Noel S. Weiss

Enhancing the Validity and Generalizability of Randomized Trials of Cancer Screening

De Gruyter | Published online: October 27, 2015

There are always some downsides to cancer screening. These include the economic costs of the screening maneuver, and of the follow-up tests of persons who screen as positive. In addition, usually there are physical costs associated with treatment of those persons diagnosed with cancer who were not in fact benefited by early detection. Therefore, it is important for studies that seek to assess the efficacy of cancer screening to document as accurately as possible not just the presence of a benefit to be weighed against these negatives, but the size of that benefit. While non-randomized studies have been successful in quantifying the large favorable impact of certain screening modalities – such as sigmoidoscopy (Selby et al., 1992) and cervical cytology (Kamineni et al., 2013; Rustagi et al., 2014) – randomized trials generally are needed when the expected benefit of screening is of relatively small or moderate magnitude. However, as Miettinen reminds us, there are features of many randomized trials of screening that can lead to a falsely low estimate of the mortality reduction associated with the introduction of a long-term screening program.

1 Noncompliance among persons assigned to be screened

Especially in studies in which participants are not restricted to those who have expressed a willingness to allow chance to dictate their screening status, the degree of non-adherence to a screening intervention can be considerable. An intent- to-screen analysis inevitably will provide a watered-down estimate of any favorable impact of the intervention in those who actually receive it. As a means of overcoming this bias, some recent trials have compared the observed mortality from breast cancer (Moss et al., 2006) or colorectal cancer (Atkin et al., 2010), for example, between screening compliers and the mortality expected among control participants who would have been screened had screening been offered to them. This expected mortality is estimated by: (a) documenting the proportion of noncompliers in the intervention arm of the trial, along with their cancer-specific mortality rate; and (b) subtracting from persons in the unscreened arm of the trial both the number of hypothetical noncompliers and the number of cancer deaths projected to have occurred in them.

2 Changes over time in the sensitivity of the screening modality or the efficacy of cancer treatment

Randomized trials of radiologic screening for lung cancer by means of chest X-ray did not observe a reduction in lung cancer mortality associated with screening (Marcus et al., 2000). In contrast, later trials of radiologic screening by means of helical CT did observe such a reduction (The National Lung Screening Trial Research Team, 2011). Almost certainly, the difference was due to the heightened sensitivity of helical CT screening in identifying small, potentially curable, bronchial lesions. Similarly, improvement in the efficacy of treatment of screen-detected lesions would have the potential to increase the beneficial impact of screening. (Correspondingly, improvement in the efficacy of treatment of clinically-detected lesions, more so than of screen-detected lesions, has the potential to diminish the efficacy of screening.) So, the results of a randomized trial done during one period of time may not be what could be expected later on.

3 An evaluation of the efficacy of the screening intervention that is too short in duration

Screening is designed to identify cancers at an early stage of their natural history, and so the deaths that a successful screening measure can prevent typically are those that would occur no sooner than several years after the tumor (or a precursor lesion) has been screen-detected. Therefore, in a comparison of mortality from a given form of cancer between persons assigned and not assigned to be screened, for a period of time (perhaps several years or more) there would not be a difference between the groups, even for a highly effective screening measure (Hanley, 2010). The absence of a mortality difference in the early years will dilute the estimate of the benefit that would be present in screenees in the longer term. For example, the estimated 15% reduction in prostate cancer mortality associated with PSA screening in the European Study of Screening for Prostate Cancer (Schroder et al., 2009) was based on an analysis in which the average duration of follow-up of participants was nine years. However, during the first seven years following the initiation of screening there was NO mortality reduction seen - many prostate cancers that ultimately prove fatal likely have a relatively long natural history. All of the mortality reduction that was observed in the trial was due to a considerably larger reduction beginning more than seven years from screening initiation. When the screening program and participant follow-up continued through 13 years, the mortality reduction (calculated across all years) grew to 21% (Schroder et al., 2014).

Miettinen bemoans the relative dearth of involvement of epidemiological researchers in randomized trials of screening for cancer, implying (to me) that he believes that the planning of the design and analysis of future trials would do well to consider issues such as 1–3 above. I would add to this the recommendation that when seeking to estimate the benefits of recommending or providing screening to members of a given population, persons INTERPRETING the results of randomized trials of cancer screening (past and future) also would do well to consider these issues.


Atkin, W. S., Edwards, R., Kralj-Hans, I. et al. (2010). Once-only flexible sigmoidoscopy screening in prevention of colorectal cancer: A multicenter randomized controlled trial. Lancet, 375:1624–1632. Search in Google Scholar

Hanley, J. A. (2010). Mortality reductions produced by sustained prostate cancer screening have been underestimated. Journal of Medical Screening, 17:147–151. Search in Google Scholar

Kamineni, A., Weinmann, S., Shy, K. K. et al. (2013). Efficacy of screening in preventing cervical cancer in older women. Cancer Causes & Control, 24:1653–1660. Search in Google Scholar

Marcus, P. M., Bergstrahl, E. J., Fagerstrom, R. M. et al. (2000). Lung cancer mortality in the mayo lung project: Impact of extended follow-up. Journal of the National Cancer Institute, 92:1308–1316. Search in Google Scholar

Moss, S. M., Cuckle, H., Evans, A. et al. (2006). Effect of mammographic screening from age 40 years on breast cancer mortality at 10 years’ follow-up: A randomized controlled trial. Lancet, 368:2053–2060. Search in Google Scholar

Rustagi, A. S., Kamineni, A., Weinmann, S. et al. (2014). Cervical screening and cervical cancer death among older women: A population-based case-control study. American Journal of Epidemiology, 179:1107–1114. Search in Google Scholar

Schroder, F. H., Hugosson, J., Roobol, M. J. et al. (2009). Screening and prostate cancer mortality in a randomized European study. The New England Journal of Medicine, 360:1320–1328. Search in Google Scholar

Schroder, F. H., Hugosson, J., Roobol, M. J. et al. (2014). Screening and prostate cancer mortality: Results of the European randomised study of screening for prostate cancer at 13 years of follow-up. Lancet, 384:2027–2035. Search in Google Scholar

Selby, J. V., Friedman, G. D., Quesenberry, C. P. and Weiss, N. S. (1992). A case-control study of screening sigmoidoscopy and mortality from colorectal cancer. The New England Journal of Medicine, 326:653–657. Search in Google Scholar

Published Online: 2015-10-27
Published in Print: 2015-12-1

©2015 by De Gruyter