Radiofrequency ablation (RFA) is a safe and effective treatment for dysplastic Barrett's esophagus (BE) (1,2). After complete eradication of intestinal metaplasia (CEIM), recurrence of BE is common, occurring in 8%–10% of treated patients per year, whereas progression to dysplasia or cancer is less common (3–7). To protect patients from recurrent neoplasia, surveillance endoscopy with retreatment of any recurrent intestinal metaplasia and neoplasia is recommended (8) and is associated with a low rate of progression to invasive esophageal adenocarcinoma (7). Surveillance endoscopy also imposes a burden of risk of adverse effects and of cost from the endoscopic procedure (9,10). To balance the benefits and risks of surveillance endoscopy, clinicians must perform this examination frequently enough to forestall progression to invasive cancer, but not more frequently as to incur unnecessary cost, patient inconvenience, and risk (11).
Although outcomes in surveillance after endoscopic eradication therapy are consistently good in observational studies (3–6,12), there are no randomized clinical trials of the effect of surveillance. Previous observational studies have important limitations to their clinical applicability. Although we ideally would compare the risks and benefits of posttreatment surveillance with subjects not undergoing such examinations, these real-world studies report the observed outcomes under a regimen of surveillance and retreatment. In the absence of randomized data, previous modeling studies seeking to understand the benefit of endoscopic surveillance after endoscopic eradication therapy have either modeled one outcome at a time, which limits the study of the rare but important outcomes such as progression to unresectable carcinoma, or they have studied a composite outcome (3–7).
To overcome these limitations and better inform clinical surveillance decisions, we constructed multistate survival models to impute the natural history of recurrence and progression after CEIM. We compared these models with the observed outcomes under surveillance and retreatment to estimate the effect of surveillance to prevent progression to invasive esophageal adenocarcinoma. We compared the effect of surveillance after treatment of low-grade dysplasia (LGD) with that after treatment of high-grade dysplasia (HGD) or intramucosal adenocarcinoma.
The US RFA Registry
The US RFA Patient Registry is a multicenter registry consisting of patients treated with RFA for BE at 148 institutions (113 community-based and 35 academic-affiliated). This observational study prospectively collected standardized details of endoscopy, histopathology, surgery, and other events in the routine clinical management starting with RFA for BE. The methods of the US RFA Registry have been described in detail in previous studies (13). All physicians participating in this registry either elected to use Western institutional review board (IRB) approval or obtained IRB approval through their respective institutions.
Cohort inclusion and definition of outcomes
Participants were included for analysis if they achieved CEIM and had at least 1 further surveillance visit with histopathology results. CEIM was defined as the first upper endoscopy after RFA treatment with histopathology results with no intestinal metaplasia or associated neoplasia in the tubular esophagus, no dysplasia or adenocarcinoma in the cardia, and no treatments given. Participants who met the definition of CEIM but had a prior histologic grade of invasive adenocarcinoma, had a prior esophagectomy, or had no history of histologic grade more advanced than nondysplastic BE were excluded. Surveillance after CEIM for nondysplastic BE was not modeled because guidelines recommend against treatment of nondysplastic BE (8). Nondysplastic intestinal metaplasia in the cardia was not considered recurrence.
