Abstract

Aims/hypothesis This study aims to develop an accessible, machine learning-derived tool for people with type 1 diabetes that predicts hypoglycaemia risk at the start of exercise, facilitating quick, clear risk assessment that can directly support safer exercise habits.

Methods We integrated data from four diverse studies encompassing 16,477 exercise sessions from 834 participants aged 12-80, using various insulin delivery methods. The XGBoost algorithm was used to develop a comprehensive and simplified model to predict hypoglycaemia during exercise, determined by continuous glucose monitor readings below 3.9 mmol/L (70 mg/dL).

Results The comprehensive model demonstrated a mean ROC AUC of 0.89, while the simplified model, relying solely on glucose levels at the start of exercise, duration of exercise and glucose rate of change arrows, achieved an ROC AUC of 0.87. This model was shown to be effective for any type of exercise and for people on a variety of insulin delivery devices. This simplified model was then translated, through collaborative efforts with type 1 diabetes participants, into “GlucoseGo,” a user-friendly, traffic-light heatmap that visually demonstrates risk of hypoglycaemia during exercise based on these three variables.

Conclusions/interpretation

The GlucoseGo heatmap offers a simple, readily available tool for predicting hypoglycaemia risk at the onset of exercise. This advancement empowers users to manage their exercise routines more safely, with potential to reduce hypoglycaemia incidents and enhancing exercise engagement among the type 1 diabetes population.

What is already known about this subject?

Exercise is crucial for managing type 1 diabetes, yet adherence to recommended guidelines is low.

Exercise-induced hypoglycaemia is a major barrier to exercise for those with type 1 diabetes.

Existing machine learning models for predicting hypoglycaemia during exercise often require complex data inputs, limiting their practical use.

What is the key question?

What are the new findings?

We developed a simplified machine learning model using only three variables; starting glucose levels, glucose rate of change arrows and exercise duration - that nearly matches the performance of more complex models, with an ROC AUC of 0.87 versus 0.89.

This model was transformed into “GlucoseGo,” user-friendly heatmaps, designed collaboratively with individuals with type 1 diabetes, that visually indicate exercise-induced hypoglycaemia risk.

Subgroup analyses show consistently good performance in predicting hypoglycaemia risk across diverse patient profiles and exercise types, validating its broad applicability.

How might this impact clinical practice in the foreseeable future?GlucoseGo offers a practical tool for safely managing exercise, potentially reducing hypoglycaemic incidents and increasing exercise participation among those with type 1 diabetes.

Competing Interest Statement

R.C.A. has received remuneration from Novo Nordisk, AstraZeneca, and Eli Lilly for conducting educational talks on diet and exercise for healthcare professionals. J.P. has received remuneration from Insulet, Abbott, Dexcom and ROCHE for conducting educational talks for healthcare professionals and attending advisory boards. John S Pemberton reports being on the advisory board for Abbott and ROCHE, and has received speaker fees from Abbott, Dexcom and Insulet in the last 3 years. A.F. has received speaker fees from Dexcom and Insulet, and consultancy fees from Insulet.

Funding Statement

This research received no specific grant from any funding agency in the public, commercial or not-for-profit sectors.

Author Declarations

I confirm all relevant ethical guidelines have been followed, and any necessary IRB and/or ethics committee approvals have been obtained.

Yes

The details of the IRB/oversight body that provided approval or exemption for the research described are given below:

The University of Exeter Research Ethics Committee waive ethical approval for this work as it involved only secondary analysis of anonymized data obtained via application from the T1DEXI and DEXIP repositories (Helmsley Charitable Trust) and from the EXTOD studies. No direct contact with human participants or access to identifiable information occurred.

I confirm that all necessary patient/participant consent has been obtained and the appropriate institutional forms have been archived, and that any patient/participant/sample identifiers included were not known to anyone (e.g., hospital staff, patients or participants themselves) outside the research group so cannot be used to identify individuals.

Yes

I understand that all clinical trials and any other prospective interventional studies must be registered with an ICMJE-approved registry, such as ClinicalTrials.gov. I confirm that any such study reported in the manuscript has been registered and the trial registration ID is provided (note: if posting a prospective study registered retrospectively, please provide a statement in the trial ID field explaining why the study was not registered in advance).

Yes

I have followed all appropriate research reporting guidelines, such as any relevant EQUATOR Network research reporting checklist(s) and other pertinent material, if applicable.

Yes

AbbreviationsCGMContinuous Glucose MonitorEXTODEXercise in Type One DiabetesEXT-101EXTOD-101 studyEXT-EDUEXTOD-Education studyIOBInsulin on Board Units per kgJCHRJaeb Center for Health ResearchMDIMultiple Daily InjectionsPPIPatient and Public InvolvementROC AUCArea Under the Receiver Operating Characteristics CurveSHAPSHapley Additive ExplanationsT1DEXIType 1 Diabetes EXercise InitiativeT1DEXIPType 1 Diabetes EXercise Initiative PediatricXGBoosteXtreme Gradient Boosting Algorithm