Conundrums in screening for cancer

Abstract

Screening for cancer has to be carefully organized for maximum effectiveness, and introduced in full understanding of the natural history of the disease. There are major potential harms as well as benefits from screening. The current state of art for breast, cervix and prostate cancer screening is reviewed, only for breast and cervix are policies of screening in the population justified.

Screening was defined by the US Commission on Chronic Illness as “the presumptive identification of unrecognized disease or defect by the application of tests, examinations, or other procedures that can be applied rapidly.”1 Although more recent definitions have been adopted,2 this one remains probably the most useful.

In screening for cancer, we hope to achieve early detection of disease, and thus change the incurable patient into one who is curable by effective therapy, resulting in a reduction in the death rate from the disease, and thus reduce health care costs. However, too often we seem to fail to screen the right people, encounter many false positive screening tests, affect the death rate from the cancer by a negligible amount, and increase health care costs.

Part of the difficulty is that for far too many cancers, we have not yet developed an effective screening test, acceptable to the population at risk, and even if we find a cancer “early,” treatment is not as effective as we hoped. However, another major difficulty is that screening is advocated under circumstances where we do not have evidence that it works, and very often, we do not understand the natural history of the disease to apply screening effectively, yet many years ago, such a requirement was recognized as a prerequisite for screening.3

In this article, I shall discuss some of the difficulties, and identify the conundrums that we should seek to resolve by more research, as well as discuss some of the issues related to screening for breast, cervix and prostate cancer.

The Process of Screening

Many of the difficulties arise because the process of screening is generally not completely understood. Screening is an activity conducted within the general population, being directed to those presumed to be at risk for the disease.

If it has been decided to initiate a screening program, the process begins by identifying the target population, e.g., women age 50–69 for breast screening. Often screening is opportunistic, i.e., it is left to (wo) men or their physicians to decide they should be screened. Many years ago, we pointed out the disadvantages of this approach, yet few countries have set up the organized program recommended.4 To identify the target population an accessible population register is required. Such a register is used for this purpose in Scandinavia while in the United Kingdom general practice registers are used.

The next step is to invite people for screening, often by letter, and if necessary followed by a reminder letter or a telephone call. Screening is a voluntary activity, thus people must understand they are potentially at risk for the disease and there are advantages to them of early diagnosis by screening if in fact they are destined to develop the disease. This will require a program of health education of the population, which should be directed not just to the target group but also to those members of their family group or friends who will influence their decisions. In some low- and middle-income countries, it may also be necessary to set up a professional education program for those in primary health care. Without such approaches adequate population coverage, and thus maximal effectiveness of the program, will not occur.

The next step is to administer the screening test, which depending on its complexity will occur in a family physician's office, or in a special facility designed for that purpose. The screening test will dichotomize the population screened into those positive to the test, and those negative. An important part of the process is ensuring that the person screened (and his or her primary care practitioner) is informed of the result as soon as possible.

The next step is to ensure diagnosis for those positive to the test. In an organized program, ensuring prompt referral to the diagnosis center is an integral component of the program—a step that if neglected, cuts right across the effectiveness of the program. The diagnostic process if efficient will result in those who have the disease being identified, and those who do not (those with a false positive screening test) being reassured. The false positives will have been disadvantaged, but with an efficient program, should be harmed as little as possible, though for some tests, a low risk of harm may be inevitable, e.g., from perforation during colonoscopy as a diagnostic test for colon cancer, while if the final approach to diagnosis requires resectional surgery, for a few harm may not be avoidable, even postoperative mortality.

The final step is to treat those who were positive to the diagnostic tests, i.e., those who truly have the disease. If the disease has been found early in its natural history, simpler treatment may be possible than if the disease had been allowed to progress and present clinically, one of the desired advantages of screening.

Unfortunately, not all the steps in this process are error free, and here we approach my first conundrum.

