P439 INTERPRETING CARDIOVASCULAR TRIALS WITH NEUTRAL OR NEGATIVE RESULTS TRIALS: A MULTISTEP APPROACH

Abstract Despite great progress in randomized controlled trial (RCT) design, several methodological issues can still be encountered when reading superiority trials published in leading journals. We propose a 10–step approach to the interpretation of neutral or negative trials to exclude important methodological issues before concluding that the treatment does not work (Figure). 1. Premature interruption Except when there is evidence of superiority at interim analysis, all premature discontinuations make the RCT underpowered. 2. Insufficient power Insufficient power may derive from lower event rates than anticipated or a smaller effect size than expected. Two other possible reasons are a high cross–over rate or a number of events “hypertrophied” by minor events or medical decisions. 3. Logistical peculiarities Logistical issues, e.g., heterogeneous patient selection and management in different geographic areas, are sometimes crucial. 4. Errors or ambiguities in endpoint adjudication Cardiovascular death and HF hospitalization are often used as endpoints. In some cases, the discrepancy between investigator–reported and centrally adjudicated events is quite prominent. Furthermore, some endpoint definitions may be open to interpretation. 5. Suboptimal choice of composite endpoints A composite endpoint should have clinical relevance and should include only elements to be impacted by the intervention. This last requirement is often not met. 6. Nosographic trap The nosographic trap occurs when the study population is heterogeneous, and the importance of the treatment target is not the same across the whole population. 7. Lack of pathophysiological prerequisites for efficacy Some RCTs may prove neutral or negative because they enrolled patients where the treatment was unlikely to be effective. 8. Algebraic cancellation Algebraic cancellation occurs when results are positive in some patients and negative in others. 9. p trap A possible instance is that p value approaches the statistical significance value, leading to uncertainties as to whether borderline significance should be interpreted. 10. The treatment does not really work When a trial is adequately powered and without major methodological flaws, but the p value for the primary endpoint is far from significance (and therefore the p trap is not an issue), the reasonable conclusion is that the treatment is not more effective than its comparator in this specific setting.