Data Challenge - Leveraging machine learning to identify patterns in the structural and social determinants of AMR

The adoption of political declaration on antimicrobial resistance (AMR) at the United Nations General Assembly highlights a renewed global commitment, with a particular emphasis on addressing the social determinants of AMR and tackling disparities in surveillance capacity and existing data gaps. The absence of linked AMR data incorporating structural and social determinants, coupled with heterogeneity in surveillance coverage and methods presents a significant challenge to generate evidence to inform design and implementation of interventions with high contextual relevance. Traditional analytic methods struggle to accommodate data with varied structure and levels of complexity. Predicting AMR trends requires versatile methods accounted for non-linearity, outliers, correlation among variables and heteroskedasticity. We aim to develop modelling methodology grounded in machine learning (ML) techniques to assess the structural and social determinants of AMR and inform data curation for future surveillance.
We will utilise a combination of conventional regression/econometric approach and ML to capture probabilistic relationships between structural and social determinants and AMR outcomes (resistance rates). Techniques including extreme bound analysis, model averaging and Bayesian network analysis will be employed. We hope to systematically evaluate how sensitive results are to changes in model specifications, combine predictions from multiple models with weights based on performance or likelihood, and enhance robustness of causal inference and trend prediction. This helps mitigate the risk of relying on a potentially biased or incomplete model and manage uncertainties in policy and health systems context.
The outcome is to develop a modelling methodology that leverages ML to assess the influence of structural and social determinants on AMR. The versatility of ML allows it to manage uncertainty and handle incomplete datasets, enabling development of a robust adaptable framework. By generating evidence on how structural and social determinants influence AMR trends, this methodology contributes to strengthening health systems and informing public health practice.