Principles, effectiveness and caveats in screening for cancer

Abstract

Background:

Cancer screening has the potential to prevent or reduce incidence and mortality of the target disease, but may also be harmful and have unwanted side-effects.

Methods:

This review explains the basic principles of cancer screening, common pitfalls in evaluation of effectiveness and harms of screening, and summarizes the evidence for effects and harms of the most commonly used cancer screening tools.

Results:

Cancer screening has either been established or is considered for breast, lung, prostate, cervical and colorectal cancer. In contrast, screening for gastrointestinal malignancies outside the large bowel is not generally accepted, available or implemented. Oesophageal and gastric carcinoma, and hepatocellular carcinoma, may be subject to screening in certain risk populations, but currently not for population screening based on available technology. Screening for colorectal cancer and cervical cancer by endoscopy and cytology respectively can decrease incidence of the target disease, whereas screening tools for lung, prostate and breast cancer detect early-stage invasive disease and thus do not decrease disease incidence. Overdiagnosis (detection of cancers that will not have become clinically apparent in the absence of screening) is a challenge in lung, prostate and breast cancer screening. The improvement of quality of clinical practice following the introduction of cancer screening programmes is an appreciated ‘side-effect’, but it is important to disentangle the effect of screening on cancer incidence and mortality from that of quality improvement of clinical services. As new, powerful screening tests emerge—particularly in molecular and genetic fields, but also in radiology and other clinical diagnostics–the basic requirements for screening evaluation and implementation must be borne in mind.

Conclusion:

Cancer screening has been established for several cancer forms in Europe. The potential for incidence and mortality reduction is good, but harms do exist that need to be addressed, and communicated to the public. Copyright © 2012 British Journal of Surgery Society Ltd. Published by John Wiley & Sons, Ltd.

Introduction

Screening in medicine is a concept to detect disease in individuals without clinical signs or symptoms of the disease to be screened for. Screening addresses presumptively healthy people. Therefore, participants in screening programmes or interventions should not be named patients, as the term patient is reserved for individuals with clinical signs or symptoms of disease. Screening is distinctly different from clinical service, as it targets people who are not knowingly sick.

Although intended to prevent disease or detect disease at a curable stage, screening for cancer is one of the most discussed topics in medicine. There is controversy about prostate-specific antigen (PSA) screening for prostate cancer and mammography screening for breast cancer, and debate also exists around screening for colorectal, lung and cervical cancer. Numerous studies have been published with often contradictory conclusions about screening. ‘The public is left totally confused’, concluded The Lancet in 20091, and probably so are many doctors and healthcare providers.

This review aims to disentangle the basic principles about cancer screening, common pitfalls in evaluation of effectiveness and harms of screening, and attempts to summarize the evidence on the effects and harms of the most commonly used cancer screening tools. Detailed information is provided for the five common cancers where screening is employed: breast, cervical, colorectal, lung and prostate. The review also gives a short overview of malignancies for which screening is not accepted at present (with a focus on gastrointestinal cancers), and the future role of screening for these conditions.

Definitions of screening

The definition of screening varies among different cultures and time periods. The Oxford English Dictionary defines ‘screen’ as ‘a system of checking a person or thing for the presence or absence of something, typically a disease’2. The verb to screen refers ‘to sift by passing through a screen’2. The verb ‘to sift’ derives from an old Dutch word (zeef), which means a ‘utensil consisting of a circular frame with a finely meshed or perforated bottom, used to separate the coarser from the finer particles of any loose material’2. Different definitions of screening in the context of health and clinical practice have been described3, 4.

Principles for screening

There are two overarching principles for screening: prevention and early detection. The idea behind the effect of screening is that early detection of disease (in an asymptomatic or precursor stage) improves prognosis. Consequently, the prognosis of the target disease must be related to stage at diagnosis (early stages must have better prognosis than late stages); treatment of early (screening detected) stages must be possible and available; and a detectable early disease stage must exist for a reasonable time duration to allow the screening test to pick it up.

