Reply

We acknowledge Dr. Agostoni and colleagues for their interest in our work, but we do not agree with their comments. Heterogeneity between studies is not a contraindication to quantitative pooling analysis. When analyzing the effect of a therapeutic mean, two different types of analysis can be performed: random-effect model (between-trial variance and within-trial variance) and fixed-effect model (1,2). The first type is used when significant heterogeneity exists, whereas the fixed-effect model assumes that between-trial variance is zero. As we clearly specified in our Methods, we used the random-effect model (3). As Agostoni and colleagues emphasize, controversy abounds concerning the best approach when heterogeneity exists, but we believe the investigators consider this methodology correct. In fact, in a very recently published meta-analysis, they concluded that radial access yields to lower procedural success than does femoral access, despite having found significant heterogeneity between studies when evaluating this end point (4).

Also, it is important to consider the causes of heterogeneity among trials. In our study, heterogeneity was not due to clinical or design discrepancies between the studies, but rather to differences in the number of patients included (treatment favored balloon only in the COMPASS trial (5) and in the study by Park et al. (6), and these had the lowest number of patients).

The second criticism raised by Agostoni and colleagues is that the relationship between risk ratio for restenosis and reference vessel diameter (RVD) should have been adjusted by the inverse of variance. Apparently they were unable to find (data not provided) any significant relation between both variables. Accordingly, we have analyzed our data after adjusting this relation for the inverse of variance (7–9). After analyzing it, the association between RVD and risk reduction remained significant, and in fact the beta-coefficient for RVD was even higher (Y = −6.514 + 2.674 [RVD] − 2.156 [varRR]); p value for RVD = 0.017).

Moreover, ever since we wrote up the study, several randomized trials comparing stent and balloon in small vessels have been reported and even published (10–12). Dr. Agostoni and colleagues only mention the recently reported ISAR SMART-2 trial, in which coronary stenting was nonsuperior to balloon in small coronary vessels. The ISAR–SMART-2, however, has one important limitation not recognized by Agostoni et al: the very high rate of cross-over from balloon to stent (>40%, in comparison with 19% in our meta-analysis) (10). Unfortunately, they do not mention the already published study by Kinsara et al. (12), in which restenosis rate decreased from 49% to 30% (balloon and stent, respectively; p = 0.009) in very small vessels (2.09 mm). The same occurs with the LASMAL-II trial, in which angiographic restenosis significantly decreased from 45% to 28%, respectively (p = 0.043) (11).

Interventional cardiology is one of the areas of medicine in which most randomized trials are being performed. Because of that, we may be tempted to use meta-analytic techniques without profound knowledge of the trials included and the methodologies employed. To avoid improper conclusions, it is necessary to understand adequately the meta-analytic techniques and their limitations, to obtain advice from expert statisticians, and to evaluate thoroughly all the trials included. This methodology allowed us to conclude that coronary stenting reduces the restenosis rate in comparison to balloon angioplasty in small coronary arteries (3).

• American College of Cardiology Foundation