Association of computed tomography screening with lung cancer stage shift and survival in the United States: quasi-experimental study
BMJ 2022; 376 doi: https://doi.org/10.1136/bmj-2021-069008 (Published 30 March 2022) Cite this as: BMJ 2022;376:e069008Editorial
Screening high risk populations for lung cancer
All rapid responses
Rapid responses are electronic comments to the editor. They enable our users to debate issues raised in articles published on bmj.com. A rapid response is first posted online. If you need the URL (web address) of an individual response, simply click on the response headline and copy the URL from the browser window. A proportion of responses will, after editing, be published online and in the print journal as letters, which are indexed in PubMed. Rapid responses are not indexed in PubMed and they are not journal articles. The BMJ reserves the right to remove responses which are being wilfully misrepresented as published articles or when it is brought to our attention that a response spreads misinformation.
From March 2022, the word limit for rapid responses will be 600 words not including references and author details. We will no longer post responses that exceed this limit.
The word limit for letters selected from posted responses remains 300 words.
Dear Editor
We would like to congratulate the authors on the publication of this well-designed study demonstrating
the impact of the introduction of lung cancer screening on lung cancer stage shift and survival at the
population level. Over the last decade, a slow but significant increase in the use of lung cancer screening
has occurred in the U.S. However, until now, it was unclear whether these efforts to screen high-risk
individuals have resulted in any measurable stage shift and improvement in survival at the population
level. For the first time, the findings of this study show that even a small amount of screening can shift
the needle to earlier diagnosis and better survival. These findings are incredibly encouraging and
demonstrate the tremendous potential for lung cancer screening to save lives.
Given the thoughtful design of this study by Potter and colleagues and the importance of its findings, we
were surprised to read the previous rapid responses (by Drs. Tao, Goh and Raffle) regarding this article.
We would like to address these rapid responses as there seem to be unwarranted concerns raised
regarding length-time bias and overdiagnosis bias.
First, to address the concern that the benefits you’re seeing from low-dose lung cancer screening is
simply from length-time bias: this issue has been definitively put to rest with the results of two large-
scale randomized controlled trials, the National Lung Cancer Screening Trial (n = 53,454) and the Dutch-
Belgian Randomized Lung Cancer Screening Trial (n = 15,789) [2,3]. Multiple papers have been written
about this already, but briefly, we will just point out that randomization has been shown to be an
effective approach to minimizing length-time bias. For example, Shlomi and colleagues have noted:
“Significant reduction in lung cancer mortality, as demonstrated exclusively by the NLST, lowers the
magnitude of lead- and length-time biases, something which the smaller randomised controlled trials
were unable to accomplish, for lack of statistical power”, further demonstrating that randomization
does, in fact, minimize such biases [4].
Second, with regard to overdiagnosis rates concerns noted by some of the rapid responses, this issue
has already been addressed thoroughly in the aforementioned randomized trials, which clearly
demonstrate the low overdiagnosis rates associated with LDCT lung cancer screening [2,3]. The
overdiagnosis rates (3.1-8.9%) reported by the NLST and NELSON trials have been shown to be on par
with those reported for other mainstream cancer screenings, including mammograms (9.7-12.4%) and
fecal occult blood tests (2-7.6%) [2,3,5,6,7].
Additionally, we would like to address the comparison made in a previous rapid response, which
attempted to compare LDCT lung cancer screening with screening for infant tumor neuroblastoma. This
comparison is flawed. The observed stage shift and improvement in survival associated with screening
for infant tumor neuroblastoma occurred before a randomized trial was conducted to assess the
effectiveness of such screening in reducing mortality [2,3]. The screening test was then found to be
ineffective in reducing mortality in a randomized trial [8]. In contrast, LDCT screening has already been
shown in two large-scale randomized trials to reduce lung cancer mortality. Even if LDCT screening
detects some indolent disease, there is no doubt that it can reduce mortality from lung cancer [2,3].
