Supervised and Unsupervised Approaches for the Real-Time Detection of Undesired Insulin Suspension Caused by Malfunctions.

BACKGROUND: Automated insulin delivery (AID) systems, permit improved treatment of type 1 diabetes (T1D). Unfortunately, malfunctioning in the insulin pump or in the infusion set can prevent insulin from being administered, reducing the AID efficacy and posing the patient at risk. Different data-driven methods available in the literature can be used to deal with the problem of automatically detecting complete insulin suspension in real-time. This article investigates both supervised and unsupervised strategies and proposes a fair comparison under either population or personalized settings. METHODS: Several algorithms are compared using data generated through the UVA/Padova T1D simulator, a computer simulator widely used to test control strategies in silico and accepted by the Food and Drugs Administration (FDA) as a substitute to animal pre-clinical trials. Two synthetic data sets, each consisting of 100 virtual subjects monitored for 1 month, were generated. Occasional faults of the insulin pump are simulated as complete occlusions by suspending the therapy administration. Personalized algorithms are investigated with unsupervised approaches only, since personalized labels are hardly available. RESULTS: In the population scenario, the supervised approach outperforms the unsupervised strategy. In particular, logistic regression and random forest achieves a recall of 72% and 82%, with 0.12 and 0.21 false positives (FP) per day, respectively. In the personalized setting scenario, the unsupervised algorithms are tailored on each patient and outperform the population ones, in particular isolation forest achieves a recall 80% and 0.06 FPs per day. CONCLUSIONS: This article suggests that unsupervised personalized approach, by addressing the large variability in glucose response among individuals with T1D, is superior to other one-fits-all approaches in detecting insulin suspensions caused by malfunctioning. Population methodologies can be effectively used while waiting to collect sufficient patient data, when the system is installed on a new patient.

Risk of hypoglycemia in type 1 diabetes management: An in-silico sensitivity analysis to assess and rank the quantitative impact of different behavioral factors.

BACKGROUND AND OBJECTIVE: In type 1 diabetes (T1D), a quantitative evaluation of the impact on hypoglycemia of suboptimal therapeutic decision (e.g. incorrect estimation of the ingested carbohydrates, inaccurate insulin timing, etc) is unavailable. Clinical trials to measure sensitivity to patient actions would be expensive, exposed to confounding factors and risky for the participants. In this work, a T1D patient decision simulator (T1D-PDS), realistically reproducing blood glucose dynamics in a large virtual population, is used to perform extensive in-silico trials and the so-derived data employed to implement a sensitivity analysis that ranks different behavioral factors based on their impact on a clinically meaningful parameter, the time below range (TBR). METHODS: Eleven behavioral factors impacting on hypoglycemia are considered. The T1D-PDS was used to perform multiple 2-week simulations involving 100 adults, by testing about 3500 different perturbations for nominal behavior. A local linear approximation of the function linking the TBR and the factors were computed to derive sensitivity indices (SIs), quantifying the impact of each factor on TBR variations. RESULTS: The obtained ranking quantifies importance of factors w.r.t. the others. Factors apparently related to hypoglycemia were correctly placed on the top of the ranking, including systematic (SI=2.05%) and random (SI=1.35%) carb-counting error, hypotreatment dose (SI=-1.21%), insulin bolus time w.r.t. mealtime (SI=1.09%). CONCLUSIONS: The obtained SIs allowed to rank behavioral factors based on their impact on TBR. The behavioral factors identified as most influential can be prioritized in patient training.

Bayesian denoising algorithm dealing with colored, non-stationary noise in continuous glucose monitoring timeseries.

