim analyses
   Interim analyses and adaptive designs: interim analyses, the resulting decisions and decision rules should be described as precisely as possible. The type 1 error resulting from the design must comply with the significance level, so that the nominal level for interim and final analyses may have to be adjusted accordingly. The adjustment must be described.
   Is the interim analysis blinded, i.e. without revealing the randomised treatments, or unblinded? In the latter case, it must be ensured that the information obtained is only accessible to the small group of people involved in the interim analysis. In particular, a confirmatory study may require an external Data Monitoring Committee (DMC). Its composition and tasks, the implementation of the decision-making rules must be described. Ideally, this should be specified in a DMC charter.
8. The criteria for discontinuing the investigation for reasons of subject safety
   e.g. accumulation of SAEs. Lack of proof of safety in the interim analysis.
9. Instructions for reporting all deviations from the original statistical plan
   Amendments must be precisely documented, justified and dated. No further amendments can be made after unblinding, i.e. database closure. The impact on the validity of the study should be described.
10. Specification of the evaluation of subgroups
11. Instructions for the documentation of all data
12. The treatment of missing, unused or falsified data, including dropouts and withdrawals of individual subjects

In the context of an ITT analysis, missing data must be replaced in a meaningful way. This should be done as "conservatively" as possible.

1. A justification for the non-consideration of individual information in the hypothesis testing, if applicable
2. In the case of multicentre trials, the minimum and maximum number of subjects to be included for each trial centre
   A minimum and maximum number cannot generally be specified. However, the design should allow for an analysis of centre effects or treatment differences between centres, countries or regions. In some cases, smaller centres may have to be pooled. The rule for this should be described in the study plan (or in the statistical analysis plan).