Table 2.

Health Economic Studies on DR Screening Using AI

Author, Year, Country	Comparators	Screening Model	Measurement of Effect	Economic Outcomes
Scotland et al,34 2007, UK	Semi-automated grading (hybrid approach) vs. manual grading alone	Digital photography and multilevel manual grading systems	The number of appropriate screening outcomes (i.e., defined as final decisions appropriate to actual grade of retinopathy present) and true referable cases detected in one year	Compared to the manual grading model, the semi-automated model led to a saving of £4088 per additional referable case detected, and of £1990 per additional appropriate screening outcome.
Tufail et al,20 2016, UK	AI-based ML tool as placement for initial manual grading (semi-automated hybrid)	AI-based (ML) two-field fundus photos	Appropriate outcomes (defined as identification of DR present vs. absent by the AI-based software)	AI-based semi-automated hybrid approach (Retmarker and EyeArt) had sufficient specificity to make them cost-effective to manual grading alone, as ICER was $18.69 and $7.14, respectively
Xie et al,50 2019, Singapore	Semi-automated hybrid approach (DLS-based) vs. manual grading alone	Retinal fundus photographs	QALYs	DLS-based (semi-automated hybrid approach) resulted in a lifetime cost-saving of $135 per patient while maintaining comparable QALYs gained.

QALYs, quality-adjusted life years;

ICER, incremental cost-effectiveness ratio;

manual grading is equivalent to human assessment.