Introduction: The retrospective analysis of continuous glucose monitoring (CGM) timeseries can be hampered by colored and non-stationary measurement noise. Here, we introduce a Bayesian denoising (BD) algorithm to address both autocorrelation of measurement noise and temporal variability of its variance. Methods: BD utilizes adaptive, a-priori models of signal and noise, whose unknown variances are derived on partially-overlapped CGM windows, via smoothing approach based on linear mean square estimation. The CGM signal and noise variability profiles are then reconstructed using a kernel smoother. BD is first assessed on two simulated datasets, DS1 and DS2. On DS1, the effectiveness of accounting for colored noise is evaluated by comparison against a literature algorithm; on DS2, the effectiveness of accounting for the noise variance temporal variability is evaluated by comparison against a Butterworth filter. BD is then evaluated on 15 CGM timeseries measured by the Dexcom G6 (DR). Results: On DS1, BD allows reducing the root-mean-square-error (RMSE) from 8.10 [6.79-9.24] mg/dL to 6.28 [5.47-7.27] mg/dL (median [IQR]); on DS2, RMSE decreases from 6.85 [5.50-8.72] mg/dL to 5.35 [4.48-6.49] mg/dL. On DR, BD performs a reasonable tracking of noise variance variability and a satisfactory denoising. Discussion: The new algorithm effectively addresses the nature of CGM measurement error, outperforming existing denoising algorithms.

The importance of interpreting machine learning models for blood glucose prediction in diabetes: an analysis using SHAP.

Machine learning has become a popular tool for learning models of complex dynamics from biomedical data. In Type 1 Diabetes (T1D) management, these models are increasingly been integrated in decision support systems (DSS) to forecast glucose levels and provide preventive therapeutic suggestions, like corrective insulin boluses (CIB), accordingly. Typically, models are chosen based on their prediction accuracy. However, since patient safety is a concern in this application, the algorithm should also be physiologically sound and its outcome should be explainable. This paper aims to discuss the importance of using tools to interpret the output of black-box models in T1D management by presenting a case-of-study on the selection of the best prediction algorithm to integrate in a DSS for CIB suggestion. By retrospectively "replaying" real patient data, we show that two long-short term memory neural networks (LSTM) (named p-LSTM and np-LSTM) with similar prediction accuracy could lead to different therapeutic decisions. An analysis with SHAP-a tool for explaining black-box models' output-unambiguously shows that only p-LSTM learnt the physiological relationship between inputs and glucose prediction, and should therefore be preferred. This is verified by showing that, when embedded in the DSS, only p-LSTM can improve patients' glycemic control.

Clinical evaluation of a decision support system for glucose infusion in hypoglycaemic clamp experiments.

AIM: To provide a preliminary evaluation of the accuracy and safety of Gluclas decision support system suggestions in a hypoglycaemic clamp study. METHODS: This analysis was performed using data from 32 participants (four groups with different glucose-insulin regulation: post Roux-en-Y gastric bypass with and without postprandial hypoglycaemia syndrome, postsleeve gastrectomy and non-operated controls) undergoing Gluclas-assisted hypoglycaemic clamps (target: 2.5 mmol/L for 20 minutes at 150 minutes after oral glucose ingestion). Gluclas provided glucose infusion rate suggestions upon manual entry of blood glucose values (every 5 minutes), which were either followed or overruled by investigators after critical review. Accuracy and safety were evaluated by mean absolute error (MAE), mean absolute percentage error (MAPE), average glucose level, coefficient of variation (CV) and minimal glucose level during the 20-minute hypoglycaemic period. RESULTS: Investigators accepted 84% of suggestions, with a mean deviation of 30.33 mg/min. During the hypoglycaemic period, the MAE was 0.16 (0.12-0.24) (median [interquartile range]) mmol/L and the MAPE was 6.12% (4.80%-9.29%). CV was 4.90% (3.58%-7.27%), with 5% considered the threshold for sufficient quality. The minimal glucose level was 2.40 (2.30-2.50) mmol/L. CONCLUSIONS: Gluclas achieved sufficiently high accuracy with minimal safety risks in a population with differences in glucose-insulin dynamics, underscoring its applicability to various patient groups.

A Markov Model of Gap Occurrence in Continuous Glucose Monitoring Data for Realistic in Silico Clinical Trials.