Constructing the postablation natural history imputation model
We constructed a model US RFA cohort before any treatment of recurrent intestinal metaplasia or neoplasia for model-based imputation of the natural history without any retreatment. We fit time-to-event, multistate models that modeled recurrence of BE, progression through stages of dysplasia, intramucosal adenocarcinoma, invasive adenocarcinoma, or death from other causes as discrete states (see Supplementary Figure 1, https://links.lww.com/AJG/C504). These states were as follows: maintaining CEIM, recurrence with only nondysplastic intestinal metaplasia, recurrence with indefinite for dysplasia or LGD, recurrence or progression to HGD, recurrence or progression to intramucosal adenocarcinoma, progression to invasive esophageal adenocarcinoma, and death from any other cause. Invasive adenocarcinoma was defined as T1b or higher stage adenocarcinoma. Recurrence or progression to adenocarcinoma and death from any cause without adenocarcinoma were modeled as absorbing states from which participants could not emerge. A passage between states in the model occurs at a constant rate known as the transition intensity, which is estimated based on the observed data. Transitions from an initial recurrence grade to a higher one are sometimes observed, but once treatment occurs after a recurrence, the natural history can no longer be observed, which introduces a potential for bias. In particular, those cases where immediate treatment was performed may be a higher risk subset than those treated after biopsies reveal recurrence or after referral to a treatment center but are not observed in the data defining the transition between intermediate states. To assess the potential impact of this bias, we performed sensitivity analyses where the transition intensities estimated from the data were increased by varying proportions up to 3-fold, representing selection bias for cases where immediate treatment is not performed. More frequent surveillance was also modeled in sensitivity analyses assuming it was 50%, 100%, and 200% more effective than what was observed. To simulate the case if the cohort was surveilled at the rate recommended by upcoming guidelines, we modeled an increase in surveillance effectiveness proportional to the increase in surveillance visits under the guidelines compared with the observed visits over 5 years. The natural history model was stratified by worst prior histologic grade before CEIM and had covariates for time-dependent age and baseline Barrett's segment length. The fit of the imputation model was assessed by comparing the cumulative incidence of invasive esophageal adenocarcinoma as simulated by the model with that observed in the actual data after CEIM but before any retreatment, i.e., during the natural history of the CEIM patients (see Supplementary Figures 2, https://links.lww.com/AJG/C505 and 3, https://links.lww.com/AJG/C506). This calibration was possible because patients in the US RFA Registry returned for endoscopic surveillance at different intervals, depending on patient compliance and provider instructions, leading to a natural experiment describing the outcomes of surveillance endoscopy performed on different intervals. Multistate survival models were fit in R version 4.0.5 and using the msm package version 1.6.8.
Imputing events after retreatment to estimate natural history
Once a cohort member is treated, their natural history can no longer be observed in the data. We used the multistate natural history model to randomly impute 1,000 natural histories after any retreatment or recurrence. We applied this model of transitions in the actual surveillance cohort to estimate the rate of transition between recurrence of various histologic grades conditional on each participant's covariates, the time, and histologic grade of recurrence. We fit Kaplan-Meier estimates of the cumulative incidence of invasive adenocarcinoma and report because the imputations estimate the mean of the imputations by time after CEIM. To estimate the 5-year risk difference for surveillance, we compared the imputed natural history estimates with as-treated Kaplan-Meier estimates. We performed 1,000 bootstrap samples to estimate the SE of the risk difference for both population sampling and the imputation processes. Intramucosal adenocarcinoma was combined with HGD for the imputation analyses because of insufficient sample size at 5 years. We calculated the number needed to surveil as 1 divided by the risk difference rounded up to the nearest whole person.
All physicians participating in this registry either elected to use Western IRB approval or obtained IRB approval through their respective institutions.
Included participants and baseline characteristics
Among 5,521 participants in the US RFA Registry, there were 1,401 who met the criteria for inclusion (Figure 1). The included participants were similar in baseline characteristics to the overall registry population (Table 1). The mean age of included participants was 64.5 years (SD 10.3), and the majority were White (94.4%) and male (82.6%). The median number of surveillance visits per included participant was 2 (interquartile range 1–4), and the median time under observation was 26.0 months (interquartile range 16.6–39.2).
Cumulative incidence of invasive adenocarcinoma in the as-treated cohort
Progression to invasive esophageal adenocarcinoma under a regimen of surveillance and retreatment after CEIM was rare (Figure 2), occurring in an estimated 0.2% (95% confidence limits 0.0–0.6) of patients with LGD and 1.5% (95% confidence limits 0.3%–2.7%) of patients with HGD or intramucosal adenocarcinoma by 5 years. The risk was 7.4 times greater for HGD or intramucosal adenocarcinoma compared with LGD.