We need to consider who benefits from a screening program. Considering the classical 4-fold table of screening (Table 1), those in the lower right quadrant, the true negatives, benefit as they have been correctly reassured that at the time of the test they do not have detectable disease. This is why most people accept screening, to receive this reassurance. The people in the upper right quadrant, the false positives, have, as indicated earlier, been disadvantaged, a few even harmed, so they do not benefit from the program—we must remember that apart from the cost of the tests themselves (often the major costs of a program), the process of identifying the false positives result in costs to the program. The people in the upper left quadrant, the true positives, if they have been found to have treatable disease, may benefit—yet there are a number of categories of true positives that I shall consider further below. The people in the lower left quadrant, the false negatives, have not benefited at all, they have been falsely reassured that they do not have disease. Yet, in the process of screening such people are not identified—they are simply lumped with the true negatives and not subjected to diagnostic tests, so they do not know who they are, and nor do we. Yet we need to identify the false negatives in order to determine the sensitivity of a screening test.

Table 1. Who benefits from a screening program?

So, here we reach my first conundrum: Without special study, the sensitivity of a screening test is unknown.

There are a number of ways we can approach the determination of sensitivity by special study,5 unfortunately most determinations of “sensitivity,” are conducted cross-sectionally, which is how one determines the sensitivity of a diagnostic test. To determine the sensitivity of a test for progressive (and potentially fatal) disease is a far more complex process. This has often not been understood, for example an evaluation of digital versus film mammography,6 was conducted by administering both tests to the same women, whereas it should have been conducted by randomly assigning women to one or other test, to enable the interval (progressive) cancers to be counted after each screening test individually.

Types of True Positives

It is useful to consider the true positives in categories in relation to whether or not screening has been performed and the eventual prognosis of the disease:

• A
  Cancers detectable by screening that are curable after clinical diagnosis.
• B
  Cancers detectable by screening that are incurable after clinical diagnosis but curable after screen detection.
• C
  Cancers detectable by screening that are incurable after both clinical diagnosis and after screen detection.

It is category B that we seek to maximize through screening, none of those cases in categories A or C benefit from screening, though they cannot be specifically identified at the time.

However, there is another category revealed by screening, which never surfaces in its absence, which we can call category O—these are cancers not destined by their inherent natural history to present clinically before death from another cause, the overdiagnosed cases.

So we reach my second conundrum: Case detection is not equivalent to efficacy. This is because the detected cancer may not be curable, not have its natural history modified by available treatment (Group C), or the detected cancer may never have become life threatening in the patient's lifetime (Groups A and O).

Biases Associated with Screening

It has been recognized for some time that there are 4 biases associated with screening, that make it impossible to use survival as a measure of the efficacy of screening: lead time, length bias, selection bias and more recently overdiagnosis.7 Lead time (the interval between the time of detection by screening and the time the disease would have been diagnosed in the absence of screening), length bias (the propensity for screening to detect more slowly progressive cancers) and selection bias (volunteers are recruited into screening, they tend to be more health conscious and have better survival) are now generally recognized as inevitably improving the survival of screen-detected cases compared to those detected clinically, even if screening does not result in a reduction in mortality from the disease. In a randomized screening trial, lead time can be eliminated by dating survival from the date of randomization rather than the date of diagnosis. Length bias is also eliminated providing all cases, both those screen-detected and those that occur in the interval between screens, are counted, and selection bias is taken care of by the internal comparison possible from the randomization.8 But overdiagnosis, though it can be detected providing follow-up is long enough to eliminate the effect of lead time and both groups have no screening after screening has ended in the intervention group, can not be eliminated, thus survival does not remain a valid evaluation measure. In the absence of a randomized control group none of the biases can be eliminated. The definition of overdiagnosis “The detection of a cancer that was not destined to present clinically in that individual in his or her lifetime” does not mean that the cancer is not a “true” cancer nor that it is necessarily slow growing, it may be histologically indistinguishable from other cancers. This is a statistical, not a biological definition, and comes about because of the inevitable presence of competing causes of mortality. This is perhaps easier to understand for lung cancer screening, when those at risk are almost invariably heavy smokers, with substantial risks of dying from other causes such as cardiovascular disease, so that even under circumstances where a screening test may advance the time of diagnosis by very little, such as chest X-ray screening, overdiagnosis has been demonstrated.9 Thus, when even smaller and earlier cancers are detected through helical chest CT screening, overdiagnosis is inevitable,10, 11 thus falsifying any attempt to evaluate efficacy through a “single arm” (survival) study in spite of the attempts by some investigators to convince us otherwise.12 Even for breast cancer screening by mammography, it has been estimated that up to 15% or more of cases may be over diagnosed.13, 14

