Author + information
- Thomas Perneger, MD, PhD∗ ()
- ↵∗Division of Clinical Epidemiology, Geneva University Hospitals, 6 rue Gabrielle-Perret-Gentil, Geneva, Geneva 1211, Switzerland
In presenting the results of the SECURITY (Second-Generation Drug-Eluting Stent Implantation Followed by 6- Versus 12-Month Dual Antiplatelet Therapy) trial, Colombo et al. (1) concluded that “6 months of DAPT appeared noninferior to a 12-month regimen with respect to the primary composite endpoint.” However, this conclusion, repeated several times in the paper, is not supported by the data.
The composite event was observed in 31 of 682 patients in the 6-month treatment arm, and in 27 of 717 patients in the 12-month arm. The risks were correctly reported as 4.5% versus 3.7%, and the risk difference as 0.8%. The 95% confidence interval was given as -2.4% to 1.7%, such that “the upper limit of the 95% CI was lower than the pre-set margin of 2%, confirming the noninferiority hypothesis.” Unfortunately, the confidence bounds are incorrect. The tell-tale sign of a problem is the asymmetry; the confidence interval should be approximately symmetric about the point estimate of 0.8%. Reanalysis shows an asymptotic 95% confidence interval of -1.3% to 2.9%, which exceeds the noninferiority margin. If an exact confidence interval is preferred, which may be a good idea given the low numbers of events, the bounds are -1.7% to 3.9%. This means that the data are compatible with an absolute excess risk of a composite event of up to 2.9% (or 3.9%) in a patient treated for 6 months, compared with a patient treated for 12 months. Noninferiority up to an excess risk of 2% does not hold.
The problem remains if a 1-sided 95% confidence interval is obtained, as it should according to the Methods section. A 1-sided upper limit on the risk difference of 0.8% is 2.5% (asymptotic method) or 3.4% (exact method). In all instances, the upper limit exceeds the noninferiority margin.
Another issue that argues against noninferiority is the rather high proportion of patients (34%) in the 6-month treatment group who continued their medications beyond the scheduled stopping time. Non-compliance with the assigned treatment dilutes the contrast between the groups. This favors the null hypothesis and causes a conservative bias in superiority trials: indeed, if a significant difference is observed between the treatment arms despite noncompliance, the true difference must be even greater. In contrast, in a noninferiority trial, the bias caused by noncompliance runs against the tested hypothesis of inferiority (that one hopes to reject to establish noninferiority). In the case of the SECURITY trial, if all patients in the 6-month arm had stopped treatment at the planned time, more composite events may have occurred and the difference with the 12-month treatment arm may have been greater.
In any case, it is unfortunate that the conclusions of this trial should contradict the results because of what appears to be an arithmetic error.
Please note: Dr. Perneger has reported that he has no relationships relevant to the contents of this paper to disclose.
- American College of Cardiology Foundation