Natural history estimates and estimated effects of surveillance
The estimated cumulative incidence of invasive adenocarcinoma without surveillance and retreatment was much higher than the as-treated cohort for HGD or intramucosal adenocarcinoma, but only slightly higher for LGD (Figure 3). The number needed to surveil to prevent one case of invasive esophageal adenocarcinoma over 5 years where surveillance performed as observed in the registry was 90 for LGD and 21 for HGD or intramucosal adenocarcinoma (Table 2). These effect estimates were moderately sensitive to the varying assumptions of underestimation of risk by our imputation model or more frequent surveillance than was performed in the cohort itself (Table 3).
Quantifying surveillance visits, treatments, and their yield
Surveillance yielded treatment of recurrent disease in an estimated cumulative 65.6% of participants by 5 years (Figure 4a). Surveillance endoscopies occurred at an overall average of 1.3 surveillance endoscopies per year. Surveillance endoscopies accumulated at a roughly linear rate and differences in the frequency of surveillance endoscopies by worst prior histologic grade were small (Figure 4b), meaning that patients treated for baseline LGD had endoscopic surveillance about as often as those treated for baseline HGD. Among surveillance retreatment endoscopies, the worst histologic grade of recurrence was higher among patients with a baseline histology of HGD or intramucosal adenocarcinoma compared with LGD.
In this observational analysis of a large, prospective study of surveillance outcomes after endoscopic eradication therapy with RFA among participants with neoplastic BE, we modeled the natural history of recurrence and progression to invasive cancer and estimated the effects of surveillance endoscopy for comparison with the risks. The natural history estimates of the incidence of surveillance invasive adenocarcinoma at 5 years were 1.3% (95% confidence limits 0.2%–4.8%) for LGD and 6.3% (2.6%–11.9%) for HGD or intramucosal adenocarcinoma. The observed rates of invasive adenocarcinoma in the cohort were 0.2% in patients with baseline LGD, and 1.5% in patients with baseline HGD or intramucosal adenocarcinoma by 5 years, suggesting a substantial protective effect of surveillance endoscopy. The estimated benefits of posttreatment surveillance in preventing invasive adenocarcinoma were 4.3-fold greater for HGD or intramucosal adenocarcinoma compared with LGD. Because endoscopic surveillance is very safe, the weight of this evidence suggests the benefits of surveillance even in LGD are greater than the low risk of serious adverse effects of surveillance endoscopy in otherwise healthy patients, but the risks may outweigh benefits in some patients at elevated risk for surveillance endoscopy.
Although surveillance was more effective for pretreatment HGD and intramucosal adenocarcinoma and less effective for pretreatment LGD, the observed frequency of surveillance in the baseline HGD/intramucosal carcinoma cohort was less frequent than guidelines recommend and similar to guideline-recommended intervals for LGD (Figure 4b) (8). Despite this pattern, more retreatments were given to patients with pretreatment HGD and intramucosal adenocarcinoma (Figure 4b), and the severity of neoplasia retreated was much greater (Figure 4c). Compared with the initial therapy of dysplastic BE (1), lesions found during endoscopic surveillance after successful ablation were predominantly of lower grade (Figure 4c) than the baseline histologic grade, which is why there is a smaller effect for endoscopic surveillance compared with initial treatment of dysplastic BE. The Barrett's segment length before ablation had a detectable effect on the natural history estimates, but the effect was too small to justify varying surveillance intervals (full data not shown).
Although this is the first analysis to estimate the natural history after successful ablative therapy, a large body of literature describes the observed outcomes in surveillance. Except for a prior report from this same registry, most studies have not reported the rate of invasive esophageal adenocarcinoma in surveillance because of insufficient sample size (7). Rates of the combined outcome of HGD, intramucosal adenocarcinoma, and invasive esophageal adenocarcinoma from other postablation cohorts are similar to the as-treated estimates of this study (14). By comparison, the estimated effects of surveillance even for HGD or intramucosal adenocarcinoma were substantially less than the benefit observed from the initial successful ablation of these lesions in randomized clinical trials of endoscopic eradication (1,2). Therefore, although endoscopic surveillance after ablation is merited, most of the protective effect is already achieved from the initial successful ablation itself.
