Estimation of Overdiagnosis of Lung Cancer in Low-Dose Computed Tomography Screening

Abstract

This study analyzes data from the Danish Lung Cancer Screening Trial to determine the amount of overdiagnosis of computed tomography–detected lung cancer.

There is uncertainty about the extent of overdiagnosis in lung cancer screening with computed tomography (CT). The National Lung Cancer Screening Trial (NLST) estimated that 18.5% of the cancers detected by CT are overdiagnosed, whereas the Italian Lung Cancer Screening Trial (ITALUNG) found no evidence of overdiagnosis. This study aimed to estimate overdiagnosis of lung cancer by screening CT in the Danish Lung Cancer Screening Trial (DLCST).

Methods

This was an unplanned, post hoc analysis of the DLCST (NCT00496977).^1,2,3 In brief, 4104 current or former smokers (≥20 pack-years; former smokers must have quit <10 years before enrollment) aged 50 to 70 years were randomized (1:1) to 5 annual low-dose CT screenings or no screening. The absolute difference in the cumulative incidence of lung cancer in the screened and control groups was assessed 5 years after the last screening round. Overdiagnosis was calculated as the ratio between this difference and the cumulative incidence of screen-detected cancers.⁴ Bootstrapping (4999 repetitions) was used to estimate the 95% CI. Participants and practitioners could not be masked to the intervention. Cancer status and chest CT use was documented from national registries. Patients were enrolled from October 1, 2004, to March 31, 2006, and the present analysis was performed on follow-up until April 7, 2015. Participant consent was not obtained, but the presented data are deidentified. Statistical analyses were performed using R, version 3.3.1 (R Foundation for Statistical Computing). The DLCST has been approved by the Danish Scientific Ethics Committee and the Danish Data Protection Agency.

Results

A total of 4104 current or former smokers (mean [SD] age, 57.3 [4.8] years; 55.3% male) participated in the study. Participants were comparable at baseline, adherence to screening was high, and there were few losses to follow-up.^1,2 From randomization until the end of follow-up, 416 participants (20.3%) in the control group had at least 1 off-protocol chest CT: 152 participants (7.4%; 357 scans) during the trial period and 264 participants (12.9%; 807 scans) during the follow-up period. In the screened group, 338 participants (16.5%; 955 scans) had at least 1 chest CT performed during the follow-up period.

The annual lung cancer incidence rate from randomization until the end of follow-up is presented in Figure, A. At the end of follow-up, 96 participants were diagnosed with lung cancer in the screened group (64 cancers were detected by screening) vs 53 participants in the control group. There was a 2.10–percentage point (95% CI, 1.0-3.2 percentage points) increase in the absolute risk of lung cancer with low-dose CT (Figure, B). Overdiagnosis was estimated at 67.2% (95% CI, 37.1%-95.4%) of the cancers detected by screening CT.

Figure. Lung Cancer Incidence in the Danish Lung Cancer Screening Trial.

Discussion

The estimate of overdiagnosis in the DLCST (67.2%) was different from the estimate in the NLST (18.5%; 95% CI, 5.4%-30.6%),⁵ and there was no overdiagnosis in the ITALUNG.⁶ The contamination of the control group was low: 7.4% until the end of screening and 20.3% at 5-year follow-up.

All estimates of overdiagnosis were calculated with similar methods and duration of follow-up. The most extreme estimates were found in the ITALUNG and the DLCST, which shared similar eligibility criteria and study design.^1,6 Thus, the differences among the trials’ results are not adequately explained by differences in participants, interventions, or comparators.

Limitations

The main limitation of the study is the possibility of higher baseline risk of lung cancer in the screened group of the DLCST. There are 2 findings that suggest this. First, in post hoc baseline comparisons, the screened group had a 3.1–percentage point higher rate of heavy smokers and a 1–percentage point lower mean ratio of forced expiratory volume in 1 second to forced vital capacity compared with the control group.³ Second, the annual lung cancer incidence in the screened group was persistently higher. After screening stops, it takes time for the cancers that were undetectable in the last screening round to grow large enough to cause symptoms. During this time, the annual cancer incidence should be lower in the formerly screened group compared with the control group.

Conclusions

The estimate of overdiagnosis in the DLCST was larger than what has been previously reported,^5,6 but the screened group could have started with a higher baseline risk of lung cancer. However, the small differences in heavy smokers and ratio of forced expiratory volume in 1 second to forced vital capacity cannot explain the 67% overdiagnosis rate. Practice should not be changed immediately; however, it is crucial that the remaining trials report their estimates of overdiagnosis because this is a critical outcome for screening participants.