Artificial Intelligence: The Future for Diabetes Care

Artificial intelligence (AI) is a fast-growing field and its applications to diabetes, a global pandemic, can reform the approach to diagnosis and management of this chronic condition. Principles of machine learning have been used to build algorithms to support predictive models for the risk of developing diabetes or its consequent complications. Digital therapeutics have proven to be an established intervention for lifestyle therapy in the management of diabetes. Patients are increasingly being empowered for self-management of diabetes, and both patients and health care professionals are benefitting from clinical decision support. AI allows a continuous and burden-free remote monitoring of the patient's symptoms and biomarkers. Further, social media and online communities enhance patient engagement in diabetes care. Technical advances have helped to optimize resource use in diabetes. Together, these intelligent technical reforms have produced better glycemic control with reductions in fasting and postprandial glucose levels, glucose excursions, and glycosylated hemoglobin. AI will introduce a paradigm shift in diabetes care from conventional management strategies to building targeted data-driven precision care.

Simple Linear Support Vector Machine Classifier Can Distinguish Impaired Glucose Tolerance Versus Type 2 Diabetes Using a Reduced Set of CGM-Based Glycemic Variability Indices

BACKGROUND

Many glycemic variability (GV) indices exist in the literature. In previous works, we demonstrated that a set of GV indices, extracted from continuous glucose monitoring (CGM) data, can distinguish between stages of diabetes progression. We showed that 25 indices driving a logistic regression classifier can differentiate between healthy and nonhealthy individuals; whereas 37 GV indices and four individual parameters, feeding a polynomial-kernel support vector machine (SVM), can further distinguish between impaired glucose tolerance (IGT) and type 2 diabetes (T2D). The latter approach has some limitations to interpretability (complex model, extensive index pool). In this article, we try to obtain the same performance with a simpler classifier and a parsimonious subset of indices.

METHODS

We analyzed the data of 62 subjects with IGT or T2D. We selected 17 interpretable GV indices and four parameters (age, sex, BMI, waist circumference). We trained a SVM on the data of a baseline visit and tested it on the follow-up visit, comparing the results with the state-of-art methods.

RESULTS

The linear SVM fed by a reduced subset of 17 GV indices and four basic parameters achieved 82.3% accuracy, only marginally worse than the reference 87.1% (41-features polynomial-kernel SVM). Cross-validation accuracies were comparable (69.6% vs 72.5%).

CONCLUSION

The proposed SVM fed by 17 GV indices and four parameters can differentiate between IGT and T2D. Using a simpler model and a parsimonious set of indices caused only a slight accuracy deterioration, with significant advantages in terms of interpretability.

LSTMs and Neural Attention Models for Blood Glucose Prediction: Comparative Experiments on Real and Synthetic Data

We have shown in previous work that LSTM networks are effective at predicting blood glucose levels in patients with type I diabetes, outperforming human experts and an SVR model trained with features computed by manually engineered physiological models. In this paper we present the results of a much larger set of experiments on real and synthetic datasets in what-if, agnostic, and inertial scenarios. Experiments on a more recent real-patient dataset, which we are releasing to the research community, demonstrate that LSTMs are robust to noise and can easily incorporate additional features, such as skin temperature, heart rate and skin conductance, without any change in the architecture. A neural attention module that we designed specifically for time series prediction improves prediction performance on synthetic data; however, the improvements do not transfer to real data. Conversely, using time of day as an additional input feature consistently improves the LSTM performance on real data but not on synthetic data. These and other differences show that behavior on synthetic data cannot be assumed to always transfer to real data, highlighting the importance of evaluating physiological models on data from real patients.

Glycaemic variability-based classification of impaired glucose tolerance vs. type 2 diabetes using continuous glucose monitoring data

Many glycaemic variability (GV) indices extracted from continuous glucose monitoring systems data have been proposed for the characterisation of various aspects of glucose concentration profile dynamics in both healthy and non-healthy individuals. However, the inter-index correlations have made it difficult to reach a consensus regarding the best applications or a subset of indices for clinical scenarios, such as distinguishing subjects according to diabetes progression stage. Recently, a logistic regression-based method was used to address the basic problem of differentiating between healthy subjects and those affected by impaired glucose tolerance (IGT) or type 2 diabetes (T2D) in a pool of 25 GV-based indices. Whereas healthy subjects were classified accurately, the distinction between patients with IGT and T2D remained critical. In the present work, by using a dataset of CGM time-series collected in 62 subjects, we developed a polynomial-kernel support vector machine-based approach and demonstrated the ability to distinguish between subjects affected by IGT and T2D based on a pool of 37 GV indices complemented by four basic parameters-age, sex, BMI, and waist circumference-with an accuracy of 87.1%.

Using LSTMs to learn physiological models of blood glucose behavior

For people with type 1 diabetes, good blood glucose control is essential to keeping serious disease complications at bay. This entails carefully monitoring blood glucose levels and taking corrective steps whenever they are too high or too low. If blood glucose levels could be accurately predicted, patients could take proactive steps to prevent blood glucose excursions from occurring. However, accurate predictions require complex physiological models of blood glucose behavior. Factors such as insulin boluses, carbohydrate intake, and exercise influence blood glucose in ways that are difficult to capture through manually engineered equations. In this paper, we describe a recursive neural network (RNN) approach that uses long short-term memory (LSTM) units to learn a physiological model of blood glucose. When trained on raw data from real patients, the LSTM networks (LSTMs) obtain results that are competitive with a previous state-of-the-art model based on manually engineered physiological equations. The RNN approach can incorporate arbitrary physiological parameters without the need for sophisticated manual engineering, thus holding the promise of further improvements in prediction accuracy.

Case-Based Reasoning for Insulin Bolus Advice

BACKGROUND

Insulin bolus calculators assist people with Type 1 diabetes (T1D) to calculate the amount of insulin required for meals to achieve optimal glucose levels but lack adaptability and personalization. We have proposed enhancing bolus calculators by the means of case-based reasoning (CBR), an established problem-solving methodology, by individualizing and optimizing insulin therapy for various meal situations. CBR learns from experiences of past similar meals, which are described in cases through a set of parameters (eg, time of meal, alcohol, exercise). This work discusses the selection, representation and effect of case parameters used for a CBR-based Advanced Bolus Calculator for Diabetes (ABC4D).

METHODS

We analyzed the usage and effect of selected parameters during a pilot study (n = 10), where participants used ABC4D for 6 weeks. Retrospectively, we evaluated the effect of glucose rate of change before the meal on the glycemic excursion. Feedback from study participants about the choice of parameters was obtained through a nonvalidated questionnaire.

RESULTS

Exercise and alcohol were the most frequently used parameters, which was congruent with the feedback from study participants, who found these parameters most useful. Furthermore, cases including either exercise or alcohol as parameter showed a trend in reduction of insulin at the end of the study. A significant difference ( P < .01) was found in glycemic outcomes for meals where glucose rate of change was rising compared to stable rate of change.

CONCLUSIONS

Results from the 6-week study indicate the potential benefit of including parameters exercise, alcohol and glucose-rate of change for insulin dosing decision support.