The Eyeball Test
Can the Blind Leading the Blind See Better Than the Statistician?
As physicians, we strive to provide the most appropriate care for our patients and to provide them with accurate assessments of both the risks and benefits they can expect to realize. Unfortunately, as humans, we are notoriously bad at estimating risk.1,2 However, as the choices of treatments for cardiac disease expand, accurate assessment of risk becomes increasingly important. A patient with coronary artery disease may have the option of medical therapy, coronary artery bypass grafting, or percutaneous intervention. Transcatheter aortic valve replacement may soon be a viable alternative to surgical aortic valve replacement for patients with aortic valve disease. For patients to be able to make the choice that best suits their goals, they must have accurate information regarding risks and benefits.
Article see p 151
In the current article, Jain et al3 compared cardiac surgeons’ assessments of mortality after coronary artery bypass grafting and valve surgery and that calculated from the Veterans Affairs Continuous Improvement in Cardiac Surgery Program (VA CICSP) risk model with actual mortality rates. The focus of the study was perioperative mortality; however, 1- and 5-year mortality rates were also considered. Although neither the statistical risk model nor surgeon estimates were perfect, the risk models performed somewhat better at all time points. Both the risk models and the surgeons’ estimates were better at estimating perioperative mortality than they were at estimating late mortality. These findings are informative and may help us to improve our estimates of procedural risk.
When examining the predictive validity of a risk model (or surgeons’ estimates), 2 factors are regularly used: calibration and discrimination. Calibration measures to what degree the model assigns risk to the population, or in a population of 100 patients with an estimated mortality risk of 3%, were there 3 observed deaths? The current article …