Furthermore, this comparison appears to misconstrue the objective of the study by Potter and
colleagues [1]. The authors’ objective was not to assess whether LDCT screening is effective in reducing
lung cancer mortality (there is no doubt of this), but to evaluate whether efforts to increase lung cancer
screening have had any measurable impact on stage shift and survival at the population level in the US.
Of note, Potter and colleagues specifically analyzed patients with indolent histologic subtypes – these
are the types of lung cancer that some skeptics out there think that patients could die “with” and not die
“from”. From 2010-2018, the authors found that “the percentage of “indolent” histologies diagnosed
decreased or did not change (supplementary table C).” Furthermore, when they only examined
aggressive histologies, they still observed an accelerated increase in the percentage of stage I disease
identified from 2010-2014. These findings clearly demonstrate that the stage shift observed is not from
overdiagnosis.
We would like to again congratulate Potter and colleagues on an outstanding study performed and
encourage further efforts to increase lung cancer screening for at-risk populations.
1. Potter, A. L., A. L. Rosenstein, M. V. Kiang, S. A. Shah, H. A. Gaissert, D. C. Chang, F. J. Fintelmann and
C. J. Yang (2022). "Association of computed tomography screening with lung cancer stage shift and
survival in the United States: quasi-experimental study." BMJ 376: e069008.
2. National Lung Screening Trial Research, T., D. R. Aberle, A. M. Adams, C. D. Berg, W. C. Black, J. D.
Clapp, R. M. Fagerstrom, I. F. Gareen, C. Gatsonis, P. M. Marcus and J. D. Sicks (2011). "Reduced
lung-cancer mortality with low-dose computed tomographic screening." N Engl J Med 365(5): 395-
409.
3. de Koning, H. J., C. M. van der Aalst, P. A. de Jong, E. T. Scholten, K. Nackaerts, M. A. Heuvelmans, J.
J. Lammers, C. Weenink, U. Yousaf-Khan, N. Horeweg, S. van 't Westeinde, M. Prokop, W. P. Mali, F.
A. A. Mohamed Hoesein, P. M. A. van Ooijen, J. Aerts, M. A. den Bakker, E. Thunnissen, J.
Verschakelen, R. Vliegenthart, J. E. Walter, K. Ten Haaf, H. J. M. Groen and M. Oudkerk (2020).
"Reduced Lung-Cancer Mortality with Volume CT Screening in a Randomized Trial." N Engl J Med
382(6): 503-513.
4. Shlomi, D., R. Ben-Avi, G. R. Balmor, A. Onn and N. Peled (2014). "Screening for lung cancer: time for
large-scale screening by chest computed tomography." Eur Respir J 44(1): 217-238.
5. Houssami, N. (2017). "Overdiagnosis of breast cancer in population screening: does it make breast
screening worthless?" Cancer Biol Med 14(1): 1-8.
6. Nelson, H. D., R. Fu, A. Cantor, M. Pappas, M. Daeges and L. Humphrey (2016). "Effectiveness of Breast Cancer Screening: Systematic Review and Meta-analysis to Update the 2009 U.S. Preventive Services Task Force Recommendation." Ann Intern Med 164(4): 244-255.
7. Wieszczy, P., M. F. Kaminski, M. Loberg, M. Bugajski, M. Bretthauer and M. Kalager (2021).
"Estimation of overdiagnosis in colorectal cancer screening with sigmoidoscopy and faecal occult
blood testing: comparison of simulation models." BMJ Open 11(4): e042158.
8. Chamberlain, J. (1994). "Screening for neuroblastoma: a review of the evidence." J Med Screen 1(3):
169-175.
Competing interests: No competing interests
Dear Editor
It is puzzling to see publication in a peer review journal of an article where 'stage shift' and survival statistics are used apparently to imply true benefit from screening. Neither of these statistics are reliable measures of the balance of benefit and harm. Both are subject to length time bias and overdiagnosis bias (1) .
Length time bias refers to the phenomenon whereby screening preferentially picks up small slow growing good prognosis lesions, and preferentially misses rapidly progressive poor prognosis lesions.
Overdiagnosis bias is the phenomenon whereby screening picks up ‘pseudodisease’ i.e. symptomless conditions that fit the case definition for the screening programme, yet in truth would never progress to causing illness during the person’s natural lifespan.
