How can machine learning help prevent hypo- and hyperglycemia?

May 26

Xbird was founded on the vision to save people’s life with technology. We know this is a bold vision and we have a long way ahead of us. Working with diabetes we were dealing with a related question: How can adverse events be predicted and even prevented? The most common adverse events in diabetes are hypo- and hyperglycemia. This means, too low or too high blood glucose levels. Blood glucose levels are directly affected by daily activities. Therefore, if we have continuously tracked blood glucose levels and have a fully detailed overview of activity we can work on predicting hypo- and hyperglycemia events by the means of machine learning algorithms. Tracking the activity manually is extremely time-consuming and as a result of that, it is mostly not consequent over long periods of time. To solve this we need technology to help us do it automatically and passively without manual documentation. We are working on a project with the aim to predict and prevent hypoglycemia and hyperglycemia. We want to share some reasons for working on it and what we have achieved so far.

Here is the why

High level of blood sugar in the body (hyperglycemia) can lead to a variety of health issues including nerve damage, eye problems, kidney and vascular damage, and problems in the extremities. Low blood sugar (hypoglycemia) is caused by overdosing of insulin, mainly by not correctly assessing the carb consumption and the effects of physical activity. Hypoglycemia is a serious issue and if not treated immediately can cause confusion, loss of consciousness, seizures, or death.

Major complications caused by diabetes in the microvascular circulatory system

Eyes

High blood glucose and high blood pressure can damage eye blood vessels, causing retinopathy, cataracts and glaucoma.

Kidney

High blood pressure damages small blood vessels and excess blood glucose overworks the kidneys, resulting in nephropathy.

Neuropathy

Hyperglycemia damages nerves in the peripheral nervous system. This may result in pain and/or numbness. Feet wounds may go undetected, get infected and lead to gangrene.

Major complications caused by diabetes in the macrovascular circulatory system

Brain

Increased risk of stroke and cerebrovascular disease, including transient ischemic attack, cognitive impairment, etc.

Heart

High blood pressure and insulin resistance increase the risk of coronary heart disease.

Extremities

Peripheral vascular disease results from the narrowing of blood vessels increasing the risk for reduced or lack of blood flow in legs. Feet wounds are likely to heal slowly contributing to gangrene and other complications.

This long list of really severe complications makes us work on a project to predict and prevent hypoglycemia and hyperglycemia. Specifically, we aim to automatically classify reasons for past hypoglycemia and hyperglycemia events, predict future similar events and notify patients in real-time so they can take action to prevent them.

We use machine learning algorithms to analyze behaviour and other data collected by sensors built into patient’s smartphones and wearables. The data is collected from a variety of phones and, if available, wearable sensors. Then it is interpreted and translated into daily behaviour and routines.

Continuous passive activity monitoring allows us to document all daily activities and they may be visualized in a timeline above. The timeline includes activities like walking or cycling and special events like waking up or coming to the office. The aim is to capture all relevant behaviour that impacts blood sugar levels.

The data and predictions based on it should be very precise and it takes a lot of testing and research and we believe that it can actually save the lives of the patients and enrich patients support with this smart digital assistance.

It is the first time that these data sets exist in this form and can be combined to be studied. Stay tuned for more results!

This project is co-financed via the European Regional Development Fund (ERDF)