BACKGROUND AND OBJECTIVE: Continuous glucose monitoring (CGM) sensors measure interstitial glucose concentration every 1-5 min for days or weeks. New CGM-based diabetes therapies are often tested in in silico clinical trials (ISCTs) using diabetes simulators. Accurate models of CGM sensor inaccuracies and failures could help improve the realism of ISCTs. However, the modeling of CGM failures has not yet been fully addressed in the literature. This work aims to develop a mathematical model of CGM gaps, i.e., occasional portions of missing data generated by temporary sensor errors (e.g., excessive noise or artifacts). METHODS: Two datasets containing CGM traces collected in 167 adults and 205 children, respectively, using the Dexcom G6 sensor (Dexcom Inc., San Diego, CA) were used. Four Markov models, of increasing complexity, were designed to describe three main characteristics: number of gaps for each sensor, gap distribution in the monitoring days, and gap duration. Each model was identified on a portion of each dataset (training set). The remaining portion of each dataset (real test set) was used to evaluate model performance through a Monte Carlo simulation approach. Each model was used to generate 100 simulated test sets with the same size as the real test set. The distributions of gap characteristics on the simulated test sets were compared with those observed on the real test set, using the two-sample Kolmogorov-Smirnov test and the Jensen-Shannon divergence. RESULTS: A six-state Markov model, having two states to describe normal sensor operation and four states to describe gap occurrence, achieved the best results. For this model, the Kolmogorov-Smirnov test found no significant differences between the distribution of simulated and real gap characteristics. Moreover, this model obtained significantly lower Jensen-Shannon divergence values than the other models. CONCLUSIONS: A Markov model describing CGM gaps was developed and validated on two real datasets. The model describes well the number of gaps for each sensor, the gap distribution over monitoring days, and the gap durations. Such a model can be integrated into existing diabetes simulators to realistically simulate CGM gaps in ISCTs and thus enable the development of more effective and robust diabetes management strategies.

Counter-regulatory responses to postprandial hypoglycaemia in patients with post-bariatric hypoglycaemia vs surgical and non-surgical control individuals.

AIMS/HYPOTHESIS: Post-bariatric hypoglycaemia is an increasingly recognised complication of bariatric surgery, manifesting particularly after Roux-en-Y gastric bypass. While hyperinsulinaemia is an established pathophysiological feature, the role of counter-regulation remains unclear. We aimed to assess counter-regulatory hormones and glucose fluxes during insulin-induced postprandial hypoglycaemia in patients with post-bariatric hypoglycaemia after Roux-en-Y gastric bypass vs surgical and non-surgical control individuals. METHODS: In this case-control study, 32 adults belonging to four groups with comparable age, sex and BMI (patients with post-bariatric hypoglycaemia, Roux-en-Y gastric bypass, sleeve gastrectomy and non-surgical control individuals) underwent a postprandial hypoglycaemic clamp in our clinical research unit to reach the glycaemic target of 2.5 mmol/l 150-170 min after ingesting 15 g of glucose. Glucose fluxes were assessed during the postprandial and hypoglycaemic period using a dual-tracer approach. The primary outcome was the incremental AUC of glucagon during hypoglycaemia. Catecholamines, cortisol, growth hormone, pancreatic polypeptide and endogenous glucose production were also analysed during hypoglycaemia. RESULTS: The rate of glucose appearance after oral administration, as well as the rates of total glucose appearance and glucose disappearance, were higher in both Roux-en-Y gastric bypass groups vs the non-surgical control group in the early postprandial period (all p<0.05). During hypoglycaemia, glucagon exposure was significantly lower in all surgical groups vs the non-surgical control group (all p<0.01). Pancreatic polypeptide levels were significantly lower in patients with post-bariatric hypoglycaemia vs the non-surgical control group (median [IQR]: 24.7 [10.9, 38.7] pmol/l vs 238.7 [186.3, 288.9] pmol/l) (p=0.005). Other hormonal responses to hypoglycaemia and endogenous glucose production did not significantly differ between the groups. CONCLUSIONS/INTERPRETATION: The glucagon response to insulin-induced postprandial hypoglycaemia is lower in post-bariatric surgery individuals compared with non-surgical control individuals, irrespective of the surgical modality. No significant differences were found between patients with post-bariatric hypoglycaemia and surgical control individuals, suggesting that impaired counter-regulation is not a root cause of post-bariatric hypoglycaemia. TRIAL REGISTRATION: ClinicalTrials.gov NCT04334161.

Combined Use of Glucose-Specific Model Identification and Alarm Strategy Based on Prediction-Funnel to Improve Online Forecasting of Hypoglycemic Events.

