Evaluating Cost‐Effectiveness of 85 Endoscopic Surveillance Strategies of Nondysplastic Barrett's Esophagus

ABSTRACT

Background

Barrett's esophagus is the known precursor to esophageal adenocarcinoma (EAC), a cancer with poor prognosis. While endoscopic surveillance detects early dysplasia and prevents progression, most Barrett's esophagus patients do not progress to EAC, leading to invasive and costly surveillance. This study aimed to identify cost‐effective endoscopic surveillance strategies by risk stratifying patients based on Barrett's esophagus segment length and sex.

Methods

A Markov cohort model was developed to simulate the natural history of Barrett's esophagus to EAC. The model assessed 85 surveillance strategies and varied endoscopy intervals from 2 to 10 years for nondysplastic Barrett's esophagus and 6–12 months for dysplasia. Risk stratification was based on segment length (≤ 2 and ≤ 3 cm) and sex. Costs, utilities and transition probabilities were derived from published literature and clinical databases. Deterministic and probabilistic sensitivity analyses were performed, and cost‐effectiveness was evaluated from a third‐party payer perspective using a threshold of AU$50 000/QALY (2023 US dollars 35 945/QALY).

Results

The most cost‐effective strategy was biennial surveillance for long‐segment BE (> 2 cm) and 12‐month surveillance for LGD, excluding surveillance in low‐risk patients (ICER US$23 737/QALY). Risk‐based surveillance consistently outperformed nonstratified strategies. Sensitivity analyses confirmed the robustness of the model, with key drivers being transition rates and endoscopy costs.

Conclusion

We identified cost‐effective risk‐stratified endoscopic surveillance strategies for Barrett's esophagus, particularly when excluding low‐risk patients. Tailored risk‐guided surveillance strategies could improve resource allocation and clinical outcomes in managing Barrett's esophagus. The conserved resources can then be utilized to identify high‐risk individuals in the community.

Abbreviations

BE

Barrett's esophagus

EAC

esophageal adenocarcinoma

ICER

incremental cost‐effectiveness ratio

US

United States

1. Background

Esophageal adenocarcinoma (EAC) is a lethal cancer arising from a precursor condition known as Barrett's esophagus (BE). BE is diagnosed through endoscopic examination and biopsy of salmon colored mucosa in the esophagus showing the presence of metaplasia on histology (i.e., columnar epithelial changes). Surveillance guidelines have been developed to detect cancer progression at earlier stages when it is amenable to minimally invasive or endoscopic interventions. Endoscopic surveillance is the cornerstone of this approach, as it allows for the detection and treatment before it advances to invasive cancer, which improves survival [1, 2]. However, the progression from nondysplastic BE (NDBE) to EAC is low [3, 4, 5, 6, 7]. It is estimated that > 80% of patients will not progress to adenocarcinoma [8] but will be subjected to invasive, expensive, and resource exhaustive surveillance.

Several cost‐effectiveness studies have concluded that current guidelines are not cost‐effective, as recently seen in a systematic review of economic evaluations [9]. It is also evident that improving cost‐effectiveness requires discriminately reducing the number of endoscopic examinations in low‐risk individuals. Factors such as the length of BE segment and sex are known to influence the risk of progression to EAC. The maximum amount of visible BE segment length is an intuitive risk factor for progression to adenocarcinoma [10, 11]. It is conceivable that a longer segment has a higher number, and thus probability, of metaplastic cells undergoing further transformation to dysplasia and adenocarcinoma. This is well supported in the literature, including several systematic reviews and meta‐analyses looking at risk factors for progression [7, 12]. Obesity, alcohol, epigenetics, and age are potentially significant risk factors for the progression of BE, but these factors are more difficult to apply clinically. Using a recently developed Markov cohort model of BE progression [13], our main aim was to find cost‐effective endoscopic surveillance strategies by varying endoscopy surveillance intervals in risk‐stratified and non–risk‐stratified cohorts of individuals with known BE. We hypothesized that reducing the frequency of endoscopic surveillance in low‐risk subgroups, while maintaining 2 yearly endoscopic surveillance in the high‐risk subgroups, will prove to be cost‐effective.

