Objective Although it is well understood that the risk of oesophageal adenocarcinoma increases with Barrett length, transition risks for cancer associated with different Barrett lengths are unknown. We aimed to estimate annual cancer transition rates for patients with long-segment (≥3 cm), short-segment (≥1 to <3 cm) and ultra-short-segment (<1 cm) Barrett's oesophagus.
Design We used three data sources to estimate the annual cancer transition rates for each Barrett length category: (1) the distribution of long, short and ultra-short Barrett's oesophagus among a large German cohort with newly diagnosed T1 oesophageal adenocarcinoma; (2) population-based German incidence of oesophageal adenocarcinoma; and (3) published estimates of the population prevalence of Barrett's oesophagus for each Barrett length category.
Results Among 1017 patients with newly diagnosed T1 oesophageal adenocarcinoma, 573 (56%) had long-segment, 240 (24%) short-segment and 204 (20%) ultra-short-segment Barrett's oesophagus. The base-case estimates for the prevalence of Barrett's oesophagus among the general population were 1.5%, 5% and 14%, respectively. The annual cancer transition rates for patients with long, short and ultra-short Barrett's oesophagus were 0.22%, 0.03% and 0.01%, respectively. To detect one cancer, 450 patients with long-segment Barrett's oesophagus would need to undergo annual surveillance endoscopy; in short segment and ultra-short segment, the corresponding numbers of patients would be 3440 and 12 364. Similar results were obtained when applying US incidence data.
Conclusions The large number of patients, who need to undergo endoscopic surveillance to detect one cancer, raises questions about the value of surveillance endoscopy in patients with short segment or ultra-short segment of Barrett's oesophagus.
- BARRETT'S OESOPHAGUS
- OESOPHAGEAL CANCER
Statistics from Altmetric.com
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.
Significance of this study
What is already known on this subject?
Risk of oesophageal adenocarcinoma increases with length of Barrett's oesophagus.
Shorter Barrett's oesophagus is more prevalent in the population than longer Barrett's oesophagus.
The transition rates for patients with different Barrett lengths are unclear.
What are the new findings?
Among patients with newly diagnosed T1 oesophageal adenocarcinoma, more than half had a long-segment Barrett's oesophagus (56%), 24% had a short- segment and 20% an ultra-short-segment Barrett's oesophagus.
When applied to the German population, the study estimated an annual cancer transition rates for patients with long, short and ultra-short Barrett's oesophagus as 0.22%, 0.03% and 0.01%, respectively.
The number of patients who would need to undergo an upper endoscopy to find one cancer per year is 450 patients with a long-segment, 3440 with a short-segment and 12 365 with an ultra-short-segment Barrett's oesophagus.
How might it impact on clinical practice in the foreseeable future?
Surveillance for patients with ultra-short-segment or short-segment Barrett's oesophagus may not be worthwhile even when considering a 5-year interval.
The alarming rise in the incidence of oesophageal adenocarcinoma in the past three decades, coupled with its dismal prognosis, has led to widespread screening and surveillance of its premalignant precursor lesion—the Barrett's oesophagus.1–3 However, there is no strong evidence that supports the benefit of screening for or surveillance of Barrett's oesophagus.3 Advances in endoscopic imaging have improved our ability to visualise mucosal abnormalities, and endoscopic resection offers an effective therapy for patients with mucosal cancer.3–5 However, only a few Barrett patients are destined to develop cancer, and it remains a challenge to identify patients at risk of progression.
