Sabrina Lyngbye Wendt

Validation of a novel automatic sleep spindle detector with high performance during sleep in middle aged subjects

Many of the automatic sleep spindle detectors currently used to analyze sleep EEG are either validated on young subjects or not validated thoroughly. The purpose of this study is to develop and validate a fast and reliable sleep spindle detector with high performance in middle aged subjects. An automatic sleep spindle detector using a bandpass filtering approach and a time varying threshold was developed. The validation was done on sleep epochs from EEG recordings with manually scored sleep spindles from 13 healthy subjects with a mean age of 57.9 ± 9.7 years. The sleep spindle detector reached a mean sensitivity of 84.6% and a mean specificity of 95.3%. The sleep spindle detector can be used to obtain measures of spindle count and density together with quantitative measures such as the mean spindle frequency, mean spindle amplitude, and mean spindle duration.

Cross-Validation of a Glucose-Insulin-Glucagon Pharmacodynamics Model for Simulation Using Data From Patients With Type 1 Diabetes:

Background: Currently, no consensus exists on a model describing endogenous glucose production (EGP) as a function of glucagon concentrations. Reliable simulations to determine the glucagon dose preventing or treating hypoglycemia or to tune a dual-hormone artificial pancreas control algorithm need a validated glucoregulatory model including the effect of glucagon. Methods: Eight type 1 diabetes (T1D) patients each received a subcutaneous (SC) bolus of insulin on four study days to induce mild hypoglycemia followed by a SC bolus of saline or 100, 200, or 300 µg of glucagon. Blood samples were analyzed for concentrations of glucagon, insulin, and glucose. We fitted pharmacokinetic (PK) models to insulin and glucagon data using maximum likelihood and maximum a posteriori estimation methods. Similarly, we fitted a pharmacodynamic (PD) model to glucose data. The PD model included multiplicative effects of insulin and glucagon on EGP. Bias and precision of PD model test fits were assessed by mean predictive error (MPE) and mean absolute predictive error (MAPE). Results: Assuming constant variables in a subject across nonoutlier visits and using thresholds of ±15% MPE and 20% MAPE, we accepted at least one and at most three PD model test fits in each of the seven subjects. Thus, we successfully validated the PD model by leave-one-out cross-validation in seven out of eight T1D patients. Conclusions: The PD model accurately simulates glucose excursions based on plasma insulin and glucagon concentrations. The reported PK/PD model including equations and fitted parameters allows for in silico experiments that may help improve diabetes treatment involving glucagon for prevention of hypoglycemia.

Comparison of three nonlinear filters for fault detection in continuous glucose monitors

The purpose of this study is to compare the performance of three nonlinear filters in online drift detection of continuous glucose monitors. The nonlinear filters are the extended Kalman filter (EKF), the unscented Kalman filter (UKF), and the particle filter (PF). They are all based on a nonlinear model of the glucose-insulin dynamics in people with type 1 diabetes. Drift is modelled by a Gaussian random walk and is detected based on the statistical tests of the 90-min prediction residuals of the filters. The unscented Kalman filter had the highest average F score of 85.9%, and the smallest average detection delay of 84.1%, with the average detection sensitivity of 82.6%, and average specificity of 91.0%.The purpose of this study is to compare the performance of three nonlinear filters in online drift detection of continuous glucose monitors. The nonlinear filters are the extended Kalman filter (EKF), the unscented Kalman filter (UKF), and the particle filter (PF). They are all based on a nonlinear model of the glucose-insulin dynamics in people with type 1 diabetes. Drift is modelled by a Gaussian random walk and is detected based on the statistical tests of the 90-min prediction residuals of the filters. The unscented Kalman filter had the highest average F score of 85.9%, and the smallest average detection delay of 84.1%, with the average detection sensitivity of 82.6%, and average specificity of 91.0%.

Relationship between Optimum Mini-doses of Glucagon and Insulin Levels when Treating Mild Hypoglycaemia in Patients with Type 1 Diabetes - A Simulation Study

Hypoglycaemia remains the main limiting factor in type 1 diabetes management. We developed an insulin‐dependent glucagon dosing regimen for treatment of mild hypoglycaemia based on simulations. A validated glucose–insulin–glucagon model was used to describe seven virtual patients with insulin pump‐treated type 1 diabetes. In each simulation, one of ten different and individualized subcutaneous insulin boluses was administered to decrease plasma glucose (PG) from 7.0 to ≤3.9 mmol/l. Insulin levels were estimated as ratio of actual to baseline serum insulin concentration (se/ba‐insulin), insulin on board (IOB) or percentage of IOB to total daily insulin dose (IOB/TDD). Insulin bolus sizes were chosen to provide pre‐defined insulin levels when PG reached 3.9 mmol/l, where one of 17 subcutaneous glucagon boluses was administered. Optimum glucagon bolus to treat mild hypoglycaemia at varying insulin levels was the lowest dose that in most patients caused PG peak between 5.0 and 10.0 mmol/l and sustained PG ≥ 3.9 mmol/l for 2 hr after the bolus. PG response to glucagon declined with increasing insulin levels. The glucagon dose to optimally treat mild hypoglycaemia depended exponentially on insulin levels, regardless of how insulin was estimated. A 125‐μg glucagon dose was needed to optimally treat mild hypoglycaemia when insulin levels were equal to baseline levels. In contrast, glucagon doses >500 μg were needed when se/ba‐insulin >2.5, IOB >2.0 U or IOB/TDD >6%. Although the proposed model‐based glucagon regimen needs confirmation in clinical trials, this is the first attempt to develop an insulin‐dependent glucagon dosing regimen for treatment of insulin‐induced mild hypoglycaemia in patients with type 1 diabetes.

O36: Sleep spindle scoring: performance of humans versus machines

Question: What is the agreement in spindle scoring within, between and among experts? How does spindle scoring by humans compare to automated spindle scoring algorithms? Methods: We crowd-sourced the collection of spindle scorings from 24 experts in a large and varied dataset of EEG (C3-M2) from 110 middle-aged sleeping subjects. Epochs were scored by an average of 5.3 unique experts. Two experts scored parts of the dataset multiple times. We developed a simple method to build a large gold standard by establishing group consensus among expert scorers. We tested the performance of six previously published automated spindle detectors against the gold standard and refined methods of performance analysis for event detection. Results: We found an interrater agreement (F1-score) of 61±6% (Cohen’s Kappa (κ): 0.52±0.07) averaged over 24 expert pairs and an intrarater agreement of 72±7% (κ: 0.66±0.07) averaged over two experts. We tested the performance of individual experts to a gold standard compiled from all the expert scorers and found average agreement of 75±6% (κ: 0.68) over the 24 experts. We recompiled the gold standard and excluded the single expert whose performance was being assessed, and found an average agreement of 67±7% (κ: 0.59). Overall, we found the performance of human experts to be significantly better than the automated sleep spindle detectors we tested (maximum F1-score of detectors: 52%). Conclusions: Sleep spindle characteristics between subjects are very diverse which makes the scoring task difficult. The low interrater reliability suggests using more than one expert when scoring a dataset.