2. Methods

A comprehensive process was undertaken to develop a clinically relevant cohort model for BE, emphasizing methodological rigor and adherence to best practices [14, 15, 16, 17]. The model was constructed using TreeAge Pro (Version 2021 R2.1, Williamstown, Massachusetts) with health states representing the progression from metaplasia (NDBE) to EAC, including low‐grade dysplasia (LGD), high‐grade dysplasia (HGD) (Figure S1), and regression from metaplasia (no BE). The starting population of the aggregate cohort was 50‐year‐old mixed sex individuals diagnosed with NDBE. Cycle length was set to 6 months, and the model's time horizon spanned 35 years, starting from an initial surveillance age of 50 years. The model terminated either at the age of 85 or death. The base case scenario was natural progression of BE (no surveillance), which was compared against each strategy. The full details of the construction, acquisition of model inputs, calibration, and validation are described elsewhere [13]. Ethical approval for the use of the Southern Adelaide Local Health Network BE surveillance database for model calibration was obtained from Southern Adelaide Human Research Ethics Committee (AUD/20/SAC/138).

2.1. Risk‐Stratified States

The model consists of two categories of health states: non–risk stratified and risk stratified. The nonstratified category represents the overall group of individuals with NDBE. The risk‐stratified category was divided into a low‐risk and a high‐risk subgroup, and together, the model outputs of these subgroups equaled the nonstratified category. Built into the model also were undetected health states, which resemble the natural progression of the disease and provide realistic simulation of “missing” progression between surveillance intervals. In the risk‐stratified strategies, patients were divided into subgroups based on BE segment length and sex.

• Segment length stratification: Patients were categorized into short‐segment BE (≤ 2 or < 3 cm, depending on the scenario) and long‐segment BE (> 2 or ≥ 3 cm). Long‐segment BE patients were considered at higher risk for progression to dysplasia and EAC.

• Sex stratification: Males were considered higher risk due to their higher incidence of both BE and progression to EAC, while females were classified as lower risk.

The combination of these risk factors and undetected states allowed us to develop 85 strategies tailored to specific patient subgroups, which included varying surveillance intervals or even excluding low‐risk patients from surveillance altogether. Movement between these states is shown in Figure 1.

FIGURE 1.

Movement through health states. Ca—cancer; EAC—esophageal adenocarcinoma; EMR—endoscopic mucosal resection; HGD—high‐grade dysplasia; LGD—low‐grade dysplasia; MDM—multidisciplinary meeting; NDBE—nondysplastic Barrett's esophagus; RFA—radiofrequency ablation.

2.2. Model Inputs

2.2.1. Transition Probabilities

Probabilities of transition between health states were derived from the literature when adequate evidence was present (Table 1). Due to the lack of high‐quality literature data informing transition probabilities between BE states (especially risk subgroups such as sex and Barrett's segment length), these inputs were calibrated to multiple model outputs, namely, overall progression to HGD, progression to esophageal adenocarcinoma, and percentage of expected cancer deaths (Table 2). The detailed process of calibration and validation has been published elsewhere [13].

TABLE 1.

Transition probabilities in common pathway from observed data in the literature.

Description	Mean value	Standard deviation	Ref
Probability of complication postendoscopic treatment (RFA or EMR)	20%	14%	[18]
Probability of requiring retreatment with RFA and/or EMR (HGD and EAC)	8.8%	11.9%	[19, 20]
Probability of HGD progression to adenocarcinoma postendoscopic treatment	1%	0.6%	[20, 21]
Probability of LGD progression to HGD postendoscopic treatment	0.5%	0.08%	[19, 21, 22]
Probability of requiring reablation of LGD lesion	9.66%	0.82%	[23, 24, 25]
Mortality postendoscopic treatment	0.6%	0.09%	[22]
Localized cancer	19%	—	[26]
Regional spread	29%	—	[26]
Unstaged cancer	20%	—	[26]
Distant spread	32%	—	[26]
Background mortality (age dependent)	Life tables		[27]

Abbreviations: EAC—esophageal adenocarcinoma; EMR—endoscopic mucosal resection; HGD—high‐grade dysplasia; LGD—low‐grade dysplasia; RFA—radiofrequency ablation.

