Evaluation and Development of the Dynamic Insulin Sensitivity and Secretion Test for Numerous Clinical Applications
Thesis DisciplineMechanical Engineering
Degree GrantorUniversity of Canterbury
Degree NameDoctor of Philosophy
Given the high and increasing social, health and economic costs of type 2 diabetes, early diagnosis and prevention are critical. Insulin sensitivity and insulin secretion are important etiological factors of type 2 diabetes and are used to define an individual’s risk or progression to the disease state. The dynamic insulin sensitivity and secretion test (DISST) concurrently measures insulin sensitivity and insulin secretion. The protocol uses glucose and insulin boluses as stimulus, and the participant response is observed during a relatively short protocol via glucose, insulin and C-peptide assays. In this research, the DISST insulin sensitivity value was successfully validated against the gold standard euglycaemic clamp with a high correlation (R=0.82), a high insulin resistance diagnostic equivalence (ROC c-unit=0.96), and low bias (-10.6%). Endogenous insulin secretion metrics obtained via the DISST were able to describe clinically important distinctions in participant physiology that were not observed with euglycaemic clamp, and are not available via most established insulin sensitivity tests. The quick dynamic insulin sensitivity test (DISTq) is a major extension of the DISST that uses the same protocol but uses only glucose assays. As glucose assays are usually available immediately, the DISTq is capable of providing insulin sensitivity results immediately after the final blood sample, creating a real-time clinical diagnostic. The DISTq correlated well with the euglycaemic clamp (R=0.76), had a high insulin resistance diagnostic equivalence (ROC c-unit=0.89), and limited bias (0.7%). These DISTq results meet or exceed the outcomes of most validation studies from established insulin sensitivity tests such as the IVGTT, HOMA and OGTT metrics. Furthermore, none of the established insulin sensitivity tests are capable of providing immediate or real-time results. Finally, and most of the established tests require considerably more intense clinical protocols than the DISTq. A range of DISST-based tests that used the DISST protocol and varying assay regimens were generated to provide optimum compromises for any given clinical or screening application. Eight DISST-based variants were postulated and assessed via their ability to replicate the fully sampled DISST results. The variants that utilised insulin assays correlated well to the fully sampled DISST insulin sensitivity values R~0.90 and the variants that assayed C-peptide produced endogenous insulin secretion metrics that correlated well to the fully-sampled DISST values (R~0.90 to 1). By taking advantage of the common clinical protocol, tests in the spectrum could be used in a hierarchical system. For example, if a DISTq result is close to a diagnostic threshold, stored samples could be re-assayed for insulin, and the insulin sensitivity value could be ‘upgraded’ without an additional protocol. Equally, adding C-peptide assays would provide additional insulin secretion information. Importantly, one clinical procedure thus yields potentially several test results. In-silico investigations were undertaken to evaluate the efficacy of two additional, specific DISTq protocol variations and to observe the pharmacokinetics of anti-diabetic drugs. The first variation combined the boluses used in the DISTq and reduced the overall test time to 20 minutes with only two glucose assays. The results of this investigation implied no significant degradation of insulin sensitivity values is caused by the change in protocol and suggested that clinical trials of this protocol are warranted. The second protocol variant added glucose content to the insulin bolus to enable observation of first phase insulin secretion concurrently with insulin sensitivity from glucose data alone. Although concurrent observation was possible without simulated assay noise, when clinically realistic noise was added, model identifiability was lost. Hence, this protocol is not recommended for clinical investigation. Similar analyses are used to apply the overall dynamic, model-based clinical test approach to other therapeutics. In-silico analysis showed that although the pharmacokinetics of insulin sensitizers drugs were described well by the dynamic protocol. However, the pharmacokinetics of insulin secretion enhancement drugs were less observable. The overall thesis is supported by a common model parameter identification method. The iterative integral parameter identification method is a development of a single, simple integral method. The iterative method was compared to the established non-linear Levenberg-Marquardt parameter identification method. Although the iterative integral method is limited in the type of models it can be used with, it is more robust, accurate and less computationally intense than the Levenberg-Marquardt method. Finally, a novel, integral-based method for the evaluation of a-priori structural model identifiability is also presented. This method differs significantly from established, derivative based approaches as it accounts for sample placement, measurement error, and probable system responses. Hence, it is capable of defining the true nature of identifiability, which is analogous, not binary as assumed by the established methods. The investigations described in this thesis were centred on model-based insulin sensitivity and secretion identification from dynamic insulin sensitivity tests with a strong focus on maximising clinical efficacy. The low intensity and informative DISST was successfully validated against the euglycaemic clamp. DISTq further reduces the clinical cost and burden, and was also validated against the euglycaemic clamp. DISTq represents a new paradigm in the field of low-cost insulin sensitivity testing as it does not require insulin assays. A number of in-silico investigations were undertaken and provided insight regarding the suitability of the methods for clinical trials. Finally, two novel mathematical methods were developed to identify model parameters and asses their identifiability, respectively.