Comparison of statistical approaches to evaluate factors associated with metabolic syndrome

In statistical analyses, metabolic syndrome as a dependent variable is often utilized in a binary form (presence/absence) where the logistic regression model is used to estimate the odds ratio as the measure of association between health-related factors and metabolic syndrome. Since metabolic syndrome is a common outcome the interpretation of odds ratio as an approximation to prevalence or risk ratio is questionable as it may overestimate its intended target. In addition, dichotomizing a variable that could potentially be treated as discrete may lead to reduced statistical power. In this paper, the authors treat metabolic syndrome as a discrete outcome by defining it as the count of syndrome components. The goal of this study is to evaluate the usefulness of alternative generalized linear models for analysis of metabolic syndrome as a count outcome and compare the results with models that utilize the binary form. Empirical data were used to examine the association between depression and metabolic syndrome. Measures of association were calculated using two approaches; models that treat metabolic syndrome as a binary outcome (the logistic, log-binomial, Poisson, and the modified Poisson regression) and models that utilize metabolic syndrome as discrete/count data (the Poisson and the negative binomial regression). The method that treats metabolic syndrome as a count outcome (Poisson/negative binomial regression model) appears more sensitive in that it is better able to detect associations and hence can serve as an alternative to analyze metabolic syndrome as count dependent variable and provide an interpretable measure of association.