In 1968, Wilson and Junger5 established principles of screening for the World Health Organization (WHO) that are still valid and should guide us in our approaches for screening today (Table 1). In addition to the three above-mentioned requirements on the part of the screening test and treatment, it is important to bear in mind that the disease to be screened for should be an important health problem (high incidence or mortality rate), and that the screening test of choice has been evaluated in its performance (sensitivity and specificity) and its ability to reduce disease incidence and/or mortality in high-quality scientific studies.

Table 1. The World Health Organization's principles of screening5

1 Screening should be directed towards an important health problem

2 There should be a simple, safe, precise and validated screening test

3 Treatment started at an early stage should be of more benefit than treatment initiated later

4 There should be evidence that the screening test is effective in reducing mortality and morbidity

5 The benefit of screening should outweigh the physical and psychological harm caused by the test, diagnostic procedures and treatment

6 The opportunity cost of the screening programme should be economically balanced in relation to expenditure on medical care as a whole

7 There should be a plan for managing and monitoring the screening programme and an agreed set of quality assurance standards

8 Potential screening participants should receive adequate information about benefits and disadvantages of participation

Effect versus harm

Unless Wilson and Junger's principles (Table 1) can be met, it may not be wise to consider screening and it is probably premature to invest in research for screening. If the principles are met, careful research is needed to measure the balance of benefits and harms, and to elucidate the best means of delivering a test, and a screening programme3. This is especially important because cancer screening is offered to presumptively healthy individuals. Therefore, the evidence for a favourable effectiveness–harm balance should be as good as possible6.

Opportunistic screening versus programme screening

Among healthcare providers, researchers and physicians, controversy exists with regard to the level of evidence required before screening for a disease should be initiated. In particular, this concerns the level of efficacy and effectiveness, as well as the knowledge of potential harmful effects that is required for any other medical intervention7. Even if a high level of evidence for efficacy and effectiveness exists, determination on how best to deliver the screening test must be met3. Guidelines and recommendations on how best to deliver screening in a population are diverse in their views of how best to implement screening8–11. In Europe and Australia, the approach is to implement public cancer screening programmes8, 10. In the USA, general recommendations are given to the population12 and screening practice is dependent on the medical insurance of individuals13. Screening organized by a healthcare provider or public service often includes personal invitation to screening for eligible individuals and a stringent organization for invitation, intervention and management of screen-detected disease, clinical surveillance and quality assurance. These programmes are often referred to as programme screening, organized screening or population-based screening. In contrast, screening strategies that include only the ability for a group of individuals or a population to be screened (for example by referral from a family doctor to a local screening facility) are termed opportunistic screening. It is believed that the former is more effective than the latter14.

Cancer screening

Cancer is the second largest cause of death in Europe (after cardiovascular disease), and accounts for about one-fifth of all deaths14. Most of the cancer burden arises from breast, lung, prostate, cervical and colorectal cancer. Both colorectal cancer and cervical cancer have well defined, benign precursor lesions (adenomas for colorectal cancer, cervical intraepithelial neoplasia for cervical cancer) that develop into malignant disease (only some benign lesions develop into cancer, but the majority of these cancers arise from the precursors) over a long time period that is accessible for screening. Lung, prostate and breast cancer do not have benign, reliably detectable, precursors, but there are screening tools to detect early-stage invasive disease, such as mammography for breast cancer and computed tomography for lung cancer. Screening for these five major cancer types has either been implemented or is subject to large-scale research in Europe and North America.

Implementing screening for common cancers in a population may be advantageous, but harbours possible drawbacks, such as producing false-positive and false-negative test results potentially leading to overtreatment or undertreatment of disease, as well as psychological distress as a side-effect of the test programme and the results. Screening can also be harmful owing to overdiagnosis. Overdiagnosis is the detection of cancers that would not have been identified clinically in someone's remaining lifetime15. As there are no markers that at the time of screening can distinguish non-lethal from lethal cancers, all individuals diagnosed with the disease screened for need to be treated. Thus, those with ‘overdiagnosed’ cancers will be treated without any benefit of the treatment, but with the risk of complications and adverse effects of the treatment, in addition to the psychological distress the patients experience due to the diagnosis.

