Prioritization of antimicrobial resistance surveillance of gram-negative pathogens - a stacked modelling approach

Antimicrobial resistance in gram-negative pathogens is a growing global health threat, particularly in low- and middle-income countries (LMICs). However, many LMICs lack the laboratory capacity needed for routine surveillance. Affordable, scalable methods are therefore required to guide the prioritisation of surveillance efforts. Although socioeconomic, governance, environmental, and health-system covariates are known to correlate with clinical AMR prevalence, these contextual covariates have not been fully utilised to estimate resistance in settings with limited data. We quantified these associations using a stacked ensemble modelling framework to generate national AMR prevalence estimates for nine priority gram-negative pathogen–antibiotic combinations from 2005 to 2021. National AMR data were obtained from the Vivli AMR Register and ResistanceMap and linked with 178 covariates describing socioeconomic conditions, governance indicators, environmental exposures, water and sanitation, and health-system performance. Child models included generalized additive models, gradient boosting machines, random forests, support vector machines, cubist models, and xgboost regressors, whose estimations were combined using spatiotemporal gaussian process regression as a meta-learner. Cross-validated performance was robust for most pathogen–antibiotic combinations, with R 2 values ranging from 0.65 to 0.93. Our modelled estimates enable the identification of priority countries for AMR surveillance by jointly considering estimated prevalence, temporal trajectories, and model performance across pathogens. We identified priority countries across multiple pathogen–antibiotic combinations, with estimated prevalence concentrated in regions such as Central Asia, the Middle East and North Africa, Latin America, and sub-Saharan Africa. Many countries within these regions remain unenrolled in the World Health Organization’s Global Antimicrobial Resistance and Use Surveillance System or contribute only sparse data, underscoring persistent gaps in surveillance where resistance is estimated to be high. By integrating AMR data with relevant contextual covariates, our framework fills temporal and geographical gaps. These estimates support evidence-based prioritisation of surveillance strengthening in resource-constrained settings.