Endoscopic Screening Program for Control of Esophageal Adenocarcinoma in Varied Populations: A Comparative Cost-Effectiveness Analysis

Abstract

BACKGROUND & AIMS:

Guidelines suggest endoscopic screening for esophageal adenocarcinoma (EAC) among individuals with symptoms of gastroesophageal reflux disease (GERD) and additional risk factors. We aimed to determine at what age to perform screening and whether sex and race should influence the decision.

METHODS:

We conducted comparative cost-effectiveness analyses using 3 independent simulation models. For each combination of sex and race (White/Black, 100,000 individuals each), we considered 41 screening strategies, including one-time or repeated screening. The optimal strategy was that with the highest effectiveness and an incremental cost-effectiveness ratio <$100,000 per quality-adjusted life-year gained.

RESULTS:

Among White men, 536 EAC deaths were projected without screening, and screening individuals with GERD twice at ages 45 and 60 years was optimal. Screening the entire White male population once at age 55 years was optimal in 26% of probabilistic sensitivity analysis runs. Black men had fewer EAC deaths without screening (n = 84), and screening those with GERD once at age 55 years was optimal. Although White women had slightly more EAC deaths (n = 103) than Black men, the optimal strategy was no screening, although screening those with GERD once at age 55 years was optimal in 29% of probabilistic sensitivity analysis runs. Black women had a very low burden of EAC deaths (n = 29), and no screening was optimal, as benefits were very small and some strategies caused net harm.

CONCLUSIONS:

The optimal strategy for screening differs by race and sex. White men with GERD symptoms can potentially be screened more intensely than is recommended currently. Screening women is not cost-effective and may cause net harm for Black women.

Endoscopic screening for esophageal adenocarcinoma (EAC) has been suggested in multiple US guidelines, along with subsequent surveillance of Barrett’s esophagus (BE) and endoscopic treatment of dysplasia.^1–5 However, the recommendations vary regarding which population should be screened. For instance, the American Gastroenterological Association suggests screening for those with multiple risk factors, such as age older than 50 years, male sex, White race, and chronic symptoms of gastroesophageal reflux disease (GERD), and recommends against screening the general population of individuals either with or without GERD. The American College of Gastroenterology recommends to consider screening men with GERD and at least 2 other risk factors, and recommends against screening in women or the general population without GERD. No randomized trials of endoscopic screening with outcomes of cancer mortality have been published, and the guidelines regarding whether to screen, at what age, and in which population are based largely on observational data and expert opinions. Randomized trials of population screening for EAC may be impractical due to cost and duration of follow-up required, particularly for comparing a large number of screening strategies. Cost-effectiveness analysis can avoid those practical limitations by mathematically comparing outcomes of competing strategies. Multiple cost-effectiveness analyses of screening for EAC have been published,^6,7 but nearly all have focused on screening older men with GERD at a specific age, and have either been limited to White race or have not specified race. To our knowledge, only 2 prior published cost-effectiveness analyses have considered endoscopic screening of the general population, irrespective of GERD symptoms or sex, and those did not provide results stratified by population.^8,9

Furthermore, published guidelines do not endorse repeat screening for BE in individuals with a prior normal endoscopy. This is due to the very low yield of BE on repeat endoscopy reported in prior studies; however, those studies are limited to relatively short follow-up of 5 years or less.^10,11 Short-term studies mostly reflect the low false negative rate of endoscopy for BE rather than any incidence of new BE. In contrast, cross-sectional observational studies demonstrate that the prevalence of BE increases with age, suggesting that there are incident cases of new BE as individuals get older, particularly at prolonged intervals (eg, 10–15 years); therefore, repeated screening might be cost-effective in certain populations.^12–16 However, we are not aware of any prior cost-effectiveness analyses that consider repeat screening.

Therefore, we aimed to fill these knowledge gaps by determining the cost-effectiveness of endoscopic screening for EAC both restricted to those with GERD symptoms and in the general population, stratified by race and sex, using a comparative modeling approach. We also aimed to identify the optimal age at which to offer screening in each population, and determine whether repeat screening at a later age among those with normal esophagus on their first screening is cost-effective.

Methods

Comparative modeling improves the credibility of simulation modeling results as the rigorous process reduces many of the concerns regarding potential biases in individual models. We used 3 independently developed simulation models of EAC natural history, screening, surveillance, and treatment. These models are part of the National Cancer Institute Cancer Intervention and Surveillance Modeling Network and provide independent results for comparison. Primary outcomes were the averaged results across the 3 models.