The combined effect of these biases means that any group of screen-detected 'cases' is, by definition, fundamentally different to a group of cases diagnosed following presentation with signs or symptoms. The stage distribution and survival rate will automatically appear more favourable, irrespective of whether screening brings benefit or not, simply because you are pulling additional innocent and easily treatable cases into the denominator.
In case anyone feels this is hair-splitting or an idle distinction, it is worth remembering experience with screening for the infant tumour neuroblastoma. Ten years of neuroblastoma screening in Japan saw dramatic improvements in survival statistics and stage shift, yet no certain change in the numbers of infants presenting with poor prognosis disease, or in the number of deaths (2). Controlled trials were needed, and two controlled trials were performed (3,4) . These revealed no improvement in neuroblastoma mortality amongst screened infants compared with those receiving usual care, and a slightly higher, though statistically non-significant rate of treatment-related harm in the screened infants.
1. Chapter 4 in Raffle, A.E. Mackie, A. Gray, J.A.M. Screening; Evidence and Practice. 2nd edition. Oxford University Press. June 2019
2. Murphy, S.B., Cohn, S.L., Craft, A.W., et al. (1991) Do children benefit from mass screening for neuroblastoma? Consensus Statement from the American Cancer Society Workshop on Neuroblastoma Screening. Lancet 337, 344–346.
3. Schilling, F.H., Spix, C., Berthold, F., et al. (2002) Neuroblastoma screening at one year of age. New England Journal of Medicine 346, 1047–1053.
4. Woods, W.G., Gao, R.N., Shuster, J., et al. (2002) Screening of infants and mortality due to neuroblastoma. New England Journal of Medicine 346, 1041–1046.
Competing interests: No competing interests
Dear Editors
My concerns about Low Dose Computerised Tomography (LDCT) for Lung Cancer (LC) screening of high risk individuals has been well documented in the BMJ responses (Ref 1-4). The matter of cost-effectiveness and true mortality rates are always the issues in contention.
Again I reiterate:
"The efficacy of LDCT in picking up early stages of lung cancer is never in doubt, nor the evidence that some lung cancer when picked up in the early stages are treatable and even curable. However the main uncertainties are what kind of screening programme, including the age group for recruitment, the frequency of screening, the outcome for those whose lung cancer are detected early versus the complication associated with false positive and ultimately the true cost effectiveness of the lung cancer screening using LDCT.
The conflict between clinicians who view reducing all cause mortality (ACM) as the main goal of any population screening programme versus those who view only disease-specific mortality (DSM) as the evidence to support any screening regimen is ongoing; the BMJ has previously featured the perspectives of the opposing sides" (Ref 5,6).
Despite attempts at suggesting that lead-time bias can be corrected in a "large randomized controlled trial which looked at differences in lung cancer mortality rates between the two arms of the trial" (ref 7), I cannot see how that can be achieved in trials like the National Lung Screening Trial (NLST) (ref 8).
In the NLST, it was found that:
"the percentage of stage IA and stage IB lung cancers was highest among cancers that were diagnosed after a positive screening test . Fewer stage IV cancers were seen in the low-dose CT group than in the radiography group at the second and third screening rounds . Low-dose CT screening identified a preponderance of adenocarcinomas, including bronchioloalveolar carcinomas"
Furthermore "in both groups, many adenocarcinomas and squamous-cell carcinomas were detected at either stage I or stage II, although the stage distribution was more favorable in the low-dose CT group than in the radiography group . Small-cell lung cancers were, in general, not detected at early stages by either low-dose CT or radiography." Certainly small cell lung cancer (SCLC - with far lower 5-year survival upon diagnosis compared to non small cell lung cancer or NSCLC, ref 9) is the main driver of the perceived bleak prognosis of "lung cancer" diagnosis amongst the medical fraternity for generations.
As the accompanying editorial acknowledges, the phenomenon of "slow natural history of NSCLC" (ref 10), in my opinion the matter of lead time bias would undoubtedly be a real concern.
