Free papers detail | 11th Asia Pacific Tele Ophthalmology Society Symposium

Title
Economic Value of AI-Driven Diabetic Retinopathy Screening in Australia: A Cost-Effectiveness Analysis for Reimbursement Decision-making

Authors
Wenyi Hu, Zhuoting Zhu

Presenting
Wenyi Hu

PURPOSE:
We aim to evaluate the cost-effectiveness of an artificial intelligence (AI)-based diabetic retinopathy (DR) screening system and estimate a reimbursement threshold for both non-Indigenous and Indigenous Australians living with diabetes.

METHODS:
A decision-analytic Markov model was developed to simulate DR progression in non-Indigenous and Indigenous Australians with diabetes aged ≥20 years over 10 years. From a health system’s perspective, we compared AI-based screening with manual screening carried out by optometrists, ophthalmologists and general practitioners. Outcomes included incremental changes in quality-adjusted life years (QALYs), blindness cases, costs, benefit-cost ratios (BCR), net monetary benefits (NMB), and cost-neutral reimbursement rates. A willingness-to-pay threshold of AU$50,000/QALY and a 3.5% discount rate were applied.

METHODS:

A decision-analytic Markov model was developed to simulate DR progression in non-Indigenous and Indigenous Australians with diabetes aged ≥20 years over 10 years. From a health system’s perspective, we compared AI-based screening with manual screening carried out by optometrists, ophthalmologists and general practitioners. Outcomes included incremental changes in quality-adjusted life years (QALYs), blindness cases, costs, benefit-cost ratios (BCR), net monetary benefits (NMB), and cost-neutral reimbursement rates. A willingness-to-pay threshold of AU$50,000/QALY and a 3.5% discount rate were applied.

RESULTS:
In the non-Indigenous population, current practice was projected to result in 20,141 blindness cases and AU$1,420.4 million in DR-related expenditure over 10 years. AI-based screening would prevent 662 cases of blindness, yield 1,930 additional QALYs, and save AU$40.6m, resulting in a BCR of 3.21 and NMB of AU$137m. A reimbursement rate of up to AU$50.1 per AI screen would remain cost-neutral. Among Indigenous Australians, current practice was projected to lead to 2,245 blindness cases and AU$179.3m in expenditure. AI-based screening would prevent 40 blindness cases, gain 110 QALYs, and save AU$2.7m, with a BCR of 3.84 and NMB of AU$8m. The cost-neutral reimbursement rate would be AU$53.3.

RESULTS:

In the non-Indigenous population, current practice was projected to result in 20,141 blindness cases and AU$1,420.4 million in DR-related expenditure over 10 years. AI-based screening would prevent 662 cases of blindness, yield 1,930 additional QALYs, and save AU$40.6m, resulting in a BCR of 3.21 and NMB of AU$137m. A reimbursement rate of up to AU$50.1 per AI screen would remain cost-neutral. Among Indigenous Australians, current practice was projected to lead to 2,245 blindness cases and AU$179.3m in expenditure. AI-based screening would prevent 40 blindness cases, gain 110 QALYs, and save AU$2.7m, with a BCR of 3.84 and NMB of AU$8m. The cost-neutral reimbursement rate would be AU$53.3.

CONCLUSIONS:
AI-based DR screening is effective and cost-saving, with estimated reimbursement thresholds comparable to the lower end of manual screening fees, suggesting that AI screening can enhance effectiveness while ensuring affordability and provider incentives.