BACKGROUND: Advanced decision support systems for type 1 diabetes (T1D) management often embed prediction modules, which allow T1D people to take preventive actions to avoid critical episodes like hypoglycemia. Real-time prediction of blood glucose (BG) concentration relies on a subject-specific model of glucose-insulin dynamics. Model parameter identification is usually based on the mean square error (MSE) cost function, and the model is usually used to predict BG at a single prediction horizon (PH). Finally, a hypo-alarm is raised if the predicted BG crosses a threshold. This work aims to show that real-time hypoglycemia forecasting can be improved by leveraging: a glucose-specific mean square error (gMSE) cost function in model's parameters identification, and a "prediction-funnel," that is, confidence intervals (CIs) for multiple PHs, within the hypo-alarm-raising strategy. METHODS: Autoregressive integrated moving average with exogenous input (ARIMAX) models are selected to illustrate the proposed solution (use of gMSE and prediction-funnel) and its assessment against the conventional approach (MSE and single PH). The gMSE penalizes the model misfit in unsafe BG ranges (e.g., hypoglycemia), and the prediction-funnel allows raising an alarm by monitoring if the CIs cross a suitable threshold. The algorithms were evaluated by measuring precision (P), recall (R), F1-score (F1), false positive per day (FP/day), and time gain (TG) on a real dataset collected in 11 T1D individuals. RESULTS: The best performance is achieved exploiting both the gMSE and the prediction-funnel: P = 65%, R = 88%, F1 = 75%, FP/day = 0.29, and mean TG = 15 minutes. CONCLUSIONS: The combined use of a glucose-specific metric and an alarm-raising strategy based on the prediction-funnel allows achieving a more effective and reliable hypoglycemia prediction algorithm.

Assessment of Seasonal Stochastic Local Models for Glucose Prediction without Meal Size Information under Free-Living Conditions.

Accurate blood glucose (BG) forecasting is key in diabetes management, as it allows preventive actions to mitigate harmful hypoglycemic/hyperglycemic episodes. Considering the encouraging results obtained by seasonal stochastic models in proof-of-concept studies, this work assesses the methodology in two datasets (open-loop and closed-loop) recorded in free-living conditions. First, similar postprandial glycemic profiles are grouped together with fuzzy C-means clustering. Then, a seasonal stochastic model is identified for each cluster. Finally, real-time BG forecasting is performed by weighting each model's prediction. The proposed methodology (named C-SARIMA) is compared to other linear and nonlinear black-box methods: autoregressive integrated moving average (ARIMA), its variant with input (ARIMAX), a feed-forward neural network (NN), and its modified version (NN-X) fed by BG, insulin, and carbohydrates (timing and dosing) information for several prediction horizons (PHs). In the open-loop dataset, C-SARIMA grants a median root-mean-squared error (RMSE) of 20.13 mg/dL (PH = 30) and 27.23 mg/dL (PH = 45), not significantly different from ARIMA and NN. Over a longer PH, C-SARIMA achieves an RMSE = 31.96 mg/dL (PH = 60) and RMSE = 33.91 mg/dL (PH = 75), significantly outperforming the ARIMA and NN, without significant differences from the ARIMAX for PH ≥ 45 and the NN-X for PH ≥ 60. Similar results hold on the closed-loop dataset: for PH = 30 and 45 min, the C-SARIMA achieves an RMSE = 21.63 mg/dL and RMSE = 29.67 mg/dL, not significantly different from the ARIMA and NN. On longer PH, the C-SARIMA outperforms the ARIMA for PH > 45 and the NN for PH > 60 without significant differences from the ARIMAX for PH ≥ 45. Although using less input information, the C-SARIMA achieves similar performance to other prediction methods such as the ARIMAX and NN-X and outperforming the CGM-only approaches on PH > 45min.

Data-Driven Supervised Compression Artifacts Detection on Continuous Glucose Sensors.

Continuous Glucose Monitoring (CGM) sensors micro-invasively provide frequent glucose readings, improving the management of Type 1 diabetic patients' life and making available reach data-sets for retrospective analysis. Unlikely, CGM sensors are subject to failures, such as compression artifacts, that might impact on both real-time and respective CGM use. In this work is focused on retrospective detection of compression artifacts. An in-silico dataset is generated using the T1D UVa/Padova simulator and compression artifacts are subsequently added in known position, thus creating a dataset with perfectly accurate faulty/not-faulty labels. The problem of compression artifact detection is then faced with supervised data-driven techniques, in particular using Random Forest algorithm. The detection performance guaranteed by the method on in-silico data is satisfactory, opening the way for further analysis on real-data.

