Generalisability of the results of the Dutch-Belgian randomised controlled lung cancer CT screening trial (NELSON): does self-selection play a role?

Abstract

The degree of self-selection in the Dutch-Belgian randomised controlled lung cancer screening trial (NELSON) was determined to assess the generalisability of the study results. 335,441 (mainly) men born in 1928-1953 received a questionnaire. Of the respondents (32%), eligible subjects were invited to participate (19%). Fifty-five percent gave informed consent and was randomised. Background characteristics were compared between male respondents on the first questionnaire (n = 92,802), eligible subjects among them (n = 18,570) and those randomised (n = 10,627) and Statistics Netherlands 2002-2005 (SN) (n = 5289) or GLOBE study-data (Dutch cohort) (n = 696). Initial respondents were less likely to be highly educated (OR(adj) = 0.84; 95% CI: 0.74-0.96) and comprised of significantly less current smokers (OR(adj) = 0.65; 95% CI: 0.61-0.69) compared to the general population. These current smokers smoked more heavily (OR(adj) = 1.23; 95% CI: 1.10-1.37), but for a shorter time-period (respondents: 31, SN: 42 years, p < 0.001). Age, general health, BMI, alcohol use and cancer prevalence were comparable. The randomised population was younger (Age 50-65) (randomised subjects: 85.3%, SN: 72% (p < 0.01)) comprised of more heavy current smokers (OR = 2.08; 95% CI: 1.75-2.44), that smoked for a shorter period of time (randomised subjects: 37, SN_selection: 42 years (p < 0.001)). Both the respondents (32%) of the first questionnaire as well as the randomised population of the NELSON trial appeared to differ slightly on smoking characteristics, but the differences were limited and probably balance each other. Results of the NELSON trial will be roughly applicable to the Dutch and probably other populations that fulfil our selection criteria.