The obvious reason for implementing screening is to reduce the incidence of or mortality from the disease targeted by the screening. Whenever screening is debated, there are always patients and doctors who tell their story about how screening saved their own or their patient's life. For many, the benefit of early detection through screening is obvious and intuitive. The thought that screening could be harmful is counterintuitive for many. We know that harm due to overdiagnosis and overtreatment exists, but, as we do not know how to differentiate between ‘overdiagnosed cancers’ and ‘dangerous cancers’, for each cancer survivor there is ‘individual evidence’ of benefit. Scientific evidence of benefits and harms is crucially important before implementing screening in a population. To persuade politicians, healthcare providers and the general public that screening has to be based on solid scientific evidence is further challenging because evaluation of a screening test in clinical trials might take at least 10–15 years. Only after a screening test is found to be more beneficial than harmful should the test be suggested for screening. Further, there should be continued evaluation of population-based screening programmes, and the results should be communicated to the public16.

Prevention versus early detection

There are two distinctively different approaches to cancer screening. One is to prevent the disease by finding and removing precursors of cancer; the other is to detect invasive cancer early in order to treat and cure the patient. The term screening is used for both preventive and early detection methods17.

Prevention

Preventive screening methods target the disease before it becomes malignant by detecting and removing precursor lesions. Examples are cervical cancer screening, which detects preinvasive cervical neoplasia, and colorectal cancer screening by endoscopy (flexible sigmoidoscopy or colonoscopy). The aim of these methods is primarily to prevent invasive disease, and thereby reduce the incidence of the disease. Reduced incidence will, as a consequence, result in reduced mortality (if fewer people get the disease, fewer will die from it). All currently used cancer screening prevention tools are also able to detect invasive tumours, in addition to detecting benign precursors of cancer. Table 2 outlines the most frequently used cancer screening tools and their conceptual ability to prevent or detect cancer early.

Table 2. Frequently used cancer screening tools and their conceptual ability to prevent cancer or detect it early

	Prevention	Early detection
MRI, magnetic resonance imaging; FOBT, faecal occult blood test; PSA, prostate-specific antigen.
Breast cancer
Mammography	No	Yes
Ultrasonography	No	Yes
MRI	No	Yes
Cervical cancer
Cervical smear testing	Yes	Yes
Colorectal cancer
FOBT	No	Yes
Colonoscopy	Yes	Yes
Flexible sigmoidoscopy	Yes (distal	Yes (distal
	colon only)	colon only)
Prostate cancer
PSA	No	Yes
Lung cancer
Chest X-ray	No	Yes
Computed tomography	No	Yes

Early detection

Screening tools such as mammography screening for breast cancer or PSA screening for prostate cancer are examples of screening methods that are intended to detect early-stage cancer. The goal in using these tools is to reduce site-specific cancer mortality as a result of early detection of the disease. Screening by methods of early detection is not able to reduce the incidence of the target disease. On the contrary, screening with these tools may result in an increase in incidence because of overdiagnosis.

Evaluation of screening

The evaluation of screening interventions demands thorough investigation of the underlying evidence, proper control groups, and knowledge of the principles of cancer screening epidemiology. Endpoints of interest in evaluation of cancer screening include disease-specific incidence and mortality, all-cause mortality, rates of overdiagnosis and adverse events, and cost-effectiveness.

Study design

Randomized controlled trials are the ‘gold standard’ for the evaluation of screening tools and methods, as they are for clinical medicine. However, randomized trials investigating incidence, mortality and long-term adverse effects of the intervention are particularly challenging to perform in screening, owing to the long time period from the screening intervention to the outcome of interest (death or cancer diagnosis), which often exceeds 10 years; the low event rate of cancer and cancer death in the target group (for example, the lifetime risk for colorectal cancer is about 5 per cent in Europe, so that 95 per cent of the population will never get the disease independent of any screening), which necessitates very large study sample sizes, often several tens of thousands; and the high costs of intervention and follow-up. Therefore, it may be intriguing to perform observational studies such as case–control studies, which are cheaper and quicker than randomized trials. However, case–control studies are problematic to use in screening, as they compare those who attend a screening with those who do not. This approach is flawed because of the fact that, often, screening compliers have a different background risk for the disease compared with non-compliers (often they are more healthy and less prone to get cancer)18. Therefore, the WHO with the International Agency for Research on Cancer concluded in their 2002 report on screening: ‘observational studies based on individual screening history, no matter how well designed and conducted, should not be regarded as providing evidence of an effect of screening’19.

