IN THE 1980S, ROSE COINED THE TERM PREVENTION PARAdox to describe the fact that a large proportion of cardiovascular disease (CVD) events occur among the many individuals with average risk factor values. He distinguished between 2 approaches to CVD prevention. The highrisk strategy, which aims to truncate the upper tail of the normal distribution of risk factors, focuses on individuals who are most likely to benefit personally from preventive treatment. By contrast, the population-based strategy aims to shift the entire risk distribution. At the time, the available lipid-lowering therapies were limited, none was well tolerated, and the risk-benefit profile for clofibrate, for instance, argued against its widespread use. Soon, the high-risk approach came to be synonymous with the use of drugs, and the population approach was identified with efforts to shift norms about diet, physical activity, or smoking. Modest lifestyle changes could be recommended to the population at large because evolutionarily sensible interventions such as a low-salt diet may be “presumed to be safe.” The real-world effects of such recommendations have been limited. Targeting high-risk individuals with preventive drug therapies optimizes the benefits for the individuals concerned but does little to reduce the overall burden of CVD in the population. In the 1990s, evidence emerged documenting the efficacy of a new class of potent low-density lipoprotein (LDL)–cholesterol-lowering agents, the 3-hydroxy-3methylglutaryl coenzyme A reductase inhibitors (statins), first in patients with coronary disease and subsequently in primary prevention. Meta-analyses of large statin trials have demonstrated an approximately 25% relative reduction in risk of CVD events in both primary and secondary prevention populations and in patients with a varying risk factor profile. For this reason, the number needed to treat to prevent 1 CVD event depends largely on the baseline absolute risk. When evidence of the benefits of statins first emerged and the potential for high-volume use was recognized, a strategy was necessary to maximize benefit and to limit any potential harm because of the high initial cost of these agents and the uncertainty about their long-term safety. In the United States, individuals considered for statin therapy were originally identified according to an agreed LDL cholesterol threshold. Because LDL cholesterol does not perform well as a CVD screening test, a threshold based on absolute risk was adopted in the United Kingdom. This absolute risk– based approach reduces the number needed to treat to prevent 1 event and maximizes the health benefits for a given cost. In the United Kingdom, it is now routine for family practitioners to use tables or computer programs based on validated risk equations to estimate individual risk at the point of care. The widely used risk models accurately assign individuals to different categories of risk in such a way that the observed event rates for the risk groups are close to the predicted rates. It is not widely appreciated, however, that risk models fail to efficiently distinguish or discriminate between patients who will or will not experience events. Recent research efforts have therefore focused on new biomarkers and vascular imaging tools to improve discrimination and risk stratification. Of these, C-reactive protein (CRP) has been associated with risk of CVD events, and a US consensus statement suggests a CRP cut point of 3 mg/L may aid the identification of high-risk patients. The fact that mendelian randomization studies suggest that CRP is not a cause of vascular disease is not important for the purpose of prediction. Nevertheless, when assessed with appropriate metrics of predictive performance, a CRP measurement, like LDL cholesterol alone, is a poor discriminator of future CVD events and only marginally enhances risk stratification using established risk factors. Vascular imaging techniques are costly and some involve radiation exposure, which may limit their applicability for primary prevention. New metrics for assessing predictive performance have been advocated based on reclassification. These assess the extent to which the addition of a new marker to an estab-