Latent Class Analysis Yields Inconsistent Support for an Unobserved Factor Linking Together Components of the Metabolic Syndrome
Abstract Number: 1003-P
Authors: EDWARD J. BOYKO, MARGUERITE J. MCNEELY, DONNA L. LEONETTI, STEVEN E. KAHN, WILFRED Y. FUJIMOTO
Institutions: Seattle, WA
Results: Associations among central adiposity, hyperglycemia, hypertension, and dyslipidemia beyond that expected by chance alone led to the recognition of the metabolic syndrome. Whether an underlying factor unites its components has not been consistently demonstrated by factor analysis, although this method is not suitable for recognition of an unobserved dichotomous condition, as it assumes continuous data. To our knowledge, latent class analysis has not been used to examine if an underlying dichotomous state can account for the inter-relationships among metabolic syndrome components. We applied this method to data from 658 subjects (53.0% male, mean age 53.3 yrs, mean BMI 24.3) from the Japanese American Community Diabetes Study who underwent a medical history, physical exam, 75 g oral glucose tolerance test after an overnight fast, and measurement of visceral fat (VF) area at L4-L5 using computed tomography. Metabolic syndrome components were defined as follows: hyperglycemia, fasting plasma > 100 mg/dl (FPG100) or 2-hr glucose > 140 mg/dl or medication treatment for diabetes; hypertension (HTN), systolic blood pressure > 135 mm Hg or diastolic > 85 mm Hg or use of antihypertensive medication; dyslipidemia (LP), triglycerides > 150 mg/dl or HDL < 40 mg/dl (men) or 50 mg/dl (women) or use of lipid-lowering medication; and waist circumference (WC) > 80 cm (women) or 90cm (men) as per recommendations for Japanese ethnicity. Alternate measures of fasting plasma glucose (> 110 mg/dl) (FPG110) and central adiposity (VF upper quartile) were also used. The goodness of fit chi-square test determined if an unobserved latent dichotomous variable accounted for associations among the observed metabolic syndrome components. Interpretation of this measure is opposite to conventional practice, in that a p-value > 0.05 indicates a good fit of the latent class model. Results are as follows: FPG100 HTN LP WC p=0.044, FPG110 HTN LP WC p=0.096, FPG100 HTN LP VF p=0.076, and FPG110 HTN LP VF p=0.025. Although results were not uniform, latent class analysis shows some promise in identifying whether an unobserved condition may account for the metabolic syndrome.