The effect of crossovers on estimates of survival in medically treated patients with coronary artery disease.

Among 1661 consecutive patients with significant coronary artery disease treated medically at the Duke University Medical Center, 239 have undergone aortocoronary bypass surgery after at least 6 months of medical management. The purpose of this investigation was twofold: (1) to identify the distinguishing characteristics of these treatment crossovers; and (2) to illustrate a method of estimating the effect that withdrawing the crossovers from the analysis would have on the survival of the medically managed patients. Of 81 baseline characteristics compared, 25 were significantly (p less than 0.05) different between crossovers and noncrossovers. These included five characteristics that had previously been determined to be independent predictors favoring improved survival in medically treated coronary artery disease. A Cox regression analysis identified six variables independently associated with the time until crossover surgery. Crossover patients were younger and had superior ventricular function. A hazard score, which summarized in a single variable the overall risk of mortality, was developed from the prognostic baseline characteristics. Crossover patients had significantly (p = 0.003) lower hazard scores and hence would have been expected to have a superior survival compared with noncrossover patients. In our series, survival in the medically treated patients would likely have been higher had the crossovers remained in the medical group.

Duke Scholars

Author Kerry L. Lee Biostatistics & Bioinformatics, Division of Biostatistics