From the *Department of Epidemiology, University of Washington, Seattle, WA; and †Division of Public Health Sciences, Fred Hutchinson Cancer Research Center, Seattle, WA.
Submitted 13 June 2007; accepted 25 September 2007.
Correspondence: Noel S. Weiss, Department of Epidemiology, University of Washington, Box 357236, Seattle, WA 98195. E-mail: firstname.lastname@example.org.
Some studies of the efficacy of cancer screening are obliged to obtain interviews from persons with cancer who are likely to die as a result of their disease, and from controls. We review the interview-based approaches that have been developed and the purpose that each one can achieve and offer suggestions for increasing the validity of these studies.
The ultimate goal of screening for a given form of cancer is the reduction of the risk of death from that cancer. Therefore, in most case-control studies that seek to evaluate cancer screening, cases are defined as persons who died of the cancer toward which screening is directed. Given this case definition, the ascertainment of screening history generally requires information from a source other than case interviews—most often review of medical records. However, outside the setting of a large prepaid health care plan with a centralized record system, relevant outpatient records may not be readily identifiable or may not be easily assembled. Even within a prepaid health care plan, the fact of testing may be reliably determined from records but the reason for testing may not. Testing may have been carried out as part of routine screening, or because of symptoms or signs. Also, some screening modalities, such as breast self-examination, are not performed by providers of health care. For these reasons, epidemiologists have designed some efficacy studies of cancer screening to obtain interviews with controls and cases (after their cancer diagnosis but before their death). We review the interview-based approaches that have been developed to indicate the purpose (s) that each approach can achieve, and to suggest means of increasing their validity.
The Efficacy of Screening in Reducing Cancer Incidence
Some screening tests have the ability to identify cancer precursors. For example, sigmoidoscopy or colonoscopy can identify colon polyps, some of which would become cancers. The Papanicolau test can identify cervical preinvasive lesions, some of which would progress to malignancy. If effective therapy is available once these premalignant lesions are found, then screening for their presence has the potential to decrease not just the mortality from that cancer, but also its incidence.
Case-control studies of screening to diminish the incidence of invasive cancer can readily employ interviews—few patients will have died of the cancer during the time after diagnosis that is needed to identify the patient and arrange for an interview. However, fairly detailed information on screening history must be sought at the interview. Crucial to the validity of the study is ascertainment of the timing of prior screening; relevant screens are only those that occurred after the first presence of a detectable preclinical lesion and before the presence of invasive disease. A means of estimating the beginning and ending dates of this critical interval has been described previously.1 However, even if these dates have been correctly gauged, a given case-control study will provide a distorted estimate of screening efficacy if the cases and controls are not accurate in reporting just when their screening activity took place.
The Efficacy of Screening in Reducing Cancer Mortality
If it is possible to identify and interview persons with advanced cancer who are highly likely to succumb to their illness, information on screening before diagnosis obtained from these cases and their controls can be compared to estimate directly the impact of screening on cancer mortality. In such a study, an effort would need to be made to identify persons with advanced cancer, no matter when during their clinical course the disease had become advanced (whether as of the time of the initial diagnosis or later).2 A history of screening in cases and matched controls would be sought for the period of time ending with the date of diagnosis (or the appearance of symptoms, whichever came first) and beginning with the presumed onset of the detectable, preclinical phase of the tumor. (If the test has the ability to identify a precursor lesion, the period would begin when the precursor was presumed to be first detectable.) However, if the quality of interview information on screening before diagnosis is anticipated to be of limited validity, or to be differentially complete between fatally ill cases and generally healthy controls, this approach will not produce an unbiased estimate of screening efficacy.
Is there anything to be learned by a comparison of screening histories of persons newly diagnosed with cancer at an advanced stage and controls? For example, Norman et al3 sought a history of screening mammography in 931 women diagnosed with stage II B or higher breast cancer and 4016 women without breast cancer. The screening histories focused on the 2-year period ending at the time of each case's diagnosis or the date on which each control was identified. However, given that all routine screening tests including mammography have the ability to identify at least some cases of cancer at an early stage, restriction to cases with a late stage at diagnosis would be expected to lead to a spuriously low proportion of screened cases, thus leading to an exaggerated estimate of efficacy. That is, even if no effective therapy were available for screen-detected breast cancer—in which case there would be no efficacy of screening against mortality from breast cancer—the proportion of women who had been screened during the 2-year period ending at the time of diagnosis would be smaller for women diagnosed with late-stage disease than for controls.
