A new study suggests that deep learning systems (DLS) would eliminate two minutes of human labour required to grade diabetic retinopathy images, however a semi-automated model is most cost-effective, saving national health systems up to 20% on screening costs.
In a cost-minimisation analysis, recently published in The Lancet Digital Health, researchers assessed data from the national diabetic retinopathy screening program in Singapore.
They aimed to evaluate the potential savings of two deep learning approaches; a semi-automated model as a triage filter before secondary human assessment; and a fully automated deep learning model without human assessment. They compared this alongside the current human assessment model.
They found the semi-automated screening model was the least expensive of the three models, at US$62 (AU$95) per patient per year. The fully automated model was $66 (AU$101) per patient, while the human assessment model was $77 (AU$118).
The findings revealed that switching to the semi-automated model could generate savings of $489 000, roughly 20% of the current annual screening cost.
“This study presents one of the first health economic evaluations of competing models for implementing a DLS designed to screen for referable diabetic retinopathy,” the researcher noted.
“From the health system perspective, the proposed semi-automated screening model could save $15 per patient as compared with the current human graders model. This saving is mainly attributable to the substantial reduction in human assessment time and workforce without sacrificing screening performance: each human grader takes roughly two minutes to grade each image, whereas the DLS drives this cost nearly to zero.”
By 2050, Singapore is projected to have 1 million people with diabetes. At this time, the annual savings were projected to be $15 million.
Further, although the fully automated model completely removes human grading, the semi-automated model, which lowers grading costs by only 74%, yields greater savings.
This is because of a higher rate of false positives, and therefore more unnecessary specialist visits, under the fully automated model. The higher costs of graders in the semi-automated model is more than offset by the lower consultation costs. However, this is based on the wages in Singapore, and might not apply to other settings.
