Key points

Introduction

The notion of endoscopic surveillance of Barrett’s esophagus with the aim of decreasing the burden of esophageal adenocarcinoma (EAC) is very attractive for several reasons. The incidence of EAC has risen dramatically over the past 4 decades in many Western nations. The mortality rate from the cancer is high. There exists a relatively easily identifiable precursor lesion in Barrett’s esophagus, and tissue can be acquired from the lesion for histology with relative ease and safety (when compared with sampling, for instance, lesions of the pancreas, kidney, liver, or lung). Although survival from even most early-stage esophageal cancers is still poor, a growing body of evidence has demonstrated the efficacy of endoscopic therapy directed at dysplasia and even adenocarcinoma confined to the mucosa (invading the lamina propria or muscularis mucosae, T1a cancers). Therefore, it would seem reasonable that an effort should be made to reduce the burden of EAC by performing periodic endoscopic surveillance in patients with Barrett’s esophagus in order to identify and intervene on patients who develop dysplasia or intramucosal cancer. Herein, the author reviews the evidence regarding the effectiveness and efficiency of surveillance, caveats in interpreting that evidence, recommendations from published guidelines, details regarding the logistics of how surveillance should be conducted, and potential future directions.

Introduction

The notion of endoscopic surveillance of Barrett’s esophagus with the aim of decreasing the burden of esophageal adenocarcinoma (EAC) is very attractive for several reasons. The incidence of EAC has risen dramatically over the past 4 decades in many Western nations. The mortality rate from the cancer is high. There exists a relatively easily identifiable precursor lesion in Barrett’s esophagus, and tissue can be acquired from the lesion for histology with relative ease and safety (when compared with sampling, for instance, lesions of the pancreas, kidney, liver, or lung). Although survival from even most early-stage esophageal cancers is still poor, a growing body of evidence has demonstrated the efficacy of endoscopic therapy directed at dysplasia and even adenocarcinoma confined to the mucosa (invading the lamina propria or muscularis mucosae, T1a cancers). Therefore, it would seem reasonable that an effort should be made to reduce the burden of EAC by performing periodic endoscopic surveillance in patients with Barrett’s esophagus in order to identify and intervene on patients who develop dysplasia or intramucosal cancer. Herein, the author reviews the evidence regarding the effectiveness and efficiency of surveillance, caveats in interpreting that evidence, recommendations from published guidelines, details regarding the logistics of how surveillance should be conducted, and potential future directions.

Incidence of cancer in Barrett’s esophagus

Whether surveillance of Barrett’s esophagus is rational depends heavily on the incidence of EAC in that setting. Some relatively small cohorts have published estimates of the incidence of EAC in nondysplastic Barrett’s esophagus exceeding 2% per year. However, there was likely a bias for disseminating the studies with the greatest observed risk. A recent meta-analysis of cohort studies has estimated the incidence at 0.33% per year. Some large cohort studies have estimated the incidence as low as 0.12% per year. When the risk of cancer is as low as 2% per decade, it is not clear that any intervention, even if it is efficacious for preventing cancer mortality, is worth the expected expense and complications. In low-grade dysplasia, the incidence of cancer may be higher, but it still ranges between 0.2% and 0.4% per year. The incidence is clearly greater when high-grade dysplasia is present, but in that case, the standard of care has become endoscopic therapy rather than surveillance.

Lead-time and length-time biases

In order to interpret the results of surveillance in Barrett’s esophagus, an understanding of lead-time and length-time biases is necessary. Studies of the effects of screening or surveillance on cancer mortality are inherently fraught with the potential for bias by lead-time and length-time effects. Lead-time effects are best understood through a thought experiment ( Fig. 1 ). A hypothetical set of twin brothers is imagined who were born on the same day, destined to develop EAC on the same day, and destined to die from EAC on the same day. Twin A does not undergo any screening or surveillance and instead is diagnosed with EAC when he presents with dysphagia, say at the age of 68 years. Assuming he dies at the age of 70 years, his survival time with EAC is 2 years. Twin B decided to undergo screening, was found to have Barrett’s esophagus, and underwent surveillance; he was diagnosed with EAC while asymptomatic, say at the age of 63 years. He undergoes esophagectomy but still ends up dying from recurrent metastatic EAC at the age of 70 years. His survival time with EAC is observed to be 7 years. In practice, it is not known when someone is destined to develop cancer or when one is destined to die (even among twins), so this difference of 5 years between the 2 patients would be mistakenly attributed as an extension of the duration of life by 5 years, when in fact all that was accomplished with screening and surveillance was increasing the proportion of life with a known diagnosis of EAC in twin B and not delaying the date of his death at all. Even if there is a true effect of screening and surveillance on survival, it can be expected that lead-time effects will bias the observation of survival toward a stronger effect.

