Article Text

Download PDFPDF

The continuing uncertainty about cancer risk in inflammatory bowel disease
  1. Hans-Olov Adami1,
  2. Michael Bretthauer2,
  3. Louise Emilsson2,
  4. Miguel A Hernán3,
  5. Mette Kalager2,
  6. Jonas F Ludvigsson4,
  7. Anders Ekbom5
  1. 1Department of Epidemiology, Harvard T.H. Chan School of Public Health, Boston, Massachusetts, USA
  2. 2Institute of Health and Society, University of Oslo, Oslo, Norway
  3. 3Department of Biostatistics, Harvard T.H. Chan School of Public Health, Boston, Massachusetts, USA
  4. 4Department of Medical Epidemiology and Biostatistics, Karolinska Institutet, Stockholm, Sweden
  5. 5Department of Medicine Solna, Clinical Epidemiology Unit T2, Karolinska Institutet, Stockholm, Sweden
  1. Correspondence to Professor Hans-Olov Adami, Department of Epidemiology, Harvard T.H. Chan School of Public Health, 677 Huntington Ave., Boston MA 02115, USA; hadami{at}hsph.harvard.edu

Statistics from Altmetric.com

Introduction

An increased risk of colorectal cancer (CRC) among individuals with IBD was suggested in the 1920s and later supported by epidemiologic studies.1 The association between IBD and CRC is not thought to arise from shared risk factors,2 but rather from a causal effect of IBD on CRC through chronic inflammation and other mechanisms that promote malignant transformation of the colonic mucosa.3 As a result, patients with IBD are considered at high risk of CRC and therefore receive anti-inflammatory treatment, colonoscopic surveillance with biopsies and prophylactic colectomy.

Multiple reviews and meta-analyses have tried to estimate CRC risk in patients with IBD (see table 1 for a summary from six such analyses4–9) and several additional studies were published during the last year.10–13 Few of these studies are large, prospective, population-based studies, which reflect the substantial methodological challenges. Here, we first outline the criteria for valid studies on IBD and cancer and then use those criteria to critically review the existing studies.

Table 1

Summary of meta-analyses of the association between IBD, UC and Crohn's disease (CD) and risk of colorectal cancer (CRC)

Methodological challenges

Defining disease onset

Failure to accurately ascertain the onset of IBD may bias the effect estimates of IBD on cancer if increasing duration of disease increases the probability of somatic mutations that lead to malignant transformation. However, IBD may only be detected years after onset during colonoscopy screening14 or after early symptoms of CRC. As a result, the relative risk of CRC may be overestimated. Unfortunately, the identification of IBD onset in existing population-based registers is challenging. Its replacement by first hospitalisation due to IBD may misclassify disease onset because outpatient management may suffice until exacerbations, surgical treatment or complications require inpatient care.

A better choice for an operational definition of disease onset would be the date of first symptoms ascertained from medical records among those with subsequent histopathological confirmation of IBD and, if reliable information on first symptoms is unavailable, the date of first histopathological confirmation itself. To explore the effect of uncertainties in the definition of onset of disease, we propose a battery of sensitivity analyses in which the estimated onset of disease is defined as the date of histopathological confirmation minus say 1, 5 and 10 years. Restricting the definition of IBD cases to those with histopathological confirmation may result in the exclusion of patients with minimal or no initial symptoms, but it ensures no misclassification of IBD among those designated as cases.

Classifying extent of disease

The extent of disease may determine the excess risk of CRC, which is particularly high in patients with pancolitis. However, determining the extent of disease and its changes during follow-up is not straightforward. Barium enema X-ray, the standard diagnostic tool before colonoscopy was available, had high specificity but low sensitivity in diagnosing pancolitis. As a result, some patients with pancolitis would be misclassified as having a more limited disease. Therefore, when applied to contemporary patients who undergo colonoscopy and biopsy, these older risk predictions may exaggerate risk in patients with left-sided colitis due to a mixture of patients who indeed have pancolitis. A detailed description of the approach to define the extent of disease is needed in future studies for readers to critically assess the study quality.

Impact of management

Severity of IBD symptoms, extent of disease, intensity of inflammation, age at onset, concomitant primary sclerosing cholangitis (PSC) and heredity for CRC may influence both risk of cancer and patient management. To reduce confounding, detailed measurement of disease features and clinical interventions is required during the entire follow-up period, followed by appropriate statistical adjustment. Abstracting detailed and reliable data during many years or decades can be accommodated only in a limited number of settings—where the relevant information can be retrieved from archived medical records—and substantial resources are required.

