The impact of postrandomization crossover of therapy in acute coronary syndromes care.

BACKGROUND In the Superior Yield of the New Strategy of Enoxaparin, Revascularization, and Glycoprotein IIb/IIIa Inhibitors (SYNERGY) study, patients assigned enoxaparin or unfractionated heparin (UFH) were treated with alternative anticoagulant therapy after randomization at physician discretion, a practice made possible because the trial was open label. Using SYNERGY as an example, we demonstrate the difficulty of evaluating the effect of postrandomization events in clinical trials and discuss possible methodology. METHODS AND RESULTS Patients with and without postrandomization crossovers were characterized and event rates analyzed. Statistical modeling was performed using inverse probability weighting and landmark analyses to evaluate the potential impact of postrandomization crossovers on event rates and treatment effect. Of 9978 SYNERGY patients, 9613 (96.3%) received at least 1 dose of randomized therapy and are included in these analyses. Of these, 740 (7.7%; 554 enoxaparin; 186 UFH) had postrandomization crossover. Crossover patients had higher unadjusted rates of 30-day death/myocardial infarction (MI) (18.9% versus 14.0%), thrombolysis in MI (TIMI) bleeding (16.9% versus 7.6%), Global Use of Strategies to Open Occluded Coronary Arteries bleeding (4.5% versus 2.3%), and transfusions (32.3% versus 15.2%). Adjustment for timing of crossover relative to the events attenuated the difference noted in death/MI but accentuated the association with TIMI bleeding. After adjustment using the inverse probability weighting technique, only a modest difference in the absolute treatment effect was observed between enoxaparin and UFH on death/MI (0.6% [unadjusted] versus 0.8% [adjusted]) and TIMI major bleeding (1.5% [unadjusted] versus 1.0% [adjusted]). The landmark analysis indicated a significant association between crossover from enoxaparin to UFH and TIMI bleeding but not in the other direction, and no crossover association was found in death/MI. CONCLUSIONS Postrandomization events in clinical trials are accompanied by substantial confounders that require careful consideration. In SYNERGY, postrandomization crossovers occurred in nearly 10% of patients, abetted by the open-label trial design. These patients had increased incidence of bleeding and death/MI, but after adjustment using several modeling techniques, only a modest impact of postrandomization crossovers on treatment effect was observed. The usual methods of analyzing end points cannot adequately address biases in changing treatment in these patients. The potential biases of membership in a postrandomization subgroup, as well as the methods used to account for the biases, should be considered when weighing the strength of results. Clinical Trial Registration- URL: http://www.clinicaltrials.gov. Unique identifier: NCT00043784.