Breast Screening

Many have felt that issues related to breast screening were solved with the IARC (2002) evaluation, in particular with the conclusion “There is sufficient evidence for the efficacy of screening women age 50–69 years by mammography as the sole screening modality in reducing mortality from breast cancer.”15 This evaluation was based upon a meta-analytic estimate from mammography alone screening trials (i.e., those from Sweden) of a rate ratio of 0.75 (95% confidence interval 0.67–0.85), i.e., a 25% reduction in breast cancer mortality. However, for women age 40–49, the conclusion was “There is limited evidence for the efficacy of screening women age 40–49 years by mammography as the sole screening modality in reducing mortality from breast cancer.” This evaluation was based upon a meta-analytic estimate from mammography alone screening trials (again those from Sweden) of a rate ratio of 0.81 (95% confidence interval 0.65–1.01), i.e., a nonsignificant 19% reduction in breast cancer mortality, which fell to 12% (RR 0.88–95% confidence interval 0.74–1.04) when all valid trials (i.e., including the HIP and Canadian trials which also used breast physical examination screening) were included.

At the time of this evaluation it was known that the UK Trial of mammography among women age 39–41 was ongoing. This trial had an innovative design, women entering their 40s were randomized 1:2 to annual mammography screening for 7 years compared to no screening, followed without screening until they reached the age of 50 and then all were invited to participate in the UK screening program of 3 yearly mammography. A total of 160,921 women were randomized, the ratio of breast cancer deaths at a mean follow-up of 10.7 years relative to the control was 0.83 (95% confidence interval 0.66–1.04),16i.e., entirely compatible with the IARC meta-analysis. Thus, although most developed countries have introduced mammography screening for women age 50–69 (usually by mammography every 2 years) most do not include younger women, as the benefit is clearly less and the cost effectiveness much lower than for older women.

It is important to recognize that although in the estimates of efficacy made by IARC15 all valid trials were included for women age 40–49, this was not so for women age 50–69 as the Canadian National Breast Screening Study 2 (CNBSS 2) was not included, as it did not compare screening with no screening. The CNBSS 2 is the only trial designed according to the recommendation of the Working Group that evaluated the US Breast Cancer Detection Demonstration Projects that “a trial to evaluate the net benefit of mammography screening should be conducted.”17 By 1980, when the trial was initiated, the HIP trial had demonstrated the efficacy of screening by mammography and breast physical examinations in women age 50–69.18 Therefore an unscreened control group in CNBSS 2 was considered unethical, so we evaluated how much mammography added to annual breast physical examinations in women age 50–59. The answer was many additional false positives, more small cancers, but no reduction in breast cancer mortality.19 The trial thus cast doubt upon the common assumption that the benefit of mammography resides in detecting impalpable breast cancers (including ductal carcinomas in situ), and it was challenged.20, 21 However, the cancer detection rates were at least as good as achieved by modern mammography, and the trial was conducted when adjuvant chemotherapy and tamoxifen were standard for stage 2 breast cancer in Canada, but were not in the Swedish Two County trial.22, 23

This leads to my third conundrum: Mammography does not add benefit to breast physical examinations if treatment is adequate.

We have to recognize that it is possible that the improvement in treatment for breast cancer has now become so great that it is no longer possible to demonstrate a benefit from breast screening, because if it becomes possible to cure all breast cancers, whatever the stage at diagnosis, screening would have no role.