Cancer Intervention and Surveillance Modeling Network–Esophageal Adenocarcinoma Models

We used the following models: Esophageal Adenocarcinoma Model (EACMo) from the Columbia University Medical Center; Microsimulation Screening Analysis Model for Esophageal Adenocarcinoma (MISCAN-EAC) from the Erasmus University Medical Center Rotterdam and the University of Utah; and Multi-Stage Clonal Expansion for Esophageal Adenocarcinoma (MSCE-EAC) model from the Fred Hutchinson Cancer Research Center. Each model created 4 cohorts of sex and race (White and Black) independently calibrated to common targets to reproduce the incidence of EAC in the United States from the Surveillance, Epidemiology, and End Results (SEER) cancer registry data until 2014.¹⁷

Each model differs in their structure and in some assumptions. MISCAN-EAC and EACMo use recursive health state transitions in a semi-Markov and Markov model, respectively. In contrast, MSCE-EAC is a biologically based stochastic model at the cellular level, ultimately leading to clinical disease in the population of individuals. In all 3 models, it was assumed that EAC develops only in patients with BE.¹⁸ The MISCAN-EAC and EACMo models assume an overall stable prevalence of symptomatic GERD across ages of approximately 20%.¹⁹ The MSCE EAC model assumes a near-linear increase in symptomatic GERD prevalence through adulthood on the basis of integrating published GERD incidence data with a prevalence reaching nearly 32% by age 85 years.^20,21 Consistent with the definitions used in those studies, GERD was defined as heartburn or regurgitation occurring at least once weekly. In all models, individuals with or without symptomatic GERD may develop nondysplastic BE, which can progress to low-grade and then high-grade dysplasia. Individuals may transition to BE and dysplasia at rates that were calibrated independently in each model for each stratum of sex and race in order to meet cancer incidence targets from SEER. EACMo and MSCE-EAC assumed a relative risk of GERD for BE of 5, and MISCAN-EAC assumed a relative risk of 6.²² MISCAN-EAC and MSCE-EAC allow for regression of dysplasia, and EACMo does not. Patients with BE with high-grade dysplasia may develop preclinical EAC, which can then progress to clinical EAC as symptoms develop. Individuals with clinical EAC may die of the disease, with probabilities dependent on age and stage. More details on the structure and parameterizations of the models have been published and are available online.^23–25

Simulated Population and Intervention

For the base case, we simulated 4 cohorts of US individuals born in 1950 stratified by sex and race (White or Black), and followed them from age 40 years until age 100 years. For each cohort, we considered 42 strategies, including no screening, one-time endoscopic screening among the entire cohort at 6 specified ages ranging from 40 to 65 years, one-time endoscopic screening among individuals with GERD symptoms at the 6 specified ages, and strategies of repeated screening in individuals with GERD at intervals of 10, 15, or 20 years starting at 1 of the 6 specified ages and stopping at ages ranging from 55 to 80 years (Supplementary Table 1 and Supplementary Figure 1).

Based on our previously published results from these 3 models identifying optimal surveillance and endoscopic treatment strategies,²⁶ endoscopic surveillance of nondysplastic BE was performed every 3 years until age 80 years. Patients who were diagnosed with low-grade dysplasia received a repeat endoscopy with biopsies after 2 months of treatment with a high-dose proton pump inhibitor to confirm low-grade dysplasia.^27,28 Patients with high-grade dysplasia or confirmed low-grade dysplasia underwent endoscopic eradication therapy (EET), followed by surveillance until death. In case of recurrence, patients received repeat EET followed by surveillance. Patients with treatment failure or recurrences more than 3 times did not receive additional EET and underwent surveillance until death. Specific surveillance intervals before and after EET are detailed in Supplementary Table 2. Supplementary Figure 2 provides an example timeline of an individual undergoing screening, surveillance, and EET. The sensitivity and specificity of endoscopy with biopsy for BE was assumed to be 87.5% and 92.5%, respectively.²⁹ Additional assumptions regarding the accuracy of diagnoses, success with treatment, and complications are presented in detail in Supplementary Tables 3 and 4.

