A team of researchers from the University of Bristol analysing automated insulin delivery data from people with Type 1 Diabetes (T1D) discovered that unexpected patterns in insulin needs are just as common as well-established ones.
The study, published today [27 November] in JMIRx Med aimed to identify patterns in changes of insulin needs and to analyse how frequently these occur in people with T1D who use the OpenAPS, a state-of-the-art, automated insulin delivery system (AID).
Lead author Isabella Degen from Bristol’s Faculty of Science and Engineering explained: “The results support our hypothesis that factors beyond carbohydrates play a substantial role in euglycemia – the state when blood glucose levels are within the standard range.
“However, without measurable information about these factors, AID systems are left to adjust insulin cautiously with the effect of blood glucose levels becoming too low or high.”
Type 1 Diabetes is a chronic condition in which the body produces too little insulin, a hormone needed to regulate blood glucose.
The principal treatment for T1D is insulin that is injected or pumped. The amount and timing of insulin must be skilfully matched to carbohydrate intake to avoid increased blood glucose levels. Beyond carbohydrates, various other factors such as exercise, hormones, and stress impact insulin needs. However, how often these factors cause significant unexpected effects on blood glucose levels has been little explored, meaning that despite all advances, insulin dosing remains a complex task that can go wrong and result in blood glucose levels outside the range that protects people with T1D from adverse health effects.
The findings highlight the complexity of glucose regulation in T1D and demonstrate the heterogeneity in insulin needs among people with T1D, underlining the need for personalised treatment approaches.
For factors beyond carbohydrates to become more systematically included in clinical practice, scientists need to find a way to measure and quantify their impact and utilise this information in insulin-dosing. This could also aid more accurate blood glucose forecasting, which the study showed is not consistently possible from information about insulin and carbohydrates alone.
Isabella added: “Our study highlights that managing Type 1 Diabetes is far more complex than counting carbs.
“The richness of insights that can be gained from studying automated insulin delivery data is worth the effort it takes to work with this type of real-life data.
“What surprised us most was the sheer variety of patterns we observed, even within our relatively small and homogenous group of participants.
“It's clear that when it comes to diabetes management, one size doesn't fit all.
“We hope our results inspire further research into lesser-explored factors that influence insulin needs to improve insulin dosing.”
The team are now working on advancing time series pattern-finding methods that can handle the diverse and complex nature of real-life medical data, including irregular sampling and missing data. Their current focus is on developing innovative segmentation and clustering techniques for multivariate time series data tailored to uncover more granular patterns and handle the challenges posed by AID data.
To support this future research, the team seeks long-term, open-access AID datasets that include a wide range of sensor measurements of possible factors and a diverse cohort of people with T1D. Additionally, they aim to collaborate with time series and machine learning experts to address technical challenges such as handling irregularly sampled data with varying intervals between variates, and uncovering causalities behind observed patterns to ultimately driving innovations in personalised care.
Paper
‘Beyond expected patterns in insulin needs of people with type 1 diabetes: temporal analysis of automated insulin delivery data’ by Isabella Degen, Kate Robson Brown, Henry W J Reeve, Zahraa Abdallah in JMIRx Med