Large-Scale Data Analysis for Glucose Variability Outcomes with Open-Source Automated Insulin Delivery Systems

The routinization of lay expertise: A diachronic account of the invention and stabilization of an open-source artificial pancreas

Embodied health movements (EHMs) advance their agendas by mediating the production, circulation, and revision of biomedical knowledge. To do this, their constituents become lay experts by blending their embodied experience of illness with self-taught technical knowledge. However, it is unclear how lay expertise is routinized within EHMs, and consequently, to what extent it can be made durable in long-term partnerships with credentialed experts. I follow the OpenAPS community-a group of people with type one diabetes who engineered an open-source 'artificial pancreas'-from their inception in the transient #WeAreNotWaiting movement to their research collaborations with endocrinologists and detente with the FDA. I argue that OpenAPS user-contributors formalized their expertise in three steps: First, they broke the OpenAPS algorithm into modules so that prospective users must become experts to assemble it. Second, they lowered this barrier to entry by facilitating the socialization of new user-contributors with a training ritual. And third, they intervened in the strained endocrinologist-patient relationship. These tactics-restricting membership, reproducing expertise, and realigning interests-won the respect of credentialled experts who saw themselves in the OpenAPS community's image. While not all EHMs follow this trajectory, this case demonstrates that lay expertise can mature and assume new institutional forms without relying on commercialization or patronage.

Glycemic Variability Assessment in Newly Treated Exocrine Pancreatic Insufficiency With Type 1 Diabetes

BACKGROUND

Thirty-nine percent of people with type 1 diabetes may have lowered pancreatic elastase levels, correlated with exocrine pancreatic insufficiency (EPI or PEI). EPI is treated with oral supplementation of pancreatic enzymes. Little is known about the glycemic impact of pancreatic enzyme replacement therapy (PERT) in people with diabetes. This article demonstrates a method of assessing glycemic variability (GV), glycemic outcomes, and other changes in an individual with type 1 diabetes using open-source automated insulin delivery (AID).

METHOD

Macronutrient, PERT intake, and EPI-related symptoms were self-tracked; diabetes data were collected automatically via an open-source AID system. Diabetes data were uploaded via Nightscout to Open Humans and downloaded for analysis alongside self-tracked data (food, PERT). Glycemic outcomes, macronutrients, PERT dosing, and a variety of GV metrics following meals were evaluated for one month before and one month after PERT commencement. Breakfast was assessed independently across both time periods.

RESULTS

In an n = 1 individual using an open-source AID, time in range was already above goal and improved further after PERT commencement. Glucose rate of change and excursions >180 mg/dL were reduced; mean high blood glucose index was reduced overall and more so specifically at breakfast following PERT commencement.

CONCLUSIONS

GV can aid in assessing response to new-onset medications, as was demonstrated in this article for n = 1 individual with type 1 diabetes (using an open-source AID) after commencing PERT for newly identified EPI. GV may be useful for evaluating the efficacy of new-onset medications for people with insulin-requiring diabetes.

Glucose Variability Analysis in Two Large-Scale and Real-World Data Sets of Open-Source Automated Insulin Delivery Systems

BACKGROUND

Open-source automated insulin delivery (OS-AID) systems combine commercially available insulin pumps and continuous glucose monitors with open-source algorithms to automate insulin dosing for people with insulin-requiring diabetes. Two data sets (OPEN and the OpenAPS Data Commons) contain anonymized OS-AID user data.

METHODS

We assessed glycemic variability (GV) outcomes in the OPEN data set and characterized it alongside a comparison to the n = 122 version of the OpenAPS Data Commons. Glucose data are analyzed using an unsupervised machine learning algorithm for clustering, and GV metrics are quantified using statistical tests for distribution comparison. Demographic data are also analyzed quantitatively.

RESULTS

The n = 75 OPEN data set contains 36 827 days worth of data. Mean TIR is 82.08% (TOR < 70: 3.66%; TOR > 180: 14.3%). LBGI (P < .05) differs by gender whereas HBGI distributions are similar (P > .05). GV metrics (except TOR < 70, LBGI) show a statistically significant difference (P < .05) between data sets.

CONCLUSIONS

Both the OPEN and OpenAPS Data Commons data sets show TOR < 70, TIR, and TOR > 180 within recommended goals, adding additional evidence of real-world efficacy of OS-AID. Future research should evaluate in more detail potential data set differences and relationships between individual patterns of user behaviors and GV outcomes.