Choosing the duration of continuous glucose monitoring for reliable assessment of time in range: A new analytical approach to overcome the limitations of correlation-based methods.

AIMS: Reliable estimation of the time spent in different glycaemic ranges (time-in-ranges) requires sufficiently long continuous glucose monitoring. In a 2019 paper (Battelino et al., Clinical targets for continuous glucose monitoring data interpretation: recommendations from the international consensus on time in range. Diabetes Care. 2019;42:1593-1603), an international panel of experts suggested using a correlation-based approach to obtain the minimum number of days for reliable time-in-ranges estimates. More recently (in Camerlingo et al., Design of clinical trials to assess diabetes treatment: minimum duration of continuous glucose monitoring data to estimate time-in-ranges with the desired precision. Diabetes Obes Metab. 2021;23:2446-2454) we presented a mathematical equation linking the number of monitoring days to the uncertainty around time-in-ranges estimates. In this work, we compare these two approaches, mainly focusing on time spent in (70-180) mg/dL range (TIR). METHODS: The first 100 and 150 days of data were extracted from study A (148 subjects, ~180 days), and the first 100, 150, 200, 250 and 300 days of data from study B (45 subjects, ~365 days). For each of these data windows, the minimum monitoring duration was computed using correlation-based and equation-based approaches. The suggestions were compared for the windows of different durations extracted from the same study, and for the windows of equal duration extracted from different studies. RESULTS: When changing the dataset duration, the correlation-based approach produces inconsistent results, ranging from 23 to 64 days, for TIR. The equation-based approach was found to be robust versus this issue, as it is affected only by the characteristics of the population being monitored. Indeed, to grant a confidence interval of 5% around TIR, it suggests 18 days for windows from study A, and 17 days for windows from study B. Similar considerations hold for other time-in-ranges. CONCLUSIONS: The equation-based approach offers advantages for the design of clinical trials having time-in-ranges as final end points, with focus on trial duration.

Machine Learning-Based Anomaly Detection Algorithms to Alert Patients Using Sensor Augmented Pump of Infusion Site Failures.

BACKGROUND: Personal insulin pumps have shown to be effective in improving the quality of therapy for people with type 1 diabetes (T1D). However, the safety of this technology is limited by the possible infusion site failures, which are linked with hyperglycemia and ketoacidosis. Thanks to the large availability of collected data provided by modern therapeutic technologies, machine learning algorithms have the potential to provide new way to identify failures early and avert adverse events. METHODS: A clinical dataset (N = 20) is used to evaluate a novel method for detecting real-time infusion site failures using unsupervised anomaly detection algorithms, previously proposed and developed on in-silico data. An adapted feature engineering procedure is introduced to make the method able to operate in the absence of a closed-loop (CL) system and meal announcements. RESULTS: In the optimal configuration, we obtained a performance of 0.75 Sensitivity (15 out of 20 total failures detected) and 0.08 FP/day, outperforming previously proposed literature algorithms. The algorithm was able to anticipate the replacement of the malfunctioning infusion sets by ~2 h on average. CONCLUSIONS: On the considered dataset, the proposed algorithm showed the potential to improve the safety of patients treated with sensor-augmented pump systems.

Impact of Carbohydrate Counting Error on Glycemic Control in Open-Loop Management of Type 1 Diabetes: Quantitative Assessment Through an In Silico Trial.

