RE: “CLUSTERING OF PROCOAGULATION, INFLAMMATION, AND FIBRINOLYSIS VARIABLES WITH METABOLIC FACTORS IN INSULIN RESISTANCE SYNDROME”

Sakkinen et al. (1) derived multiple factors of insulin resistance syndrome through factor analysis of 21 metabolic and hemostatic variables. They used orthogonal transformation (varimax rotation in SAS computer software) to rotate the factors in order to achieve clearer interpretability. Thus, the derived factors including body mass, insulin/ glucose, lipids, blood pressure, and so on are “uncorrelated.” While their approach is statistically sound, the logic is not consistent with the theory of insulin resistance syndrome, which postulates a common underlying biologic process for the close interrelation among obesity, hyperinsulinemia, glucose intolerance, dyslipidemia, and other metabolic disorders (2). There are substantial data that these components are all intercorrelated, both statistically and biologically. Thus, the “uncorrelated” factors identified from this study and several previous analyses may merely reflect a statistical artifact rather than a biologic reality. To test the clustering of the components as well as a unified mechanism that underlies various metabolic abnormalities, an alternative rotation method, an oblique rotation (promax rotation in SAS computer software), can be used to produce correlated factors (3). Then second-order factor(s) can be derived by factor analyzing the correlation matrix of the common factors obtained from the first step. This can also be achieved by confirmatory factor analysis (4), which is a theory-testing method as opposed to a datadriven method like explanatory factor analysis. This modelfitting procedure allows one to test the ability of the hypothesized factor structure to account for the observed covariance by examining the overall fit of the model. The analyses can be carried out using SAS PROC CALIS (5) or specialized computer programs, such as LISREL 8 (6). In addition, the confirmatory factor analysis procedure allows for a test of the equality of factor structure between different groups (e.g., male and female) by comparing the model fit of competing models with and without certain constraints on factor loadings (6).