Reduced Lung-Cancer Mortality with Volume CT Screening in a Randomized Trial
2020 · paper · Clinical trials
The NELSON randomized trial of volume-based CT lung screening in 13,195 men and 2594 women who smoked, versus no screening. At 10 years, lung-cancer mortality was lower in screened men (rate ratio 0.76), with similar suggestive findings in women.
3 findings
Substance Introduction: the NLST result came "after a median follow-up of 6.5 years, and the trial recently confirmed that mortality at a median follow-up of 5.5 and 6.0 years was as much as 19% lower" (ref 6, titled "...with extended follow-up..."). The extended-follow-up medians are shorter than the original median.
- p. 2 Introduction, second paragraph
the trial recently confirmed that mortality at a median follow-up of 5.5 and 6.0 years was as much as 19% lower
- p. 10 Reference 6
Lung cancer incidence and mortality with extended follow-up in the National Lung Screening Trial
- p. 2 Introduction, paragraph 2
after a median follow-up of 6.5 years, and the trial recently confirmed that mortality at a median follow-up of 5.5 and 6.0 years was as much as 19% lower
Only the authors can say what is right here.
Judge's note: Rests on outside knowledge: that the extended NLST follow-up report (ref 6, J Thorac Oncol 2019) used longer medians than 6.5 years (from memory about 11.3 and 12.3 years, with a smaller reduction). Inside the files: page 2 gives the original 6.5-year median and then a 'recent confirmation' at 5.5 and 6.0 years, citing refs 5 and 6. Ref 6 on page 10 is titled 'Lung cancer incidence and mortality with extended follow-up in the National Lung Screening Trial'. A median shorter than the original cannot describe extended follow-up, so the durations, and possibly the 19%, look misstated. The authors should check them against ref 6.
Second check: another model agreed
Report a problem with this finding
Substance Table 2, Round 3, Detection of Lung Cancer: "65/5758 (1.1)" uses denominator 5758, while every other round-3 cell uses 5768 men screened.
- p. 6 Table 2, Round 3, Detection of Lung Cancer column
65/5758 (1.1)
- p. 6 Table 2, Round 3, uptake/indeterminate/positive columns
385/5768 (6.7)
Judge's note: Checked Table 2 on page 6. In round 3, uptake is 5768/6285, indeterminate 385/5768 and positive 136/5768. The total denominator 22,600 equals 6309+6086+5768+4437. No separate denominator for detection is defined anywhere. The percentage rounds to 1.1 either way, so only the denominator is inconsistent.
Second check: another model agreed
Report a problem with this finding
Reference or label "screening uptake was on average 90.0% (95% confidence interval [CI], 76.9 to 95.8)": the bounds are the lowest (round 4, 76.9) and highest (round 1, 95.8) round-specific uptakes in Table 2, i.e. a range, not a 95% CI.
- p. 4 Results, Screening Results in Male Participants
screening uptake was on average 90.0% (95% confidence interval [CI], 76.9 to 95.8) (Table 2)
- p. 6 Table 2, Screening Uptake, Men Eligible for Screening
4437/5771 (76.9)
Judge's note: Table 2 on page 6 gives uptake among eligible men as 95.8, 94.2, 91.8 and 76.9 by round, and 22,600/25,098 (90.0) in total. The printed bounds match the minimum and maximum across rounds exactly. A 95% CI for a pooled proportion with this denominator would be far narrower. The likely fix is to label it 'range'.
Second check: another model agreed
Report a problem with this finding
7 minor (typos and the like; most checked by the judge only)
Wording Abstract says "9.2% of the screened participants underwent at least one additional CT scan", but Results and Table 2 give 9.2% as a share of scans: "In 9.2% of the scans (2069 of 22,600)".
- p. 1 Abstract, Results
On average, 9.2% of the screened participants underwent at least one additional CT scan
- p. 4 Results, Screening Results
In 9.2% of the scans (2069 of 22,600)
Judge's note: Checked the page 1 abstract and the page 4 Results. Table 2 on page 6 has a Total row of 2069/22,600 (9.2), and 22,600 is the number of CT scans across four rounds, not the number of screened men (6583 randomized to screening). The abstract wording describes a per-participant quantity that the paper does not report. The figure itself is right as a per-scan rate.
Second check: another model agreed
Report a problem with this finding
Wording Abstract says "A minimum follow-up of 10 years until December 31, 2015, was completed for all participants", but randomization ran "from December 2003 through July 2006" and censoring was at "December 31, 2015, or 10 years of follow-up since randomization (whichever came first)".
- p. 1 Abstract, Methods
A minimum follow-up of 10 years until December 31, 2015, was completed for all participants.
