Table 1
AI applications, evidence maturity, and equity risks across global health domains.
| DOMAIN | EXAMPLE AI APPLICATIONS | EVIDENCE MATURITY | PRIMARY EQUITY RISKS | KEY GAPS |
|---|---|---|---|---|
| Education | Intelligent tutoring systems, adaptive learning platforms, virtual simulation, curriculum analytics and administration | Low–moderate | Digital divide, unequal access to hardware and connectivity | Faculty training, contextualized curricula, evaluation of learning outcomes |
| Epidemiology and surveillance | Outbreak prediction, contact tracing, pathogen modeling | Moderate | Surveillance overreach, data misuse, limited transparency | Governance frameworks, cross‑border data stewardship |
| Clinical medicine | Diagnostic imaging, clinical decision support, telemedicine tools | Variable | Algorithmic bias, dataset mismatch, unclear accountability | Outcome‑based evaluation, cost‑effectiveness data |
| Data equity | Bias audits, fairness metrics, participatory design approaches | Emerging | Tokenistic inclusion, lack of enforcement | Regulatory authority, standardized equity benchmarks |
Table 2
Illustrative AI case studies presented at the CUGH 2025 pre‑conference session.
| USE CASE | GEOGRAPHIC CONTEXT | AI FUNCTION | PRIMARY BARRIER IDENTIFIED | KEY LESSON |
|---|---|---|---|---|
| Cancer screening | Multiple LMIC settings | Image classification | Limited annotated datasets | Data inequity constrains scalability |
| Mobile health applications | Multiple regions | Health service access optimization | Governance and regulation | Policy frameworks are essential for safe deployment |
| Tuberculosis diagnosis | Uganda | Imaging‑based diagnostic support | Infrastructure dependence | Contextual readiness outweighs model performance |
| Risk‑stratified care | Mixed settings | Clinical decision support | Workflow integration | Implementation challenges dominate impact |
| Appropriate technology selection | Fragile settings | Tool adaptation | Misaligned assumptions | Local stakeholder engagement is critical |
Table 3
What evidence is most urgently needed?
| RESEARCH PRIORITY | DESCRIPTION | KEY METRIC/OUTCOME |
|---|---|---|
| Longitudinal performance and drift | Studies assessing model drift and performance degradation over time | Sustained accuracy, reliability, time to failure/recalibration |
| Cost‑effectiveness analyses (CEA) | Comparing AI‑enabled pathways to standard care | Incremental cost‑effectiveness ratio, return on investment, total cost of ownership |
| Hybrid effectiveness‑implementation trials | Examining adoption, retention, feasibility, fidelity, and unintended consequences of AI tools in practice | Reach, effectiveness, adoption, implementation, maintenance (RE‑AIM), unintended social/workflow consequences |
| Comparative multi‑country validation | Studies assessing the generalizability and transferability of AI models across diverse geographical and demographic populations | Cross‑contextual performance stability, external validity metrics |
| Adaptive informed consent research | Research on informed consent models appropriate for AI‑supported care, especially regarding data use and algorithmic decision‑making | Patient understanding, ethical approval rates, perceived trust, and transparency |
Table 4
Strategic imperatives for equitable AI in global health and responsible actors.
| STRATEGIC IMPERATIVE | PRIMARY ACTORS | SUPPORTING ACTORS | PERSISTENT GAPS |
|---|---|---|---|
| Contextualized governance and policy | Ministries of health, WHO, regional bodies | Legal experts, civil society | Enforcement capacity |
| Capacity‑building | Universities, training institutions | Donors, NGOs | Long‑term sustainability |
| Evidence generation | Researchers, implementers | Funders, policymakers | Real‑world trials and outcome data |
| Open knowledge | Academic consortia, platforms | Philanthropy | Data sovereignty protections |
| Community engagement | Civil society, communities | Implementers, regulators | Power‑sharing and accountability |