Understanding temporal changes and seasonal variations in glycemic trends using wearable data

Seasonal variations in glycemic trends remain largely unstudied despite the growing prevalence of diabetes. To address this gap, our objective is to investigate temporal changes in glycemic trends by analyzing intensively sampled blood glucose data from 137 patients (ages 2 to 76, primarily type 1 diabetes) over the course of 9 months to 4.5 years. From over 91,000 days of continuous glucose monitor data, we found that glycemic control decreases significantly around the holidays, with the largest decline observed on New Year's Day among the patients with already poor glycemic control (i.e., <55% time in the target range). We also observed seasonal variations in glycemic trends, with patients having worse glycemic control in the months of November to February (i.e., mid-fall and winter, in the United States), and better control in the months of April to August (i.e., mid-spring and summer). These insights are critical to inform targeted interventions that can improve diabetes outcomes.

Long-Term Glucose Forecasting for Open-Source Automated Insulin Delivery Systems: A Machine Learning Study with Real-World Variability Analysis

Glucose forecasting serves as a backbone for several healthcare applications, including real-time insulin dosing in people with diabetes and physical activity optimization. This paper presents a study on the use of machine learning (ML) and deep learning (DL) methods for predicting glucose variability (GV) in individuals with open-source automated insulin delivery systems (AID). A three-stage experimental framework is employed in this work to systematically implement and evaluate ML/DL methods on a large-scale diabetes dataset collected from individuals with open-source AID. The first stage involves data collection, the second stage involves data preparation and exploratory analysis, and the third stage involves developing, fine-tuning, and evaluating ML/DL models. The performance and resource costs of the models are evaluated alongside relative and proportional errors for 17 GV metrics. Evaluation of fine-tuned ML/DL models shows considerable accuracy in glucose forecasting and variability analysis up to 48 h in advance. The average MAE ranges from 2.50 mg/dL for long short-term memory models (LSTM) to 4.94 mg/dL for autoregressive integrated moving average (ARIMA) models, and the RMSE ranges from 3.7 mg/dL for LSTM to 7.67 mg/dL for ARIMA. Model execution time is proportional to the amount of data used for training, with long short-term memory models having the lowest execution time but the highest memory consumption compared to other models. This work successfully incorporates the use of appropriate programming frameworks, concurrency-enhancing tools, and resource and storage cost estimators to encourage the sustainable use of ML/DL in real-world AID systems.

Prediction of Blood Risk Score in Diabetes Using Deep Neural Networks

Improving the prediction of blood glucose concentration may improve the quality of life of people living with type 1 diabetes by enabling them to better manage their care. Given the anticipated benefits of such a prediction, numerous methods have been proposed. Rather than attempting to predict glucose concentration, a deep learning framework for prediction is proposed in which prediction is performed using a scale for hypo- and hyper-glycemia risk. Using the blood glucose risk score formula proposed by Kovatchev et al., models with different architectures were trained, including, a recurrent neural network (RNN), a gated recurrent unit (GRU), a long short-term memory (LSTM) network, and an encoder-like convolutional neural network (CNN). The models were trained using the OpenAPS Data Commons data set, comprising 139 individuals, each with tens of thousands of continuous glucose monitor (CGM) data points. The training set was composed of 7% of the data set, while the remaining was used for testing. Performance comparisons between the different architectures are presented and discussed. To evaluate these predictions, performance results are compared with the last measurement (LM) prediction, through a sample-and-hold approach continuing the last known measurement forward. The results obtained are competitive when compared to other deep learning methods. A root mean squared error (RMSE) of 16 mg/dL, 24 mg/dL, and 37 mg/dL were obtained for CNN prediction horizons of 15, 30, and 60 min, respectively. However, no significant improvements were found for the deep learning models compared to LM prediction. Performance was found to be highly dependent on architecture and the prediction horizon. Lastly, a metric to assess model performance by weighing each prediction point error with the corresponding blood glucose risk score is proposed. Two main conclusions are drawn. Firstly, going forward, there is a need to benchmark model performance using LM prediction to enable the comparison between results obtained from different data sets. Secondly, model-agnostic data-driven deep learning models may only be meaningful when combined with mechanistic physiological models; here, it is argued that neural ordinary differential equations may combine the best of both approaches. These findings are based on the OpenAPS Data Commons data set and are to be validated in other independent data sets.