The results of the US National Lung Screening Trial (NLST) were published in 2011 and are considered a landmark event in lung cancer research (1). This randomized study of 53,454 individuals showed that computed tomography (CT) scans are able to reduce lung cancer mortality by 20 percent through early detection, although with important cost and morbidity due to overdiagnosis and treatment of benign nodules. (2) Several European lung cancer screening trials have also been initiated, with the largest being the NELSON trial in the Netherlands (3) and the plan is to pool data from a number of European trials (4). Clearly, screening tools that are able to identify lung cancers at an early stage have much potential to reduce the enormous burden on lung cancer mortality (5). There are now discussions on how implementation may be put in place across the world, within differing health care systems (6). The success of lung cancer screening will be dependent upon identifying populations at sufficient risk in order to maximize the benefit-to-harm ratio of the intervention.The recommendation from the US Preventive Task Forces is based on the US NLST trial, includes screening all individuals between the ages of 55 and 80 with a smoking history of 30 pack-years or more (one pack-year is 20 cigarettes/day for one year or 10 cigarettes/day for two years, etc.) (7). An in-depth analysis of the NLST showed that there were significant differences in the number of lung cancer cases detected based on underlying risk, even though all participants had satisfied the criteria for participation in the study and were considered at high risk: 60 percent of participants at highest risk for lung-cancer death (quintiles 3 through 5) accounted for 88 percent of the prevented deaths, whereas the 20 percent of participants at lowest risk (quintile 1) accounted for only 1 percent of prevented lung-cancer deaths (8).
Risk Prediction ModelsThus, accurate selection of high-risk individuals for lung cancer screening requires robust methods for risk prediction (Fig. 1). The discriminative performance of a risk model depends