Analysis of multiple biomarkers using structural equation modeling

Objectives: When examining the relationship between smoking intensity and toxicant exposure biomarkers, to understand the potential risk for smoking-related disease, individual biomarkers may not be strongly associated with smoking intensity because of the inherent variability in biomarkers. Structural equation modeling (SEM) offers a powerful solution by modeling the relationship between smoking intensity and multiple biomarkers through a latent variable. Methods: We used baseline data from a randomized trial (N = 1250) to estimate the relationship between smoking intensity and a latent toxicant exposure variable summarizing 5 volatile organic compound biomarkers. We analyzed 2 variables of smoking intensity: the self-report cigarettes smoked per day and total nicotine equivalents in urine. SEM was compared with linear regression with each biomarker analyzed individually or with the sum score of the 5 biomarkers. Results: SEM models showed strong relationships between smoking intensity and the latent toxicant exposure variable, and the relationship was stronger than its counterparts in linear regression with each biomarker analyzed separately or with the sum score. Conclusions: SEM is a powerful multivariate statistical method for studying multiple biomarkers assessing the same class of harmful constituents. This method could be used to evaluate exposure from different combusted tobacco products.