Nevertheless, while a higher level of screening among controls than among persons newly diagnosed with late-stage cancer does not necessarily imply that the screening modality can reduce mortality, such evidence would suggest that use of the test in question can lead to early detection—as long as the screening history of cases and controls is examined for the period of time that ends when the case's illness is judged to have progressed to advanced stage disease. This could be weeks to years before diagnosis, depending on the type of cancer. Screening has the potential to avert late-stage disease only if performed when the cancer is at an early stage. However, because the point in time at which the disease transitions from early- to late-stage can be guessed for a given case, such an assessment will necessarily be imprecise.
Researchers have conducted case-control studies of screening efficacy that sought to interview all newly diagnosed cases, but then analyzed only those whose cases went on to die as a result of their cancer during a specified follow-up period.4,5 For example, in an interview-based study of 706 men diagnosed with prostate cancer in the years 1993–1996, Agalliu et al5 ascertained a history of screening for prostate cancer (by means of prostate-specific antigen testing or a digital rectal examination) in the 5 years before diagnosis. Screening histories of the 44 cases who subsequently died of their disease during a follow-up period of 10.5 to 14.5 years were compared with those of 645 men of similar ages who had been selected at random to serve as controls and interviewed during the period of initial case identification in the 1990s. This approach offers the dual advantages of considering the outcome of greatest importance—mortality—and ascertaining screening history relatively close in time to the period of interest. However, as pointed out by Weiss and Lazovich,6 studies using this design have the potential to provide an unbiased result only if they are able to meet the following 2 criteria:
1. Achieve a high level of response among the cancer cases. Nonresponse typically is greatest in persons with advanced disease at the time of diagnosis, and these are the persons least likely to have been screened. The expected impact of nonresponse would therefore be a falsely high odds ratio associated with screening, and therefore a falsely low estimate of screening efficacy.
2. Achieve a duration of follow-up that is long enough to identify almost all of the deaths (from the cancer in question) that ultimately will occur among the newly identified cancer cases. If the follow-up period is too short, there will be a preferential omission of deaths that occur among persons whose cancer is identified relatively early in its natural history, ie, persons with a screen-detected cancer. The proportion of deceased cases included in the study who had been screened during the relevant period would be falsely low, leading in turn to a falsely low odds ratio and a falsely high estimate of screening efficacy.
Obtaining an unbiased result also requires that the fact and timing of testing, and the reason for testing (screening vs. other), can be ascertained accurately by means of interviews in both cases and controls. The level of accuracy of interview-ascertained screening histories no doubt varies by the type of test and by characteristics of the study population. Unfortunately, it is highly unusual to have a “gold standard” available against which to gauge the accuracy of these histories, given the earlier-noted limitations of medical records.
Interview-based case-control studies have the potential in some instances to provide useful information regarding the efficacy of a given form of screening for cancer. However, in order for the results of such a study to be valid and for the appropriate conclusions to be drawn from these results, the comparison of screening histories of cases and controls must accommodate the specific means by which the cases have been defined.
ABOUT THE AUTHOR
NOEL S. WEISS is a member of the faculties of the Department of Epidemiology, University of Washington School of Public Health and Community Medicine, and the Division of Public Health Sciences, Fred Hutchinson Cancer Research Center, in Seattle, Washington. CLARA BODELON is a mathematician enrolled in the graduate program in epidemiology at the University of Washington.
1. Weiss NS. Case-control studies of the efficacy of screening tests designed to prevent the incidence of cancer. Am J Epidemiol. 1999;149:1–4.
2. Morrison AS. Case definition in case-control studies of the efficacy of screening. Am J Epidemiol. 1982;115:6–8.
3. Norman SA, Localio AR, Zhou L, et al. Benefit of screening mammography in reducing the rate of late-stage breast cancer diagnosis (United States). Cancer Causes Control. 2006;17:921–929.
4. Norman SA, Localio AR, Weber AL, et al. Protection of mammography screening against death from breast cancer in women. Cancer Causes Control. 2007;18:909–918.
5. Agalliu I, Weiss NS, Lin DW, et al. Prostate-cancer mortality in relation to screening by prostate-specific antigen testing and digital rectal examination: a population-based study in middle-aged men. Cancer Causes Control. 2007;18:931–937.
6. Weiss NS, Lazovich D. Case-control studies of screening efficacy: The use of persons newly diagnosed with cancer who later sustain an unfavorable outcome. Am J Epidemiol. 1996;143:319–322.
© 2008 Lippincott Williams & Wilkins, Inc.