This analysis has important limitations. Although the analysis aims to describe the natural history effects of surveillance, the study design has inherent limitations to describe this. At the point of retreatment of recurrent disease in surveillance, the natural history and treated history diverge, and the natural history that would have occurred after treatment is not observed. This missing person-time, which we must impute using a multistate model, may be systematically different from the observed person-time for progression between intermediate states of recurrence and thus may introduce a bias. This bias may underestimate natural history risk by modeling more cases of microscopic rather than macroscopic recurrence and of low-volume rather than high-volume centers. Our sensitivity analyses suggest that the stronger the bias of this type, the greater the benefit of surveillance for HGD and intramucosal adenocarcinoma compared with LGD, and the benefit for LGD would remain modest (Table 3). This bias could only be resolved completely with randomization. The base case analysis compares surveillance because it was performed in the registry, which was generally less frequently than is recommended by guidelines, with the natural history estimate. Although this approach has the advantage of using real-world data instead of theoretical “perfectly performed” surveillance, it may thus underestimate the benefit of surveillance at the recommended intervals, which may be greater. Estimates extrapolating the effect of more frequent surveillance, such as is recommended in forthcoming guidelines (15), are presented in Table 3.
The strengths of this analysis are the focus on a clinically important but rarely studied outcome, progression to invasive adenocarcinoma, and the estimation of the effects of surveillance. Although there are limitations to the analysis' ability to quantify the effects of surveillance on posttreatment BE compared with a randomized clinical trial, such a study has not been reported. The US RFA Registry is a sample of community and academic ablation centers and has similar rates of recurrence and progression compared with other US studies.
This analysis of a large, multicenter, US cohort in surveillance after endoscopic eradication therapy used imputation with multistate survival models to model the natural history of recurrence and progression and thus estimate the protective effect of endoscopic surveillance after successful ablation of dysplastic BE. We found the benefit of surveillance after successful initial treatment is concentrated in patients with baseline HGD and intramucosal adenocarcinoma, whereas benefit was marginal in patients with baseline LGD. Our analysis suggests that substantial efforts are merited to retain those with baseline HGD or intramucosal carcinoma in endoscopic surveillance. Conversely, in patients with LGD and elevated endoscopy risk, the benefits of surveillance should be carefully considered against the risk, and such examinations might provide modest or no survival advantage, especially late in life, when competing causes of mortality rise. These findings reinforce the findings of our prior analysis (11) and support forthcoming surveillance recommendations (15) that will decrease surveillance frequency after successful ablation of baseline LGD and maintain aggressive surveillance after successful ablation of baseline HGD. Further studies should investigate the cost effectiveness of surveillance of LGD.
CONFLICTS OF INTEREST
Guarantor of the article: Aaron P. Thrift, PhD.
Specific author contributions: C.C.C.: manuscript writing and critical review, access to the study data, and decision to publish. N.J.S.: study design and data collection, access to the study data, critical review of manuscript, and decision to publish. A.P.T.: manuscript writing and critical review, access to the study data, and decision to publish.
Financial support: This work was funded by T32 DK007634.
Potential competing interests: C.C.C. none declared. N.J.S. has received research funding from Medtronic, Pentax, Steris, CDx Medical, Lucid, and Interpace Diagnostics and has worked as a consultant for Boston Scientific, Cernostics, Cook Medical, Aqua, Exact Sciences, and Phathom. A.P.T. none declared.
WHAT IS KNOWN
- ✓ Radiofrequency ablation is a safe and effective treatment for dysplastic Barrett's esophagus.
- ✓ Surveillance after complete eradication of intestinal metaplasia yields a clinically significant rate of recurrent disease that is usually amenable to further endoscopic treatment.
