'Conflicting' CT lung trial data can be explained

In the past few months, results of two studies evaluating CT screening for lung cancer have been published in prestigious medical journals, both getting enormous amounts of coverage in the lay press.

The first was the I-ELCAP (International Early Lung Cancer Action Program), which reported on the curability of lung cancer, using 10-year survival rates in order to estimate the rate of "cure"1 attributable to the early treatment allowed by early diagnosis. The results showed that deaths from lung cancer could be reduced by 80% if people at a suitable level of risk enrolled in a screening program.

The study used a mathematical model based on retrospective data in a nonscreened cohort to predict lung cancer deaths, compared this with three small CT screening studies, and, based on five years of follow-up, found no statistically significant evidence for a reduction in mortality, although the study suggested that a benefit up to 30% reduction was possible.2 Nevertheless, the lead author subsequently described this in the lay press as, "not a trace of evidence" for a benefit associated with CT screening.3

These "conflicting" results have now led many experts to suggest that the question of benefit will be settled only when the results from the National Lung Screening Trial (NLST) are available in 2010. In the meantime, the experts suggest that the best way to avoid lung cancer is to quit smoking.3

Most-we hope all-medical experts agree that smoking causes lung cancer and that stopping smoking will reduce the chance of dying from it. While that relationship of smoking to death from lung cancer is not seriously questioned, its basis does not come from randomized controlled trials (RCTs). Consider the idea of proving that smoking cessation saves lives by reducing deaths from lung cancer and how this might be accomplished through the use of an RCT.

It could be designed as follows: Smokers would be randomized to two groups, one that stops smoking, and one that continues to smoke. The groups are followed for five years, and deaths from lung cancer in each group are compared. Traditional estimates of mortality reduction could be calculated, and the two groups compared to see if smoking cessation was beneficial in terms of saving lives.

However, if this were actually done, there would probably be little, if any, difference between the two groups (in terms of death from lung cancer), leading to the conclusion that the RCT showed no benefit for smoking cessation. Allowing the group that stopped smoking to start smoking again after a short period, say three years, and then continuing to compare the groups over a longer follow-up period would result in an even more dramatic failure to show benefit.

Such a hypothetical RCT would fail to show a lung cancer-specific mortality reduction even though we know that the intervention does indeed work because the lung cancer deaths prevented by smoking cessation would have occurred well into the future, so focusing on those early years would show nothing. The benefit associated with the intervention obviously becomes diluted by allowing the smoking cessation group to start smoking again, so that no amount of follow-up would be useful. This is not to say that a benefit could not be demonstrated through an RCT. But attention to fundamental principles of screening RCT design would need to be followed.

In 2002, Miettinen et al described these principles for proper performance of RCTs in regard to breast cancer screening.4 He explained that two conditions need to be met in order to produce meaningful results. The screening must continue long enough for the reduction in deaths to become fully manifest, and, since the deaths averted as a result of the early treatment associated with screening are well into the future, the mortality comparison must occur at an appropriate, suitably delayed, interval of time after the screening's initiation to document the full reduction in mortality.

At the time Miettenen's study was published, there was a serious, highly publicized, and confusing debate regarding the usefulness of breast cancer screening. The prestigious Cochrane Group had concluded that of seven completed breast cancer screening RCTs, only two had sufficiently eliminated enough biases to be useful. Both of these studies were interpreted as showing no benefit from mammography.5 This view was supported by the editor of The Lancet and by the screening and prevention board of the Physician Data Query, an expert advisory panel to the National Cancer Institute.6 The U.S. Preventive Services Task Force, while not agreeing fully with those views, nevertheless lowered its support regarding the strength of evidence in support of mammography from an A to a B. There were even concerns that official guidelines regarding breast cancer screening were going to be changed.

Miettinen et al applied those principals described above to the Malmo study, one of the two studies considered of good quality in that Cochrane review, and the only one that had actually continued to provide screening long enough for the principles to be applied. That study had initially been interpreted as showing no benefit because the interpretation included the deaths in the early years, when no reduction in mortality was expected. Miettinen et al focused on the deaths in the later years and demonstrated a significant-55%-reduction in mortality.4 The full reduction in death rate became manifest after approximately seven years of continuous screening. This extraordinary reversal in the interpretation of that study is the point at which the tide began to turn in regard to understanding the true benefit associated with breast cancer screening and helped to resolve the confusion.6

It is therefore quite understandable that the I-ELCAP study could demonstrate a high cure rate for lung cancer at the same time a study focusing on early deaths would not be able to demonstrate a benefit even though it exists.

These considerations bring to light a serious concern in regard to the NLST. It involves only three rounds of screening with relatively short-term follow-up (up to six years from its initiation) and therefore focuses only on early deaths, even though it is quite clear that any deaths that might be averted as a result of screening would be well into the future, just as with mammography. Therefore, it is predictable that the NLST will seriously underestimate the benefit associated with CT screening. It also has the disadvantage of not having an unscreened control arm, which will further dilute the apparent benefit. For the NLST to be able to demonstrate a significant benefit, as all are hoping for, the screening would have to be continued for several additional rounds.

The time has come for serious and open debate of these issues. Allowing design errors to predictably permit an underestimation of the benefit associated with CT screening for lung cancer is totally unacceptable. When it came to breast cancer screening, congressional hearings helped resolve these issues and prevent a state of confusion from continuing to be played out in the media.7 Surely the same can be said for lung cancer. Perhaps there is still time to salvage a meaningful result from the NLST.

Dr. Yankelevitz is a professor of radiology at Weill Cornell Medical College and an attending radiologist at New York-Presbyterian Hospital. He is a consultant for and owns stock in PneumRx.

References

• International Early Lung Cancer Action Program Investigators; Henschke CI, Yankelevitz DF, Libby DM, et al. Survival of patients with stage I lung cancer detected on CT screening. NEJM 2006;355(17):1763-1761.
• Bach PB, Jett JR, Pastorino U, et al. Computed tomography screening and lung cancer outcomes. JAMA 2007;297(9):953-961.
• Associated Press. CT scans may not prevent lung cancer deaths. MSNBC www.msnbc.msn.com/id/17484563/.
• Miettinen OS, Henschke CI, Pasmantier MW, et al. Mammographic screening: no reliable supporting evidence? Lancet 2002;359(9304):404-405.
• Olsen O, Gotzsche PC. Cochrane review on screening for breast cancer with mammography. Lancet 2001;358(9290):1340-1342.
• Jackson VP. Screening mammography: controversies and headlines. Radiology 2002;225(2):323-326.
• Making sense of the mammography controversy: what women need to know. Committee on Appropriations, U.S. Senate. One Hundred Seventh Congress, Second Session. frwebgate.access.gpo.gov/cgi-bin/getdoc.cgi?dbname=107_senate_hearings&docid=f:78085.pdf.