By Michael Crawford, MD
Professor of Medicine, Chief of Clinical Cardiology, University of California, San Francisco
Dr. Crawford reports no financial relationships relevant to this field of study.
SYNOPSIS: An evaluation of the predictive accuracy of the new pooled Cohort Risk Equation in > 300,000 subjects without heart disease or diabetes and a low-density lipoprotein cholesterol level between 70-189 mg/dL followed for five years showed that the new equation markedly overestimated the observed risk of cardiovascular events.
SOURCES: Rana JS, Tabada GH, Solomon MD, et al. Accuracy of the atherosclerotic cardiovascular risk equation in a large contemporary, multiethnic population. J Am Coll Cardiol 2016;67:2118-2130.
Blaha MJ. The critical importance of risk score calibration: Time for transformative approach to risk score validation? J Am Coll Cardiol 2016;67:2131-2134.
The 2013 American College of Cardiology/American Heart Association Pooled Cohort Risk Equation (PRE) has been criticized for its basis in study populations conducted in the 1990s with limited ethnic diversity and age range. Clinicians believe it has limited generalizability to contemporary patients seen in clinical practice. Thus, investigators evaluated the large, contemporary, multiethnic population of Kaiser Permanente Northern California comparing the risk equation-derived five-year risk of atherosclerotic cardiovascular (CV) events with the observed rate. The population selected was > 21 years of age and had a low-density lipoprotein (LDL) cholesterol level between 70 and 189 mg/dL. The study excluded patients with known CV disease, those who had received a statin prescription within five years of enrollment, and those who used statins for primary prevention during the follow-up period. CV events included myocardial infarction, ischemic stroke, and cardiac death. The subjects were subcategorized for diabetes status. More than 941,000 patients met the initial entry criteria and after applying the exclusion criteria, 311,833 patients between the ages of 40-75 years were enrolled — 307,591 without diabetes and 4,242 with diabetes. The main analysis focused on non-diabetics, of which 62% were women, 22,283 were black, 52,917 were Asian/Pacific Islander, and 18,745 were Hispanic. In this study group, there were 2,061 CV events during 1,515,142 person years of follow-up. At all levels of risk predicted, observed CV events were substantially lower and the difference was greater at higher calculated risk scores. For example, in those with a predicted risk of > 5% in five years, the mean predicted risk was 8.72% vs. an observed rate of 1.85%. The correlation between expected and observed rates was better in diabetics, but researchers observed overestimation, especially at higher predicted risk. In the high-risk group, predicted CV events were 13.38% vs. 5.5% observed. Similar results were noted in the ethnic subgroups. The authors concluded that in a large real-world population, the Pooled Cohort Risk Equation substantially overestimated five-year risk of CV events in adults without diabetes and that ethnicity did not affect this result.
Other investigators have evaluated the PRE in several research populations that were not used in the creation of the PRE and found conflicting results. Interestingly, in cohorts gathered in the 1990s, the correlation between observed and predicted CV events by the PRE was better than in more contemporary populations. Some have argued that more contemporary populations include more patients on aspirin and statins, less smoking, and generally better lifestyles. Consequently, for this study, investigators excluded patients on statins before or during the study period. One could argue that this removed the higher-risk patients, but sensitivity analyses did not support this idea. Also, this study enrolled patients and followed them from 2008 until 2013, so it was not influenced by the new PRE. Other strengths included the large population (> 300,000) and the inclusion of reasonable numbers of all four major ethnic or racial groups in the United States. In addition, the patients were part of an integrated health system in which clinical data collection was of high quality.
One potential weakness of this study was the assumption that patients with health insurance are of higher socio-economic status and probably engage in better health habits than other populations. This may be true, but these are the majority of patients practitioners see. Additionally, data on diabetics were inadequate because researchers excluded diabetics from the main analysis. This was necessary because in their system, most diabetics were receiving statins. However, researchers’ analysis of a diabetic subgroup that was not on statins confirmed overestimation of events by the PRE, but less so than observed in non-diabetics. An accompanying editorial noted that event rates were very low in this study in part because softer endpoints, such as revascularization, were not included.
Even though this study seems like a step in the direction of adjusting the risk equation to reflect contemporary practice, this issue remains highly controversial because of the treatment implication of a “high” risk score. Using the PRE will probably result in over-treatment with statins, which will lead to high costs for any health system. Using other considerations, such as a CT calcium score, family history, or high-sensitivity C-reactive protein, have their proponents, but these practices have not been well validated. It would be nice to use this study to calibrate the PRE, but this study evaluated five-year risk, and the PRE predicts 10-year risk. At this time, we are back to using physicians’ judgment considering all the data we have at our disposal, including the PRE and perhaps even the old LDL-cholesterol targets.