WHAT IS NEW HERE
- ✓ We used imputation approaches to model the benefit of surveillance rather than the yield.
- ✓ We found a clinically significant benefit of surveillance and retreatment after endoscopic eradication of high-grade dysplasia, but a more modest benefit in the setting of low-grade dysplasia.
- ✓ A single surveillance endoscopy performed after successful ablation of high-grade dysplasia prevents 4 times as much invasive cancer than surveillance examination performed after successful ablation of low-grade dysplasia.
We thank the University of North Carolina at Chapel Hill and the Research Computing group for providing computational resources and support that have contributed to these research results.
1. Shaheen NJ, Sharma P, Overholt BF, et al. Radiofrequency ablation in Barrett's esophagus with dysplasia. N Engl J Med 2009;360:2277–88.
2. Phoa KN, van Vilsteren FG, Weusten BL, et al. Radiofrequency ablation vs endoscopic surveillance for patients with Barrett esophagus and low-grade dysplasia: A randomized clinical trial. JAMA 2014;311:1209–17.
3. Guthikonda A, Cotton CC, Madanick RD, et al. Clinical outcomes following recurrence of intestinal metaplasia after successful treatment of Barrett's esophagus with radiofrequency ablation. Am J Gastroenterol 2017;112:87–94.
4. Cotton CC, Wolf WA, Overholt BF, et al. Late recurrence of Barrett's esophagus after complete eradication of intestinal metaplasia is rare: Final report from ablation in intestinal metaplasia containing dysplasia trial. Gastroenterology 2017;153:681–8.e2.
5. Gupta M, Iyer PG, Lutzke L, et al. Recurrence of esophageal intestinal metaplasia after endoscopic mucosal resection and radiofrequency ablation of Barrett's esophagus: Results from a US multicenter consortium. Gastroenterology 2013;145:79–86.e1.
6. Sami SS, Ravindran A, Kahn A, et al. Timeline and location of recurrence following successful ablation in Barrett's oesophagus: An international multicentre study. Gut 2019;68:1379–85.
7. Wolf WA, Pasricha S, Cotton C, et al. Incidence of esophageal adenocarcinoma and causes of mortality after radiofrequency ablation of Barrett's esophagus. Gastroenterology 2015;149:1752–61.e1.
8. Shaheen NJ, Falk GW, Iyer PG, et al. ACG clinical guideline: Diagnosis and management of Barrett's esophagus. Am J Gastroenterol 2016;111:30–51; quiz 51.
9. Quine MA, Bell GD, McCloy RF, et al. Prospective audit of upper gastrointestinal endoscopy in two regions of England: Safety, staffing, and sedation methods. Gut 1995;36:462–7.
10. ASoP Committee, Ben-Menachem T, Decker GA, et al. Adverse events of upper GI endoscopy. Gastrointest Endosc 2012;76:707–18.
11. Cotton CC, Haidry R, Thrift AP, et al. Development of evidence-based surveillance intervals after radiofrequency ablation of Barrett's esophagus. Gastroenterology 2018;155:316–26.e6.
12. Komanduri S, Kahrilas PJ, Krishnan K, et al. Recurrence of Barrett's esophagus is rare following endoscopic eradication therapy coupled with effective reflux control. Am J Gastroenterol 2017;112:556–66.
13. Shaheen NJ, Kim HP, Bulsiewicz WJ, et al. Prior fundoplication does not improve safety or efficacy outcomes of radiofrequency ablation: Results from the U.S. RFA registry. J Gastrointest Surg 2013;17:21–9; discussion 28-9.
14. Krishnamoorthi R, Singh S, Ragunathan K, et al. Risk of recurrence of Barrett's esophagus after successful endoscopic therapy. Gastrointest Endosc 2016;83:1090–106.e3.
15. Shaheen NJ, Falk GW, Iyer PG, et al. Guideline to Practice: Diagnosis and Management of Barrett's Esophagus: An Updated ACG Guideline. Am J Gastroenterol 2022. Online ahead of print.