On the use of Structural Equation Modeling in Infectious Disease Epidemiology: a systematic and critical review

Structural Equation Modeling (SEM) is a multivariate statistical technique increasingly used in epidemiology to analyze complex causal relationships. This paper focuses on the practical use of the SEM technique and analyze applications in terms of model design, model development, and model evaluation to address epidemics-related problems based on articles published from 2013 to 2022. The selection of articles was based on the PRISMA methodology. The review was based on 111 scientific papers after applying the exclusion criteria. Data on research design, sample size, software, estimation methods and evaluation methods were extracted.Findings reveal a significant rise in SEM applications over the past decade, particularly during the COVID-19 pandemic, with 70.45 % of the studies focused on COVID-19-related issues. Partial Least Squares SEM (PLS-SEM) was the most frequently applied estimation method (50 %), followed by Maximum Likelihood Estimation (29.17 %) and Bayesian approaches (8.33 %). The most commonly used software included AMOS (25.68 %), SPSS (24.32 %), and Smart-PLS (14.86 %). Model evaluation relied on absolute fit indices such as Root Mean Square Error of Approximation (RMSEA; 14.47 %), Chi-square (7 %), and Standardized Root Mean Square Residual (SRMR ; 7.23 %), as well as incremental indices like Comparative Fit Index (CFI; 14.47%) and Tucker-Lewis Index (TLI; 7.55 %). Only 3.85 % of the studies had sample sizes below 100, while 24.03 % had 1,000 or more. An illustrative example of SEM applied to COVID-19 data is provided. This review underscores the growing relevance of SEM in epidemiology, emphasizing both its potential and the methodological considerations needed to ensure robust and interpretable results in public health research.