Colonoscopy is the backbone of contemporary algorithms for CRC surveillance in IBD.3 Beside diagnosing invasive cancer at early stages, colonoscopy can detect premalignant lesions and provide a biopsy for histopathological examination. In the general population, cancer risk following adenoma removal has probably been overestimated.15 Whether this pertains also to patients with IBD is currently unknown. A meta-analysis estimated the absolute CRC risk after resection of polypoid dysplasia in patients with long-standing colitis16 but it is not clear how this evidence can guide surveillance strategies. Future studies need to monitor biopsies and histopathological classification of presumed cancer precursor lesions until we know more about their natural history and utility as risk predictors.

Incidence versus mortality as outcome

CRC incidence may be used to predict mortality under the assumption that the prognosis of IBD-associated CRC is similar to that of sporadic CRC—which is currently debated.17 ,18 However, CRC incidence is a less reliable outcome variable than CRC mortality.15

On the other hand, CRC mortality may introduce bias when rates from the general population are used to estimate the standardised mortality ratio (the ratio between observed and expected number of deaths from CRC): When a cohort of patients with incident IBD is empanelled, any individual with a prevalent CRC would be excluded. Hence, the prevalence of diagnosed CRC at start of follow-up of patients with IBD is zero and all subsequent deaths from this malignancy are due to cases detected after follow-up began. In contrast, when deriving the expected number of deaths from CRC among patients with IBD from rates in the general population, some of these deaths are caused by cancers diagnosed before the study cohort began. The expected number of deaths in the study cohort will then be exaggerated and the standardised mortality ratio will be biased downward.

To eliminate such bias, only deaths occurring among patients diagnosed with CRC after follow-up of the IBD cohort began should be considered in the calculation of mortality rates.19 This incidence-based mortality is unique for each study and therefore not available in population registers.

Lead-time bias

Closer surveillance of patients with IBD may entail earlier diagnosis (lead time) of cases and overdiagnosis of CRC and that would not otherwise have surfaced clinically. Lead time and overdiagnosis would make survival analyses, which compare prognosis between IBD-related and IBD-unrelated CRC difficult to interpret.20 Therefore, the extent of colonoscopic surveillance—organised or opportunistic—should be reliably reported in all studies.

Generalisability and selection bias

Because IBD is a heterogeneous disease both with regard to clinical manifestations and risk of CRC, a population-based study would be ideal. Patients enrolled from tertiary referral centres may be highly selected with regard to clinical characteristics, prior treatment and cancer risk. Such selectivity may not necessarily jeopardise the internal validity but hampers generalisability of risk and effect estimates. While careful measurement of clinical management and population characteristics may attenuate this problem, results from one study cannot be readily generalised to populations where patients are managed according to different guidelines.

Prophylactic colectomy complicates generalisability of risk estimates even further. An individual's follow-up must be censored at colectomy because he is no longer at risk of developing cancer. Such censoring removes from the study cohort those most likely to develop CRC because colectomy is recommended to patients considered to be at high risk. Therefore, we cannot apply estimates from one particular study to other settings without knowing the indications for and the rates of prophylactic colectomy during follow-up.

Obsolete versus contemporary risk estimates

Since clinical management of IBD changes over time, studies of cancer risk face a dilemma. On the one hand, we would like to know the benefits of adherence to contemporary guidelines. On the other hand, because cancer development takes years or decades and because IBD is typically diagnosed at young ages (before age 40 in about 60% of all cases1), studies to test the hypothesis that IBD causes CRC need a long follow-up. Furthermore, the incidence of CRC, which determines the statistical power to precisely quantify any excess risk, is low.

Therefore, a cohort of unrealistic size would be required to provide informative risk estimates reflecting surveillance guidelines that prevailed 10–20 years ago (plus the time needed to ascertain cancer outcome). Such a study design would still have limited statistical power to quantify risk in subgroups defined by sex, age at onset, extent of disease, and calendar period (as surrogate measure of management strategies).