Costs and Utilities

The costs of endoscopies and EETs were estimated using the 2020 reimbursement rates from the Centers for Medicare and Medicaid Services.^30–32 We assumed that 50% of screening and surveillance endoscopies were performed with endoscopist-directed sedation and 50% were performed with anesthesia assistance,³³ each lasting 22 minutes, and all were performed in ambulatory surgery centers. We assumed that all EET and dilations were performed in hospital outpatient departments using anesthesia assistance (60 minutes for endoscopic mucosal resection, 45 minutes for radiofrequency ablation, and 30 minutes for dilation). The costs and utilities of cancer care by stage at diagnosis and those of complications due to endoscopy and EET were estimated using data in the published literature (Supplementary Table 4).^34–40 All costs and utilities were discounted at an annual rate of 3%.

Outcomes and Analysis

In each model, we computed the health outcomes of each strategy, including the incidence of clinical (ie, symptomatic) and surveillance-detected EAC, EAC mortality, complications of endoscopies and treatments, and life-years and quality-adjusted life-years (QALYs) per 100,000 individuals. In addition, we calculated the number of endoscopies, EETs, radiofrequency ablation touch-ups, complications, and cancer care years to estimate the total costs of surveillance and treatment per strategy. The outcomes were analyzed for each model separately, then the average results of the 3 models were presented per strategy as the base case. The optimal strategy for each cohort was identified through cost-effectiveness analysis from a third-party payer perspective using the average results of the 3 models with an assumed willingness-to-pay threshold (WTP) of $100,000/QALY gained. The optimal strategy was defined as the one providing the greatest QALYs at an incremental cost-effectiveness ratio (ICER) compared with the next most effective strategy that does not exceed the WTP threshold.

Sensitivity Analysis

We considered the separate results of each model as a sensitivity analysis. In addition, we performed a one-way sensitivity analysis in each model on the effectiveness of EET (Supplementary Table 5). Finally, we performed probabilistic sensitivity analyses, simultaneously varying each cost and utility parameter, as well as probabilities of complications from endoscopy and EET, with specified distributions (Supplementary Table 6) for 1000 runs in each model, with results averaged across models. We considered WTP thresholds from $50,000 to $250,000 per QALY gained. When possible, parameter distributions were estimated according to published literature, otherwise, expert opinion was employed.^41–43

Results

Without endoscopic screening, the 3 models were each well-calibrated to observed SEER age-specific EAC incidence by sex and race (Supplementary Figure 3). All models predicted similar sex- and race-specific incidence of EAC among those without GERD symptoms (Supplementary Figure 4), but MSCE-EAC predicted lower incidence among those with GERD symptoms than EACMo and MISCAN-EAC. All models predicted rising prevalence of BE with age, particularly among those with GERD symptoms, but MISCAN-EAC predicted a plateauing of BE prevalence after age 60 years (Supplementary Figure 5).

White Men

Without endoscopic screening, the averaged results of the 3 models predicted a lifetime incidence of 654 cases of EAC and 536 EAC deaths with a total cost of $20.5 million for the care of incident EAC cases per 100,000 White men. Screening prevented 10%–33% of EAC cases and 13%–53% of EAC deaths, at an increase in costs of $12.3 to $94.4 million, depending on the screening strategy (Table 1 and Supplementary Table 7).

Table 1.

Outcomes of Screening Strategies on Efficiency Frontier

Variable	Clinical EAC, n	Screen-detected EAC, n	Total EAC, n	% EAC prevented	EAC deaths, n	% EAC deaths prevented	EGDs, n	Initial EET, n	Touch-up EET, n	Cost, $	Life-years	QALYs	ICER, $/QALY
White men
NH	654	0	654	0.0	536	0.0	0	0	0	20,457,639	2,198,517	2,198,365	Reference
1GERD55a	462	77	540	17.5	409	23.6	36,010	593	376	36,561,587	2,199,071	2,198,917	29,185
15yGERD45–60b	419	83	502	23.3	374	30.1	60,262	781	530	50,073,982	2,199,306	2,199,144	59,363
1ALL45c	267	191	459	29.9	289	46.0	136,427	922	586	104,143,978	2,199,686	2,199,470	166,017
Black men
NH	105	0	105	0.0	84	0.0	0	0	0	3,239,940	1,943,313	1,943,288	Reference
1GERD55	75	11	86	17.4	64	23.6	19,867	109	72	12,816,031	1,943,433	1,943,400	85,418
15yGERD40–55d	72	12	83	20.3	62	26.6	37,456	120	91	26,280,546	1,943,496	1,943,450	270,556
White women
NH	125	0	125	0.0	103	0.0	0	0	0	3,406,619	2,361,638	2,361,612	Reference
1GERD55	88	16	104	17.1	78	24.3	24,018	157	111	15,033,836	2,361,752	2,361,715	112,311
15yGERD45–75e	77	18	96	23.5	70	31.8	62,437	232	147	31,689,611	2,361,799	2,361,745	557,979
1ALL50	37	44	81	35.0	45	56.1	108,820	296	193	64,712,882	2,361,859	2,361,771	1,296,411
Black women
NH	35	0	35	0.0	29	0.0	0	0	0	924,042	2,215,983	2,215,976	Reference
1GERD65f	26	7	34	4.9	23	21.4	19,906	71	31	8,358,918	2,216,042	2,216,027	143,781
15yGERD45–60b	24	6	30	14.6	21	25.5	38,841	78	60	21,827,244	2,216,079	2,216,050	596,950