Survival bias in screening studies

It is important to emphasize that survival (a commonly used and valid endpoint in clinical medicine) is not a valid endpoint in cancer screening. This is due to the so-called lead-time bias, which is evident in all cancer screening approaches (Fig. 1). Lead time is defined as the time interval from the diagnosis of a screening-detected cancer to the time point at which the cancer would have been detected clinically. If screening is not prolonging life (the patient dies at the same point in time no matter whether the cancer was detected clinically or by screening), owing to lead-time bias, survival estimates would still favour screening because the cancer has been detected earlier than without screening. However, in this situation screening does not provide any benefit (Fig. 1). Therefore, careful analyses taking into account lead-time bias are essential to prevent communication of presumed effects of screening when there really are none. Further, other biases such as selection bias, healthy screenee bias and overdiagnosis will affect survival estimates in cancer screening. Thus, survival is not a valid endpoint in cancer screening.

Figure 1. Lead-time bias. Lead time affects survival and incidence among patients who are diagnosed at screening, as time of diagnosis is advanced in time and the disease is diagnosed earlier than usual diagnosis without screening. The figure shows the calculation of survival time for a a patient diagnosed clinically and b an asymptomatic individual diagnosed by screening. The early diagnosis in b has not postponed the time of death. Although survival time has increased, there is no benefit for the screen-detected patient. This is called lead-time bias. (Adapted from Dickman and Adami20)

Mortality as an endpoint in screening

Mortality is not affected by lead-time bias, and can thus be used to evaluate screening. All current cancer screening tools aim to reduce cancer mortality, either through early detection of cancer or indirectly by reducing incidence. Owing to the relatively low contribution of each cancer to all-cause mortality in the population, no cancer screening intervention has been able to show a reduction in all-cause mortality. This means that none of the currently available cancer screening tools has been shown to have a ‘life-saving’ (life-prolonging) effect. But if screening works you can exclude the disease you want to die from by participating in screening (as screening reduces cancer-specific death, but not overall death). This may not be a bad deal, and is important to communicate to the public to prevent false assumptions about screening.

Although some have argued that all-cause mortality should be the overarching goal of cancer screening programmes21, this is unrealistic given the magnitude of effect of the currently used screening tools and the small proportion of each cancer to all-cause mortality. However, all-cause mortality should be monitored carefully in every screening programme, because it is important that screening does not increase all-cause mortality. If the programme were to increase all-cause mortality, this would be disqualifying and certainly should lead to its abandonment. Increased all-cause mortality is not entirely far-fetched, as there is evidence for unfavourable lifestyle changes related to screening participation, which could lead to increased mortality from cardiovascular disease22, 23.

Influence of screening on incidence and overdiagnosis

Conceptually, preventive cancer screening is more attractive than early-detection cancer screening, because it prevents people from getting a disease and thereby also reduces the mortality rate from the disease. For the screening tools designed to detect benign precursor lesions of cancer, cancer-specific incidence is a valid endpoint. However, these tools are available today for only two cancer forms: colorectal and cervical cancer (Table 2). For screening tools that are designed to detect disease early, incidence will often increase with screening as a result of overdiagnosis. This is a serious drawback of screening, and should be considered carefully. In prostate cancer screening, overdiagnosis has been estimated to be as high as 50 per cent, which, together with its modest effect on mortality, has led to recent recommendations against PSA screening24.

Screening and treatment

As mentioned above, screening is effective only when early-stage cancer has a better prognosis after treatment compared with prognosis after treatment of late-stage disease. As treatment of symptomatic cancer becomes more successful and more patients with advanced disease survive, the effect of early detection through screening is reduced. This may be the case for breast cancer screening: owing to the dramatic improvements in breast cancer treatment and awareness, screening mammography might not be as effective today as it was 20 years ago25, 26.

