The impact of LC screening on a population level has been scarcely investigated. The generally low participation rate in LC screenings (11-35%) [17
] and the volunteer effect bias [17
] raise concern about the generalizability to the population of interest at community level of the results of randomized LC screening trials of highly selected volunteers. As an example, the recently completed National Lung Screening Trial, (NLST) showed 20.3% mortality reduction after CT screening in selected volunteers; however only about 7% of smokers in the United States would meet the NLST criteria [28
]. Here we analyze the LC detection results and the LC survival rates at community level, in the setting of CXR screening offered to a clearly defined population-based cohort of smokers. Only 21% of this cohort participated in screening, a low attendance rate similar to that observed in mass screening for LC in Japan [29
]. Analyzing the demographic features and risk factors of the whole screening-invited cohort, we previously reported that participation was possibly prompted by increased awareness of LC risk, while it was not related to educational level [17
Overall, the LC yield of our CXR screening (0.61%) was similar to that recorded in the PLCO study (0.7%) [31
] and in the Lung Screening Study (0.68%) [32
]; the slightly lower LC yield in our screening was likely due to lower pack-years and lower compliance. In participants, we found that only about half (51%) of LCs were screen-detected, a proportion similar to that recorded in the PLCO trial (53%) [31
]. In the other half of participants, LCs were symptom-detected and survival was very poor. Overall in screening participants the rates of stage I LC diagnosis, LC resection and 5-year survival were nearly twice as high as in nonparticipants, confirming the significantly increased resectability and survival achievable with CXR screening [7
]. Notably, the Kaplan-Meier 10-year LC survival in the whole cohort was significantly greater than in the control group, and by multivariate Cox analysis this survival difference was independently related to CXR screening exposure, also after adjusting for age, gender, pack-years and histology. Improved LC survival persisting over 10 years in the population-based cohort of this study suggests mortality reduction, however no conclusion can be drawn about the effectiveness of CXR screening, because survival may be biased by lead-time, selection, length-time and overdiagnosis.
We addressed the potential impact of these biases. In the context of our study the impact of lead-time bias on long-term LC specific survival of participants seems negligible, because the LC specific survival curve of screening participants plateaued after 5 years (Figure ). Enrolment of asymptomatic individuals in the cohort was a source of healthy selection. The latter however did not influence the long-term LC survival, as shown by comparison of the Kaplan-Meier LC survival curve of cohort's nonparticipants and of control group (Figure ); the LC survival was initially greater in nonparticipants, likely due to healthy selection, but after 5 years from diagnosis it was virtually identical to that of control group (Figure and Table ). The latter finding is also consistent with the similar LC resection rate (27% vs. 26%) and similar proportion of histological diagnoses and of LC stage distribution in the nonparticipants and in the control group (Table and Table ). To evaluate length-time bias and overdiagnosis bias in our study, we focused on the VDT of screen-detected LCs, that is also an indicator of tumor aggressiveness [33
]. Incidence screen-detected LCs had short median VDT (80 days), and only one of these cancers had VDT > 300 days, indicating that most of them grew rapidly and unlikely were overdiagnosed, in agreement with the observations of other authors about CXR screen-detected LCs [22
]. Moreover, among screen-detected cancers we found no cases of bronchioloalveolar carcinoma, a slow-growing subtype that may be overdiagnosed. Furthermore, we previously showed that the LC incidence standardized rate ratio in the whole cohort was 1.07 [17
], suggesting that the number of possibly overdiagnosed LCs was minimal.
This study has limitations. The long duration of participants' enrolment may slightly underestimate the nonparticipants' survival [17
]. Compliance with annual screening progressively decreased, more markedly than in other CXR screening trials. By year 3 we recorded 59% adherence, while 79% was observed in the PLCO radiography screening [16
] and about 80% in the Mayo Lung Project [36
]. In our study the median number of CXR screenings done by each participant was four, instead of five expected. Sub-optimal compliance along with low participation rate possibly compromised the effectiveness of screening, an issue that will be addressed in a separate paper. Another limitation is that the investigators assigning the cause of death in the cohort and in the control LC group were not blinded to mode of LC detection. Sensitivity of LC death certificates however was shown to be high and similar in screening participants and nonparticipants [17
]; therefore, selective misclassification of the cause of death unlikely occurred. A relevant question is whether the 156 patients of the control LC group are an appropriate control for the LC patients found in the cohort. The control LC group source were all smokers resident in the Varese district who were diagnosed with LC during the calendar year 2000 and who met the screening criteria as of July 1997; therefore comparison with the cohort, enrolled in 1997 and representing well the Varese smokers population [17
] appears meaningful. We ruled out the possibility of period-effect for the year 2000 LC survival, because differences of LC survival in the cohort by pair-wise years of cancer diagnosis (1997-1998, 1999-2000, 2001-2002, 2003-2004, 2005-2006) were not significant (log-rank, P
= 0.401). Indeed the control group LCs closely represented the cohort LCs as for age, gender, pack-years, proportion of histologically confirmed LCs, distribution of LC types and stages, and duration of follow-up (Table ). Moreover, for the evaluation of long-term survival of the cohort, the control group seems appropriate because the cohort's nonparticipants and the controls (neither group was screened) had similar long-term survival (Table ). Another limitation is the small number of screen-detected LC cases; their survival rates therefore must be interpreted with caution. Strong features of this study are the completeness of the list of all eligible smokers resident in the 50 widely scattered communities invited to screening, and the completeness of follow-up of all LCs of the cohort and of the control group.