Observational cohort and case–control studies have shown that the risk of transition to cancer increases with Barrett length.6–11 One study has suggested a 17% increased risk with every 1 cm increase in Barrett length,7 and the majority of patients who transition to cancer during an observation period have long-segment Barrett's oesophagus.10 ,12 Furthermore, short-segment Barrett's oesophagus is more prevalent in the population than long-segment Barrett's oesophagus and therefore involves a larger population at risk. These observations point at a varying risk of transition to cancer based on the extent of Barrett metaplasia. Recently, revised British and German surveillance guidelines have considered Barrett length. Patients with a long-segment Barrett's oesophagus may be followed every three years, and those with a short segment every three to five years. No surveillance is recommended for those with a segment of <1 cm.13 ,14 Barrett length has not been considered in current US Barrett surveillance guidelines, and all non-dysplastic Barrett patients are recommended to undergo a surveillance upper endoscopy every three to five years.2 ,3
We were interested in quantifying the transition risk from non-dysplastic Barrett's oesophagus to cancer dependent on the Barrett length to provide supporting or refuting evidence to the Barrett guidelines. Assuming that Barrett length does not change once it has developed and that all cancers develop from Barrett's oesophagus, the transition risk can be modelled by considering two factors: (a) the incidence of oesophageal adenocarcinoma by Barrett length, and (b) the prevalence of Barrett's oesophagus by Barrett length. Our goal was to estimate annual cancer transition rates for patients with long-segment (≥3 cm), short-segment (≥1 to <3 cm) and ultra-short-segment (<1 cm) Barrett's oesophagus.
The study used three different data sources: (1) a large German cohort of Barrett patients with T1 oesophageal adenocarcinoma; (2) published data on the prevalence of Barrett's oesophagus in the general population stratified by Barrett length; and (3) the German cancer registry on the incidence of oesophageal adenocarcinoma.
Patients with T1 oesophageal adenocarcinoma
We identified all patients who were evaluated for a new diagnosis of Barrett neoplasia at a referral centre for endoscopic treatment of Barrett neoplasia (HSK Wiesbaden, Germany) from January 2003 until December 2011. Patients were included if they had a T1 oesophageal adenocarcinoma of the oesophagus within a Barrett segment. We did not include patients with more advanced cancer because tumour may obscure possible underlying shorter segment Barrett's oesophagus and erroneous estimation of Barrett length. All patients underwent an upper endoscopy with documentation of the Barrett length and a staging endoscopic ultrasound, and either endoscopic mucosal resection or surgical resection if endoscopic resection was not feasible. Final T-staging was based on the pathology specimen. Two expert pathologists (MV and MS) assessed all biopsies and resection specimens for the degree of neoplasia and the presence of intestinal metaplasia. Patients were only included if they had biopsy-proven Barrett's intestinal metaplasia. The main outcome of interest from this cohort was the proportion of patients with oesophageal adenocarcinoma, who had ultra-short Barrett's oesophagus (<1 cm), short-segment Barrett's oesophagus (1 to <3 cm) and long-segment Barrett's oesophagus (≥3 cm).
Oesophageal adenocarcinoma incidence
In the primary analysis, incidence data for 2010 were obtained from population-based cancer registries in Germany for oesophageal adenocarcinoma in men and women aged 50 years or older and adjusted to the European Standard Population.15–17 Oesophageal adenocarcinoma was defined anatomically as being located in the oesophagus (International Classification of Diseases for Oncology, third edition (ICD-O-3) topography code C15) and histologically as adenocarcinoma (ICD-O-3 morphology codes 814-838).
In a secondary analysis, we also applied incidence data obtained from the 2010 US cancer registry Surveillance Epidemiology and End Results (SEER) 9 database of the National Cancer Institute. Similarly to the German incidence data, we obtained the oesophageal adenocarcinoma incidence in men and women aged 50 or older. The SEER data were age-adjusted and sex-adjusted to the European Standard Population using the method of direct standardisation.
Prevalence of Barrett's oesophagus in the population by Barrett length
Estimates on the population prevalence of Barrett's oesophagus by Barrett length were extracted from the literature. Studies examined the prevalence of Barrett's oesophagus in three different populations: the general population,18 ,19 patients referred for an upper endoscopy20 ,21 and patients undergoing a colorectal cancer screening colonoscopy.22–24 We summarised the number of all patients in each Barrett length category (numerator) and the number of all patients (denominator) from all studies to calculate the prevalence for the base-case analysis.