Another challenge has recently become apparent in prostate cancer screening: radical prostatectomy is the standard treatment for localized prostate cancer. For clinically detected tumours, there is good evidence that prostatectomy reduces prostate cancer-related mortality compared with watchful waiting27. However, in a recent randomized trial that compared radical prostatectomy with watchful waiting in screening-detected localized prostate cancer, prostatectomy did not reduce prostate cancer mortality28. Although this study had important limitations, it illustrates an important general caveat: screening is effective only if treatment of screen-detected lesions is superior to no treatment.

Effectiveness and harms of the most common screening tools

The most common screening tools and their approximated effects on cancer incidence and mortality are illustrated in Fig. 2. None of the tests reduces all-cause mortality.

Figure 2. Different screening tests and their relative effect (reduction or increase) on target cancer-specific mortality and incidence. The presented estimates are based on approximations performed by the authors and should be used accordingly. They are derived from randomized trials for lung cancer screening (chest radiography and computed tomography (CT)), colorectal cancer screening (faecal occult blood testing (FOBT) and flexible sigmoidoscopy) and prostate-specific antigen (PSA) screening for prostate cancer. For mammography screening, estimates are derived from both randomized trials and observational studies after the introduction of screening programmes, and for cervical cancer screening they are derived only from observational studies

Breast cancer

Mammography is the predominantly used screening tool for breast cancer for average-risk individuals. The idea behind mammography screening is to reduce breast cancer mortality through detection of breast cancer in an early curable stage.

Several randomized controlled trials of mammography screening were performed before widespread implementation of the method. In the first randomized mammography screening trial, the Health Insurance Plan (HIP) study in New York in 1963–1966, mammography screening was compared with no screening among women aged 40–64 years29. Other randomized studies followed30, 31, and a review by the WHO concluded that mammography screening for women in the age group 50–69 years reduced the breast cancer mortality rate by 25 per cent19. For younger women (aged 40–49 years), a 15 per cent non-significant mortality rate reduction was found32. Nevertheless, mammography screening is still debated. The size of the mortality reduction is considered uncertain, chiefly due to alleged methodological limitations in some of the randomized trials33. A Cochrane review of the effect of mammography screening estimated the breast cancer mortality reduction for women aged 50–69 years to be 15 per cent33, considerably less than reported by the earlier studies.

During the past two decades, mortality from breast cancer has decreased in many Western countries34, and it is intriguing to attribute this favourable trend to mammography screening. However, recent studies investigating the effect of mammography screening programmes have indicated that a considerable amount of the mortality reduction observed after the introduction of widespread mammography screening is not due to the screening itself25, 35, 36, but to improvement in clinical practice related to the screening activity, such as improved diagnostics, treatment and surveillance, and increased patient awareness.

Overdiagnosis of breast cancer by mammography screening has been estimated over a wide range, from 5 to 50 per cent15. For illustration, a 20 per cent rate of overdiagnosis was chosen in Fig. 2.

Cervical cancer

There are no large-scale randomized trials on the effectiveness of cervical cancer screening. The most frequently used test is cytology screening. According to the US Preventive Services Task Force, the introduction of population screening reduced cervical cancer incidence by at least 60 per cent, and cervical cancer death by 20–60 per cent37. There seems to be little overdiagnosis in cervical cancer screening. However, all screening will lead to some overdiagnosis. A quantification of the amount of overdiagnosis in cervical cancer screening is difficult (as it is for colorectal cancer screening tests such as flexible sigmoidoscopy and colonoscopy), because cytology screening is primarily a cancer prevention test. Thus, the primary effect is through prevention of the disease, and any possible overdiagnosis would be counteracted by the effect of incidence reduction.

Colorectal cancer

Several tools for colorectal cancer screening are available. They can be divided into early detection tests (such as faecal occult blood testing, FOBT) and cancer prevention tests (such as flexible sigmoidoscopy and colonoscopy). Colonoscopy is the predominant screening test in the USA and in some European countries38. It has high sensitivity for both adenomas and cancer, and is used as follow-up intervention for individuals screening positive with one of the other colorectal cancer screening modalities. However, colonoscopy has never been investigated in a randomized trial with regard to incidence and mortality of colorectal cancer. Therefore, the magnitude of efficacy for colonoscopy on colorectal cancer incidence and mortality is currently unknown. Colonoscopy is not recommended as a primary colorectal cancer screening tool by the European Union.