Lead-time bias. Each of a pair of twins is destined to develop cancer on the same day and die from cancer on the same day. One twin does not undergo screening and is diagnosed with cancer because of symptoms. The other twin is diagnosed with cancer when he undergoes screening. If it is not known that they are destined to live the exact same number of days in their entire life, it would be observed that the second twin survives longer with his cancer than the first twin. But in fact, screening did not extend his life; it only increased the proportion of his life with a diagnosis of cancer.

Length-time effects refer to the predilection of screening and surveillance examinations to identify indolent disease. Cancers are heterogeneous, with some progressing quickly and others slowly ( Fig. 2 ). Any screening or surveillance examination is more likely to identify a slow-growing tumor than a fast-growing one because there is more time available to detect the cancer between onset and death. A reasonable assumption is that a slow-growing tumor is also less likely to be fatal than a fast-growing tumor; so any screening or surveillance examination is more likely to detect cancers that are less fatal. The observed longer survival in surveillance-detected cases could be entirely because of the differences in the biology of the tumors detected by surveillance versus symptom-detected tumors and not a causal effect of surveillance. The potential for length-time bias can be aggravated by the manner in which cases and controls are classified. For instance, in one analysis of the effect of surveillance of Barrett’s esophagus, the cases were defined as cancers detected by surveillance and controls as cancers detected by symptoms in the interval between surveillance. Such a design inherently compares slow-growing to fast-growing cancers; the investigators found a strong inverse association between surveillance (and likely preferential detection of slow-growing cancers) and mortality. Analyzing the data from the same population in a different manner wherein patients with fatal cancer were compared with those with Barrett’s esophagus for the presence of a prior endoscopy, the same investigators strikingly found no evidence of a protective effect of endoscopy. The only way to entirely avoid length-time effects in studies of surveillance (even in randomized trials) is to perform surveillance so frequently that all tumors are detected by surveillance and not in the interval between surveillance (for instance, in an absurd scenario, with weekly or even daily endoscopies).

Length-time bias. Cancers are heterogeneous. Slow-growing (less-fatal) cancers are represented by long bars (duration of time a patient is alive with the cancer). Fast-growing (more fatal) cancers are represented by short bars. In this example, the prevalence of slow-growing and fast-growing cancers is equal (there are 6 of each type). Surveillance is performed at the 4 times represented by the dashed lines. Because the slow-growing cancers are present for a longer period, surveillance is inherently more likely to identify slow-growing tumors (all 6 in this example), but miss fast-growing tumors (only 2 of 6 are detected in this example).

Empirical studies of the effectiveness of surveillance endoscopy

There have not been any published randomized trials of surveillance in Barrett’s esophagus. Several retrospective studies have suggested that prior upper endoscopy among patients with EAC is associated with earlier stage of cancer at the time of initial diagnosis and improved survival. One must pay careful attention to the design of those studies in order to interpret the results. For instance, in a retrospective review of 589 patients with adenocarcinoma of either the esophagus or the gastroesophageal junction diagnosed within an integrated health system of Kaiser Permanente Northern California, Corley and colleagues found 23 patients who had been diagnosed with Barrett’s esophagus at least 6 months before the diagnosis of cancer. Of these 23 patients, 15 had their cancer detected during a surveillance endoscopy, and 8 patients were detected due to symptoms, none of whom had undergone any surveillance endoscopies. The age at diagnosis of patients in whom cancer was detected by surveillance was similar to that of patients without a prior diagnosis of Barrett’s esophagus, but less than that of those with a prior Barrett’s esophagus diagnosis who were not undergoing surveillance. The cancers in patients with a prior diagnosis of Barrett’s esophagus who were not undergoing surveillance were diagnosed at a later stage than the surveillance-detected cancers ( P = .02; the latter group with none worse than stage IIA). Adjusting for age, the patients with surveillance-detected cancer survived longer than those with the symptom-detected cancer with prior Barrett’s esophagus (hazard ratio [HR], 0.25; 95% confidence interval [CI], 0.06–1.0). However, defining cases and controls based on surveillance detection of cancer inherently aggravates the potential for length-time bias. After 11 years, Corley and colleagues updated the review of their health system’s data to assess the role of surveillance in cancer mortality; they identified 38 deaths from adenocarcinoma of the esophagus or gastroesophageal junction in patients with a diagnosis of Barrett’s esophagus at least 6 months before their cancer diagnosis and compared these to 101 patients with Barrett’s esophagus matched for age, sex, and year of Barrett’s esophagus diagnosis. In this analysis, they did not find any association between cancer mortality and receipt of surveillance endoscopy within 3 years prior (odds ratio [OR], 0.99, 95% CI, 0.36–2.75).

