The hazards of approximations and extrapolations: Insights from the ATLANTIC study.

2015; 114: 7–8. The ATLANTIC study demonstrated that administration of the potent P2Y12 antagonist ticagrelor shortly before primary percutaneous coronary intervention (PCI) of STEMI did not improve reperfusion of the culprit artery before the procedure (primary objective) but was safe (1). The prespecified secondary endpoint of post-procedural definite stent thrombosis was reduced significantly with the prehospital administration of ticagrelor, these results being timely consistent with the pharmacodynamic and electrocardiographic findings suggesting that maximal effect of the pre-hospital administration of ticagrelor occurred shortly after the end of the procedure. All the other pre-specified clinical endpoints were not significantly affected by the pre-hospital administration of ticagrelor. In this issue of Thrombosis and Haemostasis, Serebruany and colleagues’ Viewpoint focuses on one of these non-significant endpoints. Consistent with his prior critiques, Dr. Serebruany relies on several approximations and extrapolations on non-significant endpoints from grossly underpowered subgroups (2–4). We answer point by point to each critique addressed to the ATLANTIC study and investigators. “ATLANTIC corresponds well with the immediate clinical outcomes including the early PCI “death paradox” in PLATOUSA patients, and lack of early ticagrelor benefit in the overall invasive PLATO cohort.” This critique is a shortcut of three ideas jumping to a wrong conclusion. First, the authors refer to an exploratory analysis which examined the subset of patients with planned invasive management who actually received PCI within 24 hours (h) of randomisation, defining a subgroup in timing for the PCI subset of PLATO (“early PCI”). They do not report the primary endpoint for this subgroup of subgroup, which was actually similar to the overall PLATO results, hazard ratio (HR)=0.85, 95 % confidence interval (CI) (0.74, 0.99) (5). There was no significant difference in cardiovascular death for this subgroup of subgroup but a numerical trend in the wrong direction for ticagrelor: thus, there was no death paradox in the “early PCI subgroup” exploratory analysis. Second, they extend this speculation to the USA patients who represented 7.6 % of the total PLATO population. In the redefined USA subset (different from the initial North America subset), the primary endpoint was not significantly different between ticagrelor and clopidogrel (HR 1.27 (0.92–1.75), p= 0.15) but the trend was not favorable to ticagrelor, going in the opposite direction than for the remaining 92.4 % of patients (p for interaction=0.009) (2). However, considering CV death in the US cohort, 24 US patients died on ticagrelor (3.4 %) vs 19 (2.7 %) on clopidogrel, p=0.45 (p value for interaction was not significant, p=0.12): thus again, there was no death paradox in the US sub-population. Interestingly, in this analysis the numerical difference in death occurred late, beyond four months, unlike in the ATLANTIC study where the numerical difference occurred very early, within the first 24 h. In PLATO, the US subpopulation was different from the rest-of-the-world population by many characteristics including more NSTEMI and more early PCI. Whether ticagrelor treatment before rapid coronary angiography/PCI in NSTEMI (like in the US population) is appropriate is another question which has not been evaluated with ticagrelor but with prasugrel in another study (6–8). Third, the lack of early ticagrelor benefit in the overall invasive PLATO cohort is well recognised with potential explanations (role of clopidogrel pre-treatment and of the selected population) (9). Thus, we believe that the significant reductions in death with ticagrelor observed in the main PLATO trial (n=18,758) and in its pre-specified PCI cohort (n=13,408) are real. There was also no effect one way or the other on early mortality in PCI patients in both PLATO and ATLANTIC. Multiplying the subgroups and analysing small subgroups expose rapidly to the risk of false positive. The next remark relates to another subgroup with a new extrapolation to mortality.