Diyabetik Retinopati Tanı Yöntemlerinin Maliyet Etkililik Analizi

Abstract

Artificial Intelligence (AI) tools are rapidly demonstrating their impact in the healthcare sector as in every part of lives. It is observed that AI technologies have significant effects in reducing costs, expanding access and improving quality in the field of health services. The importance of AI is increasing due to its faster and more accurate diagnosis of diseases, prevention of disease, and early treatment opportunities. One of the significant health problems observed in approximately 537 million people worldwide is diabetes. As the duration of diabetes increases, the risk of developing diabetic retinopathy (DR) also rises. Considering the increase in the prevalence of diabetes and the aging population, early diagnosis of DR is a fundamental requirement for diabetic patients to combat the complications of diabetes and to use resources effectively. In this context, the main purpose of the study is to analyze the cost-effectiveness of the AI diagnostic method and the classical diagnostic method (diagnosis by an ophthalmologist) in diagnosing and classifying DR. For this purpose, the analysis will be carried out with the Markov model. The second purpose of the study is to evaluate the DR diagnostic and classification performance of the AI software developed using convolutional neural networks. In line with the second purpose, 547 fundus images from 275 patients were obtained from the database of Ankara Etlik City Hospital, and the performance of the AI software in diagnosing and classifying DR was evaluated. The study results indicated that the AI diagnostic method performed a high performance with an average sensitivity of 90.58% and specificity of 96.07%. In accordance with the main purpose, the study was conducted with 113 diabetic patients aged 18 and over who accepted the study and applied to Ankara Etlik City Hospital Eye Diseases Polyclinic. Costs were determined from the perspective of the reimbursement institution using a method of payment per service and the time-driven activity-based costing method for healthcare personnel costs. The effectiveness data of the study were collected using the EQ-5D-5L scale. According to the study findings, the AI method provided 13,812 QALYs at a cost of 105.166 TL, while the ophthalmologist method provided 13,794 QALYs at a cost of 106.121 TL. As a result of the cost-effectiveness analysis, the AI method was found to be less costly and more effective than the classical method in diagnosing DR, since the ICER value (-54,687 TL) remained below the determined threshold values and was located in the south-east quadrant in the additional cost-effectiveness plane. However, the fact that the INMB values (6,388.71 TL and 3,301.64 TL) were positive according to the determined threshold values showed that the AI method was cost effective compared to the classical method. The research emphasizes the potential of AI tools in the field of healthcare and recommends the use of AI software in diagnosing DR in the future. In developing countries such as Turkey, increasing the potential of AI applications in the health sector to cope with problems such as aging population and resource constraints, providing early diagnosis and treatment of chronic diseases, facilitating access to health services in disadvantaged regions, and demonstrating the benefits of health information technologies in healthcare management and health economics will help improve the quality of health services in the long term.