Exploring the correlates of COVID-19 vaccination inequity: a global analysis using machine learning from a health economic lens.

Introduction: The COVID-19 pandemic has reduced system resilience, caused economic welfare loss, and widened disparities in healthcare access. This study aims to identify the macro-level factors contributing to COVID-19 vaccination inequity by proposing an AI-driven monitoring framework using classical statistical modelling and machine learning (ML) classifiers from a pooled dataset (WHO, World Bank, and our World in Data) covering 195 countries. Methods: Daily vaccination data and national-level indicators were investigated using concentration indices, classical odds ratios across global regions and seven ML classifiers-logistic regression, naïve Bayes, decision tree, random forest, LightGBM, extra trees, and XGBoost-to identify significant predictors of vaccination inequity with higher accuracy. Results and discussion: Findings show vaccine coverage remains pro-rich in Africa, North America, Europe, and Oceania, whereas Southeast Asia depicts pro-poor vaccine uptake. Capacity of health system, availability of hygiene infrastructure, prioritization of people suffering from noncommunicable diseases, and exposure to behavioral risk factors were strongly associated with pro-poor distribution of vaccine access in low-income (LIC) and lower-middle income countries (LMIC) in Asia and Africa and improves over time. Among ML models, random forest and XGBoost achieved the highest performance. As the final best model, random forest (RF) is selected with highest weighted score (98.5%), AUROC (99.9%). XGBoost is the second-best model attained the second highest weighted score (97.9%), AUROC (99.8%), and both attained good 5-fold cross validation standard deviation (0.009 for RF and 0.014 for XGBoost) allowing temporal stratification of folds. Results justify the superiority of ensemble ML models over single learners in predicting inequity in vaccine uptake. This study proves that machine learning outperforms conventional predictive analysis and more suitable for monitoring inequity in COVID-19 vaccination access to inform global health policy towards intelligent pandemic preparedness. Therefore, to reduce regional inequity in vaccine uptake, the regional-structural nonlinearities in LICs and LMICs should be adjusted through accurate and robust integration of artificial intelligence in monitoring systems.