Virtual Library

Start Your Search

Urania Dafni

Author of

  • +

    YI02 - Clinical Trials (ID 108)

    • Event: WCLC 2019
    • Type: Young Investigator Session
    • Track: Young Investigators
    • Presentations: 1
    • Now Available
    • +

      YI02.03 - Statistical Pitfalls in Clinical Trial Design (Now Available) (ID 3698)

      09:00 - 10:30  |  Presenting Author(s): Urania Dafni

      • Abstract
      • Presentation
      • Slides


      This topic will be discussed in the context of the following principles that clinical trials should adhere to.

      A clinical trial is an experiment testing medical treatments on human subjects.

      The process of evaluating medical treatment in humans starts with a Phase I clinical trial, followed by a Phase II and a Phase III trial towards regulatory approval.

      Phase III

      Phase III clinical trials are the gold standard for the evaluation of therapeutic interventions’ efficacy. The goals of a Phase III clinical trial include minimization of random error, elimination of systematic error (bias) and ensuring the generalizability of study results. The clinical trial design is the methodology for achieving these goals.

      Randomization, always present in phase III trials, provides a treatment assignment that is independent of outcome and patient/disease features, thus balancing treatment groups on known and unknown factors associated with outcome. Further, the intention-to-treat (ITT) analysis approach is the gold standard for all phase III randomized, controlled clinical trials. It analyzes all patients in the treatment groups as randomized without regard to treatment actually received. The systematic error (Bias), which is any effect rendering the observed results not representative of the treatment effect, is addressed through randomization and corresponding ITT analysis. Minimization of random error is addressed through the use of adequately large sample size.

      Phase II

      Phase II clinical trials can be either randomized (screening, selection, randomized discontinuation designs) or non-randomized. The latter may be seriously misleading since the impact of prognostic factors is usually far larger than that of treatment, while known prognostic factors may explain little variance.

      Randomized phase II trials allow control of selection bias and simultaneous testing of several new treatments, combinations, doses etc. They are preferable than non-randomized trials (some degree of control is better than none!) but they could be misleading due to the small sample sizes. They cannot replace phase III trials. In Phase II trials, the objective is to select active drugs for further testing and document toxicity, but not to provide definite estimate of new drugs’ efficacy, wh8ch is achieved only though a well-powered Phase III clinical trial.

      Hypothesis Testing

      In any experiment, two hypotheses covering all possible outcomes are pitted against each other. Only one of them can be true. The alternative hypothesis is the statement we would like to prove, and the null hypothesis, the statement we would like to reject. The only possible conclusions in hypothesis testing are: 1. Reject the Null Hypothesis, and thus prove the desired alternative hypothesis (positive trial), or 2. Not able to reject the null hypothesis (negative trial).

      Any trial needs to be designed in such a way that it is known a-priori what the errors relative with each of the two possible conclusions will be. Rejecting the null wrongly (false positive result), is subject to Type-I error (alpha), or significance level, while not rejecting the null hypothesis wrongly (false negative result), is subject to Type-II error (beta). The sample size of the trial is decided at the design stage to guarantee that these two errors remain below pre-defined bounds. These bounds are usually 5% for Type-I error, and 20% for type II error.

      One more important design characteristic is power, which is equal to (1-beta) and it is the probability of correctly rejecting the null hypothesis (values usually set above 80%).

      Superiority vs Equivalence/Non-inferiority

      Clear distinction should be made between superiority and equivalence/non-inferiority Phase III trials, with each testing a different type of null hypothesis. In a superiority trial we aim to reject a null hypothesis of “no effect” or “no difference”, while in an equivalence trial we aim to reject a null hypothesis of “different effect”.

      More particularly, in a superiority trial we aim to demonstrate the superiority of a new therapy compared to an established therapy or placebo. In this case, the determination of the sample size takes into account the clinical significance (by how much the new therapy should be better than the established one), the power and the significance level of the test, as well as the magnitude of the variation of the corresponding measure of interest.

      In equivalence trials, the objective is to demonstrate that a new treatment is equivalent to a standard therapy with regards to a specific clinical end point, while it has an intrinsic benefit for other clinical end points, while in non-inferiority trials, to evaluate whether the new treatment is not inferior to or as effective as the standard therapy for a particular end point. In this case, a tolerance and a non-inferiority margin must be predefined, along with the power and the significance level of the test.

      Failure to reject the Null hypothesis should not be confused with acceptance of the Null hypothesis.

      Subgroup Analysis

      Another issue that should be taken into account in the design of a trial is subgroup analysis, involving multiplicity issues. It is common practice to perform multiple subgroup analyses, but the probability of a false positive finding (type-I error) increases as the number of subgroup analyses increases (curse of multiplicity).

      Prognostic vs Predictive Marker

      The correlation between a biomarker and a true clinical endpoint corresponds to a prognostic marker, but not to a predictive one. It is the statistically significant difference in treatment effect between the levels of the biomarker (treatment by group interaction), that characterizes a predictive biomarker.

      Only Members that have purchased this event or have registered via an access code will be able to view this content. To view this presentation, please login, select "Add to Cart" and proceed to checkout. If you would like to become a member of IASLC, please click here.

      Only Active Members that have purchased this event or have registered via an access code will be able to view this content. To view this presentation, please login or select "Add to Cart" and proceed to checkout.