In a case-control study within the US Veterans Health Administration, Kearney and colleagues identified 245 cases of death from adenocarcinoma of the esophagus or gastric cardia in patients diagnosed with gastroesophageal reflux disease (GERD) using International Classification of Diseases diagnostic billing codes, and compared them to 980 controls with GERD, matched for age, sex, and race. Receipt of a prior upper endoscopy at least 1 year prior was associated with a decreased odds of death from cancer (OR, 0.66; 95% CI, 0.45–0.96), but the investigators noted that none of the controls had been diagnosed with EAC and none had undergone esophagectomy, raising the likelihood that the results were biased in some way. For instance, those who underwent upper endoscopy may have been more conscious of their health and may have had other health-conscious habits such as avoidance of tobacco, healthy diet, and exercise, which were not measured. The author’s group also examined the effect of prior upper endoscopy on patients with EAC with GERD using the same database. Differences in the analysis were that data from 4 additional years were included, all subjects were required to have at least 1 encounter with the Veterans Health Administration in each of the 5 years preceding the diagnosis of cancer, and electronic medical records were reviewed to verify case status, dates of endoscopies, stage at diagnosis, and follow-up. From 311 potential subjects using administrative data, 155 subjects with EAC and a prior diagnosis of GERD were verified. Of the 155 subjects with EAC, 25 cases had undergone upper endoscopy between 1 and 5 years before the diagnosis of cancer. Those undergoing prior upper endoscopy were diagnosed at an earlier stage ( P = .03), but there was no survival benefit (HR, 0.93; 95% CI, 0.58–1.50, adjusting for age, comorbidities, and year of diagnosis). Adherence with the recommended interval of surveillance (eg, 3 years for nondysplastic Barrett’s esophagus), trended toward an association with improved survival (adjusted HR, 0.52; 95% CI, 0.24–1.12), but even among patients who did not undergo cancer resection, there seemed to be an association between adherence and survival (adjusted HR, 0.42; 95% CI, 0.16–1.09). This observation suggests that even if there had been a statistically significant association between surveillance and survival from EAC, it may not be a causal relation, but instead the result of bias such as selection of health-conscious patients for endoscopy and other biases described below.

In an analysis of the US Surveillance, Epidemiology, and End Results (SEER) cancer registry linked to Medicare data, Cooper and colleagues identified 777 patients with EAC who were at least 70 years old at diagnosis. An upper endoscopy was performed at least 1 year before cancer diagnosis in 13%, but Barrett’s esophagus was coded in only 7%. The stage of cancer was earlier in those with a prior upper endoscopy (62% vs 35% local cancer or dysplasia; notably, approximately 12% of the patients with a prior upper endoscopy and classified as having cancer by the study actually only had dysplasia). Patients classified with EAC with a prior upper endoscopy had longer median survival time (7 months vs 5 months, P <.01). Adjusting for age, sex, race, and comorbidities, prior upper endoscopy was associated with improved survival (HR, 0.73; 95% CI, 0.57–0.93). Results were similar for patients who not only underwent a prior upper endoscopy but also were diagnosed with Barrett’s esophagus.

Cooper and colleagues updated their analysis of the SEER-Medicare linked database 7 years later but limited the newer analysis to the receipt of an upper endoscopy between 6 months and 3 years before the diagnosis of EAC and lowered the minimum age to 68 years. A total of 2754 patients with EAC were identified, 11.5% of whom had undergone upper endoscopy within the defined time, and Barrett’s esophagus was coded in 8.1%. Among those who had undergone prior upper endoscopy, the median number of prior endoscopies was 4. A prior diagnosis of Barrett’s esophagus was more strongly associated with both earlier stage and improved survival than merely a prior upper endoscopy (for survival: prior Barrett’s esophagus HR, 0.45; 95% CI, 0.25–0.80 vs prior endoscopy HR, 0.66; 95% CI, 0.47–0.93).

