The evolution of sampling in epidemiology: from classical probability models to AI-enhanced recruitment and active learning.

Sampling is a foundational element of epidemiological research because it determines the validity, efficiency, and generalizability of population-based inferences. Classical probability and non-probability sampling methods have long underpinned public health study design by providing robust frameworks for unbiased estimation and causal inference. However, the rapid expansion of digital health data, electronic health records, and large-scale online cohorts has exposed important limitations in these largely static sampling frameworks. This narrative review traces the evolution of sampling in epidemiology from traditional probability-based designs to contemporary artificial intelligence-enhanced recruitment strategies, with particular emphasis on active learning and adaptive sampling. By synthesizing classical statistical theory with emerging machine-learning approaches and recent methodological advances, this review argues that the future of epidemiological sampling lies in hybrid frameworks that preserve statistical rigor while leveraging algorithm-driven adaptability. Methodological opportunities, ethical risks, and implications for low-income and middle-income countries are critically examined.