Independent cohort cross-validation of the real-time DISTq estimation of insulin sensitivity
Introduction
Insulin resistance (IR) has been widely accepted as a strong indicator of an individual's risk of type 2 diabetes (T2DM) [1], [2]. A longitudinal study of the pathogenesis of T2DM has shown that those subjects who were diagnosed with T2DM had a 60% higher IR than average when assessed 10 years earlier [3]. IR is thus a strong predictor of T2DM risk and cardiovascular disease [4]. Therefore, low-cost, accurate estimation of IR could be used to screen patients, monitor interventional lifestyle changes, and to guide other therapies that could drastically reduce the incidence and cost associated with T2DM [5].
The various tests used to estimate insulin sensitivity (SI, SI = IR−1) use various methods to provoke and measure the subject's glycaemic responses [6], [7]. The euglycemic hyperinsulinaemic clamp (EIC) aims to suppress endogenous glucose production (EGP) and significantly suppress endogenous insulin production (Uen) to assess tissue sensitivity to exogenous insulin [8]. In contrast, the intravenous glucose tolerance test (IVGTT) stimulates Uen to measure insulin sensitivity [9]. Hence, while the metrics obtained by these tests are similar, they are not equivalent. An ideal metric for clinical or diagnostic use would measure the efficiency of insulin to dispose of glucose to the periphery at physiologically relevant glucose and insulin concentrations.
The gold standard for SI testing is the EIC. It measures the rate of glucose disposal at basal glucose, driven by hyper-physiological insulin concentrations designed to suppress Uen. The EIC is accurate and repeatable [8], but takes 4–5 h and approximately 6 clinician hours to perform. The time, intensity and cost of the EIC prohibit its use in many clinical situations. A reliable result is not necessarily guaranteed with an inexperienced clinician.
The IVGTT measures the subject's response to a 20–25 g intravenous (IV) glucose bolus with very frequent 1–3 min sampling. Some protocols modify the response with a 2–3 U IV bolus of insulin following the glucose bolus (IM-IVGTT) [7]. SI is then typically obtained by fitting the minimal model [10] to the sampled data. The boluses in this test tend to be supra-physiological and the trial generally runs for 2–3 h requiring significant clinical effort due to the frequent sampling. Model parameters are often unidentifiable, in particular in subjects with low SI values [11], [12].
Lower cost and lower intensity surrogate tests include fasting glucose, 2-h oral glucose tolerance (2hr OGTT) and the homeostasis model assessment (HOMA). Fasting glucose allows a diagnosis of T2DM [13], but does not offer an estimate of SI. Elevated fasting glucose is a resulting symptom of significant IR and the inability to maintain glycaemic homeostasis. Once elevated fasting glucose is detected, significant, and often irreversible beta-cell damage has already occurred [14]. Hence, fasting glucose is not an effective screening tool for early risk diagnosis to prevent further disease development.
The 2hr OGTT measures the subject's ability to dispose a 75 g oral glucose load. Two hours after ingestion of the glucose load the blood glucose concentration is measured for a T2DM diagnostic. The accuracy of the 2hr OGTT is questionable, with studies finding intra-subject reclassification of diagnosis rates of 50–60% [15], [16], [17]. Similarly to a fasting glucose level, early diagnosis of risk factors prior to the development of T2DM is difficult with the OGTT.
HOMA multiplies the fasting insulin and glucose assay values from a single blood test to produce a surrogate estimation of IR. The underlying assumption is that subjects with low sensitivity will require more insulin to maintain glycaemic homeostasis, elevating fasting levels of glucose and/or insulin. This test has an inconsistent correlation with the clamp (R = −0.19 → R = −0.82) [18], [19], does not track changes from intervention well [20], and does not fully represent insulin–glucose dynamics at physiologically relevant concentrations.
The dynamic insulin sensitivity test (DIST) is a short, infrequently sampled, low-dose intravenous glucose tolerance test. The test takes 30–45 min to administer. Glucose, insulin and C-peptide data is used with a clinically validated physiological model [20], [21] to provide accurate estimates for Uen, insulin clearance rate, and SI. The DIST has shown good correlation to the EIC in virtual trials (R = 0.93) [20], and high repeatability in a clinical pilot study (Δ = 6%) [21], with a validation study ongoing.
