The Role of Artificial Intelligence in Diabetes Management: UK Innovations

According to the International Diabetes Federation (IDF), 10.5% of the adult population (20-79 years) has diabetes. This percentage translates to 540 million people worldwide.

The UK, in particular, is estimated to have approximately five million cases of diabetes, with more and more people getting diagnosed with type 2 diabetes before turning 40!

Readably, the chronic metabolic condition is escalating at a frightening rate, highlighting the importance of learning about diabetes care and staying up to date on recent innovations.

One innovation deserving of all the attention it can get is using artificial intelligence (AI) in diabetes management and its influence on data-driven precision care.

From predictive population risk stratification to patient self-management, the utility of AI has revolutionized the way diabetes is prevented, diagnosed, and managed, as you’re about to learn.

AI is the simulation of human intelligence processes by machines. It’s a wide-ranging branch of computer science concerned with creating systems that can handle and analyse data with great efficiency.

Now, within the healthcare sector, AI-powered systems have been used for disease diagnosis, drug discovery and development, health monitoring, and medical research, to name a few applications.

The technology has been particularly promising when it comes to diabetes care. It introduced new dimensions of patient self-care, quick and reliable decision-making, and optimised resource utilisation.

Different entities from around the world have recognised and embraced the potential of AI in supporting diabetes patients and helping them take control of their health.

For instance, the National Institute for Health and Care Research (NIHR) has helped finance the development of ROMI, an advanced conversational AI system designed to support type 2 diabetes patients.

Cloud-based reasoning (CBR) is an AI technique used to process and analyse large amounts of data in the cloud. It can solve new problems by learning from older encounters.

With regard to diabetes management, CBR can provide real-time feedback on blood glucose levels. What’s more, it’s used to offer personalised insulin therapy recommendations for diabetes patients.

An example of CBR in diabetes care is the 4 Diabetes Support System, a case-based reasoning system designed to help patients with type 1 diabetes achieve and maintain good blood glucose control.

Machine learning and deep learning have been integral in developing personalised diabetes management plans, predicting the risk of diabetes-related complications, and optimising the use of diabetes medications.

Machine learning programs can identify people at a high risk of diabetes based on genetic and metabolic factors. Further, machine learning principles have been used for automated blood glucose variability screening.

Support vector regression (SVR) is a type of machine learning algorithm that has been utilised in the diabetes management sector to create a hypoglycemia predictor that issues an alert when the patient’s blood glucose levels are low.

The AI method has also been used to identify the factors that influence blood glucose control, as well as estimate the insulin dose needed to achieve target blood glucose levels.

Neural networks are utilised for pattern recognition and prediction. They can link and analyse disparate data and then provide personalised solutions based on their analysis.

In the context of diabetes management, artificial neural networks (ANNs) can be trained on historical blood glucose data to identify patterns that can be used to predict future diabetes trends.

What’s more, ANNs can be used to develop personalised insulin dosing regimens, with factors such as meal size, insulin sensitivity, and weight considered.

AI technology has been used in the development of clinical decision-support tools that grant medical professionals access to in-depth information and insights about individual diabetes cases.

With the aid of these medical devices, healthcare providers can make quick yet highly informed decisions about their patients’ care. This, in turn, improves the quality of life for people living with diabetes.

To be more specific, supervised decision-support tools that are based on machine learning principles have been developed to predict HbA1c response post-insulin initiation in type 2 diabetes patients.

The same decision-support tools can also pinpoint the clinical factors that can influence HbA1c response—both short and long-term—in patients.

AI technology, particularly machine learning, has also been used to create an intuitive approach to predicting the risk of hospitalisation in diabetics and customising medication adherence interventions.

AI-based retinal screening has displayed remarkable feasibility and accuracy in the detection and monitoring of diabetic retinopathy, with a high specificity and sensitivity of 93.7% and 92.3%, respectively.

The linked-to study also highlights high patient satisfaction, where 96% of the patients reported being satisfied or very satisfied with automated retinal screening.

Just so you know, Automated retinopathy screening is based primarily on deep learning algorithms.

Also retinopathy-related is using convolutional neural network technology (CNN), a subset of machine learning, to develop lesion-specific probability maps for microaneurysms, haemorrhages, exudates, and more.

