A Model-Based Insulin Dose Optimization Algorithm for People With Type 1 Diabetes on Multiple Daily Injections Therapy

Methods in dynamic treatment regimens using observational healthcare data: A systematic review

We present a systematic review of methods used to estimate Dynamic Treatment Regimens (DTR) using observational healthcare data and provide a brief summary of their strengths and weaknesses, evaluation metrics, and suitable research problem settings. We considered all observational studies identified in PubMed or EMBASE between January 1950 until January 2022, including only studies that evaluated medical treatments or interventions as exposure and/or outcome in patients and where DTRs were estimated. 83 studies met our inclusion criteria; 44.6% estimating DTR utilizing reinforcement learning, 18.1% utilizing counterfactual-based models, 12.1% utilizing classification-based methods, and 9.6% utilized g-methods. Among the studies analyzed, 28.9% aimed to replicate human expert DTRs, while 71.1% aimed to refine and improve existing DTRs. Approximately two-thirds of studies (65.1%) reported the assumptions required for their applied methods, such as exchangeability, positivity, consistency, and Markov property. Most of the studies (83.1%) estimated DTRs with more than two treatment options; 50.6% mentioned time-varying confounders, only a few estimated conditional average treatment effects (7.2%). Most (85.5%) validated their methods, with 32.5% using expected outcomes (e.g., survival rates), 26.5% employing simulated data, and 25.3% conducting direct comparisons with observational data.

Distribution-based sub-population selection (DSPS): A method for in-silico reproduction of clinical trials outcomes

Diabetes presents a significant challenge to healthcare due to the short- and long-term complications associated with poor blood sugar control. Computer simulation platforms have emerged as promising tools for advancing diabetes therapy by simulating patient responses to treatments in a virtual environment. The University of Virginia Virtual Lab (UVLab) is a new simulation platform engineered to mimic the metabolic behavior of individuals with type 2 diabetes (T2D) using a mathematical model of glucose homeostasis in T2D and a large population of 6062 virtual subjects. This work proposes a statistical method - the Distribution-based sub-population selection (DSPS) method - for selecting subsets of virtual subjects from this large initial pool, ensuring that the selected group possesses the desired characteristics necessary to reproduce and predict the outcomes of a clinical trial. DSPS formulates the sub-population selection as a linear programming problem, identifying the largest virtual cohort to closely resemble the statistical properties (moments) of key outcomes from real-world clinical trials. The method was applied to the insulin degludec arm of a 26-week phase 3 clinical trial, evaluating the efficacy and safety of insulin degludec and liraglutide combination therapy. DSPS selected a sub-population that mirrored clinical trial data across key metrics, including glycemic efficacy, insulin dosages, and cumulative hypoglycemia events, with a relative sum of square errors of 0.33 and a percentage error of 1.07 %. This approach bridges the gap between large population simulation platforms and clinical trials, enabling the selection of virtual sub-populations with specific properties required for targeted studies.

Postprandial glucose-management strategies in type 1 diabetes: Current approaches and prospects with precision medicine and artificial intelligence

Postprandial glucose control can be challenging for individuals with type 1 diabetes, and this can be attributed to many factors, including suboptimal therapy parameters (carbohydrate ratios, correction factors, basal doses) because of physiological changes, meal macronutrients and engagement in postprandial physical activity. This narrative review aims to examine the current postprandial glucose-management strategies tested in clinical trials, including adjusting therapy settings, bolusing for meal macronutrients, adjusting pre-exercise and postexercise meal boluses for postprandial physical activity, and other therapeutic options, for individuals on open-loop and closed-loop therapies. Then we discuss their challenges and future avenues. Despite advancements in insulin delivery devices such as closed-loop systems and decision-support systems, many individuals with type 1 diabetes still struggle to manage their glucose levels. The main challenge is the lack of personalized recommendations, causing suboptimal postprandial glucose control. We suggest that postprandial glucose control can be improved by (i) providing personalized recommendations for meal macronutrients and postprandial activity; (ii) including behavioural recommendations; (iii) using other personalized therapeutic approaches (e.g. glucagon-like peptide-1 receptor agonists, sodium-glucose co-transporter inhibitors, amylin analogues, inhaled insulin) in addition to insulin therapy; and (iv) integrating an interpretability report to explain to individuals about changes in treatment therapy and behavioural recommendations. In addition, we suggest a future avenue to implement precision recommendations for individuals with type 1 diabetes utilizing the potential of deep reinforcement learning and foundation models (such as GPT and BERT), employing different modalities of data including diabetes-related and external background factors (i.e. behavioural, environmental, biological and abnormal events).

