Covariate Selection for the IVGTT Minimal Model of Glucose Disappearance
Paolo Denti (1), Alessandra Bertoldo (1), Paolo Vicini (2,3), Claudio Cobelli (1)
(1)Department of Information Engineering, University of Padova, Padova 35131 Italy; (2)Department of Bioengineering, University of Washington, Seattle, WA 98195; USA (3)Current address: Pfizer Global Research and Development, San Diego, CA 92121, USA
Objectives: Nonlinear mixed-effects modelling and its advantages have recently been discussed in the context of glucose-insulin metabolism [1,2,3] The research presented here aims at extending the use of these techniques, by proposing the introduction of covariates in the analysis of the IVGTT minimal model of glucose disappearance [4].
Methods: The dataset consists of IM-IVGTT on 204 healthy subjects (mean age 56 yrs, range 18-87; mean BMI 27 kg/m2, range 20-35) [5]. Besides blood samples, additional information about the patients was collected including age, gender, height, weight, body fat amount and distribution, basal glycemia and insulinemia. Given the high number of potential covariates, a standard step-wise procedure would have been unduly time-consuming, so a hybrid selection method was used. In a first step, the individual parameters provided by the base model (without covariates) were regressed on the covariates with a traditional linear regression to narrow down the pool of candidate models. In a further step, the most promising models were implemented in SPK [6] and ranked analysing the value of the objective function.
Results: Our method selects age, visceral abdominal fat and basal insulinemia as predictors for Insulin Sensitivity (SI), and age, total abdominal fat and basal insulinemia for the insulin kinetics parameter (P2). The predictors for the volume of distribution are age, gender, percentage of total body fat and basal glycemia. For glucose effectiveness (SG) our method selects height, weight and body surface area; however the actual physiologic significance of these covariates is not obvious. An alternative model for SG uses age and basal glycemia as predictors. For SI and P2 in particular, the incorporation of covariates results in a significant shrinking of the BSV. (from 70% to 44% and 51% to 39% respectively).
Conclusions: Our results support the hypothesis that the overall predictive power of the minimal model can be increased by the incorporation of easily, inexpensively and non-invasively collectible physiological information. We offer a starting point for further investigation about the significance of the relationships detected; issues that remain to be investigated include the role of collinearity in the predictors, especially for SG. Ultimately, this approach would provide a tool to allow the design of less invasive and less expensive protocols for epidemiological studies of the glucose disposal metabolic system.
References:
[1] Denti, P., Bertoldo, A., Vicini, P., Cobelli, C., Nonlinear Mixed Effects To Improve Glucose Minimal Model Parameter Estimation: A Simulation Study In Intensive and Sparse Sampling, IEEE Transactions on Biomedical Engineering, Accepted.
[2] Silber, H.E., Jauslin, P.M., Frey, N., Gieschke, R., Simonsson, U.S., and Karlsson, M.O., An integrated model for glucose and insulin regulation in healthy volunteers and type 2 diabetic patients following intravenous glucose provocations. J Clin Pharmacol, 2007. 47(9): p. 1159-71.
[3]Krudys, K.M., Kahn, S.E., and Vicini, P., Population approaches to estimate minimal model indexes of insulin sensitivity and glucose effectiveness using full and reduced sampling schedules. Am J Physiol Endocrinol Metab, 2006. 291(4): p. E716-23.
[4] Bergman, R.N., Ider, Y.Z., Bowden, C.R., and Cobelli, C., Quantitative estimation of insulin sensitivity. Am J Physiol, 1979. 236(6): p. E667-77.
[5] Basu, R., Dalla Man, C., Campioni, M., Basu, A., Klee, G., Toffolo, G., Cobelli, C., and Rizza, R.A., Effects of age and sex on postprandial glucose metabolism: differences in glucose turnover, insulin secretion, insulin action, and hepatic insulin extraction. Diabetes, 2006. 55(7): p. 2001-14.
[6] The RFPK team - University of Washington. System for Population Kinetics (SPK). Available from: http://spk.rfpk.washington.edu/.
