A predictive method for estimating the late angiographic results of coronary intervention despite incomplete ascertainment.

BACKGROUND Investigations of coronary restenosis typically use late (4-6-month) angiographic end points. Since only 50-80% of patients generally undergo repeat angiography, however, restenosis for the population as a whole is usually estimated by assuming that nonrestudied and restudied patients are similar. If restudied and nonrestudied patients differ, incomplete angiographic follow-up can yield an erroneous estimate of restenosis. No suitable method has yet been devised to detect and correct these errors. METHODS AND RESULTS We studied the clinical indications for angiographic restudy in an actual series of 301 treated lesions in 267 consecutive patients who underwent either Palmaz-Schatz stenting (126 patients) or directional coronary atherectomy (141 patients) at our institution. While only 249 (83%) treated segments underwent 4-6-month angiographic follow-up, all had clinical follow-up that described whether specific indications for restudy were present. Patients who had no clinical indications for such restudy were designated as having elective follow-up. In contrast, patients who had recurrent symptoms or positive exercise studies and were scheduled for repeat angiography at the independent recommendation of their referring cardiologist were designated as having nonelective follow-up. Mean late percent stenosis or binary restenosis rate (> 50% diameter stenosis) was determined for elective versus nonelective lesions that underwent follow-up angiography. These values were then used to input the behavior of the nonrestudied lesions according to their clinical status. From these imputations, a "predictive" model was developed to estimate the mean restenosis values that would have been found had the entire population actually undergone angiographic follow-up. Comparisons between the estimates of this predictive method and the traditional method that uses only the actual angiographic data demonstrate how alterations in various parameters influence the selection bias caused by incomplete angiographic follow-up. Of the 301 lesions available for follow-up, 100 of the 103 (97%) nonelective versus 149 of the 198 (75%, p < 0.001) elective lesions actually underwent angiographic follow-up. Mean follow-up percent stenosis (50% versus 27%) and the binary restenosis rate (53% versus 13%) differed significantly for the nonelective versus the elective lesions, respectively (both p < 0.001). Even at the fairly high (83%) angiographic follow-up rate, elective versus nonelective status was thus a confounder that caused differences between the restenosis rate estimated by the traditional (29.1%; 95% CI: 23.4, 34.7) and the predictive methods (26.3%; 95% CI: 21.4, 31.1). Larger (and even statistically significant) differences may be present under the conditions that exist in many current studies. CONCLUSIONS Restenosis trials with < 90% angiographic follow-up suffer from selection bias. Traditional methods that analyze only the restudied patients fail to correct for the important confounding influence of the clinical status of the nonrestudied patients. By using this readily available clinical information about the nonrestudied patients, a predictive method may be developed that provides a closer estimate of the true restenosis behavior for the population as a whole.