UK innovations for AI-powered retinopathy screening include:

• Moorfields Eye Hospital: Ophthalmologists from Moorfields Eye Hospital in London have been part of a research project that utilises AI to detect eye disease among diabetics. The research project validated that AI software can reduce the need to grade diabetic eye screening images by over 5 million images per year.
• RetinaLyze, Retmarker, and EyeArt: These are AI-based retinopathy screening software currently being used by eye doctors in the UK to screen patients for diabetic retinopathy.

Another noteworthy application of AI in diabetes management is risk prediction. By analysing the patient’s physical and mental health, lifestyle, and social network activities, machine learning models can predict the risk of diabetes.

Many of these models have been developed to not only detect long-term complications of diabetes, including renal, cardiovascular, and retinal, but they also can detect short-term complications like hypoglycemia.

On top of that, mobile applications have been developed to scan and interpret images of feet to help users keep an eye on the development of diabetic foot ulcers. A good example is the MyFootCare app.

The UK has contributed to this field of AI-powered diabetes care by introducing the QDiabetes risk calculator, a tool doctors and academics in the NHS developed to predict the risk of type 2 diabetes in the next five years.

QDiabetes is now integrated into leading general practice computer systems in the UK and is recommended in both the NICE type 2 diabetes prevention guide and NHS Health Checks best practice guideline.

Also new to the world of diabetes care—courtesy of artificial intelligence—is the introduction of advanced genomics, digital biomarkers, epigenetic alterations, and molecular phenotyping.

These techniques can be quite handy when analysing massive data sets, which is often the case when addressing diabetes, considering the disease’s heterogeneous and chronic nature.

For example, researchers have been able to build a repository of microbial marker genes by analysing microbiome data. This has helped in the treatment of patients with confirmed type 2 diabetes.

Moreover, over 400 distinct association signals with type 2 diabetes were found in some of the latest genome-wide association studies, which could potentially establish the genetic predisposition to diabetes.

FYI: The UK, along with Denmark and the Netherlands, took part in the ADDITION trial, which aimed to determine whether screening for undiagnosed type 2 diabetes mellitus is feasible with the aid of genomic data.

Empowering diabetics to become experts on their condition and be able to self-manage is another major benefit of the use of AI in diabetes management.

AI-powered digital platforms have made it easy to access targeted education as a diabetes patient. Such platforms, which include web tools and mobile apps, provide awareness about diet and activity patterns.

This targeted form of diabetes education has proven especially beneficial among pregnant women with gestational diabetes mellitus (GDM), helping broaden their grasp of the condition.

In addition to providing targeted education, AI-powered self-management tools have enabled diabetics to self-treat by assessing their food intake and making appropriate diet and activity decisions.

One of the most prominent AI self-management tools is the One Drop mobile app. It’s designed to help individuals with type 1 and 2 diabetes access data-driven insights and stats, schedule medication reminders, and track health outcomes.

Since AI algorithms are trained on data sets, the quantity and quality of data available can impact the overall performance of AI models.

The quantity aspect is pretty self-explanatory. The quality aspect refers to how mature and structured the data is. The more high-quality the data is, the more it will help digital applications come up with impactful solutions.

But that’s not all!

Concerns about security and data protection are brought up whenever data is brought up. Such regulatory concerns can limit the adoption of AI technology in diabetes management.

AI models can be biased, which is no surprise, considering different models are trained on different data. This may lead to inaccurate predictions for certain groups of people.

Another limitation is that AI models and applications are developed and tested using retrospective data. In other words, they’re validated using data from the past.

This can be limiting because diabetes management is an ongoing process, and relying solely on historical data won’t help account for future variations in patients’ conditions.

The solution here is to move toward prospective validation, where AI models are tested in real time and data is collected during the validation process.

Quite a few human factors can limit the application of AI in diabetes management. One such factor that called for a meta-analysis of 14 randomised control studies is age.

The results of that meta-analysis were interesting, showing that younger patients are more likely to benefit from the use of diabetes apps than older patients.

Dependence is another factor we must account for. Relying on AI can lead to de-skilling physicians, which, in itself, can hinder the adoption of AI since the technology requires periodic refinement by experts.

From predictive risk stratification and remote monitoring to clinical support and patient self-management, the use of AI in diabetes care has helped redefine our approach toward prevention and management.

With AI, healthcare professionals can now provide accurate and timely intervention, and patients are empowered to take matters into their own hands by managing their own health.

There’s no telling how the continuous use and improvement of AI will change the healthcare industry decades from now, but it will surely make it easier to cope with chronic conditions like diabetes.