Computed tomography screening for lung cancer: Results of ten years of annual screening and validation of cosmos prediction model

Abstract Introduction It is unclear how long low-dose computed tomographic (LDCT) screening should continue in populations at high risk of lung cancer. We assessed outcomes and the predictive ability of the COSMOS prediction model in volunteers screened for 10 years. Materials and methods Smokers and former smokers (>20 pack-years), >50 years, were enrolled over one year (2000–2001), receiving annual LDCT for 10 years. The frequency of screening-detected lung cancers was compared with COSMOS and Bach risk model estimates. Results Among 1035 recruited volunteers (71% men, mean age 58 years) compliance was 65% at study end. Seventy-one (6.95%) lung cancers were diagnosed, 12 at baseline. Disease stage was: IA in 48 (66.6%); IB in 6; IIA in 5; IIB in 2; IIIA in 5; IIIB in 1; IV in 5; and limited small cell cancer in 3. Five- and ten-year survival were 64% and 57%, respectively, 84% and 65% for stage I. Ten (12.1%) received surgery for a benign lesion. The number of lung cancers detected during the first two screening rounds was close to that predicted by the COSMOS model, while the Bach model accurately predicted frequency from the third year on. Conclusions Neither cancer frequency nor proportion at stage I decreased over 10 years, indicating that screening should not be discontinued. Most cancers were early stage, and overall survival was high. Only a limited number of invasive procedures for benign disease were performed. The Bach model – designed to predict symptomatic cancers – accurately predicted cancer frequency from the third year, suggesting that overdiagnosis is a minor problem in lung cancer screening. The COSMOS model – designed to estimate screening-detected lung cancers – accurately predicted cancer frequency at baseline and second screening round.