In order to bring you the best possible user experience, this site uses Javascript. If you are seeing this message, it is likely that the Javascript option in your browser is disabled. For optimal viewing of this site, please ensure that Javascript is enabled for your browser.
Did you know that your browser is out of date? To get the best experience using our website we recommend that you upgrade to a newer version. Learn more.

We use cookies to optimise the design of this website and make continuous improvement. By continuing your visit, you consent to the use of cookies. Learn more

What went wrong with statistical analysis in major clinical trials? A session for the clinical cardiologist…

Major randomised controlled clinical trials (RCT) are considered to be the gold standard for providing clinical cardiologists with the evidence basis for decision-making. Interpretation of the results of these trials can be challenging, especially regarding the statistical analyses, which are sometimes of great complexity. On Monday 2nd September, we had a very interesting, educative, and interactive session where the audience was enlightened on the tricks and pitfalls of planning, conducting, and interpreting results from RCTs.
MPH Sabina Murphy, Boston, US, first gave an interesting talk on interpreting superiority versus non-inferiority (NI) trials. Basically, superiority trials are designed to show that one treatment is better than the other, and are, in general, more simple in their design and analysis methods than non-inferiority trials, which are designed to show that a new treatment is “not unacceptably worse” than current therapy.
Given this term “not unacceptably worse”, it is evident that a great deal of subjectivity has to be dealt with in choosing a clinically meaningful margin when designing the trial, since validity of any conclusions of NI trials depends on the choice of margins. Sabina Murphy provided us with some tricks on how to define these margins and gave examples of why NI trials, in spite of their complexity, unpopularity among some statisticians, and challenges, can and should be used (when placebo trials are non-ethical due to established gold standard therapy, to show that new treatment is safer, cheaper, more convenient). One other advantage of NI trials is that once non-inferiority is established, the design of the trials makes it feasible to evaluate for superiority.
Another very important issue in analysing CRTs is choosing the right treatment analysis model. Prof. Jouven, Paris, France, gave a fine overview of intention to treat (ITT) versus per protocol (PP) analysis approaches. Dr. Jouven gave us several hypothetical examples illustrating what the two methods can provide. In general, ITT analysis better reflects the “real clinical world” and preserves the strength of randomisation of the patients in CRTs. Conversely, PP analysis – only analysing data from the patients fully adherent to their randomised treatment – does not maintain comparability of the patient groups generated from randomisation. Furthermore, we were presented with quite thought-provoking data showing that a large proportion of published trials that declare having used ITT actually do not implement this type of analysis when publishing the trial results, and the methods used to handle missing data are often not clarified. Missing outcome data is indeed quite challenging, but during his talk, Prof. Jouven gave some possible strategies on how to deal with this problem (statistical models, imputation, extreme case analysis etc).
Finally, Prof. Pocock, London, GB, gave an illustrative presentation addressing endpoint analysis focusing on time-to-first-event versus all-event analysis. As an example, analyses of time to first heart failure hospitalisation versus all heart failure hospitalisations in some of the great trials evaluating the effect of treatment with statins and ARBs were used to show how alternative endpoint analyses can give rise to different interpretations of treatment effect. Using all-events, a gain in statistical power is achieved in trials with “repeat events” allowing a smaller sample size. An overview of the different statistical methods for analysing trials with repeat events was given and Prof. Pocock rounded off by pointing out the need for better agreement on methods of analysis, better and more graphical display and stressed the importance of sensitivity analysis and correct power calculations.
A very enlightening session with complex topics made less abstruse. Look for yourself on ESC Congress 365.


Session Title: What went wrong with statistical analysis in major clinical trials? A session for the clinical cardiologist…

The content of this article reflects the personal opinion of the author/s and is not necessarily the official position of the European Society of Cardiology.