NOTE. All results are per 100,000 individuals. Initial EET incudes 3.55 radiofrequency ablation sessions and 0.55 endoscopic resection sessions. Touch-up EET is 1 radiofrequency ablation session. Costs, life-years, and QALYs are discounted. ICERs are compared with the next most effective strategy.

EGD, esophagogastroduodenoscopy; NH, natural history.

^aScreening among those with GERD symptoms at age 55 years.

^bScreening among those with GERD symptoms at ages 45 and 60.

^cScreening entire population at age 45 years.

^dScreening among those with GERD symptoms at ages 40 and 55.

^eScreening among those with GERD symptoms at ages 45, 60, and 75.

^fScreening among those with GERD symptoms at age 65 years.

Averaged over all 3 models, the strategies that entailed screening only once among those with GERD (Supplementary Figure 6) were less costly than most other strategies, but also less effective than most other strategies. The most effective, but also the most costly, strategies were those that screened the entire White male population, including those with and without GERD symptoms, at a single age between 40 and 55 years. The cost-effectiveness of screening strategies tended to cluster around the same starting age, with strategies of screening at younger ages being both more effective and more costly. A few strategies involving repeated screening of White men with GERD symptoms were efficient or nearly efficient, and the optimal strategy overall was screening twice among those with GERD symptoms at ages 45 and 60 years with an ICER of $59,363 per QALY gained compared with the next most cost-effective strategy (screening individuals with GERD only once at age 55 years, Figure 1A). This strategy of repeated screening prevented 23.3% of EAC cases and 30.1% of EAC deaths, and required 60,262 endoscopies among the 100,000 White men (Table 1).

Figure 1.

Cost-effectiveness plane of screening. (A) White men, (B) Black men, (C) White women, and (D) Black women. Results are averaged across the 3 models. Black lines connect the strategies that lie on the efficiency frontier. The slope of the line represents the inverse of the ICER. Only those strategies on the efficiency frontier are displayed. Figures with the full set of strategies are displayed in Supplementary Figures 6, 7, 8, and 10.

Screening the entire population of White men once at age 45 years was also on the efficiency frontier, but at an ICER that was more expensive than the WTP threshold compared with the optimal strategy ($166,017 per QALY gained). This strategy prevented 29.9% of EAC cases and 46.0% of EAC deaths, but required 136,427 endoscopies among the 100,000 White men.

There was some variation between the 3 individual models regarding the optimal screening strategy for White men (Table 2, Supplementary Figure 6). The MISCAN-EAC model identified repeated screening among men with GERD symptoms at ages 50 and 60 years as the optimal strategy and MSCE-EAC identified screening all men regardless of GERD symptoms at age 45 years as the optimal strategy. In all 3 models, however, the overall optimal strategy of screening White men with GERD twice at ages 45 and 60 years was very close to the efficiency frontier. The optimal strategy identified by MISCAN-EAC—screening White men with GERD twice at ages 50 and 60 years—was also close to the efficiency frontier in the other 2 models. In 1-way sensitivity analyses, the overall optimal strategy remained screening twice among White men with GERD symptoms at ages 45 and 60 years regardless of whether EET was very effective or very ineffective (Table 2).

Table 2.

Optimal Screening Strategies and in Sensitivity Analyses

Variable	White men	Black men	White women	Black women
Overall	15yGERD45–60a	1GERD55b	NH (no screening)	NH (no screening)
Sensitivity analyses: individual models
EACMo	15yGERD45–60	NH (no screening)	NH (no screening)	NH (no screening)
MISCAN-EAC	10yGERD50–60c	NH (no screening)	NH (no screening)	NH (no screening)
MSCE-EAC	1ALL45d	1GERD45e	1GERD40f	1GERD65g
1-way sensitivity analysis: effectiveness of EET
Low effectiveness	15yGERD45–60	1GERD60h	NH (no screening)	NH (no screening)
High effectiveness	15yGERD45–60	1GERD55i	NH (no screening)	NH (no screening)

NH, natural history.