Meta-analyses of randomized trials show that FOBT screening reduces colorectal cancer mortality by 16 per cent39, but FOBT does not reduce the incidence of colorectal cancer. Flexible sigmoidoscopy screening reduces colorectal cancer incidence by about 20 per cent and colorectal cancer mortality by approximately 30 per cent40–43. FOBT screening is cheap and non-invasive, but results in large numbers of false-positive tests and needs to be repeated frequently. Flexible sigmoidoscopy is more invasive, but can be applied as a once-only screening38.

Lung cancer

Recently, two large randomized trials from the USA have been concluded on lung cancer screening; the Prostate, Lung, Colorectal, and Ovarian (PLCO) trial showed that annual screening with chest radiography does not reduce lung cancer mortality compared with no screening44. The National Lung Screening Trial (NLST) reported that annual computed tomography in smokers or former smokers reduced lung cancer mortality by 20 per cent compared with annual chest radiography45. Overdiagnosis does occur with both methods. Although the exact magnitude is difficult to quantify, it is probably in the range of 10–15 per cent44–47.

Prostate cancer

Two large-scale randomized trials exist of PSA screening for prostate cancer. The prostate cancer arm of the US PLCO study showed no effect of PSA screening on prostate cancer mortality48. A considerable amount of screening in the control group may have affected these results. A European multicentre trial demonstrated a 21 per cent reduction in prostate cancer mortality49. Taking into account the two trial results and their limitations, in Fig. 2 the effect has been set to 15 per cent, which is between the two study results. Overdiagnosis is a significant problem in PSA screening. Rates of overdiagnosis were estimated to be as high as 50 per cent in the European trial, and 17–30 per cent in the PLCO trial (averaged to 30 per cent in Fig. 2).

Role of screening in other cancers

In addition to the five most important cancers with regard to general population screening (Fig. 2), there has been interest in screening for other malignancies, particularly gastrointestinal cancers. However, due to lack of one or several of Wilson and Junger's basic requirements5 (Table 1), none of these cancers is considered a valid target for general population screening in most countries. They may have a low incidence compared with other diseases; adequate screening tools may not be available, are too expensive or too invasive; or the potential for cure may not be sufficiently dependent on stage at diagnosis. However, some cancers may be worthwhile to consider for screening in certain areas of the world, or in particular risk groups. The following paragraphs highlight some of these cancers.

Oesophageal cancer

Oesophageal cancer needs to be divided into two entities, adenocarcinomas and squamous cell cancers, and the potential role of screening is different for the two. Although the incidence of squamous cell cancer has been decreasing substantially in many Western countries, it constitutes a major health risk in Asia and some regions of Africa50. Adenocarcinoma of the oesophagus, in contrast, shows an increasing burden in the West. It is associated with Barrett's oesophagus, for which risk factors (obesity, reflux disease) are on the rise in many countries50.

Oesophageal adenocarcinoma is not a candidate for general population screening, because the incidence in most countries is low compared with that of the five cancers discussed above. However, screening for patients with Barrett's oesophagus is widely performed to detect high-grade dysplasia or early cancer, although there have been no randomized trials investigating whether this approach is effective. Recent evidence has shown that the absolute risk for cancer in patients with Barrett's oesophagus is lower than previously thought (the new estimates are approximately 0·13 per cent each year, compared with 0·5–3 per cent per year)51. Therefore, screening and surveillance for oesophageal cancer in Barrett's patients may be questionable.

Squamous cell cancer screening is discussed in high-incidence areas such as parts of China or Singapore, but there are considerable problems with screening for the disease even in these endemic areas, such as the cost and availability of screening tests (endoscopy, capsule endoscopy), and false-negative sampling (dysplasia and early cancer are difficult to detect with current methods)52. Further, randomized trials investigating the effectiveness of screening are lacking (although some are currently under way53).

