DIET AND INSULIN SENSITIVITY: STATISTICAL METHODS

点击翻译按钮获取中文摘要

DESCRIPTION: Assessment of determinants of insulin sensitivity is a basic objective of research into diabetes and cardiovascular disease risk. Trade-offs between invasiveness and indirectness of insulin sensitivity (SI) measurement are a challenge for nutritional epidemiology and clinical trials. While the glucose clamp is highly accurate, it is too invasive and expensive t use in large studies. The oral glucose tolerance test (OGTT) is much more practical but requires complex modeling with numerous assumptions to yield commonly used estimates of insulin sensitivity. Various OGTT SI algorithms based on compartmental models and associated differential equations have been published, but there are indications that widely cited methods cannot be used effectively in the context of ongoing or incipient metabolic disease. The NHLBI OMNI-Carb study was a controlled feeding study of 163 overweight or obese participants at elevated risk for diabetes or CVD in Baltimore and Boston. All participants received a random sequence of four five-week diets in a crossover factorial design with high or low glycemic index and high or low carbohydrate defining the main experimental factors; a two-hour, seven sample OGTT protocol was performed at baseline and at the end of each diet period. Full OGTT series on all diets were obtained for 135 participants. Application of the oral minimal model based estimation (Della Man et al AJP 2004) indicated that estimation of SI was numerically unstable on a nontrivial fraction of participants in the sense that estimates were dependent on convergence of solution processes that were sensitive to arbitrarily chosen starting values; convergence to biologically plausible solutions was also not achievable for a nontrivial fraction of series. The proposed project will use new procedures for assessing "practical" nonidentifiability of key parameters of the minimal model, and will apply newly developed Bayesian inference techniques to properly quantify uncertainty in inference on SI. We will also create and illustrate analytical methods that employ newly developed concepts in multivariate time series analysis to more powerfully discriminate features of OGTT series that correlate with exposures or health conditions. This work will help fully harvest existing and new OGTT databases and will facilitate harmonization of the growing body of research results that use OGTT to address the consequences of obesity, diabetes and cardiovascular disease epidemics by providing transparent computational solutions and more flexible and accessible analytic frameworks.