Statistical Methods in Epidemiology

Authors: Hon Keung Tony Ng (Bentley University, USA) and Miodrag Lovric (Radford University, USA)

This article provides a comprehensive overview of the essential statistical methods used in epidemiology, with a focus on study design, risk measurement, and modern causal inference techniques. Epidemiology, the science of understanding the distribution and determinants of disease in populations, relies heavily on statistical reasoning to infer conclusions from data collected across different health outcomes and populations.

The paper begins by outlining the major types of epidemiologic studies: cohort, case-control, and cross-sectional. It explains how incidence rates, relative risks (RR), attributable risks (AR), and odds ratios (OR) are calculated and interpreted in various study designs. For instance, cohort studies typically use Poisson or binomial models, while case-control studies use odds ratios, especially when disease incidence is low.

A major contribution of the article is its exposition of advanced statistical methods in causal inference. The authors discuss modern tools like:

• Propensity Score Methods: Balancing covariates between treatment groups using matching, stratification, and weighting.
• Inverse Probability Weighting (IPW): Adjusting for confounding and missing data by reweighting samples based on exposure probabilities.
• Doubly Robust Estimation: Combining exposure and outcome modeling to provide valid estimates even if one model is misspecified.
• G-Computation: A structural modeling method based on counterfactuals to assess the effect of exposures.
• Marginal Structural Models (MSMs): Handling time-varying exposures and confounders in longitudinal data.
• Targeted Learning: Machine learning-driven approaches (e.g., TMLE) to optimize estimation of causal effects.

The article emphasizes that modern epidemiology goes beyond descriptive statistics and traditional inference, embracing more complex methodologies to approximate causal effects in the presence of confounding, selection bias, and time-varying dynamics.

For full mathematical explanations, practical examples, and reference studies, see the complete article in the International Encyclopedia of Statistical Science.