Abstract
Purpose :
While alternatives to standard ophthalmoscopy for screening premature infants for retinopathy of prematurity (ROP) have been developed and shown to be efficacious, little information exists regarding the cost-utility of these screening programs. We performed a cost-utility analysis for three different ROP screening techniques: standard ophthalmoscopy, telemedicine, and an automated deep convolutional neural network (artificial intelligence).
Methods :
A decision tree was created to model each ROP screening method as it would occur in practice, where premature infants weighing under 1,500g enter one of the three arms and follow different paths until arriving at a terminal node. A positive screen was defined as pre-plus disease or worse. Sensitivity and specificity of each screening modality from published data, in addition to incidence of treatment-requiring ROP (TR-ROP)—defined as prethreshold disease or worse—generated the probabilities of entering each node.
Costs were assigned to each pathway using 2019 Centers for Medicare and Medicaid Services reimbursement rates for Current Procedural Terminology (CPT) codes germane to each arm. Utility values were assigned based on visual acuities for outcome possibilities—whether patients were (in)correctly (un)treated—from previously published research. For each arm, expected utility values and costs were compared against no intervention or screening, and total cost per quality-adjusted life year (QALY) was discounted at 3% for an average 77.5 year lifespan based on standard methods. One-way sensitivity analyses for sensitivity and specificity of each screening technique (75-100%), and incidence of TR-ROP (1-25%), were also compared.
Results :
The cost per QALY was $6,355 for standard ophthalmoscopy screening, $1,558 for telemedicine screening, and $1,305 for artificial intelligence. One-way sensitivity analysis showed the cost per QALY ranged from $5,647 to $7,237 for ophthalmoscopy, $848 to $2,418 for telemedicine, and $574 to $2,182 for artificial intelligence.
Conclusions :
While both telemedicine and artificial intelligence were more cost-effective screening modalities for ROP than standard ophthalmoscopy, artificial intelligence cost the least per QALY and thus may be of benefit in some settings—especially where traditional screenings are either expensive or difficult to obtain.
This is a 2020 ARVO Annual Meeting abstract.