^aScreening among those with GERD symptoms at ages 45 and 60.

^bScreening among those with GERD symptoms at age 55 years.

^cScreening among those with GERD symptoms at ages 50 and 60.

^dScreening entire population at age 45 years.

^eScreening among those with GERD symptoms at age 45.

^fScreening among those with GERD symptoms at age 40.

^gScreening among those with GERD symptoms at age 65.

^hScreening among those with GERD symptoms at age 60.

ⁱScreening among those with GERD symptoms at age 55.

In the probabilistic sensitivity analyses at a WTP threshold of $100,000 per QALY, the strategy of screening White men with GERD symptoms at ages 45 and 60 years was optimal in 100% of runs (Figure 2A).

Figure 2.

Probabilistic sensitivity analyses. (A) White men, (B) Black men, (C) White women, and (D) Black women. Results for each demographic are averaged over the 3 models. The horizontal axis displays the WTP threshold, and the vertical axis displays the proportion of model runs in which a strategy is optimal at that WTP (only 4 strategies were ever optimal under any threshold among the different demographics). The dashed line represents the base-case WTP threshold of $100,000 per QALY. NH, natural history. 1GERD55, screening among those with GERD symptoms at age 55; 15yGERD45_60, screening among those with GERD symptoms at ages 45 and 60; 1ALL45, screening entire population at age 45; 1GERD45, screening among those with GERD symptoms at age 45; 15yGERD40_50, screening among those with GERD symptoms at ages 40 and 55.

We compared screening strategies to understand the differing mechanisms of failure to prevent more EAC deaths (Supplementary Results and Supplementary Table 8), finding that, depending on the strategy, the primary cause was either lack of screening or EAC arising from BE that developed after screening, with failure of EET or surveillance accounting for only small proportions of failures.

Black Men

Without endoscopic screening, the average results of the 3 models predicted many fewer EAC cases and EAC deaths per 100,000 Black men than among White men (Table 1 and Supplementary Table 9). However, on a relative scale, screening Black men still prevented 8%–32% of EAC cases and 11%–50% of EAC deaths (Supplementary Table 9), similar to the proportions among White men.

Averaged over all 3 models, similar patterns regarding clustering and ordering of strategies were present as in White men (Supplementary Figure 7). However, due to the overall lower burden of EAC among Black men, the optimal strategy in Black men was once-only screening among those with GERD symptoms at age 55 years rather than repeated screening (Figure 1B). A strategy of repeat screening among those with GERD symptoms at ages 40 and 55 years was also on the efficiency frontier, but was much more expensive than the WTP ($270,556 per QALY, Table 1). No strategy of screening the entire population of Black men was on the efficiency frontier.

Once again, there was some variation across the 3 individual models regarding the optimal screening strategy in Black men (Table 2, Supplementary Figure 7). In MSCE-EAC, the optimal strategy was screening Black men with GERD symptoms once at age 45 years. The optimal strategy in both EACMo and MISCAN-EAC was no screening, but a strategy of screening Black men with GERD symptoms at age 60 years was borderline cost-effective in both models ($104,459 and $108,084 per QALY, respectively). The overall optimal strategy of screening Black men with GERD symptoms once at age 55 years was on the efficiency frontier only in the MISCAN-EAC model, but very close to the frontier in the other 2 models.

In 1-way sensitivity analyses, the overall optimal strategy changed to screening once among Black men at age 60 years rather than age 55 years if EET was at the lower bounds of effectiveness (Table 2). In the probabilistic sensitivity analysis, screening Black men with GERD symptoms once at age 55 years remained the optimal strategy overall in 99.6% of runs at a WTP of $100,000 per QALY (Figure 2B).

White Women

Without endoscopic screening, the averaged result of the 3 models predicted slightly more EAC cases and EAC deaths per 100,000 White women than among Black men (Table 1, Supplementary Table 10), and screening prevented similar proportions of EAC cases and EAC deaths as in White men and Black men.