This study presents the DISTq (quick DIST) which is an alternative method for solving DIST data using only glucose samples and the subjects’ physical attributes (height, weight, sex, and age). Glucose samples can be assayed at the test station during sampling, enabling low-cost test analysis. As no insulin or C-peptide assays are required, the DISTq can effectively provide SI immediately in “real-time”. To remove the need for insulin and C-peptide assays, the insulin concentrations in plasma and interstitium must be estimated using knowledge available at testing. Parameter relationships derived from the fully sampled clinical DIST pilot study data [21] can be used to generate the required estimates [22]. These parameter estimates are used with the physiological model shown in Fig. 1 to simulate an interstitial insulin profile with sufficient accuracy to identify SI (R = 0.86 to the fully sampled DIST) [22].
In this research study, the validity of the DISTq assumptions is tested on two separate cohorts. One cohort is used to generate the insulin estimation functions, which are then tested on both cohorts. The goal is to assess how applicable and valid these estimations are across cohorts, and thus estimate or identify any additional errors in using this.
Section snippets
Model
The DISTq method utilises only the glucose and anatomical data (height, weight, sex and age) from each subject as used in the previously published DIST protocol [20], [21] to identify model-based insulin sensitivity (SI). The model is defined:where k1, k2, k3, nK, nL, and nC are transport rate parameters [min−1]; nI, is the transport rate between plasma and interstitium [L
Pilot cohort derived parameter estimates
The pilot data yielded parameter-estimation graphs sufficiently accurate to enable a correlation of R = 0.890 between the fully sampled DIST method and the DISTq method. Furthermore, the correlation of DIST and DISTq SI for the second cohort using these same estimations was R = 0.825. The pilot cohort showed good equivalence with fully sampled data sets with a gradient of 1.049, whereas the second cohort gradient showed a significant shift or bias in magnitude (grad = 1.507). Fig. 6 shows the
Discussion
The pilot-derived correlation (R = 0.890) confirms the primary assertion that a physiologically relevant SI metric can be identified using only anatomical data and glucose samples, without requiring insulin and C-peptide measurements. This approach enables relatively low-cost, immediate or real-time identification of SI. Furthermore, the correlation between DIST SI and DISTq SI improved compared to the previously presented DISTq method [22] which used an ad-hoc method of Uen simulation. Overall,
Conclusion
The DISTq method allows real-time, low-cost SI prediction using participant anatomical data (height, weight sex and age), the DIST protocol, a series of population-based parameter-estimation equations, and the iterative integral identification method. The method produces SI metrics that highly correlate with the fully sampled DIST test.
The addition of the second cohort confirmed the applicability of the test in cohorts isolated from the development cohort via cross-validation. Cross-validation
References (39)
- et al.
Role of glucose and insulin resistance in development of type 2 diabetes mellitus: results of a 25-year follow-up study
Lancet
(1992) - et al.
Methods for clinical assessment of insulin sensitivity and beta-cell function
Best Pract. Res. Clin. Endocrinol. Metab.
(2003) - et al.
MINMOD: a computer program to calculate insulin sensitivity and pancreatic responsivity from the frequently sampled intravenous glucose tolerance test
Comput. Methods Programs Biomed.
(1986) - et al.
Monte Carlo analysis of a new model-based method for insulin sensitivity testing
Comput. Methods Programs Biomed.
(2008) - et al.
Integral-based parameter identification for long-term dynamic verification of a glucose–insulin system model
Comput. Methods Programs Biomed.
(2005) - et al.
Are there time and cost savings by using telemanagement for patients on intensified insulin therapy? A randomised, controlled trial
Comput. Methods Programs Biomed.
(2002) - et al.
Validation of the insulin sensitivity index (ISI0,120): comparison with other measures
Diabetes Res. Clin. Pract.
(2000) - et al.
Insulin release and peripheral sensitivity at the oral glucose tolerance test
Diabetes Res. Clin. Pract.