TABLE 2.

Transition probabilities of progression in Barrett's esophagus stages for aggregate cohort as well as subgroups. Mean values shown in annual percentages from the best fitting model.

Risk‐related health states	No surveillance	Female	Male	Short (≤ 2 cm)	Long (> 2 cm)	Short (< 3 cm)	Long (≥ 3 cm)
No Barrett's to NDBE	21.47%	22.05%	21.94%	20.57%	31.19%	20.88%	31.67%
No Barrett's to LGD	0.18%	0.27%	0.24%	0.07%	0.56%	0.13%	0.79%
No Barrett's to HGD	0.03%	0.04%	0.03%	0.02%	0.07%	0.01%	0.10%
NDBE to No Barrett's	17.24%	17.32%	17.37%	14.04%	19.91%	14.02%	22.84%
NDBE to LGD	4.19%	4.04%	4.54%	1.16%	6.45%	1.32%	7.30%
NDBE to HGD	0.11%	0.16%	0.20%	0.10%	0.31%	0.10%	1.18%
NDBE to EAC	0.09%	0.10%	0.09%	0.04%	0.17%	0.04%	0.53%
LGD to HGD	5.89%	4.65%	5.80%	2.24%	4.61%	2.35%	7.82%
LGD to EAC	1.26%	0.93%	1.34%	0.36%	1.22%	0.40%	1.57%
LGD to NDBE	15.80%	20.47%	16.64%	19.48%	0.16%	18.90%	0.15%
HGD to EAC	3.80%	—	—	—	—	—	—

Abbreviations: EAC—esophageal adenocarcinoma; HGD—high‐grade dysplasia; LGD—low‐grade dysplasia; NDBE—nondysplastic Barrett's esophagus; No Barrett's—nil intestinal metaplasia.

2.2.2. Costs and Utilities

Costs were categorized as either “per cycle” or “per event” and reported in 2023 US dollars. Cost for endoscopic examination and treatment as a per‐event cost was generated from the Southern Adelaide Local Health Network BE surveillance database. Cancer‐related expenses, as well as the cost of maintaining a database of BE patients for surveillance, were sourced from a prior study conducted at our institution [28, 29, 30] and adjusted to reflect their 2023 US dollar equivalents using the consumer price index [31]. The perspective was third‐party payer (local health network).

Utility (quality‐adjusted life years [QALYs]) values were assigned to health states, whereas disutility values were assigned to events. Stages of BE were considered asymptomatic, so they were assumed to have the same quality of life as background utility. Background utility was derived from Viney et al. [32] and was adjusted for age (weighted for ~60% male population). In cancer states, the utility depended on the age and stage of cancer. Temporary disutility values were assigned to endoscopic treatments and complications associated with all treatments. Both costs and utilities were discounted annually at 5%. These values can be found in Table 3.

TABLE 3.

Costs and utilities. Costs are in US$ (2023) annual amounts. The utility values are 6 monthly subtracted per cycle from the health state.