Advanced decision support system for individuals with diabetes on multiple daily injections therapy using reinforcement learning and nearest-neighbors: In-silico and clinical results

Many individuals with diabetes on multiple daily insulin injections therapy use carbohydrate ratios (CRs) and correction factors (CFs) to determine mealtime and correction insulin boluses. The CRs and CFs vary over time due to physiological changes in individuals' response to insulin. Errors in insulin dosing can lead to life-threatening abnormal glucose levels, increasing the risk of retinopathy, neuropathy, and nephropathy. Here, we present a novel learning algorithm that uses Q-learning to track optimal CRs and uses nearest-neighbors based Q-learning to track optimal CFs. The learning algorithm was compared with the run-to-run algorithm A and the run-to-run algorithm B, both proposed in the literature, over an 8-week period using a validated simulator with a realistic scenario created with suboptimal CRs and CFs values, carbohydrate counting errors, and random meals sizes at random ingestion times. From Week 1 to Week 8, the learning algorithm increased the percentage of time spent in target glucose range (4.0 to 10.0 mmol/L) from 51 % to 64 % compared to 61 % and 58 % with the run-to-run algorithm A and the run-to-run algorithm B, respectively. The learning algorithm decreased the percentage of time spent below 4.0 mmol/L from 9 % to 1.9 % compared to 3.4 % and 2.3 % with the run-to-run algorithm A and the run-to-run algorithm B, respectively. The algorithm was also assessed by comparing its recommendations with (i) the endocrinologist's recommendations on two type 1 diabetes individuals over a 16-week period and (ii) real-world individuals' therapy settings changes of 23 individuals (19 type 2 and 4 type 1) over an 8-week period using the commercial Bigfoot Unity Diabetes Management System. The full agreements (i) were 89 % and 76 % for CRs and CFs for the type 1 diabetes individuals and (ii) was 62 % for mealtime doses for the individuals on the commercial Bigfoot system. Therefore, the proposed algorithm has the potential to improve glucose control in individuals with type 1 and type 2 diabetes.

Estimation and Prediction of Glucose Appearance Rate for Use in a Fully Closed-Loop Dual-Hormone Intraperitoneal Artificial Pancreas

OBJECTIVE

A fully automated artificial pancreas requires a meal estimator and predictions of blood glucose levels (BGL) to handle disturbances during meal times, all without relying on manual meal announcements and user interventions. This study introduces a technique for estimating the glucose appearance rate (GAR) and predicting BGL in people with type 1 diabetes and insulin and glucagon administration. It is demonstrated for intraperitoneal insulin and glucagon delivery but may be adapted to other delivery sites.

METHOD

The estimator is designed based on the moving horizon estimation (MHE) approach, where the underlying cost function incorporates prior statistical information on the GAR in subjects over the course of a day. The proposed prediction scheme is developed to predict GAR using estimated states and an intestinal model, which is then used to predict BGL with the help of an animal glucose metabolic model.

RESULTS

The intraperitoneal dual-hormone estimator was evaluated on three anesthetized animals, achieving a 21.8% mean absolute percentage error (MAPE) for GAR estimation and a 10.0% MAPE for BGL prediction when the future GAR is known. For a 120-minute prediction horizon, the proposed predictor achieved an 18.0% MAPE for GAR and a 28.4% MAPE for BGL.