We have reported the results of a model-based evaluation of the likely efficacy of the breast physical examinations in the CNBSS, which suggested a 20% reduction in breast cancer mortality compared to no screening.24 Subsequently an evaluation of the cost-effectiveness of biennial breast examinations compared to biennial mammography based upon the same data set has indicated a more than 2-fold advantage to breast examinations.25 Increasingly, studies in developing countries are providing data that shows a stage shift consequent on breast examinations.26 This recently led the Eastern Mediterranean Region of WHO to propose breast examination screening policies; Morocco is taking the lead, other countries are following suit.

Figure 1 depicts trends in breast cancer mortality in several developed countries. Starting about 1990, there has been a remarkable reduction, though in a few (especially the United Kingdom) this was preceded by an unexplained rise. Burton (personal communication, 2008) has shown that in Australia, the maximum percentage reduction occurred among women age 40–49 (29%) for whom breast screening was not recommended, compared to a 23% reduction for women age 50–69, for whom screening was recommended. Thus the timing of recent falls in breast cancer mortality is compatible with improvements of therapy, while the timing in some countries is not compatible with an effect of mammography screening.

Figure 1. Trends in mortality from breast cancer. (Source: WHO Mortality Data Base).

Which leads to my fourth conundrum: The fall in breast cancer mortality in many countries seems attributable to improved treatment, not to screening.

Cervix Screening

Cytology screening for cervical cancer was introduced because of its ability to identify precursors of invasive cancer. Figure 2 illustrates the reductions in mortality from cervix cancer that have resulted from applying cytology. The contrast between Finland and Norway pointed out many years ago27 and the later adoption by the United Kingdom of organized programs with a major impact28 is well shown. Both Canada and the United States have done almost as well as Finland (reduction in mortality of 83% and 81% compared to 84%, respectively, from 1950 to 2005) but with a far greater utilization of resources, with annual to 3 yearly screening at ages 20–64 for Canada, largely annual screening at ages 18–69 for the United States, but 5 yearly screening at ages 30–59 for Finland.

Figure 2. Trends in mortality from cancer of the cervix. (Source: WHO Mortality Data Base).

What is often not appreciated is the substantial probability of regression of even carcinoma in situ and also of cytologically diagnosed dysplasia that is a dominant feature of the natural history of the disease29, 30 so that programs designed, as in the United States, to maximize detection of what is now called high grade disease (in fact largely CIN 2, with a strong probability of regression) is responsible for the overtreatment that occurs. An early IARC study, based on data from programs in Canada and Europe, showed the futility of annual screening starting at age 18 or 20.31 A more recent IARC working group confirmed this, recommending screening should not start before age 25, should be 3 yearly until about 40 and 5 yearly thereafter and stop at age 65, all within the context of an organized program.32

We are now entering an era when cytology screening will gradually switch to HPV testing, with different approaches to those prophylactically vaccinated and those not.33 An advantage of HPV testing is that those negative over the age of 35 can, unless their lifestyles change, safely delay re-screening for at least 5 or even 10 years.34 A disadvantage is its inapplicability in women under the age of 30, among whom self-limiting HPV infections, which produce temporary cytological changes, are so common. Further, like cytology, the test is unable to distinguish progressive disease.

So we reach my fifth conundrum: The majority of precursors detected by screening regress. It should be noted that this is probably a general phenomenon, when cancer precursors can be identified. It almost certainly applies to the adenomatous polyps identified by endoscopy screening for colorectal cancer. The implication for cervical screening is that we are over-treating many low-risk women, while allowing many high-risk women to slip through the screening net, because of failures of organization of screening. As we enter the HPV vaccine era we must increase the efficiency of the organization of cervical screening, to retain its effectiveness, both for those vaccinated but infected with types other than 16 or 18, and for those unvaccinated.

Prostate Screening

In many countries, there has been a dramatic increase in the incidence of prostate cancer, largely because of PSA screening, while mortality from the disease has usually risen a little, only to fall to previous levels. Figure 3 illustrates the trends in mortality in many developed countries, it is striking that the fall has been greatest in the United States.

