Cancer incidence in Barrett's esophagus: does it really matter, and who's counting anyway?
Article Outline
- How should we interpret summaries of the data?
- Why were the results heterogeneous?
- Do the incidence rates have any meaning?
- Disclosure
- References
- Copyright
Abbreviations: BE, Barrett's esophagus, VA, Veterans Affairs
All dysplasia may not be created equal.
The cancer incidence in BE (and in dysplastic BE) has assumed considerable importance. If the incidence is extremely low, neither physicians nor patients would care about the disorder. However, the ambiguity of the cancer risk, combined with the specter of a highly lethal malignancy, cause clinicians and patients substantial uncertainty and anxiety. This confusion is reflected in published cost-effectiveness analyses. These analyses, which labor to pull together the data regarding risks and benefits of endoscopic screening or surveillance, suggest that surveillance may be a reasonable and cost-effective choice if the cancer incidence is high (eg, one cancer for every 100-200 patients per year), but would be costly if the incidence is low (eg, one cancer for every 400 patients per year).1 Other authors suggest that after someone is screened just once, subsequent surveillance is costly (relative to the lives saved) if the cancer incidence is less than one cancer for every 54 patients per year.2
In this issue of Gastrointestinal Endoscopy, Rastogi et al3 have pulled together the existing literature to evaluate part of the puzzle: what is the cancer incidence in the group with the highest risk—those with high-grade dysplasia? For this study they performed a systematic review of numerous data sources, established criteria for study inclusion, and integrated the results from the different studies. The authors appropriately acknowledge several key limitations in the existing data. Foremost among these is the relatively small number of patients that met the entry criteria: 236 patients with high-grade dysplasia with a total of 1241 patient years of follow-up (an average of 5.26 years of follow-up per patient). The small number of patients is partially reflective of the study's inclusion criteria, which required at least one follow-up endoscopic examination within 6 months to exclude malignancies missed on the first examination.
How should we interpret summaries of the data?
The statistical synthesis of different study populations can be challenging. The authors' systematic review used several databases to identify all readily available studies. They then conducted a meta-analysis that pooled these results into a single “best” estimate with the hope that this would create a more accurate “average”; this estimate, however, assumes that the studies' results differ by no more than random chance. The evaluation of whether studies differ by more than chance uses statistical and graphical tests; if they differ, the results are considered “heterogeneous.” Effectively, the results cannot be meaningfully combined into a single average because the studies are like plums and apricots: the combination creates a “pluot,” but the “pluot” is not an accurate “average” of either fruit by itself. A prior systematic review of cancer risk in BE found heterogeneous results that the authors attributed to publication bias.4 The main result for the current review is also heterogeneous, making it difficult to interpret the summary estimates. The authors evaluated for the source of variation by excluding individual publications and seeing if the homogeneity improved, although the exclusion of the one study they believe was an “outlier” still provided fairly heterogeneous results (P = .1), given the low power of these statistical tests to detect heterogeneity.
Why were the results heterogeneous?
One reason the results were heterogeneous is that all dysplasia may not be created equal. The authors noted that the studies did not uniformly report for each subject whether dysplasia was shown in several samples (multifocal dysplasia) or only in a single sample. It is quite possible that subjects with multiple areas of dysplastic tissue have a higher likelihood of malignancy than subjects with only a single small area (which may even be completely removed with the biopsy).
A second reason may be the differences between the study populations, which gets to the question of who is doing the counting. The study by Reid et al5 occurred at a well-established referral center for patients with BE. In this study, the counting was done by experienced investigators performing detailed serial examinations in what is, presumably, a high-risk population.
The results from Overholt et al6 consisted of patients in the “no ablation” arm of a randomized high-grade dysplasia ablation trial. These patients were an extremely select group of referral subjects willing to undergo an experimental ablation therapy. It is quite possible they were also at higher-than-average risk of cancer, possibly explaining this group's relatively high cancer incidence rate (annual rate 10.27, 95% CI, 6.27-15.86). This rate is particularly remarkable because the average follow-up for this group, as listed by the authors, was only 1.5 years.
In contrast, the study by Weston7 specifically evaluated only patients with unifocal high-grade dysplasia (dysplasia confined to a single place in the esophagus). Although the original article does not specify how the patients were recruited, its 15 eligible persons were presumably identified during routine clinical practice or from clinical databases at a Veterans Administration hospital.
Finally, the study by Schnell et al8 at the Edward Hines, Jr. Veterans Affairs (VA) Hospital systematically identified all patients with high-grade dysplasia within a defined veteran community, made systematic efforts to exclude prevalent disease, and then followed the cases for a mean of 7.3 years. Only 16% of the study's 75 initially “cancer-free” patients ultimately developed esophageal adenocarcinoma during this extended follow-up, and only 1 patient died from cancer. This study population was the only one that clearly captured the full spectrum of patients with BE within a given population, although the cancer incidence in the VA population may still differ in important ways from other general populations.
Do the incidence rates have any meaning?
An average incidence rate suggests that the likelihood of cancer detection is constant over time. Under this assumption, a patient's chance of having a cancer detected in the second year after diagnosis is the same as his chance in the 10th year after diagnosis. As the authors note, this assumption is unproven. This assumption is likely false for several reasons, making a single average incidence rate difficult to interpret.