CONCLUSION

The findings demonstrate the effectiveness and reliability of the proposed estimator and its potential for use in a fully automated artificial pancreas and reducing user interventions.

SIGNIFICANCE

This study represents advancements toward the development of a fully automated artificial pancreas, ultimately enhancing the quality of life for people with type 1 diabetes.

A simple modeling framework for prediction in the human glucose-insulin system

Forecasting blood glucose (BG) levels with routinely collected data is useful for glycemic management. BG dynamics are nonlinear, complex, and nonstationary, which can be represented by nonlinear models. However, the sparsity of routinely collected data creates parameter identifiability issues when high-fidelity complex models are used, thereby resulting in inaccurate forecasts. One can use models with reduced physiological fidelity for robust and accurate parameter estimation and forecasting with sparse data. For this purpose, we approximate the nonlinear dynamics of BG regulation by a linear stochastic differential equation: we develop a linear stochastic model, which can be specialized to different settings: type 2 diabetes mellitus (T2DM) and intensive care unit (ICU), with different choices of appropriate model functions. The model includes deterministic terms quantifying glucose removal from the bloodstream through the glycemic regulation system and representing the effect of nutrition and externally delivered insulin. The stochastic term encapsulates the BG oscillations. The model output is in the form of an expected value accompanied by a band around this value. The model parameters are estimated patient-specifically, leading to personalized models. The forecasts consist of values for BG mean and variation, quantifying possible high and low BG levels. Such predictions have potential use for glycemic management as part of control systems. We present experimental results on parameter estimation and forecasting in T2DM and ICU settings. We compare the model's predictive capability with two different nonlinear models built for T2DM and ICU contexts to have a sense of the level of prediction achieved by this model.

Titration of Long-Acting Insulin Using Continuous Glucose Monitoring and Smart Insulin Pens in Type 1 Diabetes: A Model-Based Carbohydrate-Free Approach

OBJECTIVE

Multiple daily injections (MDI) therapy is the most common treatment for type 1 diabetes (T1D), consisting of long-acting insulin to cover fasting conditions and rapid-acting insulin to cover meals. Titration of long-acting insulin is needed to achieve satisfactory glycemia but is challenging due to inter-and intra-individual metabolic variability. In this work, a novel titration algorithm for long-acting insulin leveraging continuous glucose monitoring (CGM) and smart insulin pens (SIP) data is proposed.

METHODS

The algorithm is based on a glucoregulatory model that describes insulin and meal effects on blood glucose fluctuations. The model is individualized on patient's data and used to extract the theoretical glucose curve in fasting conditions; the individualization step does not require any carbohydrate records. A cost function is employed to search for the optimal long-acting insulin dose to achieve the desired glycemic target in the fasting state. The algorithm was tested in two virtual studies performed within a validated T1D simulation platform, deploying different levels of metabolic variability (nominal and variance). The performance of the method was compared to that achieved with two published titration algorithms based on self-measured blood glucose (SMBG) records. The sensitivity of the algorithm to carbohydrate records was also analyzed.

RESULTS

The proposed method outperformed SMBG-based methods in terms of reduction of exposure to hypoglycemia, especially during the night period (0 am-6 am). In the variance scenario, during the night, an improvement in the time in the target glycemic range (70-180 mg/dL) from 69.0% to 86.4% and a decrease in the time in hypoglycemia (<70 mg/dL) from 10.7% to 2.6% was observed. Robustness analysis showed that the method performance is non-sensitive to carbohydrate records.

CONCLUSION

The use of CGM and SIP in people with T1D using MDI therapy has the potential to inform smart insulin titration algorithms that improve glycemic control. Clinical studies in real-world settings are warranted to further test the proposed titration algorithm.

SIGNIFICANCE

This algorithm is a step towards a decision support system that improves glycemic control and potentially the quality of life, in a population of individuals with T1D who cannot benefit from the artificial pancreas system.