Abstract P082: Estimating Equation for Computed Tomography Derived Visceral Fat: The Multi-Ethnic Study of Atherosclerosis

Abstract

Background: The only way to determine visceral fat separately from subcutaneous fat is through magnetic resonance imaging (MRI) or computed tomography (CT) scans. The use of MRI or CT scans is unacceptable for screening in the general population. Given this, estimating equations for visceral fat that use commonly collected clinical variables have potential utility. Methods: Data are from 1842 participants, aged 45 to 84, from the Multi-Ethnic Study of Atherosclerosis with visceral fat assessed by CT scan at study visit 2 or 3, and anthropometry measured at matched visits. We excluded 19 participants with reported thiazolidinedione use, 11 with waist circumference greater than 140cm, and 5 with BMI greater than 45kg/m 2 . Visceral fat area, measured in cm 2 , was calculated as the average area from two CT scan slices at the L4/L5 vertebrae. Visceral fat was naturally log transformed to account for non-normality. Anthropometric indices included: height(cm), weight(kg), BMI(kg/m 2 ), waist circumference(cm), hip circumference(cm), waist to hip ratio, and waist to height ratio. Other variables included sex, Black, Asian, and Hispanic race/ethnicities, and age. Data were separated into training and testing datasets containing 2/3 and 1/3 of the data, respectively. Multivariable linear models were used to generate coefficients for the estimating equations and included non-linear and interaction terms. The PRESS statistic, R 2 , and RMSE were used to determine goodness of fit and likelihood ratio tests were used for variable and model selection. Results: Models including multiple measures of anthropometry performed better than models including only waist circumference or BMI. Models including non-linear terms for anthropometry and interaction by sex and race performed better than simple linear models. After analysis, the final estimating equation was: Ln Visceral Fat = -8.64 + 0.006*age - 8.42*sex - 8.53*black - 0.099*weight + 0.41*bmi + 0.022*waist + 0.029*hip + 8.58*waist/hip - 0.00396*bmi 2 +0.00015*waist 2 - 0.0003*hip 2 - 4.89*(waist/hip) 2 - 0.014*weight*sex - 0.066*waist*sex + 0.076*hip*sex + 8.57*(waist/hip)*sex - 0.097*waist*black + 0.088*hip*black + 10.31*(waist/hip)*black. The R 2 in the testing dataset for this model was 0.62. The estimates of visceral fat from the final equation provided estimates within 10% of measured values for nearly 90% of the observations. Conclusion: Our final model, that predicts visceral fat from CT scans, is highly parsimonious involving only seven variables that can be easily collected in the clinical setting. This practical equation might help improve the estimation of visceral fat in order to determine who is at greatest risk for CVD and who might benefit most from weight loss interventions.