Screening: when things go wrong

Patient case scenarios

How can screening for disease cause adverse outcomes?

Table 1 shows what happens in the common situation of a diagnostic or screening test. Most people who are screened are grateful; if disease is detected, they are thankful that it was found, enabling treatment. We cannot know who among the diagnosed group benefited. Most assume they did, as do those found to have a false-positive result after investigation. This has been called the popularity paradox of screening, as shown in Figure 1.^5,6 People who have a negative result feel even better.

• Download figure
• Open in new tab

Figure 1.

The paradox of screening

Taking a different view, Calonge⁷ describes that in any screening intervention there are 5 potential outcomes—4 of which are bad. Table 2 illustrates these 5 outcomes of applying a screening test and the order of magnitude of each possibility for prostate cancer screening.^8–10 Taking this contrary viewpoint, only 1 man in 1000 benefits from the prostate-specific antigen test; all other results either cause harm or represent wasted effort. Because many of the diagnoses occur shortly after screening, many men who do not benefit are diagnosed earlier than they otherwise would be, so are “sick” for longer than if they had waited until they presented clinically—the lead-time bias.¹¹

Understanding the detection of disease in relation to screening

To detect disease earlier in its course, a screening test is usually set with a threshold more sensitive than a diagnostic test, but it consequently often picks up changes that would never go on to cause trouble. This is demonstrated in Figure 2,⁶ illustrating the varied natural history of pathology that causes positive screening test results. “Fast” disease advances rapidly, so people might present clinically before screening tests can detect disease. Such disease is often rapidly fatal, so not helped by screening. Screening can detect “slow” disease early, so treatment can improve the outcome for at least a proportion of patients. “Very slow” disease does not need to be detected, as it will not be fatal before some other disease kills the person. “Nonprogressive” disease has a different natural history and does not progress or might even regress, so never needed detection, much less treatment. While it might be surprising that some cancers regress, there is increasing evidence for this phenomenon. These concepts can be applied not only to cancer, but also to metabolic diseases, such as renal failure or diabetes, and cardiovascular disease.

• Download figure
• Open in new tab

Figure 2.

Heterogeneity of cancer progression

Reproduced from Welch and Black with permission.⁶

The proportion of each type of disease and outcome varies with the disease and potentially the circumstances of the person—the development of early changes in younger people might vastly differ from those detected in older people. These differences in the stages of diseases detected and variation in their natural history mean that many “cases” detected by screening tests are overdiagnosed. Overdiagnosis is defined as the detection of an asymptomatic “abnormality” or “condition” that would ultimately not go on to cause symptoms or death.² The difficulty is that individual overdiagnosed cases (“very slow” or “nonprogressive”) cannot be distinguished from “slow” diseases. The number of overdiagnosed cases can only be estimated through long-term follow-up in randomized trials.^2,6

Outcomes of screening

Let’s consider each outcome in Table 2 in more detail.^8–10

Availability of evidence on harms

To assess the harms of screening requires evidence on false positives, overdiagnosis, and complications of diagnostic investigations and treatment^6,12,16; however, obtaining accurate estimates can be challenging. Investigators, evaluators, and program managers seldom systematically collect evidence on harms. A systematic review of 57 cancer screening trials found that the harms from overdiagnosis were reported in 4 and those from false positives in 2 of the trials evaluated.¹⁶ A review of US cancer screening guidelines found that 69% did not quantify benefits and harms or presented them in an asymmetric manner that favoured the benefits.¹⁷ Not surprisingly physicians and patients rarely have accurate expectations of the outcomes of treatments, screening, and tests, generally overestimating benefits and underestimating harms.^18,19

Application to the cases

Mr Brown had an interval cancer and he might have developed colorectal cancer after the colonoscopy. Alternatively, he might have had a false-negative result (ie, disease that was missed at his colonoscopy). There will always be a small error rate in endoscopy, which varies with operator skill and attention to detail.^20,21 In an ideal world, family physicians would like to know the error rates for the specialists or teams we refer to, but those data are only analyzed and returned in a few Canadian centres. Most colonoscopists do not know their own error and complication rates or those of surrogate markers (adenomatous polyp detection rate, rates of colonoscopy completed to the cecum, average time taken to do the examination).²² Even when collected, these data are usually only made available to the providers doing the procedures, without consideration of possible benefits from sharing the information with the family physicians or patients.

Patients have emotional responses of regret or even anger; physicians have such responses too. Mr Brown was very angry, asserting that something must have failed in the screening process. When he learned that a single colonoscopy screening reduces death from colon cancer by perhaps only 40% to 65%,²³ he felt there was misrepresentation, as the information given to him implied that screening would prevent deaths due to colon cancers, without providing clear information that it has only partial effect. His wrath encompassed his family physician, as well as the rest of the “system” that he believed had failed him.

Mr Smith was following the widespread messages about the value of prostate cancer screening, which seldom mention the potential for harm. He took the risk of being screened, perhaps swayed by emotion, and was diagnosed as not having cancer, but he sustained the serious harm of infection. This is a recognized complication of biopsy, occurring in at least 1% of men, and more recently 3.6%, with 1 in 1000 biopsied men dying as a result.²⁴ Transperineal biopsies have a lower complication rate than transrectal biopsies do, so while MRI-guided prostate biopsies might have a greater probability of finding cancer, their complication rate appears to be related to the access route.²⁴ By comparison, the optimistic estimate of living longer from prostate screening is around 1 per 1000 over 10 to 15 years,²⁵ while the risk of overdiagnosis is estimated as 33 to 50 out of 1000 men screened after the same period,⁸ with many of these receiving overtreatment. This was discussed with Mr Smith. He was aware that he chose to be screened anyway and he regretted that decision, but was grateful that his family physician had at least taken the time to discuss his choices.

How can the family physician mitigate the harms associated with screening?

Because of the inherent potential for harms after deciding to screen or not, family physicians must consider how best to limit or mitigate them. To reduce the adverse outcomes, several possible strategies can be employed. These are listed in Table 3.^33–38

Screening is generally performed in “low-risk populations,” but extending screening tests into populations with very low prevalence of disease has low possibility of benefit. Therefore, most positive results are false positives. Such populations include, for example, people below the recommended age for screening or those for whom the test is repeated at a shorter interval than recommended. In the extreme case, when there is no true disease in the population, all positive results are overdiagnosis or false positives. For the 5% of screening mammograms performed in Canada on women younger than age 50³¹ and for those getting annual breast or cervical screening or cervical screening before age 25, these are the most likely outcomes. As many of these people have concerns raised but investigation subsequently shows no disease, they are grateful to be given the “all clear”—again, the popularity paradox.

Quality of screening

In any judgment-based tests, such as pathology slide and x-ray scan readings or observing abnormalities in the mucosa of internal organs, there is variation in the ability to detect disease. Such variability is generally highest when trying to discriminate small changes in early disease. Therefore, reading screening results is difficult and quality needs to be regularly checked. It is hard for family physicians to assess the quality of specialists’ performance—still greater for screening tests. However, it is reasonable to ask what quality improvement processes they participate in and whether they perform to accepted standards.

While we all would like more sensitive tests to detect every cancer, generally tests are more sensitive at the cost of lesser specificity, thus causing many more false positives with associated harm. Thus, we must be cautious about services that offer more sensitive tests, without informing both referring physicians and patients about the concomitant potential harms caused by false positives and overdiagnosis. Examples include new technologies for breast screening and human papillomavirus testing for cervical screening, and lowering the thresholds for diagnosing renal failure or essential hypertension.