BACKGROUND: In the management of type 1 diabetes (T1D), systematic and random errors in carb-counting can have an adverse effect on glycemic control. In this study, we performed an in silico trial aiming at quantifying the impact of different levels of carb-counting error on glycemic control. METHODS: The T1D patient decision simulator was used to simulate 7-day glycemic profiles of 100 adults using open-loop therapy. The simulation was repeated for different values of systematic and random carb-counting errors, generated with Gaussian distribution varying the error mean from -10% to +10% and standard deviation (SD) from 0% to 50%. The effect of the error was evaluated by computing the difference of time inside (∆TIR), above (∆TAR) and below (∆TBR) the target glycemic range (70-180mg/dl) compared to the reference case, that is, absence of error. Finally, 3 linear regression models were developed to mathematically describe how error mean and SD variations result in ∆TIR, ∆TAR, and ∆TBR changes. RESULTS: Random errors globally deteriorate the glycemic control; systematic underestimations lead to, on average, up to 5.2% more TAR than the reference case, while systematic overestimation results in up to 0.8% more TBR. The different time in range metrics were linearly related with error mean and SD (R2>0.95), with slopes of ßMEAN=0.21, ßSD=-0.07 for ∆TIR, ßMEAN=-0.25, ßSD=+0.06 for ∆TAR, and ßMEAN=0.05, ßSD=+0.01 for ∆TBR. CONCLUSIONS: The quantification of carb-counting error impact performed in this work may be useful understanding causes of glycemic variability and the impact of possible therapy adjustments or behavior changes in different glucose metrics.

Linear Model Identification for Personalized Prediction and Control in Diabetes.

OBJECTIVE: Type-1 diabetes (T1D) is a disease characterized by impaired blood glucose (BG) regulation, forcing patients to multiple daily therapeutic actions, including insulin administration. T1D management could considerably benefit of accurate BG predictions and automated insulin delivery. For both tasks, the large inter- and intra-individual variability in glucose response represents a major challenge. This work investigates different techniques to learn individualized linear models of glucose response to insulin and meal, suitable for model-based prediction and control. METHODS: We focus on data-driven techniques for linear model-learning and compare the state-of-art parametric pipeline with a novel non-parametric approach based on Gaussian regression and Stable-Spline kernel. On data collected by 11 T1D individuals, the effectiveness of different models was evaluated by measuring root mean squared error (RMSE), coefficient of determination (COD), and time gain associated with BG predictors. RESULTS: Among the tested approaches, the non-parametric technique results in the best prediction performance: median RMSE = 29.8 mg/dL, and median COD = 57.4%, for a prediction horizon (PH) of 60 min. With respect to the state-of-the-art parametric techniques, the non-parametric approach grants a COD improvement of about 2%, 7%, 21%, and 41% for PH = 30, 60, 90, and 120 min (paired-sample t-test p ≤ 0.001, p = 0.003, p = 0.03, and p = 0.07 respectively). CONCLUSION: Non-parametric linear model-learning grants statistically significant improvement with respect to the state-of-art parametric approach. The higher PH, the more pronounced the improvement. SIGNIFICANCE: The use of a linear non-parametric model-learning approach for model-based prediction and control could bring to a more prompt, safe and effective T1D management.

Mathematical modelling of SigE regulatory network reveals new insights into bistability of mycobacterial stress response.

BACKGROUND: The ability to rapidly adapt to adverse environmental conditions represents the key of success of many pathogens and, in particular, of Mycobacterium tuberculosis. Upon exposition to heat shock, antibiotics or other sources of stress, appropriate responses in terms of genes transcription and proteins activity are activated leading part of a genetically identical bacterial population to express a different phenotype, namely to develop persistence. When the stress response network is mathematically described by an ordinary differential equations model, development of persistence in the bacterial population is associated with bistability of the model, since different emerging phenotypes are represented by different stable steady states. RESULTS: In this work, we develop a mathematical model of SigE stress response network that incorporates interactions not considered in mathematical models currently available in the literature. We provide, through involved analytical computations, accurate approximations of the system's nullclines, and exploit the obtained expressions to determine, in a reliable though computationally efficient way, the number of equilibrium points of the system. CONCLUSIONS: Theoretical analysis and perturbation experiments point out the crucial role played by the degradation pathway involving RseA, the anti-sigma factor of SigE, for coexistence of two stable equilibria and the emergence of bistability. Our results also indicate that a fine control on RseA concentration is a necessary requirement in order for the system to exhibit bistability.

Toward an Optimal Definition of Hypoglycemia with Continuous Glucose Monitoring.