Figure 3. Trends in mortality from cancer of the prostate. (Source: WHO Mortality Data Base).

Many believe that these falls are largely a consequence of PSA screening, though others point out that they could equally be due to improved therapy, as if life is prolonged by therapy, there is a greater probability that other competing causes of death (heart disease, other cancers) will cause death.35 The majority of men with prostate cancer die with, not from their disease.

There are two large randomized screening trials evaluating prostate screening, the Prostate, Lung, Colon and Ovary (PLCO) trial in the US and the European Randomized study of Screening for Prostate Cancer (ERSPC). Both have reported results on prostate cancer mortality, but with disparate findings.36, 37 The PLCO trial was conducted on a background of persistent, long-term advocacy of PSA screening for prostate cancer.38, 39 This resulted in a substantial proportion of those who entered the trial having been screened at least once in the previous 3 years, and about half of the control (usual care) group continuing to have screening PSA tests during the period when screening was promulgated in the intervention group in the trial.

In contrast, in the ERSPC trial, PSA screening was infrequent when the trial was initiated in the 7 countries for which results have been reported, though that situation probably changed during the course of the trial. One difficulty with the ERSPC report is that when the analysis includes all men randomized the RR is 0.85 (95% CI 0.73–1.00), it is only when the results are restricted to men age 55–69 that the RR falls to 0.80 (95% CI 0.65–0.98). There are some other problems, for example there is no table confirming the comparability of the randomized groups (apart from the data on all cause mortality, which is not broken down by cause). This is presumably because in the 3 countries who invited men for screening from population registries with no consent before randomization they could not collect the required data on prostate cancer risk factors from the control group (who were not contacted individually but followed by record linkage with cancer registries).

The PLCO trial results indicate clearly that following a form of organized screening, when nearly 90% of those in the intervention group received PSA tests annually for 6 years, there was no indication of a benefit in terms of reduction of mortality from prostate cancer through to 10 years.

However, there is a strong possibility that the reduction in prostate cancer mortality seen in the screened group in the ERSPC trial was due not to the screening per se, but to better therapy of the detected case. The policy in the PLCO trial not to mandate specific therapies after screen detection resulted in substantial similarity in treatment by stage between the two arms. In the ERSPC trial, however, trial arm was associated with treatment choice, especially in men with high-risk prostate cancer. A control subject with high-risk prostate cancer was more likely than a screen subject to receive radiotherapy (OR 1.43, 95% CI 1.01–2.05), expectant management (OR 2.92, 95% CI 1.33–6.42) or hormonal treatment (OR 1.77, 95% CI 1.07–2.94) instead of radical prostatectomy.40

The harms from prostate screening are considerable. In addition to the complications associated with false positive diagnoses, and the risk of postoperative mortality in elderly men subjected to prostatectomy, there is evidence of substantial overdiagnosis, estimated in ERSPC to be 27% from a single screening test at age 55 and 56% for a single screening test at age 75.41 Although the situation is improving with surgery performed in specialist centers, substantial numbers of men treated for prostate cancer will become incontinent (up to 50%) and impotent (on average 70%).42 Raffle and Gray43 have coined the term “The Popularity Paradox” for this situation: “The greater the harm from overdiagnosis and overtreatment from screening, the more people there are who believe they owe their health, or even their life, to the program.”

So, I reach my final conundrum: There is probably no early benefit from PSA screening, yet screening results in substantial overdiagnosis, and other harms.

Conclusions

The assumption that earlier diagnosis must be beneficial is not correct.

• Effective treatment is essential for effective screening, while effective screening is dependent on efficient organization.

• Improvements in treatment make it difficult to use observational data to confirm effectiveness, while as treatment improves the benefit from screening will fall.

• Establishing benefit from screening is not simple. The only unbiased method is a randomized screening trial, with death from cancer as the endpoint.

• Screening is an expensive use of health care resources. Screening cannot abolish mortality from cancer, and people who accept screening should not be deceived that it will.