Prevalence and Impact of Treatment Crossover in Cardiac Surgery Randomized Trials: A Meta-Epidemiologic Study
Background: Crossover dilutes treatment effect and reduces statistical power of intention-to-treat analysis. We examined incidence and impact on cardiac surgery randomized controlled trial (RCT) outcomes of crossover from experimental to control interventions, or vice versa. Methods and Results: MEDLINE, EMBASE, and Cochrane Library were searched, and RCTs (≥100 patients) comparing ≥2 adult cardiac surgical interventions were included. Crossover from the initial treatment assignment and relative risks (RRs) for each trial's primary end point and mortality at longest available follow-up were extracted. All RRs were calculated as >1 favored control group and <1 favored experimental arm. Primary outcome was the effect estimate for primary end point of each RCT, and secondary outcome was all-cause mortality; both were appraised as RR at the longest follow-up available. Sixty articles reporting on 47 RCTs (25 440 patients) were identified. Median crossover rate from experimental to control group was 7.0% (first quartile, 2.0%; third quartile, 9.7%), whereas from control to experimental group, the rate was 1.3% (first quartile, 0%; third quartile, 3.6%). RRs for primary end point and mortality were higher in RCTs with higher crossover rate from experimental to control group (RR, 1.01 [95% CI, 0.94–1.07] versus RR, 0.80 [95% CI, 0.66–0.97] and RR, 1.02 [95% CI, 0.95–1.11] versus RR, 0.94 [95% CI, 0.82–1.07], respectively). Crossover from control to experimental group did not alter effect estimates for primary end point or mortality (RR, 0.82 [95% CI, 0.63–1.05] versus RR, 0.95 [95% CI, 0.86–1.04] and RR, 0.88 [95% CI, 0.73–1.07] versus RR, 1.02 [95% CI, 0.95–1.09], respectively). Conclusions: Crossover from experimental to control group is associated with outcomes of cardiac surgery RCTs. Crossover should be minimized at designing stage and carefully appraised after study completion.