Hepatocellular cancer

Hepatocellular carcinoma (HCC) is not a candidate for general population screening, but owing to the high risk in patients with chronic liver disease and cirrhosis (70–90 per cent of HCCs arise in patients with chronic liver disease) screening of these patients is indicated54. Both US and European guidelines recommend ultrasonography every 6 months in these patients to detect HCC early55, 56. α-Fetoprotein has a suboptimal performance in this setting and is therefore not recommended for HCC screening in chronic liver disease56.

Gastric cancer

Gastric cancer has diminished in importance in the West owing to a significant fall in incidence, whereas the disease is still a major cancer burden in other regions of the world, such as South America and Japan. The value of screening for gastric cancer is controversial57. Population screening programmes have been established in some high-incidence countries (Japan has a mass screening programme), but high-quality studies on its effectiveness are lacking. The important role of Helicobacter pylori eradication in prevention of gastric cancer (in competition to screening by endoscopy) has recently been demonstrated in a randomized trial from China, where the odds of developing gastric cancer were reduced by 39 per cent in patients who received treatment for H. pylori compared with the placebo group after 14 years of follow-up (absolute risk 3·0 per cent versus 4·6 per cent; odds ratio 0·61, 95 per cent confidence interval 0·38 to 0·96; P = 0·03)58.

Cancer screening programmes improve clinical practice

Cancer screening programmes for the five ‘big’ cancers focused on in this review have been established in many countries around the world. After the start of mammography screening in many countries in the late 1980s and the 1990s, colorectal cancer screening programmes were initiated in most European countries after the year 2000. PSA screening for prostate cancer is also prevalent in most countries, although often not organized as a public programme. It is important to recognize that the establishment of screening programmes is changing the field of medicine in which they are operating (most often in a positive direction), because such change can greatly affect the effectiveness of the screening.

Breast cancer mortality in Norway has been reduced by 30 per cent since the introduction of the Norwegian breast cancer screening programme in 199625. However, most of this mortality reduction is not attributable to the screening (taking radiographs of women's breasts). At least two-thirds of the observed 30 per cent mortality reduction is because of improved patient care by rigorous reorganization and quality improvement of breast cancer care in Norway when the screening programme was introduced16.

In the UK, a large survey in 1999 revealed poor performance of colonoscopy services59. Average caecum intubation rates were only 57 per cent, although most guidelines define 90 per cent or above as adequate performance. Driven largely by the planned implementation of the national bowel cancer screening programme (BCSP) (a pilot had been concluded by 2002), a comprehensive endoscopy training programme was established and quality improvement programmes were implemented throughout the country60. Today, all endoscopy services participating in the BCSP must adhere to a stringent quality assurance system, organized and overseen by the Joint Advisory Group on Gastrointestinal Endoscopy (JAG). It is too early to evaluate the BCSP with regard to the effect of screening on colorectal cancer incidence and mortality, but it will be interesting to determine how much of the expected benefit is due to screening and how much is related to the improved quality of the UK endoscopy services.

The improvement of quality of clinical practice following the introduction of cancer screening programmes is a much appreciated ‘side-effect’. However, it is essential to be able to disentangle the effect of screening from that of quality improvement of the clinical services on cancer incidence and mortality, because the former comes with potential harm and the latter does not.

Future perspectives

Screening for cancer is a double-edged sword: the potential to do good is great because of the enormous disease burden and the increasing ability for early detection and prevention using advanced diagnostic tools. However, adverse effects are not negotiable, as none of the screening tests is entirely free from complications, the risk of overdiagnosis is substantial, and the costs of screening must be reasonable with regard to the benefits. As new powerful screening tests emerge (especially in the molecular and genetic area, but also in radiology and other clinical diagnostics), it is important to have in mind the basic requirements for screening. Even though it may appear attractive to jump on new exciting tools, high-quality randomized trials should be our guiding tool critically to evaluate effectiveness, cost-effectiveness and harms of new screening approaches, to prevent the population from potential harm and the society from unnecessary expenses.

Disclosure

The authors declare no conflict of interest.