The ability to predict which patient will obtain maximum benefit from treatment depends not only on the inherent potential of the treatment but also on the underlying risk of the disease. Given the current plethora of diagnostic and therapeutic options in cardiovascular medicine, an accurate assessment of the risk for cardiovascular death or major illness is vital. In this issue of Annals, Pryor and coworkers [1] develop a cogent argument for the application of first principles—history taking, physical examination, and routine laboratory information—to the stratification of risk in a population of patients with suspected coronary heart disease. Cardiovascular medicine has evolved from an empiric discipline of inspection, palpation, and auscultation to one of procedures and labor-intense technology. Emphasis has shifted from diagnosis and natural history to intervention and, sometimes, cure. Driven by dramatic clinical presentations and equally impressive mortality rates, cardiologists and cardiovascular surgeons have developed considerable expertise in the management of patients with coronary heart disease. These efforts have not been without reward. Consider the effect of coronary artery bypass surgery on survival in patients with significant left main coronary artery obstruction or the administration of thrombolytic therapy in patients with acute myocardial infarction. These two examples are but the tip of the iceberg. However, patients with coronary heart disease represent a considerable spectrum of anatomic, physiologic, and functional diversity. The entry of these patients into the "system"—that is, their coming to medical attention—is usually through the development of symptoms suggestive of myocardial ischemia. The physician's responsibility at this point is to characterize the individual patient's risk for subsequent cardiac morbidity and mortality and to decide on an appropriate treatment course in which the potential for benefit is optimized.

Much of our current understanding of prognosis in and risk for development of coronary heart disease is derived from population studies [2-5] and randomized, controlled trials of specific therapies [6, 7]. However, the relative strength of a given prognostic feature will vary as a function of the population in which it is examined. For example, gender has a substantial role in the characterization of an individual's risk for developing coronary heart disease, but the effect of gender is attenuated in patients with extant disease. Further, it is the summation of multiple individual factors that best serves to identify patients at high risk. Not surprisingly, given the diversity of studies from which these data derive, prognosis is based on clinical, historical, physiologic (stress testing), and invasively obtained (cardiac catheterization) data. Given the current high level of utilization of procedures and invasive approaches to the diagnosis and management of patients with suspected, or proven, coronary heart disease [8-10], the importance of clinical and historical information has been relegated to a secondary role.

Using previously developed and validated predictive models from the Duke Database for Cardiovascular Disease, Pryor and colleagues [1] derived a new set of regression coefficients from easily obtained clinical and laboratory information in a group of outpatients referred for exercise testing. Demographic features, anginal characteristics, surface electrocardiographic information, and the presence of a ventricular gallop provided a basis for patient-specific estimates of the likelihood of significant disease or cardiovascular death. They found good agreement between predicted and observed outcomes (presence of significant disease and 3-year survival). The elegance and relative simplicity of this approach is attractive. The study conclusions are broad, however, and in some respects are not completely supported by the data. A truly unselected group of patients may be impossible to obtain. The fact that the 1030 outpatients were referred by outside physicians as well as local Duke cardiologists affects not only the generalizability of their findings but also raises the concern of ascertainment bias. That these patients were "referred" for exercise testing because of "suspected" coronary heart disease indicates an unidentified and perhaps important step in the sequence of events leading from presentation to clinical decision making. Unfortunately, even the rigorous method used (dividing the population into a "test" set and a "training" set) cannot obviate the effects of such potential bias. Only the application of their findings to other populations in other settings can validate these data. Another limitation of the study is the small group of patients undergoing coronary angiography. Although accurate identification of patients with "significant" obstructive disease was possible, substantial overlap and loss of discrimination were noted in the two subgroups of greatest clinical importance—those with "severe" disease and those with significant left main coronary artery obstruction. A similar limitation exists in the interpretation of their survival data. Three percent of patients suffered a cardiovascular death over the 3-year follow-up period. A worst-case scenario would be that 110 patients had "severe" coronary heart disease (62 requiring bypass grafting, 18 requiring percutaneous transluminal coronary angioplasty, and 30 dying of cardiovascular causes). Thus, only 10% of the original group would have had clinically important coronary heart disease. Because of such low event rates in a relatively "healthy" population, the conclusions of Pryor and colleagues must be viewed with circumspection.

Despite these caveats, a key feature of the approach of Pryor and colleagues is the translation of their findings into a cost–benefit perspective. Ideally, accurate and reliable prognostication would eliminate the need for further testing or treatment (or both) in low-risk groups, thereby shifting these resources to the higher-risk groups. Using a stringent definition of disease severity and an acceptable misclassification rate, almost one third of all changes were attributed to the low-risk subgroup. The shift of these resources (and costs, assuming a 2:1 charge-cost ratio) would result in the ability to screen 16 additional patients for every 100 with a 2% or less chance of having significant left main coronary artery obstruction. These crude extrapolations are even more impressive using a more liberal (although perhaps less acceptable) misclassification rate of 5% or less. Under these circumstances, the re-allocation of resources could allow for the screening of 25 additional patients. Their analysis is sobering, albeit convincing, and may, it is hoped, provide greater impetus for the refinement and further validation of their method.

The report of Pryor and colleagues [1] arrives at a critical time. The simplicity of their suggested approach to risk stratification is refreshing. The use and validation of "soft" end points are important reminders of essential aspects of clinical care. It is in the initial physician-patient encounter that vital prognostic aspects of the history are elicited, pertinent findings on physical examination are noted, and supporting electrocardiographic and roentgenographic data are reviewed. The importance of the determination of prognosis from this initial encounter has been alluded to previously [11]. Pryor and colleagues have begun to demonstrate the quantitative effect of these variables and, in the appropriate population, their predominant role.