The null hypothesis, denoted as H0, is the initial assumption about a treatment effect formulated in the statistical hypothesis test. In a comparative clinical trial, it frequently describes that there is no difference between the treatments in the pre-defined endpoint or that a specified limit is not exceeded.
Function in the hypothesis test
A hypothesis test links the pre-formulated null hypothesis with a test statistic and a decision rule. The observed study data are subsequently evaluated as to how well they are compatible with H0. If the result lies within the pre-determined rejection region, H0 is rejected. This does not mean that the alternative hypothesis is proven in an absolute sense, but that the data provide sufficient evidence against H0 under the test logic.
The exact formulation depends on the study objective. In a superiority study, H0 can express that there is no advantage of the investigational treatment over the control. In non-inferiority or equivalence studies, the comparison is defined by a clinically and statistically justified margin. Hypotheses, endpoints, analysis populations, evaluation methods, and rules for multiple testing must be described in the protocol and in the statistical analysis plan prior to knowledge of the results.
Error probability, p-value and effect size
The significance level determines in advance what the maximum probability of a type I error should be, i.e., falsely rejecting H0. It is a threshold of the decision rule, not the null hypothesis itself. In the case of multiple endpoints, doses, comparisons, or interim analyses, an appropriate control of the overall error probability is required. An adaptive design must not subsequently lose this control.
The p-value is likewise neither H0 nor the significance level. It describes, under the assumptions of the test and under H0, how unusual an at least equally extreme result would be. A p-value alone measures neither the size nor the clinical relevance of a treatment effect. Therefore, estimates of the effect and confidence intervals should be considered together with the hypothesis test. Sensitivity analyses additionally show whether the conclusion depends on essential assumptions.
Differentiation from hypothesis and significance level
Hypothesis is the generic term for a testable assumption or statement. The null hypothesis is a special type of hypothesis that forms the reference for the formal test. In many studies, it is contrasted by an alternative hypothesis, which describes a difference or a directional effect, for example. The separate entry hypothesis covers the generic term; this entry explains the specific role of H0 in the statistical test.
The significance level, in turn, is neither a statement about the treatment nor an alternative to H0. It is the pre-selected limit for the risk of a type I error within the test strategy. H0 answers what is being tested; the significance level determines how strict the evidence threshold for rejecting H0 is. This conceptual separation prevents an imprecise interpretation of statistical results.
The test strategy must additionally be suitable for the type of data. Different statistical methods can be considered for binary, continuous, or time-to-event endpoints. The choice of the method follows the pre-defined hypothesis and the endpoint; it must not be made only after reviewing favorable results. In the case of multiple target variables, a hierarchical or other pre-defined strategy regulates the order in which hypotheses may be tested.
The statistical analysis plan must define the null hypothesis, the primary endpoint, the significance level, the test method, and, if applicable, the multiplicity strategy before data unblinding. This test strategy determines under which conditions H0 is rejected and protects the error probability from result-driven changes to the statistical conclusion.
Relevance for clinical trials
In clinical trials, the null hypothesis is part of a comprehensive evidence strategy. It must be appropriate for the clinical question, the primary endpoint, the control group, and the estimand. A subsequent modification of the test hypothesis or the analysis pathway after knowledge of unblinded results can jeopardize the validity of the study. The decisive factor is the pre-traceable connection between the medical objective and the statistical evaluation.
Full-service CROs such as Mediconomics support the translation of clinical objectives into hypotheses, endpoints, and statistical test strategies. This includes the preparation and review of the statistical analysis plan, the planning of multiplicity control and interim analyses, statistical programming, as well as the presentation of effect sizes, confidence intervals, and hypothesis tests in the clinical study report.
Frequently Asked Questions (FAQ)
Does rejecting H0 mean that a medicinal product is clinically relevant?
No. It shows evidence against H0 according to the pre-defined test rule. Clinical relevance additionally requires the evaluation of effect size, uncertainty, safety, and patient relevance.
Is H0 always “no difference”?
Frequently, but not exclusively. In non-inferiority or equivalence studies, H0 refers to a pre-defined margin and the respective study question.
Why must H0 be formulated in advance?
The pre-specification protects against result-driven analyses and enables a transparent control of the error probability. It is part of a traceable trial and analysis concept.
Regulatory references
- ICH E9 “Statistical Principles for Clinical Trials” – covers hypothesis tests, error probabilities, and confidence intervals.
- ICH E9(R1) “Addendum on Estimands and Sensitivity Analysis” – maps hypothesis tests to the precise treatment effect question.
- ICH E8(R1) “General Considerations for Clinical Studies” – requires the pre-specification of hypothesis tests and analyses.
- EMA Reflection Paper CHMP/EWP/2459/02 – requires full control of the type I error in adaptive designs.