Ascertainment of data on the prevalence of Barrett's oesophagus by Barrett length is limited by an unclear definition of the minimum Barrett length. While a cut-off of >3 cm of salmon-coloured epithelium was used in the 1990s,25 most studies required the presence of specialised intestinal metaplasia in a visible tongue of ‘salmon-coloured mucosa’ for the diagnosis of Barrett's oesophagus;18–20 ,22 ,24 however, the minimum length of such tongue is unclear. A visible mucosal extension with intestinal metaplasia above the gastroesophageal junction of <1 cm can be found in 16% of patients older than 40 years of age.23 ,26 A similar proportion was reported in studies that examined the prevalence of intestinal metaplasia at the gastroesophageal junction without providing details on the appearance of the Z-line.21 ,22 ,27 Because the distinction between an irregular Z-line and a minimum Barrett length is subjective, we assumed that oesophageal adenocarcinomas may arise in the presence of intestinal metaplasia in any mucosal extension of <1 cm above the gastroesophageal junction.
We contacted individual authors of studies on Barrett prevalence to obtain detailed information on Barrett prevalence by age and gender. Unfortunately, this information was limited and we could therefore not apply age-specific and gender-specific rates in the analysis.
To examine the distribution of patients with T1 oesophageal adenocarcinoma by Barrett length, we calculated the proportions of patients with ultra-short-segment, short-segment and long-segment Barrett's oesophagus among all patients with T1 oesophageal adenocarcinoma. To obtain the incidence of oesophageal adenocarcinoma by Barrett length, we multiplied the proportion of Barrett length categories among patients with cancer with the overall incidence of oesophageal adenocarcinoma in a population of 100 000 individuals age 50 or older.
Annual transition rates among patients with ultra-short-segment, short-segment and long-segment Barrett's oesophagus were calculated as the ratio between the incidence of oesophageal adenocarcinoma and the prevalence of Barrett's oesophagus in each Barrett length category. This approach assumes that Barrett length remains constant once it has developed, and that Barrett's oesophagus is the premalignant condition for all oesophageal adenocarcinomas. These assumptions are generally accepted as characteristics of Barrett's oesophagus and the pathophysiology of oesophageal adenocarcinoma.2 ,28–30 Discussions on mosaic distribution of different types of epithelia or segments of columnar metaplasia above the GI junction with no goblet cells were left aside since data are still controversial.
The number of patients needed to be examined to identify one patient with cancer per year was obtained by calculating the reciprocal of the transition rate. Considering a similar transition risk over time, we calculated the number of patients who would need to be examined to find one cancer dependent on the time interval between surveillance examinations.
We varied Barrett prevalence estimates over a range of plausible assumptions in one-way sensitivity analysis to assess the robustness of our analysis. We selected estimates to best reflect ranges published by the studies on Barrett prevalence.
Distribution of Barrett length among patients with T1 oesophageal adenocarcinoma
Table 1 shows the distribution of Barrett length among a cohort of 1017 patients with early oesophageal adenocarcinoma. The mean age of patients was 69 years, and 85% were men. The mean Barrett length was 5.7 cm. In total, 573 patients (56%) had long-segment Barrett's oesophagus, 240 patients (24%) had short-segment and 204 patients (20%) had ultra-short-segment Barrett's oesophagus.
Reported population prevalence of Barrett's oesophagus are presented in table 2 stratified by Barrett length and shown with their corresponding ranges. Among persons 50 years of age or older, we considered a prevalence of 1.5% for long-segment Barrett's oesophagus, 5% for short-segment and 14% an ultra-short-segment Barrett's oesophagus.
Annual cancer incidence by Barrett length categories in the general population
Based on the German cancer registry, the incidence of oesophageal adenocarcinoma for patients 50 years of age or older was 5.8 per 100 000 in 2010. Applying the distribution of Barrett length among patients with cancer to the cancer incidence (table 3), we obtained annual cancer incidence rates by Barrett length. Accordingly, of 5.8 per 100 000 patients with newly diagnosed oesophageal adenocarcinoma, 3.3 arose in patients with long-segment Barrett's oesophagus, 1.4 in patients with short-segment and 1.2 in patients with ultra-short-segment Barrett's oesophagus.