- p. 3 Methods, Recruitment
from December 2003 through July 2006 (median randomization date, November 2004)
- p. 4 Methods, Statistical Analysis
December 31, 2015, or 10 years of follow-up since randomization (whichever came first)
Only the authors can say what is right here.
Judge's note: Page 3 (Recruitment) gives randomization from Dec 2003 through July 2006. Page 4 (Statistical Analysis) gives censoring at 31 Dec 2015 or 10 years, whichever came first. Anyone randomized after 31 Dec 2005 had less than 10 years of follow-up by the cutoff, so a minimum of 10 years for all participants cannot hold under the stated censoring rule. Only the authors can say which statement is intended.
Second check: another model agreed
Report a problem with this finding
Wording "Similar rate ratios ... were observed at years 8, 9, and 11 (Fig. 1 and Table S3)": Fig. 1 ends at year 10, year 11 is reported separately as a sensitivity analysis, and the abstract lists only "years 8 and 9".
- p. 6 Mortality
were observed at years 8, 9, and 11
- p. 1 Abstract
similar to the values at years 8 and 9
- p. 6 Mortality, second paragraph
were observed at years 8, 9, and 11 (Fig. 1 and
- p. 7 Sensitivity Analyses
At the 11-year follow-up (up to December 2016)
Judge's note: Fig. 1 on page 6 has x-axes running 0–10. Page 7 (Sensitivity Analyses) reports year 11 (to Dec 2016) as RR 0.78 (95% CI, 0.63 to 0.95), so the year-11 claim is supported elsewhere in the paper. The reviewers' suggested replacements ('7, 8, 9' or '8, 9, 10') are not supported. The defect is only that the pointer to Fig. 1 cannot show year 11, plus the mismatch with the abstract. Table S3 is not in the files.
Second check: another model agreed
Report a problem with this finding
Wording "264 of 22,600 screened participants over all rounds (1.2%) had a false positive test": 22,600 is the number of CT scans (Table 2), not participants (6583 men in the screening group).
- p. 4 Results, Screening Results in Male Participants
264 of 22,600 screened participants over all rounds (1.2%) had a false positive test
Judge's note: Page 4 says just before this that '22,600 CT scans were performed', and that '467 of 22,600 CT scans (2.1%) were test-positive'. 467 − 203 = 264, so the arithmetic is right. Only the unit (participants instead of scans) is misdescribed.
Second check: another model agreed
Report a problem with this finding
Wording "Most (screening-detected) lung cancers were adenocarcinomas (52.0% in the screening group and 43.8% in the control group)": 52.0% is for any lung cancer in the screening group (179/344); screening-detected cancers are 123/203 (60.6%).
- p. 6 Lung Cancer in Male Participants
Most (screening- detected) lung cancers were adenocarcinomas (52.0% in the screening group and 43.8% in the control group).
- p. 7 Table 3
123 (60.6)
- p. 6 Lung Cancer in Male Participants
Most (screening-detected) lung cancers were adenocarcinomas (52.0% in the screening group
Judge's note: Table 3 on page 7 gives adenocarcinoma as 123 (60.6) for screening-detected, 179 (52.0) for any cancer in the screening group, and 133 (43.8) for the control group. The control group has no screening-detected cancers. So the quoted percentages are for all cancers, and the parenthetical 'screening-detected' misdescribes them.
Second check: another model agreed
Report a problem with this finding
Wording Discussion: "Important differences were seen in screening results at baseline ... the percentage of patients with a positive test was 2.1% in the NELSON trial ... and the positive predictive value was 43.5%". These are the all-round totals of Table 2; baseline (round 1) values are 2.3% (147/6309) and 38.1%.
- p. 9 Discussion, first paragraph
Important differences were seen in screening results at baseline in the NELSON trial
- p. 9 Discussion
the percentage of patients with a positive test was 2.1% in the NELSON trial and 24% in the NLST, and the positive predictive value was 43.5% and 3.8%
- p. 6 Table 2, Round 1 and Total rows
147/6309 (2.3)
Only the authors can say what is right here.
Judge's note: Table 2 on page 6 gives Round 1 as 147/6309 (2.3) with PPV 38.1, and Total as 467/22,600 (2.1) with PPV 43.5. The figures labelled 'at baseline' match the Total row, which is also per scan rather than per patient. Only the authors can say whether 'at baseline' or the numbers are intended.
Second check: another model agreed
Report a problem with this finding
Typo Table 1, Duration of smoking: categories "41–45 yr" and "≥45 yr" overlap at 45 years.
- p. 5 Table 1, Duration of smoking
≥45 yr
Judge's note: Checked Table 1 on page 5. The other bins do not overlap (≤25, 26–30, 31–35, 36–40, 41–45), so the last bin should presumably be >45 or ≥46. The counts sum to the denominator (25+657+1652+2030+1451+748 = 6563), so this is a label defect only.
Second check: none (judge only)