Predictive performance for population models using stochastic differential equations applied on data from an oral glucose tolerance test

Jonas Bech Møller, R.V. Overgaard, Henrik Madsen, T. Hansen, O. Pedersen, S.H. Ingwersen

    Research output: Contribution to journalJournal articleResearchpeer-review


    Several articles have investigated stochastic differential equations (SDEs) in PK/PD models, but few have quantitatively investigated the benefits to predictive performance of models based on real data. Estimation of first phase insulin secretion which reflects beta-cell function using models of the OGTT is a difficult problem in need of further investigation. The present work aimed at investigating the power of SDEs to predict the first phase insulin secretion (AIR(0-8)) in the IVGTT based on parameters obtained from the minimal model of the OGTT, published by Breda et al. (Diabetes 50(1): 150-158, 2001). In total 174 subjects underwent both an OGTT and a tolbutamide modified IVGTT. Estimation of parameters in the oral minimal model (OMM) was performed using the FOCE-method in NONMEM VI on insulin and C-peptide measurements. The suggested SDE models were based on a continuous AR(1) process, i.e. the Ornstein-Uhlenbeck process, and the extended Kalman filter was implemented in order to estimate the parameters of the models. Inclusion of the Ornstein-Uhlenbeck (OU) process caused improved description of the variation in the data as measured by the auto-correlation function (ACF) of one-step prediction errors. A main result was that application of SDE models improved the correlation between the individual first phase indexes obtained from OGTT and AIR(0-8) (r = 0.36 to r = 0.49 and r = 0.32 to r = 0.47 with C-peptide and insulin measurements, respectively). In addition to the increased correlation also the properties of the indexes obtained using the SDE models more correctly assessed the properties of the first phase indexes obtained from the IVGTT. In general it is concluded that the presented SDE approach not only caused autocorrelation of errors to decrease but also improved estimation of clinical measures obtained from the glucose tolerance tests. Since, the estimation time of extended models was not heavily increased compared to basic models, the applied method is concluded to have high relevance not only in theory but also in practice.
    Original languageEnglish
    JournalJournal of Pharmacokinetics and Pharmacodynamics
    Issue number1
    Pages (from-to)85-98
    Publication statusPublished - 2010


    • Oral minimal model (OMM)
    • Pharmacokinetic (PK)
    • Extended Kalman filter (EKF)
    • Ornstein-Uhlenbeck (OU)
    • Acute insulin response (AIR)
    • Stochastic differential equations (SDEs)
    • Autocorrelation function (ACF)
    • Intravenous glucose tolerance test (IVGTT)
    • Oral glucose tolerance test (OGTT)

    Fingerprint Dive into the research topics of 'Predictive performance for population models using stochastic differential equations applied on data from an oral glucose tolerance test'. Together they form a unique fingerprint.

    Cite this