BACKGROUND AND OBJECTIVE: As continuous glucose monitoring (CGM) becomes common in research and clinical practice, there is a need to understand how CGM-based hypoglycemia relates to hypoglycemia episodes defined conventionally as patient reported hypoglycemia (PRH). Data show that CGM identify many episodes of low interstitial glucose (LIG) that are not experienced by patients, and so the aim of this study is to use different PRH simulations to optimize CGM parameters of threshold (h) and duration (d) to provide the best PRH detection performance. METHODS: The algorithm uses particle Markov chain Monte Carlo optimization to identify the optimal h and d which maximize an objective function for detecting PRH. We tested our algorithm by creating three different cases of PRH simulations. RESULTS: We added three types of simulated PRH events to 10 weeks of anonymized CGM data from 96 type 1 diabetes people to see if the algorithm can detect the optimal parameters set out in the simulations. In simulation 1, we changed the locations of PRHs with respect to LIG episodes in the CGM signal to simulate random optimal LIG parameters for every individual. In simulation 2, the PRHs are CGM glucose <3.9 mmol/L followed by at least 20 min of rise > 0.11 mmol/L/min. Simulation 3 is like simulation 2 but with glucose threshold of 3.0 mmol/L. The median [interquartile range] of deviation between the optimized (found by the algorithm) and the optimal (known) h and d are -0.07% [-0.4, 1.9] and -1.3% [-5.9, 6.8], respectively across the subjects for simulation 1. The mean [min max] of the optimized LIG parameters are h = 3.8 [3.7, 3.8] mmol/L and d = 12 [10, 14] min for simulation 2 and they are h = 3.0 [2.9, 3] mmol/L and d = 10 [8, 14] min for simulation 3 across a 10-fold cross validation. CONCLUSIONS: This work demonstrates the feasibility of the algorithm to find the best-fit definition of CGM-based hypoglycemia for PRH detection. In a prospective clinical study collecting CGM and PRH, the current algorithm will be used to optimize the definition of hypoglycemia with respect to PRH with the ambition of using the resulted definition as a surrogate for PRH in clinical practice.

Design of clinical trials to assess diabetes treatment: Minimum duration of continuous glucose monitoring data to estimate time-in-ranges with the desired precision.

AIM: To compute the uncertainty of time-in-ranges, such as time in range (TIR), time in tight range (TITR), time below range (TBR) and time above range (TAR), to evaluate glucose control and to determine the minimum duration of a trial to achieve the desired precision. MATERIALS AND METHODS: Four formulas for the aforementioned time-in-ranges were obtained by estimating the equation's parameters on a training set extracted from study A (226 subjects, ~180 days, 5-minute Dexcom G4 Platinum sensor). The formulas were then validated on the remaining data. We also illustrate how to adjust the parameters for sensors with different sampling rates. Finally, we used study B (45 subjects, ~365 days, 15-minute Abbott Freestyle Libre sensor) to further validate our results. RESULTS: Our approach was effective in predicting the uncertainty when time-in-ranges are estimated using n days of continuous glucose monitoring (CGM), matching the variability observed in the data. As an example, monitoring a population with TIR = 70%, TITR = 50%, TBR = 5% and TAR = 25% for 30 days warrants a precision of ±3.50%, ±3.68%, ±1.33% and ±3.66%, respectively. CONCLUSIONS: The presented approach can be used to both compute the uncertainty of time-in-ranges and determine the minimum duration of a trial to achieve the desired precision. An online tool to facilitate its implementation is made freely available to the clinical investigator.

Forecasting of Glucose Levels and Hypoglycemic Events: Head-to-Head Comparison of Linear and Nonlinear Data-Driven Algorithms Based on Continuous Glucose Monitoring Data Only.

