Table 5 Decision model parameters
From: Cost-effectiveness of AI for pediatric diabetic eye exams from a health system perspective
Parameter Description | Base-case estimate | Low | High | Distribution | Sources |
|---|---|---|---|---|---|
Population-level metrics | |||||
Prevalence of DR in T1D | 5.6% | 3.4% | 20.1% | Beta | |
Prevalence of DR in T2D | 9.1% | 6.0% | 51.0% | Beta | |
Diagnostic accuracy metrics | |||||
Sensitivity of ECP | 33% | 0% | 100% | Beta | |
Specificity of ECP | 95% | 70% | 100% | Beta | |
Sensitivity of autonomous AI | 89.3% | 42.1% | 99.6% | Beta | |
Specificity of autonomous AI | 88.2% | 74.3% | 93.7% | Beta | |
 Diagnosability of autonomous AI | 96.8% | 94.6% | 97.3% | Beta | |
Human-behavior factors | |||||
Probability of patient going for initial ECP-based screening | 52.0% | 15.3% | 72% | Beta | |
Probability of patient accepting AI- based screening | 95-96.4% | 50% | 100% | Beta | |
Probability of patient following up with ECP after ECP screening yielded positive result | 29% | 0 | 100% | Beta | |
Probability of patient following up with ECP after AI autonomous screening yielded positive result | 65% | 55.4% | 100% | Beta | |
Autonomous AI Costs to the Health System | |||||
Acquisition (equipment, camera, system) per site | $10,000 | $1000 | $100,000 | Log normal | Stakeholder interviews |
IT integration (including health system and vendor) per site | $3000 | $1000 | $20,000 | Log normal | Stakeholder interviews |
Ongoing support fee (including health system and vendor)/ per site | $10,000 | $1000 | $100,000 | Log normal | Stakeholder interviews |
Facility Space Allocation Cost for AI Device per site | $2800 | $0 | $8500 | Log normal | |
tion | $33.00/hour | $7.25/hour | $42.00/hour | Gamma | |
Optician Productivity Rate | 2 patients screened/hour | 1 patient screened/hour | 4 patients screened/hour | Poisson | |
Standard of Care (ECP) Costs to the Health System | |||||
Cost of service per patient | $172 | $110 | $240 | Log normal | |