	Cost (US$ annual)	Ref
Cost of an endoscopy	$1000	SA database
Cost of maintaining database (per individual annually)	$135	[29] Gordon 2014 (CPI adjusted)
Cost of radiofrequency ablation	$7189	SA database
Cost of endoscopic mucosal resection	$10 065	SA database
Cost of complication postendoscopic intervention	$8843	SA database
Cost of curative treatment (trimodality)	$68 620	[28, 29]
Cost of palliation	$11 287	[28, 29]

	QALYs	Ref
Background utility	Age dependent	[29, 32] Viney 30
Surveillance detected cancer	Background utility—0.08	[29, 33] Sullivan 34
Localized cancer (no nodal spread)	Background utility—0.168	[29, 34] Garside 32, Gerson 33
Regional spread	Background utility—0.235	[29, 33, 34] de Boer 31
Metastatic cancer	Background utility—0.3	[29, 33, 34] de Boer 31
Unstaged cancer	Background utility—0.235	[29]
Disutility for complication postendoscopic intervention (1 cycle)	−0.05	[29]
Disutility associated with endoscopic intervention (1 cycle)	−0.035	[29]

2.3. Deterministic and Probabilistic Analysis

Costs and QALYs were generated by TreeAge Pro as a mean value per cohort of one person. The main outcome variable was the incremental cost‐effectiveness ratio (ICER). A willingness‐to‐pay threshold of US$35 945/QALY (AU$50 000/QALY) was considered cost‐effective [35, 36]. The standard of care was natural progression of BE (no surveillance) which was compared against each strategy. One‐way sensitivity analyses were performed for all noncalibrated variables (Supplemental Table S1). A cost‐efficiency frontier graph was constructed to illustrate the relationship between surveillance strategy costs and QALYs. Undominated strategies, representing the most efficient trade‐offs between cost and effectiveness, were plotted and connected as a line.

Probabilistic sensitivity analysis (1000 Monte Carlo simulations) was performed by random variation based on specified distribution of model inputs (program evaluation and revision technique [PERT] distribution). Calibrated transition probabilities of BE were varied in matched sets to ensure compatible cumulative probabilities in tree branches did not exceed 100% and to also ensure ICER values between strategies were attributable to the treatment/strategy itself rather than variation in key model outputs, described in detail in a previous publication [13]. Noncalibrated transition probabilities and utility values were assigned beta distribution, whereas costs were assigned gamma distribution. All post hoc analyses and graphing were performed using RStudio (R Version 4.3.1).

3. Results

3.1. Notation of Surveillance Strategies

When alluding to a strategy, square brackets were used with three characteristics of the strategy separated by a semicolon, example below:

$$\left[10\text{yearly SSBE}\left(\le 2\mathrm{cm}\right);2\text{yearly LSBE};\mathrm{LGD}12\text{monthly}\right]$$

85 surveillance stratagies (Tables S2 and S3) were developed by varying (a) risk subgroups and (b) varying the frequency of endoscopic surveillance (2–10 yearly). Australian guideline surveillance was considered as follows:

$$\left[5\text{yearly SSBE}\left(\le 2\mathrm{cm}\right);2\text{yearly LSBE};\mathrm{LGD}6\text{monthly}\right]$$

3.2. Nondiscriminate Reduction in Endoscopic Surveillance Frequency (19 Strategies)

The endoscopic surveillance frequency was modified from every 2–10 years for NDBE and 6 monthly or 12 monthly for LGD. An additional two strategies tested dysplasia only surveillance, yielding a total of 18 strategies. None were found to be cost‐effective.

3.3. Risk‐Stratified Reduction in Endoscopic Surveillance Frequency

3.3.1. Sex‐Based Endoscopic Surveillance Frequency (22 Strategies)

Among 22 sex‐based strategies, none were cost‐effective. The most favorable approach ([0 yearly female; 2 yearly male; LGD 12 monthly]) excluded surveillance for low‐risk females while maintaining 2 yearly intervals for males (ICER = US$46 613/QALY), 52.3% reduction in endoscopies, and missing 28.6% of advanced EAC.