Annual cancer transition rates by Barrett length
Table 3 details the results of cancer transition rates among Barrett patients by Barrett length category. In the base-case analysis, the annual cancer risk was 0.22% for patients with long-segment Barrett's oesophagus, 0.03% for patients with short-segment Barrett's oesophagus and 0.01% for those with ultra-short-segment Barrett's oesophagus.
The corresponding numbers of patients that would need to undergo an endoscopy to detect one cancer per year are 450 for long-segment Barrett's oesophagus, 3440 for short-segment and 12 364 for ultra-short-segment Barrett's oesophagus, respectively.
In a secondary analysis, we also calculated transition rates for each Barrett length category in a US population. With a slightly higher incidence of oesophageal adenocarcinoma (8.2 per 100 000), the transition rates for long-segment, short-segment and ultra-short-segment Barrett's oesophagus were 3.1, 0.4 and 0.1 per 1000 patients, respectively (table 4). To find one cancer per year, 318 patients with long-segment Barrett's oesophagus would need to undergo endoscopy; the corresponding number of patients with short-segment and ultra-short-segment Barrett's oesophagus are 2433 and 8744, respectively.
Figure 1 shows the number of Barrett patients that would need to undergo endoscopy to find one cancer based on the surveillance interval. For instance, considering a 3-year interval, the corresponding number of patients with long-segment, short-segment and ultra-short-segment Barrett's oesophagus would be 150, 1147 and 4121, respectively. Such calculation is based on the simplifying assumption that all surveillance intervals would be equally efficacious and that by extending the surveillance interval alone one could easily reduce the number of necessary endoscopies.
Within plausible ranges of Barrett prevalence rates, the annual transition rates varied little. The maximum annual transition rate for patients with a long segment Barrett's oesophagus was 0.33%, and for patients with a short and ultra-short segment these were 0.07% and 0.01%, respectively (table 5). The corresponding minimum number of patients that needed to undergo a surveillance examination to detect one cancer were 307, 1464 and 6888 patients with long-segment, short-segment and ultra-short-segment Barrett's oesophagus, respectively.
This study estimated transition rates for patients with Barrett's oesophagus of different Barrett lengths. We found that patients with long-segment Barrett's oesophagus have an annual transition rate of 0.22%. Patients with short-segment (1–3 cm) or ultra-short-segment (<1 cm) Barrett's oesophagus have a much lower annual risk of 0.03% and 0.01%, respectively. When applying a 3-year surveillance interval, the number of patients who needed to be examined to find one cancer in the German population ranges from 150 for patients with a long-segment Barrett's oesophagus to approximately 1000 for those with a short-segment and >4000 for those with an ultra-short-segment Barrett's oesophagus. When applying the slightly higher SEER incidence data, the number of patients who needed to be tested to find one cancer in the US population is still fairly high for short-segment Barrett's oesophagus (at least 800 patients) and for ultra-short-segment Barrett's oesophagus (almost 3000 patients). These results question current surveillance guidelines and call for a different approach to surveillance that takes the underlying Barrett length into account.
Reported estimates of cancer risks for patients with Barrett's oesophagus have decreased over the past two decades. A recent meta-analysis summarised 57 studies and found a 0.33% annual cancer transition rate.12 Population-based studies from the USA, Denmark, Ireland and the Netherlands, which were not included in the aforementioned meta-analysis, reported annual transition rates ranging between 0.12% and 0.4%.10 ,31–33 Our results suggest that the overall transition risk may be even lower when patients with an ultra-short Barrett segment of <1 cm are added to the Barrett's population at risk. It is likely that previous studies captured these patients inconsistently or not at all, because of the unclear definition of what constitutes a minimum Barrett length.2 ,3
Cancer risk increases with the length of the Barrett segment as shown in prior cohort and case–control studies.6–10 Our results are similar to those of prior studies that assessed cancer risk in patients with short-segment and long-segment Barrett's oesophagus.6 ,10 ,12 A recent retrospective study from Ireland reviewed 3148 hospital records of Barrett patients and found that the risk of progression was seven times greater for patients with long than short Barrett segment,6 which is similar to the 7.7-fold difference of our study. It is interesting to note that almost all cancers in observational cohort studies occurred in patients with a long-segment Barrett's oesophagus. In a large cohort study from the USA, no patient with a short-segment Barrett's oesophagus developed cancer during a mean follow-up of 5 years.10 Most patients who developed cancer had a Barrett segment of 6 cm or longer, and the annual transition rate for patients with a segment of <6 cm was 0.09%. These observations corroborate our findings of a low transition risk for patients with a short-segment Barrett's oesophagus.
