Abstract
Purpose :
Optical coherence tomography (OCT) shows high sensitivity for detecting glaucoma in specialized clinics. However, its interpretation still needs expertise in both glaucoma and OCT, which limits its application in glaucoma screening. We aim to determine the long-term cost-effectiveness of an established artificial intelligence (AI) model analyzing OCT volumetric raw scans for glaucoma screening in comparison to glaucoma specialists’ interpretation of OCT reports in Hong Kong (HK).
Methods :
We developed a Markov model from a societal perspective. A Markov cycle tree was built to model 6 health states, including healthy, mild glaucoma, moderate glaucoma, severe glaucoma, unilateral blindness, and bilateral blindness (using TreeAge Pro 2022). A hypothetical cohort was followed in the model from age 45 years through a total of 50 1-year Markov cycles. The model assumed that patients who were screened correctly as glaucoma positive with either the AI model or specialists’ interpretation of OCT reports. Effectiveness was measured in Quality Adjusted Life Years (QALYs). Costs in HK dollars (HKD) were obtained from a private clinic in HK and converted to US dollars (USD) at a rate of 7.81 HKD per USD (the year 2022) and with a 3% discount rate per year. The prevalence, transitional probabilities, and utilities of different staged glaucoma were obtained from existing literature. The main outcome was the incremental cost-effectiveness ratio (ICER) per QALYs. We performed one-way and probabilistic sensitivity analyses to assess associated factors affecting the ICER.
Results :
For each subject, the AI model had an ICER of -5.58 USD per QALYs (95% confidence interval, -6.76 USD to -4.56 USD) compared with the human interpretation of OCT reports, which was cost-effective at any willingness-to-pay thresholds (Fig.1). Over the 50 years, a total of 369.88 USD was saved by the AI model for each screened person compared with specialists’ interpretation of OCT reports. In the sensitivity analysis, the sensitivities of glaucoma detection (either by specialists or the AI model) and the discount rate influenced the ICER (Fig.2).
Conclusions :
The AI model is likely to be cost-effective in HK for glaucoma screening in long term. Future studies are warranted to investigate the cost-effectiveness of the AI model in less developed regions and establish a feasible program for glaucoma screening in populations.
This abstract was presented at the 2023 ARVO Annual Meeting, held in New Orleans, LA, April 23-27, 2023.