By focusing on NSCLC survival (hence excluding the SCLC), the apparent favourable results exaggerates the positive effect of LDCT lung cancer screening which was to detect both SCLC and NSCLC. The dramatic effect of LDCT the authors attempt to persuade with the change in all cause survival post 2013 is deceiving, considering that the "USPSTF first recommended low dose computed tomography screening for people at high risk in December 2013" without any formal screening program implemented then. While one interpretation of this result can be attributed to true longer survival of those with NSCLC, another interpretation would also be that there is an increase in diagnosis of early stage NSCLC with high survival rate with or without early detection (and treatment) consistent with lead time bias as suggested by another respondent (ref 11)
This study using big data and all cause survival may help in understanding the effect of LDCT lung cancer screening on early diagnosis of NSCLC, however it does not convince me of the true cost-effectiveness and its effect on mortality rate of lung cancer.
References
1. https://www.bmj.com/content/359/bmj.j5450/rr-0
2. https://www.bmj.com/content/364/bmj.l54/rr
3. https://www.bmj.com/content/365/bmj.l1416/rr-0
4. https://www.bmj.com/content/359/bmj.j5742/rr-1
5. https://www.bmj.com/content/352/bmj.h6080
6. https://www.bmj.com/bmj/section-pdf/187371?path=/bmj/343/7830/Head_to_He...
7. https://www.bmj.com/content/359/bmj.j5742/rr-2
8. https://www.nejm.org/doi/full/10.1056/nejmoa1102873
9. https://www.webmd.com/lung-cancer/guide/lung-cancer-survival-rates
10. https://www.bmj.com/content/376/bmj.o666
11. https://www.bmj.com/content/376/bmj-2021-069008/rr
Competing interests: No competing interests
Dear editor,
I read with great interest this article[1]。The authors provided an in-depth analysis of the population benefits of Large-scale screening.
While we see significant improvements in the median all cause survival for lung cancer due to screening, it should be noted that this median all cause survival may be overestimated due to lead time bias.
When the difference between screening diagnostic time and clinical diagnostic time is interpreted as the prolonged survival time of screening, there exists lead time bias. This apparently prolonged survival time is actually a bias caused by screening leading to an earlier diagnosis. To take an extreme example, a lung cancer patient's tumor formation time is 2010, the clinical diagnosis time is 2012, the final death time is 2018, and the survival time is 6 years. If the patient is diagnosed earlier in 2011 because of screening, even if the death time is still 2018, the survival time will be calculated as 7 years, but the seemingly prolonged one-year survival time cannot be interpreted as a benefit due to screening. The results should therefore be interpreted more cautiously.
Besides, lung cancer overdiagnosis and the resulting overtreatment should be taken into consideration. And a 16 additional years of follow-up study by Pamela M Marcus et al. confirm that this concern does exist[2]. Excessive medical treatment not only wastes precious medical resources, but also brings enormous mental pressure to patients, seriously affecting the quality of life.
Screening does bring great benefits to tumor survival, but its value should be more objectively estimated, so that healthcare resources could be used wisely and result in greater benefits.
[1] POTTER A L, ROSENSTEIN A L, KIANG M V, et al. Association of computed tomography screening with lung cancer stage shift and survival in the United States: quasi-experimental study [J]. 2022, 376(e069008.
[2] MARCUS P M, BERGSTRALH E J, ZWEIG M H, et al. Extended lung cancer incidence follow-up in the Mayo Lung Project and overdiagnosis [J]. Journal of the National Cancer Institute, 2006, 98(11): 748-56.
Competing interests: No competing interests
Misdirection by invoking other studies does not absolve this study's potential flaws
Dear Editors
I refer to Venkateswaran & Liou’s response (ref 1).
Drs. Tao, Raffle and myself (ref 2-4) raised 3 issues regarding Lung Cancer Screening using LDCT (LCS-LDCT):
1. Lead-time bias (see https://online.stat.psu.edu/stat507/lesson/10/10.6)
2. Length-time bias including overdiagnosis
3. Cost-effectiveness
Venkateswaran & Liou addressed mostly the issue of length-time bias/overdiagnosis, justifying their stance that benefits seen from LCS-LDCT cannot be due to length-time bias by pointing to 2 RCTs, NLST and NELSON.