3.3.2. Barrett's Segment Length–Based Endoscopic Surveillance Frequency (3‐cm Threshold—22 Strategies)

Four out of 22 endoscopic surveillance strategies were found to be cost‐effective. The most favorable [0 yearly SSBE (< 3 cm); 2 yearly LSBE; LGD 12 monthly] with ICER = US$22 437/QALY, achieving a 50.4% reduction in endoscopic procedures and missing 19.6% of advanced EAC. Of note, current Australian guidelines for surveillance [5 yearly SSBE (< 3 cm); 2 yearly LSBE; LGD 6 monthly] had ICER of US$51 307/QALY, missing 8.7% of advanced EAC.

3.3.3. Barrett's Segment Length–Based Endoscopic Surveillance Frequency (2‐cm Threshold—22 Strategies)

Nine out of 22 endoscopic surveillance strategies were found to be cost‐effective. [0 yearly SSBE; 2 yearly LSBE; LGD 12 monthly] was the most cost‐effective approach of all 85 strategies ICER = US$23 737/QALY, reducing endoscopic procedures by 23.4% compared with guideline surveillance and missing 16.4% of advanced EAC. The next most cost‐effective endoscopic surveillance strategy was [10 yearly SSBE (≤ 2 cm); 2 yearly LSBE; LGD 12 monthly] with ICER value of US$30 438/QALY, reducing 19.9% of endoscopies and missing 12.5% of advanced EAC.

4. Sensitivity Analyses

4.1. Cost‐Efficiency Frontier

Only three were undominated strategies were #1—Natural history/No surveillance; #64—[0 yearly SSBE (< 3 cm); 2 yearly LSBE; LGD 12 monthly]; and #42—[0 yearly SSBE (≤ 2 cm); 2 yearly LSBE; LGD 12 monthly]. Eleven strategies had ICER values below the WTP threshold but externally dominated (Figure 2).

FIGURE 2.

The cost‐efficiency frontier shows undominated strategies (green line and points) representing optimal trade‐offs between costs (AUD) and QALYs. Absolutely dominated (gray) and extended dominated (red) strategies are inefficient. Shapes differentiate risk stratification groups. The x axis represents QALYs, and the y axis shows costs. Numbered strategies are #1—Natural history; #42—[0 yearly SSBE (≤ 2 cm); 2 yearly LSBE; LGD 12 monthly]; #64—[0 yearly SSBE (< 3 cm); 2 yearly LSBE; LGD 12 monthly].

4.2. One‐Way and Probabilistic Sensitivity Analysis

Eleven variables were tested in one‐ and two‐way sensitivity analyses (20 intervals for each variable) for all strategies. The key drivers of the model were: Probability of natural progression of HGD to EAC, cost of an endoscopic examination, and cost of database maintenance.

Probabilistic sensitivity analysis (1000 Monte Carlo simulations) compared 84 strategies (18 indiscriminate and 66 risk stratified) to “no surveillance/natural history,” the strategy “[0 yearly SSBE (≤ 2cm); 2 yearly LSBE; LGD 12 monthly]” was the most cost‐effective strategy in 93.7% of the simulations (undominated). All other strategies seen to be cost‐effective in the base case analysis were either externally dominated or absolutely dominated in comparison to this strategy (Figure 2). When strategies that exclude the low‐risk subgroup were removed from the analysis, the strategy [10 yearly SSBE (≤ 2 cm); 2 yearly LSBE; LGD 12 monthly] was the most cost‐effective strategy in ~80% of the simulations with an ICER value of US$30 438/QALY (Figure 3).

FIGURE 3.

Probabilistic analysis. Cost‐effectiveness acceptability curve of 85 risk‐stratified and non–risk‐stratified strategies showing percent of simulations (y axis) each strategy was cost‐effective at willingness‐to‐pay thresholds between 0 and US$100 000/WTP (x axis). LGD—low‐grade dysplasia; LSBE—long‐segment Barrett's esophagus; SSBE—short‐segment Barrett's esophagus.

5. Discussion

Our study presents 85 strategies of which 66 were risk stratified and 19 were non–risk stratified. None of the non–risk‐stratified strategies nor the 22 sex‐stratified strategies were seen to be cost‐effective. In contrast, 13 out of the 44 length‐based strategies were found to be cost‐effective.