Averaged over all 3 models, there was no screening strategy among White women that was cost-effective at the WTP of $100,000 per QALY (Figure 1C, Table 1). A strategy of once-only screening among White women with GERD symptoms at age 55 years was the most cost-effective screening strategy, but was slightly more expensive than the WTP threshold ($112,311 per QALY). A strategy of repeated screening among White women with GERD symptoms at ages 45, 60, and 75 years and a strategy of screening the entire population at age 50 years were both on the efficiency frontier, but were very expensive. The MSCE-EAC model identified one-time screening of White women with GERD symptoms at age 40 years as the optimal strategy. The EACMo and MISCAN-EAC models found no screening strategy to be cost-effective at the base-case WTP threshold, but strategies of screening White women with GERD symptoms at ages 50 and 55 years, respectively, were on the efficiency frontier (Table 2, Supplementary Figure 8). In the 1-way sensitivity analysis of effectiveness of EET, the optimal strategy remained to not screen any White women, even with highly effective EET (Table 2).

In the probabilistic sensitivity analysis, at a WTP of $100,000 per QALY, the strategy of not screening White women was optimal in 70.8% of runs overall, but a strategy of screening White women with GERD symptoms at age 55 years was optimal in 29.2% of runs (Figure 2C).

Analyses explaining why screening White women was not cost-effective, although they have a slightly greater burden of EAC than Black men and screening was found to be cost-effective among Black men, are presented in the Supplementary Results and Supplementary Figure 9.

Black Women

Without endoscopic screening, the averaged result of the 3 models predicted many fewer EAC cases and EAC deaths per 100,000 Black women than in other demographic groups (Table 1, Supplementary Table 11). Although screening prevented up to 27.9% of EAC cases and 56.9% of EAC deaths, the maximum absolute numbers of cases and deaths prevented were only 9 and 17, respectively.

Averaged over all 3 models, there was no cost-effective screening strategy among Black women (Figure 1D, Table 1). A strategy of once-only screening among Black women with GERD symptoms at age 65 years was the most cost-effective screening strategy, but was considerably more expensive than the WTP threshold ($143,781 per QALY). A strategy of repeated screening among Black women with GERD symptoms at ages 45 and 60 years was also on the efficiency frontier, but very expensive. Screening strategies in Black women led to smaller increases in life-years compared with other populations, and 10 screening strategies actually led to fewer QALYs than no screening, due to the disutility related to endoscopy, EET, and iatrogenic complications.

The MSCE-EAC model identified one-time screening of Black women with GERD symptoms at age 65 years as the optimal strategy. The EACMo and MISCAN-EAC models found no screening strategy to be cost-effective at the base-case WTP threshold (Table 2, Supplementary Figure 10). In the 1-way sensitivity analysis of effectiveness of EET, the optimal strategy remained to not screen any Black women, even with highly effective EET. In the probabilistic sensitivity analysis, the optimal strategy in Black women was no screening in 100% of runs at a WTP threshold of $100,000 per QALY.

Discussion

We conducted extensive comparative analyses assessing the cost-effectiveness of endoscopic screening to reduce mortality from esophageal adenocarcinoma. To our knowledge, this is the first such study assessing the cost-effectiveness of screening in specific demographics by age, sex, and race. We also considered screening the general population irrespective of GERD symptoms and, for the first time, considered repeated screening after a normal index screening. Overall, we found that a strategy of repeat screening at ages 45 and 60 years among White men with GERD symptoms was optimal, and we demonstrated for the first time the cost-effectiveness of screening among Black men, finding that screening once at age 55 years among Black men with GERD symptoms was optimal. In contrast, we found that screening women (whether White or Black, including restricted to those with GERD symptoms) was not optimal at any age. In particular, among Black women, screening is, at best, very expensive with little benefit, and some strategies cause net harm. Even among White men with GERD, the majority of individuals undergoing screening and, to a lesser extent, those undergoing EET, would not benefit from the efforts because they would not be destined to die from EAC. However, just as in other types of cancer screening, we found that the benefits accrued to the few who are otherwise destined to die from EAC appear to be worth the cost and effort of screening and EET.