One reason the assumption is unlikely to hold is the performance characteristics of the test used to detect dysplasia and cancer—endoscopy. In most patients with BE, surveillance biopsies randomly sample only a small proportion of the available metaplastic tissue; the sensitivity is unknown and is likely substantially less than 100%. As a simple example, let us assume that the sensitivity for detecting high-grade dysplasia in any single endoscopy is 60%. Now let us perform annual surveillance on a new population of 1000 patients with BE, 100 of which have high-grade dysplasia. In the first year, 60 persons would be diagnosed with high-grade dysplasia (60% of the 100). Assuming that the high-grade dysplasia does not spontaneously resolve, the following year an additional 24 cases of high-grade dysplasia would be diagnosed among the original patients with dysplasia (60% of the remaining 40 patients with high-grade dysplasia). After an additional year, an additional 10 cases would be diagnosed among the original patients with dysplasia (60% of the remaining 16 patients). Thus, the average incidence over these 3 years would appear to be 94/1000 (9.4%), with an average annual incidence of 3.1% per year, even though all 100 subjects had prevalent high-grade dysplasia from the very beginning, and the apparent incidence of diagnosis in the first year was 6% (60/1000), in the second year was 2.6% (24/940), and in the third year was 1.1% (10/916). Although this example is highly simplified, particularly because it does not include the development of truly new cases and uses annual surveillance, it illustrates how an imperfect test sensitivity and the number of follow-up examinations performed can markedly influence average incidence rates of diagnosis. Similar principles apply to the diagnosis of early cancers in patients with high-grade dysplasia, given it may take many months or years for small cancers to be endoscopically visible. Thus, a new high-risk group of patients who are followed-up for a short period of time with intensive surveillance (eg, a referral endoscopic ablation study) might readily appear to have a higher incidence rate than a prevalent group of subjects followed-up for an extended period (eg, the Hines VA study).
Interestingly, the current study supports the concept that the apparent incidence rate of diagnosis decreases with extended follow-up. The cancer incidence rates in the studies listed show a progressive decline in reported incidence with increasing follow-up. The study with the shortest average follow-up (1.5 years) had the highest cancer incidence rate (incidence 10.27, 95% CI, 6.27-15.86), the study with the longest follow-up (7.3 years) had the lowest cancer incidence (2.27 95% CI, 1.17-3.96), and the other studies had intermediate follow-up periods and intermediate incidence rates.
The use of test performance characteristics is also crucial for deciding surveillance intervals. Decision analyses, given their inherent complexity, tend to use average incidence rates. However, the conclusions reached with this approach do not optimally inform clinical decision, as surveillance intervals for patients with BE may need to be shorter initially to exclude prevalent dysplasia or cancer, and longer later to minimize unnecessary testing. Although there are few real-world data to guide evidence-based decisions regarding surveillance intervals for BE, at least one recent practice guideline supports the concept of more intensive initial surveillance.7
The main question most clinicians want an answer to is this: If I have a patient with BE with a new diagnosis of high-grade dysplasia, what is that patient's risk of cancer? The answer is that it depends on who does the counting. The current data synthesis by Rastogi et al3 contributes to our understanding of cancer risk in high-grade dysplasia, but the diverse nature of the studies involved suggests that pooled incidence figures should be used cautiously. If your practice is at a referral center, and the referral patient you are seeing has undergone few sessions of intensive surveillance, then the average cancer incidence you find may well be higher than the pooled incidence figures, especially during the first couple of years of surveillance. Conversely, if you follow a diverse group of patients with high-grade dysplasia in a population of veterans for a long time, such as the patients from Edward Hines, Jr. Veterans Affairs Hospital, the average risk may be substantially lower than the pooled estimate. The risks of dysplasia and cancer in community-based settings are unknown (given the paucity of studies in these populations), but they may more closely approximate the figures of the fixed VA populations than the numbers seen from referral centers. Tip O'Neill, the former Speaker of the United States House of Representatives, famously noted that “all politics is local”; the cancer incidence in BE, too, likely depends on your local practice setting.
Disclosure
The author reports that there are no disclosures relevant to this publication.
References
- . Barrett's esophagus: a new look at surveillance based on emerging estimates of cancer risk. Am J Gastroenterol. 1999;94:2043–2053
- Screening and surveillance for Barrett esophagus in high-risk groups: a cost-utility analysis. Ann Intern Med. 2003;138:176–186
- Incidence of esophageal adenocarcinoma in patients with Barrett's esophagus and high-grade dysplasia: a meta-analysis. Gastrointest Endosc. 2008;67:394–398
- Is there publication bias in the reporting of cancer risk in Barrett's esophagus?. Gastroenterology. 2000;119:333–338
- Predictors of progression to cancer in Barrett's esophagus: baseline histology and flow cytometry identify low- and high-risk patient subsets. Am J Gastroenterol. 2000;95:1669–1676
- Photodynamic therapy with porfimer sodium for ablation of high-grade dysplasia in Barrett's esophagus: international, partially blinded, randomized phase III trial. Gastrointest Endosc. 2005;62:488–498
- Long-term follow-up of Barrett's high-grade dysplasia. Am J Gastroenterol. 2000;95:1888–1893
- Long-term nonsurgical management of Barrett's esophagus with high-grade dysplasia. Gastroenterology. 2001;120:1607–1619
See CME section; p. 509.
PII: S0016-5107(07)02728-9
doi:10.1016/j.gie.2007.09.029
© 2008 American Society for Gastrointestinal Endoscopy. Published by Elsevier Inc. All rights reserved.