Artificial Intelligence, or AI, aims to enable a machine (or computer) to think and learn like humans. First conceptualised in the 1950s, AI is widely used by companies like Facebook and Google. The healthcare industry, however, has been slow to adopt AI. But, with research and startups exploding around the world, change is on the anvil. As the disease of out times, diabetes has come under the spotlight. Could AI be the next big breakthrough in diabetes?
AI in health
The statistics are stark and alarming: about 10 per cent of the global population has diabetes. In absolute terms, the number has zoomed from 151 million in 2000 to 463 million now. India is home to over 72 million people with diabetes. With the economic burden of diabetes care going out of control, scientists and policymakers are looking at AI, machine learning algorithms and Big data, to lessen the world’s burden of diabetes. The efficient computer-based tool has been found to augment human intelligence in healthcare, especially with algorithms in the analysis of complex medical data.
AI solutions are being developed to automate image analysis and diagnosis in radiology; to identify new potential therapies from vast databases of information on existing medicines for faster drug discovery; to provide real-time support to clinicians to help identify at-risk patients by analyzing vast amounts of historic patient data; to address patient solutions to triage with quick, scalable access to basic questions and medical issues.
The diabetes pipeline
Digital health companies are coming up for diabetes care. Programs, apps, and other solutions are being introduced. Innovate solutions—from non-invasive insulin delivery systems to continuous glucose monitoring devices, and digital diabetes management platforms—are emerging. AI is finding widespread use inpatient self-management tools. Machine learning algorithms are being built to predict the risk of developing diabetes as well as its complications. Both patients and healthcare professionals are getting empowered by AI driven clinical decision support, that helps gathers and store more data, observe trends and be pointed to better decisions. The industry is growing at a fast clip. A number of creative solutions are emerging for using AI in diabetes management.
Managing chronic diabetes
Self-management is crucial in treating diabetes. Digital healthcare services are leveraging Big data with the promise that people with diabetes may not need to consult a doctor for every medical issue. Just a connected device and some well-timed advice may help them manage their conditions better. What’s emerging is a range of cell phone-enabled blood-glucose meters, cuff-to-cloud blood-pressure devices, wireless-enabled scales and so on. They allow the users to instantly upload their blood sugar data onto the analytic engine of the healthcare company.
This data is then used for generating insights to help them manage their condition. AI is also used to continuously analyze and agglomerate all the data generated by a person with diabetes, to create a personalized feedback loop of advice and adjustments, as needed. For instance, if one’s blood sugar test results are outside normal parameters, a call or text from a diabetes consultant follows, to assess the situation and suggest the next steps.
Prick to scan
Glucose monitoring is one of the most intrusive parts of living with diabetes. AI-powered mobile phone apps are about to change the way you check blood glucose levels. Instead of finger-prick tests, one can simply wave a compatible smartphone over a sensor on the arm, to get real-time blood sugar levels. It will be a real boon for those who have lived with type 1 diabetes for many years, as fingers can become hard and calloused, making it harder to do finger-prick checks. Integrating diabetes management tools into mobile phone technologies is an advancement that promises to reduce the burden of people living with diabetes, make it easier to better self-manage their diabetes and help to reduce diabetes-related complications.
Automated diabetes kit
AI is paving the way for fully automated diabetes kits, too. For people with type-1 diabetes, missing a meal or an unplanned physical exertion can dangerously reduce blood glucose levels. Addressing such risks are electronic devices for continuous glucose monitor (CGM) in real-time. An enhanced algorithm also allows it to guide a programmable insulin pump for automated delivery.
In response to changes in blood glucose levels, the insulin pump delivers tiny amounts of insulin into the tissue under the skin. If the CGM picks up that blood sugar is dropping below a certain point, the pump stops delivering insulin until the sugar level rises again and it is safe to continue. Such systems can improve sugar levels, prevent highs and lows, and reduce the need for constant adjustments.
Similar functions can also be provided by devices like Android phone apps, designed to work with commercially available insulin pumps and CGMs. There are DIY systems in the offing. They use algorithms to control insulin-dosing, based on CGM data. US-based diabetes management companies are also working to develop automated insulin delivery systems along with artificial pancreas. The main barriers to the widespread adoption of such automated systems, however, are cost and complexity. The DIY route, in particular, requires technical know-how.
Diabetic retinopathy diagnosis
As diabetes continues to be the leading cause of blindness among adults, the power of Big data and predictive algorithms are being leveraged to automate the diagnosis of diabetic retinopathy, the complication associated with diabetes that damages eyesight leading even to vision loss. Here, AI-based screening is used to detect and monitor incidences of diabetic retinopathy.
In a paper published in Diabetes Care (Jan 5, 2021), however, researchers compared the algorithms against the diagnostic expertise of retina specialists and found that the former don’t perform as well as human screeners. The researchers claimed that differences in camera equipment and technique could be one explanation. The algorithms are designed to work with a minimum quality of images.
Diabetes risk modeling
Risk prediction models are widely used in clinical practice. They use traditional statistical techniques to analyze information about people—age, lifestyle, disease patterns, family history—to identify those at high risk of developing an illness and make decisions about their care. Traditional risk prediction models perform well at the population level, but not so much at the level of individuals. Now healthcare organizations are using machine learning to model the risk of diabetes by analyzing their details, including physical and mental health, and social media activity.
Beyond the debate
Can medicine really be transformed through AI-enabled healthcare pathways? Can it allow a clinician to push a button and provide patient-specific predictions of expected outcomes if no treatment is provided? Can such systems support the clinician and patient in making what may be life-or-death decisions? With health systems in many countries investing millions in AI, such debates have broken out among scientists.
Multiple papers suggest that machine learning outperforms statistical models of disease risk prediction, identify patients most at risk, allow for earlier diagnosis and cheaper, more focused, personalised prevention. Others point to the challenges of widespread adoption of AI systems: from the complexity and quality of healthcare data, the need to consider multiple interactions between diverse events over a patient’s life, the difficulty in estimating counterfactual outcomes, the need for cross-disciplinary collaboration, the problem of delivering AI solutions at scale and so on.
While the debates continue, Artificial Intelligence is making its presence felt increasingly. A McKinsey review (Artificial intelligence: The time to act is now, 2018) predicts healthcare as one of the top five industries that would involve AI in the coming decades. Instead of opposing the game-changing technology, scientists and clinicians might need to think how they can benefit the most from it.