(1990) - et al.
Model-based glycaemic control in critical care—a review of the state of the possible
Biomed. Signal Process. Control
(2006) - et al.
Insulin resistance. A multifaceted syndrome responsible for NIDDM, obesity, hypertension, dyslipidemia, and atherosclerotic cardiovascular disease
Diabetes Care
(1991)
Insulin resistance is central to the burden of diabetes
Diabetes Metab. Rev.
Heterogeneity in prevalence of risk factors for cardiovascular disease and type 2 diabetes in obese individuals: impact of differences in insulin sensitivity
Arch. Int. Physiol. Biochim. Biophys.
Diagnosis, prognosis, and treatment of impaired glucose tolerance and impaired fasting glucose
Evid. Rep. Technol. Assess.
How to measure insulin sensitivity
J. Hypertens.
Glucose clamp technique: a method for quantifying insulin secretion and resistance
Am. J. Physiol.
Quantitative estimation of insulin sensitivity
Am. J. Physiol.
Undermodeling affects minimal model indexes: insights from a two-compartment model
Am. J. Physiol.
Minimal model S(I) = 0 problem in NIDDM subjects: nonzero Bayesian estimates with credible confidence intervals
Am. J. Physiol. Endocrinol. Metab.
Diagnostic criteria for diabetes mellitus
Rinsho Byori.
Cited by (23)
The goldilocks problem: Nutrition and its impact on glycaemic control
2021, Clinical NutritionCitation Excerpt :Per-patient BG outcomes are also reported in places, to remove any bias in overall result from patients who received insulin for longer periods. Finally, since STAR uses a personalised, time-varying and clinically well-validated insulin sensitivity metric (SI) to guide care [19,44–46], cohorts are compared in terms of SI (1/SI = insulin resistance) to quantify and account for differences in insulin sensitivity between cohorts. Cumulative distribution functions (CDFs) compare outcomes.
High Inter-Patient Variability in Sepsis Evolution: A Hidden Markov Model Analysis
2021, Computer Methods and Programs in BiomedicineCitation Excerpt :The patient-specific hour metabolic biomarker used has shown proven value in this type of analysis using kernel density estimates and other formats, such as neural networks [16,42,43,45]. In particular, the SI insulin sensitivity metric used here captures patient condition changes and variability due to changes inflammation analysis impacting metabolism but due to organ failure, inflammation, and sepsis (e.g. [59-67]). This metric thus has the potential to aid diagnosis and better discriminate the sepsis states considered when compared to using the bedside physiological signals alone [16].
Quantifying misclassification and bias errors due to hierarchical sepsis scores in real-time sepsis diagnosis
2020, Biomedical Signal Processing and ControlA subcutaneous insulin pharmacokinetic model for insulin Detemir
2019, Computer Methods and Programs in BiomedicineGlycemic control in the intensive care unit: A control systems perspective
2019, Annual Reviews in ControlCitation Excerpt :It also has a neonatal ICU (NICU) version based on the same dynamics scaled for the specific case (Dickson et al., 2016; Le Compte, Chase, et al., 2011; Le Compte, Chase, et al., 2010). It is driven by an insulin sensitivity parameter identifiable from bedside data (Docherty, Chase, & David, 2012; Docherty, Chase, Lotz, & Desaive, 2011), which can be used to monitor patient condition and its evolution over time (Docherty, Chase, Lotz, Hann, et al., 2011; Pretty et al., 2012; Sah Pri et al., 2014). It has been used in virtual patients (Chase, Suhaimi, et al., 2010; J. L. Dickson et al., 2018; Lonergan, LeCompte, et al., 2006), GC design (Fisk et al., 2012; Lonergan, LeCompte, et al., 2006; Lonergan, Compte, et al., 2006) and real-time GC (Dickson, Lynn, Shaw, & Chase, 2019; Fisk et al., 2012; Lonergan, LeCompte, et al., 2006; Lonergan, Compte, et al., 2006; Stewart et al., 2016).
Comparison of modulated and clinically set nutrition protocol's for STAR
2018, IFAC-PapersOnLine