While randomisation can reduce the effects of length-time bias in a trial, two additional criteria were required:
The primary outcome measures must involve mortality rates rather than survival rates
The time origin is taken as the point of randomization, not the point of diagnosis of lung cancer
NLST and NELSON appear to have fulfilled these criteria.
However those were experimental trials rather than observational "real-world" studies, which is what this retrospective analysis of "quasi-experimental study" was.
If RCTs are the final word for any cancer screening program, there is no need for observational "real-world" studies since everything that is not satisfactory addressed in these kinds of studies can be whitewashed by referring to the RCTs.
Each study should be judged on its merit; any flaw should be acknowledged. This study has characteristics which cannot address length-time bias:
1. It is retrospective
2. It is not randomised
3. It uses survival rates rather than mortality rates
Unless correction were made accounting for lead- and length-time bias (eg ref 5), otherwise concerns about bias are justified.
Despite obvious strong advocacy for LCS-LDCT by various researchers (ref 6,7), they also acknowledged the issue of high rates of false-positives (96.4% in NLST compared to 94.5% in comparison radiography group) and subsequently the cost-effectiveness of LCS-LDST using NLST screening criteria (which the USPSTF recommendations used in this article is based on) is still contentious.
Venkateswaran & Liou quoted Shlomi et al to support their idea that "randomization does, in fact, minimize such biases". Shlomi et al actually have further comments providing context:
"Significant reduction in lung cancer mortality, as demonstrated exclusively by the NLST, lowers the magnitude of lead- and length-time biases, something which the smaller randomised controlled trials were unable to accomplish, for lack of statistical power. Disease-specific mortality reduction is the most conclusive indicator of the benefits of screening. The mortality reduction measure provides the net effect of screening regardless of the length of time recorded from diagnosis to death. Randomisation of a large population together with long-term follow-up can scientifically lower selection biases, although the magnitude of this reduction is currently unknown."
"We saw earlier that, as attested by the NLST, mortality rate reduction lowered lead- and length-time biases, including the extreme aspect of length-time bias, the overdiagnosis. These biases were not overcome by other, smaller randomised controlled trials, where no mortality reduction was demonstrated. The exact impact of mortality reduction in diminishing the effect of those biases is yet to be gauged. In any case, a benefit produced for a specific population is not synonymous with a benefit enjoyed by the individual patient. It is essential that the system be able to identify those patients in whom indolent cancer may never develop to cause them real harm thus rendering treatment unnecessary. Screening, especially with a tool as sensitive as CT, identifies small lesions of uncertain significance. All lung cancer screening trials with low-dose CT found suspected nodules in a quarter to a third of examinations on average. As mentioned above, at least 90% of CT findings in most trials were false-positive results. However, even in cases of cancer diagnosis, death is not inevitable. While highly aggressive cancer will cause death regardless of screening, some very indolent cancers, particularly in older patients with co-morbidities, will not necessarily be the cause of death. Treatment of indolent cancers could end up in complications and mortality. Conversely, treatment of indolent lung cancers biases the reported results in such a way as to show higher survival rates and lower death rates of lung cancer in the screened population."
Hardly an unreserved endorsement of randomisation addressing length-time bias.
Thus Lead-time bias, Length-time bias (including overdiagnosis) and cost-effectiveness not adequately addressed in this study, and hence concerns were justified.
References
1. https://www.bmj.com/content/376/bmj-2021-069008/rr-2
2. https://www.bmj.com/content/376/bmj-2021-069008/rr
3. https://www.bmj.com/content/376/bmj-2021-069008/rr-0
4. https://www.bmj.com/content/376/bmj-2021-069008/rr-1
5. https://doi.org/10.1093/aje/kwn120
6. DOI: 10.1056/NEJMoa1102873
7. http://dx.doi.org/10.1016/j.acra.2014.10.011
8. DOI: 10.1183/09031936.00164513
Competing interests: No competing interests