Verbeek and colleagues analyzed data from the Netherlands National Cancer Registry linked to a nationwide pathology registry. The investigators identified 9780 cases of EAC, 791 (8%) of whom had a diagnosis of Barrett’s esophagus at least 1 year before the cancer diagnosis, including 452 undergoing adequate surveillance (defined as at least 1 additional set of biopsies performed within 1.5 times the recommended interval; eg, 4.5 years for nondysplastic Barrett’s esophagus). Those classified as undergoing adequate surveillance were more likely to be detected at stage I (47% vs 18% for inadequate surveillance, 17% for no surveillance but prior diagnosis of Barrett’s esophagus, and 6% for no known prior Barrett’s esophagus). Adequate surveillance was associated with improved 5-year survival compared with those without a prior diagnosis of Barrett’s esophagus (OR, 0.74; 95% CI, 0.58–0.94, adjusting for sex, age, year of diagnosis, academic vs community hospital, location of tumor, stage, differentiation, and type of treatment). Although Barrett’s esophagus with inadequate surveillance or without surveillance were both associated with improved survival in univariate analysis, there were no associations in the adjusted analyses. The definition for adequate surveillance used in this study aggravates the potential for length-time bias because patients with only 1 endoscopy before the diagnosis of EAC would be classified as no surveillance, even if that procedure had been performed within 4.5 years of the cancer diagnosis. This fact likely explains why patients classified as not undergoing surveillance had a shorter interval between diagnosis of Barrett’s esophagus and EAC than those classified as adequate surveillance (median 5.2 years vs 7.5 years). Patients with inadequate surveillance were older at the time of diagnosis than those with adequate surveillance but had longer interval between diagnosis of Barrett’s esophagus and EAC (median 12.6 years), suggesting that their age at diagnosis of Barrett’s esophagus may have been similar. As the survival analyses were conducted starting at the age of diagnosis of cancer, this suggests that at least some of the observed association between adequate surveillance and survival is likely because of lead-time bias. Strangely, the association of adequate surveillance with survival persisted after adjusting for stage and treatment modality; if the effect of surveillance is mediated through earlier stage at diagnosis and greater likelihood of surgical resection, then it should be expected that association with adequate surveillance should no longer be evident after adjusting for those factors. This suggests that at least some of the observed association between surveillance and survival is biased by length-time effects and/or confounded by other unmeasured factors such as comorbidities or health-conscious behaviors.

Recently, Bhat and colleagues reported results from a population-based study in Northern Ireland. The investigators identified 716 cases of EAC, 52 (7.3%) of whom had a diagnosis of Barrett’s esophagus at least 6 months prior. Those with a prior diagnosis of Barrett’s esophagus were more likely to be diagnosed at stage I or II (44.2% vs 11.1%) and more likely to undergo surgical resection (50.0% vs 25.5%). Patients with EAC with a prior diagnosis of Barrett’s esophagus seemed to survive longer (HR, 0.44; 95% CI, 0.30–0.64; adjusted for age, sex, and tumor grade). However, there was a survival benefit of prior Barrett’s esophagus even in patients who did not undergo esophagectomy (HR, 0.50; 95% CI, 0.32–0.79), suggesting that the estimate of the effect was likely biased. The investigators attempted to adjust for the potential of lead-time bias by estimating the duration of asymptomatic, but clinically detectable cancer (the sojourn time) by comparing the average age at cancer diagnosis between those with and without prior Barrett’s esophagus diagnosis (67.4 vs 69.9 years, a difference of 2.5 years). In the analysis adjusting for lead-time bias, the effect of a prior diagnosis of Barrett’s esophagus on survival was attenuated (HR, 0.65; 95% CI, 0.45–0.95). The investigators also attempted to adjust for the potential of length-time bias by accounting for the relative proportions of low-grade differentiated tumors in both groups, and the expected survival difference between grades of differentiation, finding that it only slightly altered the estimate for the effect of prior diagnosis of Barrett’s esophagus on survival.

In summary, several retrospective studies have observed apparent associations between surveillance and improved survival from EAC, but the results have not been consistent and the studies are prone to several important potential biases.