In type 1 diabetes management, the availability of algorithms capable of accurately forecasting future blood glucose (BG) concentrations and hypoglycemic episodes could enable proactive therapeutic actions, e.g., the consumption of carbohydrates to mitigate, or even avoid, an impending critical event. The only input of this kind of algorithm is often continuous glucose monitoring (CGM) sensor data, because other signals (such as injected insulin, ingested carbs, and physical activity) are frequently unavailable. Several predictive algorithms fed by CGM data only have been proposed in the literature, but they were assessed using datasets originated by different experimental protocols, making a comparison of their relative merits difficult. The aim of the present work was to perform a head-to-head comparison of thirty different linear and nonlinear predictive algorithms using the same dataset, given by 124 CGM traces collected over 10 days with the newest Dexcom G6 sensor available on the market and considering a 30-min prediction horizon. We considered the state-of-the art methods, investigating, in particular, linear black-box methods (autoregressive; autoregressive moving-average; and autoregressive integrated moving-average, ARIMA) and nonlinear machine-learning methods (support vector regression, SVR; regression random forest; feed-forward neural network, fNN; and long short-term memory neural network). For each method, the prediction accuracy and hypoglycemia detection capabilities were assessed using either population or individualized model parameters. As far as prediction accuracy is concerned, the results show that the best linear algorithm (individualized ARIMA) provides accuracy comparable to that of the best nonlinear algorithm (individualized fNN), with root mean square errors of 22.15 and 21.52 mg/dL, respectively. As far as hypoglycemia detection is concerned, the best linear algorithm (individualized ARIMA) provided precision = 64%, recall = 82%, and one false alarm/day, comparable to the best nonlinear technique (population SVR): precision = 63%, recall = 69%, and 0.5 false alarms/day. In general, the head-to-head comparison of the thirty algorithms fed by CGM data only made using a wide dataset shows that individualized linear models are more effective than population ones, while no significant advantages seem to emerge when employing nonlinear methodologies.

Mathematical Models of Meal Amount and Timing Variability With Implementation in the Type-1 Diabetes Patient Decision Simulator.

BACKGROUND: In type 1 diabetes (T1D) research, in-silico clinical trials (ISCTs) have proven effective in accelerating the development of new therapies. However, published simulators lack a realistic description of some aspects of patient lifestyle which can remarkably affect glucose control. In this paper, we develop a mathematical description of meal carbohydrates (CHO) amount and timing, with the aim to improve the meal generation module in the T1D Patient Decision Simulator (T1D-PDS) published in Vettoretti et al. METHODS: Data of 32 T1D subjects under free-living conditions for 4874 days were used. Univariate probability density function (PDF) parametric models with different candidate shapes were fitted, individually, against sample distributions of: CHO amounts of breakfast (CHOB), lunch (CHOL), dinner (CHOD), and snack (CHOS); breakfast timing (TB); and time between breakfast-lunch (TBL) and between lunch-dinner (TLD). Furthermore, a support vector machine (SVM) classifier was developed to predict the occurrence of a snack in future fixed-length time windows. Once embedded inside the T1D-PDS, an ISCT was performed. RESULTS: Resulting PDF models were: gamma (CHOB, CHOS), lognormal (CHOL, TB), loglogistic (CHOD), and generalized-extreme-values (TBL, TLD). The SVM showed a classification accuracy of 0.8 over the test set. The distributions of simulated meal data were not statistically different from the distributions of the real data used to develop the models (α = 0.05). CONCLUSIONS: The models of meal amount and timing variability developed are suitable for describing real data. Their inclusion in modules that describe patient behavior in the T1D-PDS can permit investigators to perform more realistic, reliable, and insightful ISCTs.

An analytical approach to determine the optimal duration of continuous glucose monitoring data required to reliably estimate time in hypoglycemia.

Diabetes is a chronic metabolic disease that causes blood glucose (BG) concentration to make dangerous excursions outside its physiological range. Measuring the fraction of time spent by BG outside this range, and, specifically, the time-below-range (TBR), is a clinically common way to quantify the effectiveness of therapies. TBR is estimated from data recorded by continuous glucose monitoring (CGM) sensors, but the duration of CGM recording guaranteeing a reliable indicator is under debate in the literature. Here we framed the problem as random variable estimation problem and studied the convergence of the estimator, deriving a formula that links the TBR estimation error variance with the CGM recording length. Validation is performed on CGM data of 148 subjects with type-1-diabetes. First, we show the ability of the formula to predict the uncertainty of the TBR estimate in a single patient, using patient-specific parameters; then, we prove its applicability on population data, without the need of parameters individualization. The approach can be straightforwardly extended to other similar metrics, such as time-in-range and time-above-range, widely adopted by clinicians. This strengthens its potential utility in diabetes research, e.g., in the design of those clinical trials where minimal CGM monitoring duration is crucial in cost-effectiveness terms.