One sensible compromise might be to enrol patients diagnosed with an incident IBD over an extended, yet relatively recent calendar period. Adequate statistical power to estimate cancer risk over the life course needs to be weighed against risk estimates influenced by outdated management strategies, for example, because the lack of thiopurines and tumour necrosis factor α (TNFα) antagonists, or of prophylactic colectomy guided by colonoscopic surveillance and chromoendoscopy. It is indeed sobering to realise that one of the largest population-based studies published to date reflects management of IBD in 1922–1983.1

What we know

Overall risk of CRC

The estimated relative risk of CRC in patients with IBD compared with the general population ranges from about 1 to 30.1 Table 1 summarises results from six systematic reviews and meta-analyses—based on partly overlapping studies—published since 2000.4–9 The substantial heterogeneity conveyed by the table might have different sources. One is differences between studies in the distribution of factors that markedly modify risk of CRC such as extent of disease, age at onset and follow-up time. Differences in drug treatment and surgical interventions among high-risk patients are probably also important.

Another source of heterogeneity is likely differences in procedures for patient enrolment, definition of disease onset, information about extent of disease and indications for and rates of prophylactic colectomy. Misclassification of age at onset, extent of disease and enrolment of prevalent cases of IBD who already have a CRC may undermine validity. Failure to censor follow-up at colectomy will overestimate person-years at risk and expected number of cases.

Extent of disease

Published studies consistently report a higher risk among patients with more extended colonic involvement (table 1). Estimates range from a 10-fold to 15-fold excess risk in pancolitis and a 2-fold or 3-fold excess risk in left-sided colitis to no significant association between isolated proctitis and CRC.1 For Crohn’s disease, an excess risk of CRC seems confined to patients with colonic involvement,21 which underlines the similarities between isolated colonic Crohn's disease and UC.22

Duration of follow-up

The early excess risk found in several studies1 ,23 might arise from inaccurate determination of IBD onset. The impact of such bias becomes even more plausible when a large excess risk within the first year vanishes during continued follow-up as recently found in a Danish study.23 Although available evidence is far from consistent, there is modest support for the biologically plausible hypothesis that the relative risk of CRC increases over follow-up time.3 The lack of reliable information on colectomies and the short duration of follow-up in recent ones further impedes any firm conclusions.

Sex and age at onset

Evidence that risk of CRC is modified by sex is mixed, with several studies showing no difference1 ,23 and another a 60% higher relative risk in men than in women.24 Several studies1 ,23 ,25 found a higher relative risk of CRC among patients diagnosed with IBD before age 15 or 20 versus later in life. In two studies with sufficient statistical power, this pattern persisted after adjustment for duration of follow-up and extent of disease, and the cumulative risk of CRC among those diagnosed with pancolitis before age 15 was 40% after 35 years.1 ,25 Similar to UC, young age at onset of Crohn's disease is associated with a higher risk.21 This is a growing clinical problem because of the increased incidence in paediatric colonic Crohn’s disease during the last two decades.26

Severity of inflammation

Three studies found an increased cancer risk associated with severity of inflammation.27–29 In a study that used a 5-level score for histological inflammation,29 each one unit increase in the score increased risk of CRC almost fivefold. In another study that used a 6-level score, each unit increase in the score increased risk of CRC about 2.5-fold.28

Temporal trends

There is no consensus that the excess incidence of CRC in patients with UC has declined over time. Data from Denmark23 and from one systematic review5 indicate such a trend, but evidence from Sweden30 and the US31 argue against it. Methodological issues such as changes over time in patient selection, diagnostic procedures and surgical interventions preclude any firm conclusions.

Assessing the overall mortality from CRC is also problematic. While some studies reported an excess mortality, others reported mortality to be lower than in the general population. However, to the best of our knowledge, no single study—including those published by ourselves1 ,30—has analysed incidence-based mortality. Hence, the validity of allegedly reduced mortality, notably among those with a recent diagnosis and/or short follow-up cannot be critically assessed. In the case of Crohn's disease, data are too scarce to allow any conclusions.

Primary sclerosing cholangitis

PSC is a dreaded complication in IBD. Formerly, PSC was believed to be confined to patients with UC but evidence shows an association also with colonic Crohn’s disease.32 The risk of CRC is about fivefold higher in patients with both PSC and IBD compared with those with IBD alone, especially for right-sided colon cancer.33 In Crohn's disease, there are no reports of an association, but data are scarce. IBD is often a silent disease in patients who also have PSC and often detected due to symptoms from the liver. A proper definition of disease onset is particularly hard among these patients.

