Surgical Therapy for Coronary Disease: Randomized and Observational Studies

To the Editor.—  A recent article from the important Duke observational study¹ of coronary bypass surgery and medical therapy claimed "a significant improvement in patient survival with contemporary surgical therapy for the majority of patients with ischemic heart disease." What is worrisome in this report based on follow-up of 5809 patients is the reliance on the Cox² model as the sole method of analysis. As in the Veterans Adminstration trial,³ the data in Fig 1 of the article seem to indicate that the proportional hazards assumption may be violated with extended follow-up. If so, the treatment effect cannot be summarized by a single hazard ratio, as in Figs 4 and 7, because the effect is not constant over time. Logically, the hazard ratio cannot be expected to remain constant. Since surgery prolongs life, particularly in high-risk patients, the overall surgical hazard would be expected to increase when