How AI is Revolutionizing Diabetes Care

For diabetics like myself, AI is already improving health outcomes and making lives easier

Photo from: https://clinicablasi.com/en/diabetes-and-gum-disease/

I was diagnosed with Type 1 Diabetes two months into my freshman year of college after fighting some mystery sickness for weeks. My fall break was spent injecting oranges with saline to learn how to give myself insulin shots, pricking my finger to check my blood sugar level, and learning my new dietary restrictions. On top of painful pricks came arduous homework. Until I was ready for the training wheels to come off, I spent the next few months recording my blood sugar, carbs, and insulin on paper forms provided by my endocrinologist. Chronic illness was a lot of work.

In the six years since my diagnosis, I no longer give myself shots or prick my finger. I certainly do not use paper logs to track my blood sugar. Instead, I track my blood sugar from my phone and rely on a machine learning algorithm to measure needed insulin. AI has revolutionized how I manage this disease.

Diabetes tech has advanced greatly, even since I was first diagnosed. And as innovation continues in the diabetes technology market, emerging technologies like AI hold inordinate potential to improve patient outcomes for diabetics and individuals with other chronic diseases. Innovation and experimentation should be encouraged in digital health technology, especially the development of artificial intelligence in medicine.

Reducing the Burden

The first major AI-powered advancement in diabetes has been the introduction of machine learning powered insulin pumps. In what’s called a “closed-loop system” these pumps take blood sugar readings from a constant glucose monitor (CGM) and utilize an algorithm to decide how much insulin is required to keep blood sugar levels in range. These closed-loop systems have been shown to reduce variation in blood sugar and avoid hypoglycemia – low blood sugar levels that if left untreated can be deadly.

My closed-loop system consists of two parts: the Dexcom G6 (CGM) and the Omnipod 5 (insulin pump). The two devices, connected by bluetooth, talk to each other and make treatment decisions on my behalf every five minutes. The Omnipod 5, fed blood sugar readings by the Dexcom G6, continually develops a predictive control algorithm to keep my blood sugar levels in target. Based on what it has learned, the system also adjusts to match predicted blood sugar in the next 60 minutes.

In the first month I’ve used a closed-loop system, the time I have spent hitting my target blood sugar has increased by seven percent – making me well on my way to the expected 90-day improvement of around nine percent. While this may seem small, improving time-in-range reduces the risk of complications and in turn, deaths. Any improvement in blood sugar levels goes a long way to ensuring that diabetics live long and healthy lives.

Improved numbers are important, but they are just one facet of how machine learning diabetic technology is improving lives. Managing diabetes takes work – as one example, diabetics make 120 more decisions per day on average than a non-diabetic person – and finding ways to reduce the mental load is important to prevent diabetes burnout. By largely removing the patient from the “loop,” closed-loop systems do just that. Using a closed-loop system is associated with lower anxiety levels about managing care, less lost sleep from diabetes management, and more “time off” from diabetes management. These mental benefits from closed-loop systems also translate to parents of young children with diabetes and other caregivers.

What’s on the Horizon and What’s at Stake

What’s currently available on the market pales in comparison to what could be coming down the pipeline. Beyond self-management tools, artificial intelligence has made great strides in the treatment and prevention of diabetes:

1. Diabetic Eye Damage
   Diabetic Retinopathy is a rare complication of diabetes that affects the blood vessels at the back of the eye and, if not treated, may cause blindness. New AI technologies can detect Diabetic Retinopathy based on photos of the back of the eye. In clinical trials, the AI algorithm was more successful than general ophthalmologists and retina specialists at catching Diabetic Retinopathy. The algorithm correctly identified Diabetic Retinopathy in 96.5 percent of cases, compared to 59.5 percent of cases by the retina specialist and 20.6 percent of cases by the general ophthalmologists.
   

2. Predicting Risk
   AI algorithms have also been used to successfully determine the risk of developing Type 2 diabetes. One Chinese study found that a machine learning algorithm correctly identified if a patient would develop Type 2 in 80 percent of cases. Similar uses of machine learning have been used to identify whether pregnant women with gestational diabetes will develop Type 2 Diabetes later in life. One model correctly identified risk in 83 percent of the cases in one group and 77 percent of the cases in another.
   

3. Fully AI Powered Pancreas
   Researchers at the University of Virginia have augmented the closed-loop system to include AI tools that can anticipate disturbances in blood sugar like meals and exercise and react automatically. In a clinical trial, the AI supported pancreas was just as successful at keeping patients in range as the standard closed-loop system. As technology develops, this could allow diabetics to live an almost normal life – as if their pancreas had never stopped working at all.
   

These are only a few examples of how artificial intelligence can improve the health and wellbeing of diabetics. But getting these and other innovations in how we treat diabetes on the market will take time. Historically, major innovations in diabetes care have taken decades – the insulin pump wasn’t available on the market until 1978, 15 years after the first model was presented. The slow introduction of life saving innovations are due to both long regulatory approval processes and long “practice gaps” – or the time between the creation of new, evidence-based practices and technologies becoming the standard of care.

Artificial intelligence powered innovations in healthcare face a unique lack of trust that may slow both their approval process and clinical adoption. The inner workings of AI systems are not understood and therefore not trusted by patients and doctors. The complex and iterative nature of AI systems also makes it harder for them to be approved by regulatory bodies. If we want to fully capture the benefits from AI, regulators should focus both on promoting innovation and getting buy-in from patients and doctors. Focusing on AI diffusion in healthcare can ensure that patients get the benefits from new healthcare innovations as quickly as possible.

Having the closed-loop system has simplified my life greatly and quickly improved my diabetes care. I’m hopeful that as innovation continues, the role I play in managing my diabetes will continue to be scaled back in favor of better and more intelligent technology.