Our finding that repeated screening in White men with GERD symptoms is cost-effective is due to de novo incident cases of BE developing after the index endoscopy. Repeat screening of 20% of the population would prevent an additional 11% of EAC deaths (that arise from incident BE) compared with one-time-only screening of White men with GERD symptoms, whereas screening 100% of the population once would prevent an additional 21% of EAC deaths compared with one-time screening of only those with GERD symptoms (ie, twice the effectiveness, but requiring 4 times the number of additional screening endoscopies) (Supplementary Table 8). Such incident cases of BE were inferred to exist in the natural history of all 3 independent models with varying magnitudes directly from calibrating to the observed EAC incidence in SEER across multiple birth cohorts. They are also supported from cross-sectional endoscopy studies demonstrating an increasing yield for BE with increasing age.^12–16 Prior longitudinal studies searching for new BE after a normal index endoscopy have found a scant number of such incident cases, and are not able to distinguish between missed BE on the index endoscopy; however, those studies have been limited to less than 5 years of follow-up.^10,11 Further empiric studies are needed to confirm the existence of incident BE at prolonged intervals (eg, 15 years) before implementation of the strategy of repeated screening identified as optimal in this cost-effectiveness analysis.

We found that screening the entire White male population once at age 45 years was more effective than screening only among those with GERD symptoms twice, but was more expensive and not cost-effective (due to an ICER greater than the WTP) in the majority of runs in the probabilistic sensitivity analysis. In 1 of the 3 models (MSCEEAC), a strategy of screening the entire White male population once was the optimal strategy demonstrating that strategies of screening White men without GERD symptoms should be considered. If empiric studies do not confirm the existence of substantial incidence of BE at prolonged intervals in middle age, then strategies with repeated screening should be omitted from consideration, and the overall optimal screening strategy among White men would be one-time screening among those with GERD symptoms at age 55 years because screening the entire population at age 45 years costs an incremental $121,679 per QALY gained. Future studies should examine the cost-effectiveness of strategies using less costly nonendoscopic modalities for screening the general population, and also incorporating other risk factors, such as obesity and tobacco use, into assessing the cost-effectiveness of screening populations without GERD symptoms.

Notably, we found that screening women was not cost-effective at any age, regardless of race or the presence of GERD symptoms. However, among White women with GERD symptoms, screening once at age 55 years was close to the WTP threshold, and costs <$100,000 per QALY in 29% of runs in the probabilistic sensitivity analysis. Although we did not explicitly model other risk factors here, it is certainly conceivable that screening White women with GERD symptoms would be cost-effective if they had additional risk factors (eg, tobacco use, obesity, and family history) and/or if screening were to be performed nonendoscopically. Future studies should determine which combinations of risk factors would make which screening modalities cost-effective among White women.

In contrast to screening in White individuals or men, screening Black women, even among those with GERD symptoms, was very expensive and some strategies caused net harm. Black women considering screening should be informed that screening would prevent very few cancer deaths and could lead to net harm from rare iatrogenic complications or due to the decrement in quality of life associated with screening, surveillance, and EET.

We certainly acknowledge the history of health care inequities, and that race is a social construct that, in the vast majority of medical contexts, has no biological basis. We are circumspect regarding making recommendations based on race or sex if environmental exposures or genetic factors on which to make to those recommendations were available.⁴⁴ First, our models are calibrated primarily to cancer registry outcomes rather than surrogate outcomes (such as BE diagnosis), which could be influenced by access to care. Studies have demonstrated that Black individuals have a lower risk of EAC than White individuals and women have a lower risk than men, including when adjusting for known risk factors of obesity and smoking, as well as for other potential confounding environmental exposures, such as alcohol and income.^45,46 However, it is certainly conceivable that other environmental exposures could be found in the future that explain the association of race and sex with EAC. Some evidence indicates that differences in circulating sex hormones may mediate the effect of sex on risk of EAC.^47,48 A prior study identified a potential genetic basis for the protective association with race in a deletion and promoter duplication in GSTT2 resulting in decreased expression in White compared with Black individuals. But that study was accomplished by recognizing that Black individuals with BE have GSTT2 expression that is more similar to White individuals than Black individuals without BE, which itself demonstrates the existing genetic heterogeneity among Black individuals and the limits of using race as a proxy for genetics.⁴⁹ Nonetheless, GSTT2 may not be the only genetic basis for the effect of race on EAC risk, differences in circulating sex hormones may not explain all of the effects of sex on EAC, and testing for GSTT2 and circulating sex hormones has not been studied or validated as a screening tool. We believe that until more accurate predictor variables based on environmental exposure or genetics are available, tailoring screening based on race and sex is suitable, and the results of this study should be used in the context of shared decision making for individualized care decisions.

