Transferability of Global AI-Based Obesity Risk Prediction Models to Sub-Saharan Africa: A Critical Analysis of Health Equity and Implementation Determinants

Background: Artificial intelligence (AI)-based obesity risk prediction models have been mainly developed and proven in high-income countries (HIC) however, their applicability to sub-Saharan Africa (SSA) remains uncertain. This study critically examines the transferability of Western-developed AI-based obesity risk prediction models to SSA, with particular attention to contextual fit and equity implications.

Method: Using the Health Equity Implementation Framework (HEIF) as a guide, this study evaluates and synthesises evidence from existing literature across epidemiological, technical, socio-cultural, and health-system dimensions. Evidence was identified from peer-reviewed and grey literature and mapped to HEIF constructs relevant to model transferability and equitable implementation.

Findings: The evidence from this study indicates that the direct deployment of AI-based obesity risk prediction models trained on non-African populations is constrained by differences in body composition profiles, disease patterns, data availability, digital infrastructure, and clinical workflows. These constraints increase the risk of misclassification, algorithmic bias, and inequitable health outcomes when models are adopted without appropriate adaptation. The study findings also indicate that AI-based prediction models operate as socio-technical interventions rather than purely technical tools.

Analysis: Overall, these findings suggest that effective implementation in SSA requires context-specific strategies, including the use of locally relevant data, model recalibration, participatory design approaches, workforce capacity building, and robust governance mechanisms. Technical methods such as transfer learning, bias auditing, and iterative validation may support model adaptation, but only when embedded within equity-oriented implementation processes.

Conclusion: This study concludes that while AI has the potential to support obesity prevention and risk stratification in SSA, its benefits are likely to remain limited without deliberate localisation and sustained institutional investment. Future research should therefore prioritise the development and evaluation of SSA-specific models, assess real-world implementation impacts over time, and explore governance arrangements that enable African institutions and communities to play leading roles in developing AI solutions for obesity risk prediction and related AI-in-Health initiatives.