The current study takes a different approach to estimating transition risks than previous studies. The analysis used three data inputs. First, it benefited from data on a large cohort of patients with newly diagnosed T1 oesophageal adenocarcinoma, who had confirmed Barrett's oesophagus—both by endoscopy and by histopathology. These data were aligned with population-based incidence data from the German cancer registry to calculate the incidence by categories of Barrett length. The third data input came from published studies on the population prevalence of Barrett's oesophagus, again by categories of Barrett length.
To our knowledge, data on the distribution of Barrett length among patients with oesophageal adenocarcinoma have not been published before. This type of information was crucial for the calculations underlying our analysis. The distribution of Barrett length among patients with T1 oesophageal adenocarcinoma may not be representative for all patients with oesophageal adenocarcinoma. The majority of these cancers are diagnosed at a later stage, and only approximately one-fourth are detected at an early stage.1 It is therefore plausible that the distribution of Barrett length in later cancer stages differs from earlier stages as a result of a different biology by Barrett length. If, for instance, long-segment Barrett's oesophagus was more aggressive than short-segment Barrett's oesophagus, it may be underrepresented in T1 cancers (less commonly detected at the T1 stage with fast growth). In that situation, however, the calculated cancer transition rates for patients with short or ultra-short Barrett segments would have been overestimated.
It should further be noted that the input data for our analysis came from different populations, which limits the validity of our analysis. The cohort of patients with T1 cancer was diagnosed at a large referral centre in Germany. Details concerning the treatment results and the long-term follow-up were recently published.5 Cancer incidence data were obtained from the German cancer registry, and population-based Barrett prevalence data were available from European countries and the USA.
A marked and similar incidence increase has been observed in most Western countries since the 1970s.34 ,35 In addition to the German incidence data, we have also calculated Barrett to cancer transition risks by Barrett length for a US population using the SEER data. This approach assumes that the distribution of Barrett length among newly diagnosed patients with cancer is similar in patients with oesophageal adenocarcinoma from the USA and Germany. Not surprisingly, the US transition risks were found to be quite similar to the German risks.
Published Barrett prevalence data range widely from 2% to 25%. The broad variation is likely related to the inclusion of different populations with different risk factors for oesophageal adenocarcinoma across these studies. While the highest prevalence was seen in a study that included predominantly older men,22 the lowest prevalence was seen in a study that examined the general population of men and women 18 years of age or older.18 Here, we calculated the transition rates for patients 50 years of age or older, because almost all oesophageal adenocarcinomas develop in patients of this age group.36 Unfortunately, we were not able to apply age-specific and gender-specific rates to our analysis because only insufficient data were available from previous studies on Barrett prevalence. We therefore applied the average Barrett prevalence of all studies to our analysis. Because the average estimate includes younger patients, it is possible that the true prevalence of Barrett's oesophagus for older persons is actually higher than estimated by the present analysis. Under such circumstances, our calculations would have overestimated the cancer transition rates.
All studies of Barrett prevalence required the presence of intestinal metaplasia for the diagnosis of Barrett's oesophagus. More recent studies suggest that the risk of neoplasia is present in any intestinal-type metaplasia.3 ,37 It is likely that the prevalence of any intestinal-type metaplasia in a mucosal extension proximal to the gastroesophageal junction is higher than reported.38 ,39 A higher Barrett prevalence would decrease the estimated cancer transition risk. To account for a possible variation, we performed a sensitivity analysis, considering plausible ranges in the population-based Barrett prevalence by Barrett length and in the length distribution of Barrett's oesophagus among patients with cancer. We found that applying broad ranges of assumptions exerted only a little influence on the main outcome of our analysis.