Chemoprevention

Early reports indicated that 5-aminosalicylic acid (ASA) has a protective effect against CRC in patients with UC.34 ,35 This association was further substantiated in a meta-analysis in 2005.36 However, more recent meta-analyses failed to detect any protective effect.37 Thus, the effectiveness of maintenance therapy with 5-ASA to prevent CRC in patients with UC remains unclear.

Thiopurine drugs have also been recommended as prophylaxis against CRC in patients with IBD. But as for 5-ASA, the results are contradictory with some reports showing a substantial protective effect38 while others failed to detect any benefit.39 Finally, the introduction of new biological treatments in the end of the 1990s has revolutionised the treatment of IBD, but if and how that will affect the risk of CRC still remains an unanswered clinical question.

Familial occurrence

A family history of IBD does not affect the risk of CRC in patients with IBD,2 which to some extent contradicts the hypothesis that young age at onset should be a marker for a subgroup of patients with IBD, who are especially vulnerable for a malignant transformation in the colon. Similar to the general population, a family history of CRC will increase the risk of CRC among patients with IBD about twofold and even more pronounced if there is a first-degree relative with a diagnosis of CRC below the age of 50, about 10-fold.40

Discussion

IBD is a life-long and relatively common disease—estimated to affect 0.4% of the population in Europe and the US3—often with onset early in life. Clinical guidelines to reduce any excess risk of CRC in patients with IBD3 are not supported by randomised evidence. In this review, we have described contradictory results and knowledge gaps that make it difficult to develop evidence-based guidelines based on the existing data.

We hope that these knowledge gaps will stimulate investigators to undertake high-quality research that overcome the methodological challenges summarised in table 2. During the last several decades, there has been little progress in the quantification of cancer risk and the possible benefit of anti-inflammatory treatment with thiopurines and TNFα antagonists. Even the evidence for sulfasalazine, which has remained first-line therapy since the 1940s, is limited.34

Table 2

Summary of methodological challenges and weaknesses in studies of IBD and risk of colorectal cancer

Because disease characteristics and perceived cancer risk influence clinical management, no single study can provide useful information by addressing an isolated part of the complex web that is likely to determine cancer risk in an individual patient. We, therefore, propose a large-scale, high-quality study providing sufficient information in various subgroups of patients for the development of a risk prediction model that accommodates all relevant clinical and treatment data longitudinally.

The best approach to design such a study is not obvious and has indeed been a topic of intense discussion among the authors. Adequate statistical power is driven by the baseline incidence of CRC in the background population, the number of person-years at risk and the magnitude of excess risk among patients with IBD. A study of IBD and cancer, therefore, needs (1) to empanel a very large cohort of patients diagnosed with IBD during recent decades; (2) to calculate person-years at risk based on reliable definition of disease onset and censoring if a patient in the analytic cohort undergoes colectomy, dies or is lost at follow-up and (3) to completely ascertain incident cases and deaths due to CRC during follow-up.

Such a study also needs to collect detailed clinical and treatment data throughout the follow-up period. If doing so from the entire cohort comprising many thousands of patients is unfeasible, a nested case–control design would be ideal because information would be abstracted only from patients diagnosed with CRC and from a sample of similar size from the entire cohort. Available resources could then be devoted to a detailed abstraction of all available data—both at first diagnosis and during follow-up—on intensity and extent of disease, pharmacologic treatment by dose and duration, colonoscopic surveillance and biopsies with histopathological examination as well as concomitant morbidity.

We believe time is ripe for a large-scale, multinational initiative in settings that can accommodate the challenging prerequisites for a high-quality study of IBD and CRC. Such a study has the potential to greatly facilitate risk prediction and management of individual patients to reduce unnecessary surveillance and treatment and yet maximise the benefit of risk reduction.

References

View Abstract

Footnotes

  • Contributors H-OA wrote the first draft of this manuscript and AE has resumed main responsibility for the literature review. Subsequently, all authors have been deeply involved in discussions and several rounds of editing of the manuscript to an extent that, beyond any doubt, justifies authorship. All authors have approved the current version.

  • Funding Karolinska Institutet Distinguished Professor Award to H-OA (Dnr: 2368/10-221), NIH grant P01 CA134294.

  • Competing interests None declared.

  • Provenance and peer review Not commissioned; externally peer reviewed.

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.