Evaluating the cost-effectiveness of artificial intelligence in Barrett's surveillance

Artificial intelligence (AI) has emerged as a promising tool to detect early dysplasia in Barrett's esophagus (BE). However, the cost-effectiveness of AI-assisted BE surveillance has not been evaluated.

A Markov model simulated 1000 Australian individuals with nondysplastic Barrett's esophagus (NDBE) undergoing surveillance from age 50 to 80 years, with follow-up until age 100. We compared AI-assisted surveillance with targeted biopsies against standard endoscopy with four-quadrant biopsies under 3-yearly and 5-yearly surveillance. The primary outcome was the incremental cost-effectiveness ratio (ICER). Secondary outcomes included cumulative incidence of high grade dysplasia (HGD)/T1 lesions and advanced esophageal adenocarcinoma (EAC), as well as their relative differences. A health care system perspective was employed, with costs and utilities discounted at an annual rate of 3%.

AI-assisted surveillance was cost effective across both intervals, with ICERs of AUD 14 039/quality-adjusted life year (QALY) (3-yearly) and 3609/QALY (5-yearly). Compared with standard surveillance, AI reduced the cumulative incidence of advanced EAC by 5 and 3 cases per 1000 people (relative reductions of 3.5% and 1.6%) for 3- and 5-yearly surveillance, respectively. Conversely, AI increased HGD/T1 detection by 27 and 38 cases per 1000 people (relative increases of 21.8% and 27.7%) for 3- and 5-yearly surveillance, respectively. Additionally, AI reduced missed HGD/T1 incidence by 37 and 45 cases per 1000 people (relative reductions of 72.9% and 72.2%) for 3- and 5-yearly surveillance, respectively.

AI-assisted endoscopic surveillance in BE was a cost-effective strategy in the Australian health care setting, reducing the cumulative incidence of advanced EAC and missed HGD/T1 lesions.