Toward More Efficient Surveillance of Barrett’s Esophagus: Identification and Exclusion of Patients at Low Risk of Cancer

Abstract

Background

Endoscopic surveillance of Barrett’s esophagus (BE) is probably not cost-effective. A sub-population with BE at increased risk of high-grade dysplasia (HGD) or esophageal adenocarcinoma (EAC) who could be targeted for cost-effective surveillance was sought.

Methods

The outcome for BE surveillance from 2003 to 2012 in a structured program was reviewed. Incidence rates and incidence rate ratios for developing HGD or EAC were calculated. Risk stratification identified individuals who could be considered for exclusion from surveillance. A health-state transition Markov cohort model evaluated the cost-effectiveness of focusing on higher-risk individuals.

Results

During 2067 person-years of follow-up of 640 patients, 17 individuals progressed to HGD or EAC (annual IR 0.8%). Individuals with columnar-lined esophagus (CLE) ≥2 cm had an annual IR of 1.2% and >8-fold increased relative risk of HGD or EAC, compared to CLE <2 cm [IR—0.14% (IRR 8.6, 95% CIs 4.5–12.8)]. Limiting the surveillance cohort after the first endoscopy to individuals with CLE ≥2 cm, or dysplasia, followed by a further restriction after the second endoscopy—exclusion of patients without intestinal metaplasia—removed 296 (46%) patients, and 767 (37%) person-years from surveillance. Limiting surveillance to the remaining individuals reduced the incremental cost-effectiveness ratio from US$60,858 to US$33,807 per quality-adjusted life year (QALY). Further restrictions were tested but failed to improve cost-effectiveness.

Conclusions

Based on stratification of risk, the number of patients requiring surveillance can be reduced by at least a third. At a willingness-to-pay threshold of US$50,000 per QALY, surveillance of higher-risk individuals becomes cost-effective.

Appendix: Cost-effectiveness analysis

In essence, a hypothetical high-risk cohort member with a starting age of 50 years old was modeled until age 80 or death (whichever came first). A total of 10 health states were considered in the model, including BE free, non-dysplastic BE, LGD, HGD, EAC (consisted of T1, T2, T3, T4 stages and distant metastases), and all-cause death (Appendix Fig. 6). Among the initial cohort members, 95% had a confirmed diagnosis of non-dysplastic BE, 4% had LGD, and 1% had HGD. The Markov cycle length (i.e., duration between health-state transitions) was 6 months. Cohort members who developed HGD or EAC were subjected to a treatment pathway (determined by cancer T stage for EAC) as described in detail in Gordon et al. [13]. Transition probabilities, cost (measured from an Australian health system perspective), and utility weights were obtained from both South Australian data (whenever possible) and the published literature (Appendix Table 5). In particular, progression rates for the high-risk individuals undergoing surveillance were estimated using a Markov Model based on data from our BE surveillance program [13]. The Markov model was initially developed in Microsoft Excel, and the surveillance progression rates were derived iteratively from reverse-model runs, starting with estimates from the surveillance data. The derived rates accurately reproduced the incidences and cumulative incidences of LGD and HGD and esophageal adenocarcinoma and gastroesophageal junction carcinoma observed in the surveillance program. The rates and cumulative incidences were then verified in the TreeAge model.

Fig. 6

Multistate Markov model, NDBE non-dysplastic Barrett’s esophagus, LGD low-grade dysplasia, HGD high-grade dysplasia, T tumor, DM distant metastases

Table 5 Model parameters

For the comparison arm of high-risk individuals without surveillance, the corresponding progression rates were estimated in the Markov model using incidences calculated from average non-surveillance transition rates sourced from the literature (see Appendix Table 5), with the assumption that this population group (≥2 cm NDBE) had the same incidences of HGD and EAC as people in the population that have non-dysplastic BE. The endoscopic surveillance intervals were based upon current UK British Society of Gastroenterology Guidelines when developed in 2012. Full details about the construction of this Markov model are described by Gordon et al. [13].