Cost-Utility Analysis of Deep Learning and Trained Human Graders for Diabetic Retinopathy Screening in a Nationwide Program

Recent evidence of economic evaluation of artificial intelligence in ophthalmology

“…There are four common methods of HEE: cost-minimization analysis (CMA) [15,16], cost-benefit analysis (CBA) [17,18], cost-effectiveness analysis (CEA) [19,20], and cost-utility analysis (CUA) [21,22 ▪ ]. The differences among these methods are their measurement and valuation of outcomes.…”

“…In the context of middle-income countries (MIC), the CUAs comparing artificial intelligence and manual grading for diabetic retinopathy screening in China [29], Thailand [22 ▪ ], and Brazil [30 ▪ ] arbitrarily set the price of artificial intelligence per patient at $1.45, $1.0, and $1.52, respectively.…”

“…Majority of them are for diabetic retinopathy screening. There are two CUA models on artificial intelligence for diabetic retinopathy screening (Brazil [30 ▪ ] and Thailand [22 ▪ ]), which applied a lifetime horizon of patients for evaluation. The study in Brazil was modeled for screening biannually, and treatment with laser photocoagulation [30 ▪ ]; the study in Thailand was for screening annually, and treatment with bevacizumab [22 ▪ ].…”

“…There are two CUA models on artificial intelligence for diabetic retinopathy screening (Brazil [30 ▪ ] and Thailand [22 ▪ ]), which applied a lifetime horizon of patients for evaluation. The study in Brazil was modeled for screening biannually, and treatment with laser photocoagulation [30 ▪ ]; the study in Thailand was for screening annually, and treatment with bevacizumab [22 ▪ ]. It was found that screening using artificial intelligence in either country, in a similar population of patients with diabetes entering the model at age 40 years, was not cost-effective compared with MG from a healthcare system perspective.…”

“…It was found that screening using artificial intelligence in either country, in a similar population of patients with diabetes entering the model at age 40 years, was not cost-effective compared with MG from a healthcare system perspective. This was likely due to artificial intelligence's ability to detect more patients for treatment [22 ▪ ]. The CUA in Thailand found further that artificial intelligence was a cost-effective alternative to manual grading from a societal perspective when opportunity costs from bilateral blindness were considered.…”