Bimodal distribution of angiographic measures of restenosis - what does it mean?

Azeem Latib ^1*, John cosgrave ² and Antonio Colombo ³

¹ San Raffaele Scientific Institute, Italy
² St James Hospital, Ireland
³ EMO-GVM Centro Cuore Columbus, Italy

^* To whom correspondence should be addressed. E-mail: alatib{at}gmail.com.

Accepted 23 June 2009

Abstract

The study by Byrne et al[1] in this edition of the Journal, analyzing the distribution of angiographic measures of restenosis after drug-eluting stent (DES) implantation, provides an opportunity for reflection and discussion of the applicability of surrogate endpoints in the evaluation of coronary stents. A surrogate endpoint is defined as a biomarker that is intended to substitute for a clinical endpoint and is expected to predict clinical outcomes based on epidemiologic, therapeutic, pathophysiologic, or other scientific evidence.[2] The concept of using physical signs, lab assays, or imaging measures as a substitute for clinical outcomes has a long history and surrogate endpoints have become frequently used in all branches of medical research. In cardiology, blood pressure, cholesterol, and haemoglobin A1C are all used as surrogates for clinical vascular events, including death, stroke, and myocardial infarction.[3] They are attractive endpoints in the design of clinical studies as they allow for smaller sample sizes. However, not all surrogates perform equally well as substitutes for a clinical endpoint. The ideal surrogate endpoint: 1) has a strong statistical correlation with the clinical endpoint of interest (Prentice criterion #1)[4, 5]; 2) however "a correlate does not a surrogate make" and replacement of the clinical outcome requires that "the effect of the intervention on the surrogate predicts the effect on the clinical outcome—a much stronger condition than correlation" (Fleming criterion)[6]; 3) captures the net effect of all of the treatment’s mechanisms of action on the clinical endpoint; thus any treatment difference in clinical outcome should be entirely explained by treatment differences in the surrogate (Prentice Criterion #2)[4, 5]; and 4) the magnitude of treatment difference in clinical outcome is clearly linked to the magnitude of treatment difference in the surrogate (Hughes criterion )[7]. Finally, a surrogate should have a known distribution in the population studied and its validity needs to be rigorously established through thorough evaluation utilizing appropriate statistical methodology.