Data Challenge: Integrating Pharmaceutical Stability and Climatic Risk Factors into Machine Learning Models of Global Beta-Lactam Resistance Trends

Background & Problem Statement: Beta-lactam antibiotics remain the most widely prescribed antibiotic class globally, yet resistance to these agents is escalating rapidly, particularly in low- and middle-income countries (LMICs). While existing AMR forecasting modelling approaches primarily rely on epidemiological data, a critical and underexplored predictor has been consistently overlooked: the environmental and pharmaceutical stability factors that may influence antibiotic effectiveness in real-world settings, particularly in tropical regions. WHO/ICH Climatic Zones Iva and IVb countries are typified by high temperature and humidity conditions known to promote beta-lactam degradation, potentially leading to subtherapeutic drug exposure and increased selective pressure for resistance. No study has integrated pharmaceutical stability science into large-scale AMR surveillance modelling. Research Question: To what extent do climatic conditions associated with beta-lactam pharmaceutical instability explain variation in resistance escalation trends across countries, and does integrating a pharmaceutical stability index improve machine learning predictions of AMR trajectories? Additionally, can such a model improve predictive performance for data-sparse countries by leveraging global resistance patterns and incorporating stability-informed climatic risk features? Methodology: Using the Vivli AMR Register’s dataset, we will: Assign each country an ICH climatic zone-derived pharmaceutical stability risk score, based on published beta-lactam degradation kinetics; Build baseline ML models predicting beta-lactam resistance trends using established predictors such as time, country, pathogen, and antibiotic class; augment models with the pharmaceutical stability index; and compare performance improvements. Apply regression analysis, XGBoost, and time-series modelling to capture both temporal and spatial resistance dynamics; Evaluate feature importance to determine the contribution of pharmaceutical stability variables to model predictions. Expected outcome: A validated AMR prediction framework incorporating pharmaceutical stability as a novel feature, alongside a comparative analysis of model performance with and without stability-informed climatic risk factors. Additionally, a policy-relevant risk mapping framework identifying countries where climatic conditions may contribute to accelerated resistance trends, particularly in LMIC settings such as Nigeria, where Zone IVb climatic conditions are prevalent. The model will also be used to explore if adding pharmaceutical stability information makes predictions more accurate in global AMR surveillance data.