Strategies that generated more QALYs were more cost‐effective than strategies that reduced endoscopic examinations. Nondiscriminate reduction in surveillance led to undetected progression to advanced EAC in both high‐risk and low‐risk groups, reducing QALYs. A risk‐based approach mitigates the hazard of missing cancers by selectively investigating high‐risk subgroups. Length‐based strategies were able to select a small group (25%–38%) of high‐risk individuals for regular endoscopic surveillance and reduce surveillance in the large group of low‐risk (62%–75%) individuals. Identification of dysplasia and early cancer led to lower morbidity and mortality, which generated higher QALYs.

Within the length‐based risk‐stratified strategies, using the 2‐cm threshold was more effective. This was surprising, as the high‐risk group with BE ≥ 3 cm was predicted to show better discrimination. However, two key factors explained this result: (1) The 2‐cm threshold captured more “progressors to cancer” (37.5%) than the 3‐cm threshold, and (2) the smaller (62.5%) short‐segment group (< 2 cm) had fewer progressors, so reducing surveillance led to fewer advanced EAC. In the 3‐cm threshold model, 75% of the cohort was in the short‐segment group, meaning reduced surveillance increased the number of advanced cancers. By contrast, the 2‐cm threshold group made up 62.5% of the cohort, and reducing surveillance resulted in fewer symptomatic cancers. Additionally, individuals with less than 2‐cm BE have a lower risk of progression than 3 cm, making less frequent surveillance in the lowest risk group more cost‐effective.

5.1. Strengths

The main strength of this study is the development of the model through a comprehensive process. It started with a systematic review that identified the need for cost‐effective endoscopic surveillance options [9]. It also identified that risk stratification was necessary, whether it was through clinical features such as male sex, long‐segment BE [30], and biochemical features such as mutational load/genomic instability [37] or epithelial/stromal abnormalities [38]. Varying surveillance intervals was done through undetected health states, which was a more realistic simulation of the natural progression of BE. This was then calibrated to key model outputs and validated in a seven‐stage process [39, 40]. All transition probabilities of both nonstratified and risk‐stratified groups/subgroups were calibrated simultaneously to the same aggregate model outputs. As a result, all costs, QALYs, ICER values, and numerous other outputs were comparable between all presented 85 strategies as well as future strategies that may be of interest. This is often not possible in economic evaluations because differences in cohort characteristics mean only the intended comparator strategies can be assessed.

5.2. Limitations

The main limitation for most economic evaluations is its model inputs. One of the challenges of this study was selecting the correct inputs to simulate the natural progression of community BE. It is impossible to estimate the unobserved progression of a silent disease such as BE without active surveillance. As such, transition probabilities were compared with the literature to ensure they were within expected ranges. Our confidence in the accuracy of transition probabilities for progression or regression of BE states was based on two factors: comparison to previous economic evaluations and model calibration [30, 41, 42, 43, 44, 45]. We were reassured when our model outputs matched other BE and EAC models in the literature [42, 46]. A recent RCT from UK Barrett's Esophagus Surveillance Study looked at the cost‐effectiveness of 2 yearly endoscopic surveillance versus at‐need surveillance, finding the 2 yearly surveillance to be not cost‐effective. Although the costs and outcomes are difficult to compare directly because of differences in starting populations, time horizon, and other variables, key transition probabilities (NDBE to HGD, NDBE to EAC, and ablation success) were equivalent to our study. The limitations of other model inputs were alleviated by performing sensitivity analyses. Extended ranges (beyond realistic imagination) were tested with 11 of these variables, which helped understand how the model functioned and whether certain pathways were unrealistically monopolizing the model outputs.