Our study had important limitations. Most importantly, our results indicate that repeated screening among White men with GERD symptoms is optimal, but the data to support such a strategy are largely from cross-sectional studies. Presumably, those individuals with GERD symptoms and a normal index endoscopy would likely be prescribed proton pump inhibitors, which could conceivably decrease the risk of developing BE later. Empiric longitudinal studies are needed to confirm whether repeated screening has a substantial yield of incident BE. In addition, although our study is the first to consider such a wide variety of ages and endoscopic screening strategies across different demographics, we did not incorporate additional risk factors, such as obesity, smoking, and family history, or nonendoscopic screening strategies. Future cost-effectiveness analyses should do so. Finally, our study was calibrated to cancer incidence among US residents and used US costs, and is not necessarily generalizable to other countries.

Our comparative modeling approach brings important strengths. Each model was independently designed with differing structures and then calibrated to the same set of observed cancers across multiple birth cohorts obtained from SEER. The specific strategy identified as optimal in each model differed. For instance, among White men, only the MSCE-EAC models identified a strategy of screening the entire population as optimal. Although both the EACMo and MISCAN-EAC models identified such a strategy on the efficiency frontier, they were not found to be optimal because both models have a greater incidence of BE at later ages in order to meet the same calibration target of EAC incidence found in SEER as with MSCE-EAC. The result is that EACMo and MISCAN-EAC found that screening only once misses individuals who later develop BE and then progress to EAC. Such differences in structure between models effectively serve as an additional layer of sensitivity analyses. However, with so many strategies considered that often differed in small degree by age of screening, the differences in incremental costs and QALYs for each strategy across models were often very small. So although a strategy may not have been on the efficiency frontier in one model, it might have still been very close to the frontier. By averaging the results across models, the strategies that had the smallest differences in outcomes between models tended to be highlighted. Indeed, the strategy identified as optimal overall for Black men was not identified as optimal in any one model, and the overall optimal strategies in White men and White women were identified as optimal in only 1 model each, but the overall optimal strategies were close to or on the efficiency frontier in each model. Strategies that might be favored by one model but are far from the efficiency frontier in another model are unlikely to be on the efficiency frontier of the averaged overall results. Thus, comparative modeling can provide reassurance regarding the robustness of the overall results. In addition, the results were robust to 1-way sensitivity analysis of the effectiveness of EET, and were also very robust to the probabilistic sensitivity analyses, with only 1 or 2 strategies identified as potentially cost-effective at the WTP threshold.

In conclusion, we conducted comparative cost-effectiveness analyses of a spectrum of possibly viable endoscopic screening strategies to reduce mortality from EAC in various populations, finding that the optimal strategy differs by sex and race due to differing risks of cancer and competing causes of death. We found that the optimal strategy is screening White men with GERD symptoms twice at ages 45 and 60 years, screening Black men with GERD symptoms once at age 55 years, and not performing screening in women. The specific strategy of screening might be different if other risk factors are included, and whether the phenomenon of incident BE after a normal endoscopy is confirmed. Future studies should assess the cost-effectiveness of screening strategies that incorporate additional risk factors and nonendoscopic screening modalities, as well as assessing the implementation of screening, including strategies stratified by race and sex.

WHAT YOU NEED TO KNOW.

BACKGROUND AND CONTEXT:

Guidelines suggest endoscopic screening for esophageal adenocarcinoma (EAC) among individuals with symptoms of gastroesophageal reflux disease (GERD) and additional risk factors.

NEW FINDINGS

In these comparative cost-effectiveness analyses, the optimal screening strategy was twice at ages 45 and 60 years for White men with GERD and once for Black men with GERD at age 55 years. The optimal strategy for women was no screening.

LIMITATIONS

There are no long-term longitudinal empiric studies to demonstrate the yield of repeat screening in White men with GERD.

IMPACT

White men can potentially be screened more intensely for EAC than currently recommended. Screening women is not cost-effective.

Abbreviations used in this paper:

BE

Barrett’s esophagus

EAC

esophageal adenocarcinoma

EACMo

Esophageal Adenocarcinoma Model

EET

endoscopic eradication therapy

GERD

gastroesophageal reflux disease

ICER

incremental cost-effectiveness ratio

MISCAN-EAC

Microsimulation Screening Analysis Model for Esophageal Adenocarcinoma

MSCE-EAC

Multi-Stage Clonal Expansion for Esophageal Adenocarcinoma Model

QALY

quality-adjusted life-year

SEER

Surveillance, Epidemiology, and End Results cancer registry

WTP

willingness-to-pay