Artificial Intelligence (AI) - Assisted Visual Impairment Screening Model: Community-based Implementation and Evaluation of Performance, Feasibility and Costs.

The primary outcome is the detection performance for VI (refractive error-related and disease-related VI) and the rate of correct referral, with reference to the expert panels' diagnosis . To assess whether the new AI-assisted model has better referral accuracy than the current traditional model, the accuracy, AUC values, sensitivity, specificity and other performance metrics of the two models will be calculated and compared.

Evaluation includes (i) average screening time per patient and (ii) average number of patients screened per session.

Patient Acceptability is assessed via a 6-item questionnaire (4-point Likert scale: 'Satisfaction' or 'Likelihood') administered by trained coordinators.

Perceptions of feasibility is evaluated through at least two focus groups with optometrists and one with service providers will be conducted, guided by the Consolidated Framework for Implementation Research (CFIR). Discussions will explore barriers and facilitators influencing the AI screening model's adoption. Data will be inductively and deductively coded by two researchers using CFIR constructs; discrepancies resolved through consensus or a third researcher.

Researchers will quantify the incremental cost savings of implementing the AI-assisted screening model over the PSS model using an Activity Based Costing (ABC) approach that quantifies all non-sunk costs (including labor, materials and supplies, and amortized technology and space utility/ rental costs) required to conduct each assessment stratified by key activities of each screening model (e.g., conduct screening examinations, operationalize the AVIRI algorithm, on-site generation of test results etc.). Fixed costs will be amortized over the inputs' expected useful life (i.e. involved fixed assets' life expectancy). For the cost of clinical assessments, researchers will use non-subsidized bill sizes as these are expected to approximate actual costs.