Guidelines from gastroenterological societies have recommended surveillance intervals tailored to segment length [47, 48, 49, 50, 51]. Most guidelines, somewhat arbitrarily, consider > 3 cm as a threshold for long‐segment BE, which requires 2–3 yearly endoscopic surveillance, while the short‐segment group receives endoscopic surveillance every 4–5 years. They have self‐acknowledged this is based on weak evidence and these recommendations were not seen to be cost‐effective in our analysis. Our study suggests that this interval could be lengthened further indefinitely or to 10 years (Figure S2), which reduces costs significantly while detecting most EAC at an early stage. However, some will progress to advanced EAC under a lengthier surveillance interval, seen in Yang et al. examining cost‐effectiveness of patients with inadequate surveillance, concluding that longer surveillance intervals may increase the chance of developing cancer and death but most BE patients die of other causes rather than cancer‐related deaths [52]. While this may be seen as ethically contentious, we are already complicit in this with current guidelines which recommend [4–5 yearly SSBE (< 3 cm); 2–3 yearly LSBE; LGD 12 monthly] which missed 8%–27% of advanced EACs in this model. Furthermore, 80%–90% of EACs arise undetected outside of surveillance programs [53], and by choosing to serve a low‐risk surveillance population, we are condemning high‐risk individuals in the community to a higher morbidity and mortality.

6. Conclusion

Limiting endoscopic examinations to long‐segment BE is cost‐effective. However, if excluding short‐segment individuals is not ethically feasible, then reducing their endoscopic frequency to 10 years is still cost‐effective. Making endoscopic surveillance programs cost‐effective is the first step in introducing a screening program, which will ultimately improve the overall survival of EAC.

Funding

Dr. Vissapragada is sponsored by a Ph.D. grant from the Hospital Research Foundation. Dr. Aoki is supported by the APAGE/JGH Foundation Clinician–Scientist Training Fellowship. Dr. Bulamu is supported by a Cancer Council South Australia Beat Cancer Early Career Research Fellowship.

Disclosure

No generative AI was used in scientific writing.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Figure S1: High‐grade dysplasia treatment. Diagnosis of high‐grade dysplasia is confirmed with an additional endoscopy within 3 months. Confirmed high‐grade dysplasia undergoes 360° radiofrequency ablation and repeat endoscopy within 6 months. If further dysplasia is found, then ablation is performed again.

Table S1: One‐way sensitivity variables

Table S2: Cost and QALY rankings table for all 85 strategies. Incremental cost‐effectiveness ratio (ICER) values are calculated using costs and utilities from “Natural history.” Costs and QALY are accumulated over time horizon for entire cohort. Costs are in US$ (2023 value) per cohort of 1.

Table S3: Model outputs for all 85 strategies explaining number of cancers developed during the time horizon along with total percentage of missed cancers and percentage reduced endoscopies compared with Australian guideline surveillance [5 yearly SSBE (< 3 cm); 2 yearly LSBE; LGD 6 monthly]. Undominated strategies are highlighted in yellow whereas externally dominated are in green.

Figure S2: Cost‐effectiveness acceptability curve of risk‐stratified and non–risk‐stratified strategies that do NOT eliminate low‐risk subgroups from surveillance (73 total). y axis shows percent of simulations each strategy was cost‐effective at willingness‐to‐pay thresholds between 0 and US$100 000/WTP (x axis). LGD—low‐grade dysplasia; LSBE—long‐segment Barrett's esophagus; SSBE—short‐segment Barrett's esophagus.

Vissapragada, R. , Bulamu, Norma B. , Yazbeck, R. , Aoki, T. , Bright, T. , Watson, David I. , and Karnon, J. (2026) Evaluating Cost‐Effectiveness of 85 Endoscopic Surveillance Strategies of Nondysplastic Barrett's Esophagus. Journal of Gastroenterology and Hepatology, 41: 992–1000. 10.1111/jgh.70238.

Data Availability Statement

Data supporting findings are provided in the manuscript, the Supporting Information, and in previously published methodological articles by authors relevant to this study. Further study materials and data can be made available if necessary.