Self-selection effects in smokers attending lung cancer screening: a 9.5-year population-based cohort study in Varese, Italy.

BACKGROUND We hypothesize that mortality risk profile of participants and nonparticipants in nonrandomized lung cancer (LC) screening of smokers may be different. METHODS In 1997, a population-based cohort of 5815 smokers of Varese Province was invited to nonrandomized LC screening by annual chest x-ray examination for 4 years. LC risk factors and screening participation rate were recorded. Except for screening, the whole cohort received usual care. After 9.5-year observation, we compared mortality of participants versus nonparticipants by assessing age-standardized all-cause mortality rate ratio (MRR) and disease group-specific MRR with 95% confidence intervals (95% CI). RESULTS Self-selected screening participants were 21% of cohort. Participants were younger (p < 0.001), were more frequently current smokers (p = 0.019), had more pack-years of smoking (p < 0.0001), and had higher rate of LC family history (p < 0.0001) and of occupational LC risk (p < 0.0001) relative to nonparticipants. In logistic regression analysis familial LC, occupational risk and pack-years smoked were significant predictors of participation in screening and of developing LC. Participants displayed a healthy effect, as shown by all-cause MRR = 0.67 (95% CI, 0.53-0.84), all cancers except LC MRR = 0.61 (95% CI, 0.41-0.91), cardiovascular diseases MRR = 0.38 (95% CI, 0.22-0.63), and noncancer disease other than cardiovascular or respiratory MRR = 0.57 (95% CI, 0.34-0.92). The LC mortality (MRR = 1.40; 95% CI, 1.03-1.91) was higher in participants relative to nonparticipants (p = 0.031). CONCLUSION The selection effect in LC screening participants was dual: healthy effect and higher LC mortality. In assessing the overall effectiveness of LC screening on a population level, a higher LC mortality risk in participants should be considered.