In light of the aforementioned limitations, it is important to note that our results should be considered estimates only. However, two arguments favour our approach: first, data on transition risk by Barrett length are limited. Second, we used a sensitivity analysis to examine the influence of variations in the baseline estimates, with the main results remaining largely unaffected by such variations. The present article may inspire other investigators to consider alternative approaches to provide additional and more precise estimates for the influence of Barrett length on cancer progression.
Since the results appear robust under a broad range of assumptions, they may have important clinical implications with respect to any recommended protocol of Barrett surveillance. Patients with long-segment Barrett's oesophagus would benefit most from surveillance. For these patients, a 3-year surveillance interval may be appropriate. In contrast, our results question the benefit of surveillance for patients with ultra-short-segment Barrett's oesophagus. Although these patients constitute one-fifth of newly diagnosed patients with cancer, the population at risk is much larger, and the overall risk of transition remains very low. Even with a 10-year surveillance interval 1000 patients would still need to undergo upper endoscopy to find only one patient with cancer. The value of surveillance examination depends on the accepted threshold of the number of patients that would need to undergo a surveillance examination.
Stratifying the Barrett population into low-risk and high-risk patients by Barrett length would resemble current colonoscopy surveillance practice, where patients with high-risk adenomas are assigned a shorter surveillance interval than patients with low-risk adenomas.40 Our study suggests that patients with a minimal Barrett tongue may not benefit from surveillance, and obtaining routine biopsies from an irregular Z-line or a <1 cm tongue should be discouraged. Patients with a short-segment Barrett's oesophagus (1 to <3 cm) may be followed every five years (or not at all), and patients with a long-segment Barrett's oesophagus may be followed every three years. Recently revised British and German guidelines implemented very similar recommendations where patients with a long segment may be followed every three years, and those with a short segment every three to five years, and no surveillance for those with a segment of <1 cm.13 ,14 Future studies may consider additional risk factors as gender, age and smoking to be included in a risk prediction model.8 ,33
In conclusion, our study applied novel data regarding the distribution of Barrett length among patients with newly diagnosed oesophageal adenocarcinoma to population-based cancer incidence data and Barrett prevalence data. Although cancer does develop in short (1 to <3 cm) and ultra-short Barrett's oesophagus (<1 cm), the overall cancer risk in such patients is very low. The large number of patients with Barrett's oesophagus who would need to undergo surveillance examination in order to find one cancer becomes prohibitive for patients with short or ultra-short Barrett's oesophagus. Our analysis suggests that the decision in favour of endoscopic surveillance of Barrett's oesophagus should be tailored according to the length of the underlying Barrett metaplasia.
We thank Lauren Gerson (California Pacific Medical Center, San Francisco, California, USA), Douglas Rex (Indiana University, Indianapolis, Indiana, USA) and Jukka Ronkainen (University of Oulu, Finland) for providing additional details on Barrett prevalence in their former study cohorts. This material is in part the result of work supported with resources and the use of facilities at the VA Medical Center, White River Junction, Vermont.
Contributors HP and OP contributed equally. HP: study concept, design, analysis and interpretation of data, manuscript draft and critical revision for important intellectual content, statistical analysis. OP: study concept, design, analysis and interpretation of data, manuscript draft and critical revision for important intellectual content. HA, HM, AM and CE: acquisition of data; critical revision of the manuscript for important intellectual content. MS: pathology review of all polyps and study biopsies; acquisition of data, critical revision of the manuscript for important intellectual content. KK: acquisition of data on incidence of oesophageal adenocarcinoma in Germany; critical revision of the manuscript for important intellectual content. AS: analysis and interpretation of data, critical revision for important intellectual content; statistical analysis.
Disclaimer The contents of this article do not represent the views of the Department of Veterans Affairs or the United States Government.
Competing interests None declared.
Ethics approval Ethics Committee Krankenhaus Wiesbaden.
Provenance and peer review Not commissioned; externally peer reviewed.