AI revolutionises diabetes management by shifting from reactive care to proactive, personalised solutions. Mark Venables discovers how Roche’s Accu-Chek SmartGuide Predict app harnesses advanced predictive algorithms to empower individuals with foresight and control, redefining the future of chronic disease care.
Managing diabetes is a relentless and complex daily task for millions worldwide. Despite the widespread availability of continuous glucose monitoring (CGM) systems, many individuals still struggle to maintain glycemic control. Dr Pau Herrero, Algorithms & Decision Support Tech Lead at Roche Diagnostics, and his team are determined to address this challenge. At the heart of Roche’s efforts lies the Accu-Chek SmartGuide Predict app, part of the Accu-Chek SmartGuide CGM solution, an AI-enabled digital tool designed to transform diabetes self-management through predictive insights and personalised interventions.
Herrero explains that the goal is to shift diabetes care from reactive responses to proactive decision-making. “With the Accu-Chek SmartGuide Predict app, we wanted to give users a clearer picture of what lies ahead. Instead of responding to fluctuations after they occur, individuals can act in advance to stabilise glucose levels. This is about empowering people with diabetes to regain control over their daily lives.”
AI at the core of personalised care
Central to the app’s functionality is a suite of machine learning models that analyse CGM data in real time. These models are designed to predict potential glycaemic events, offering a two-hour window for glucose fluctuations, a 30-minute warning for low glucose, and insights into overnight risks of hypoglycaemia. Herrero highlights the significance of this approach, noting, “Prediction of biomarkers, particularly glucose, in healthcare is incredibly complex because human biology is not static. Factors such as diet, physical activity, hormonal fluctuations, and individual metabolic differences contribute to this complexity. Our models incorporate present and past glucose readings and variables like insulin dosage, carbohydrate intake, and even the time of day. This allows for a dynamic and personalised prediction.”
The app’s predictive algorithms use state-of-the-art machine learning techniques, including deep recurrent neural networks and gradient-boosting models. The former powers the app’s two-hour glucose forecast, while the latter underpins its hypoglycaemia prediction functionalities. “Each model has been rigorously trained and validated using diverse datasets to ensure robustness,” Herrero elaborates. “We have evaluated performance across individuals with type 1 and type 2 diabetes on different insulin therapies, using real-world data and clinical trial datasets. This breadth of testing ensures that the app is designed to be sound and practical in everyday use.”
The results speak for themselves. During evaluations, the app’s 2-hour glucose prediction model achieved 98.7 per cent of forecasted values within clinically acceptable accuracy for mid-term horizons. The 30-minute warning for low glucose levels demonstrated an average accuracy of 98.9 per cent. In detecting nighttime hypoglycaemia risk, the model achieved an average accuracy of 86.5 per cent. Herrero underscores the implications of these figures: “What we are delivering is reliability. Users can trust that the app’s predictions are grounded in evidence and validated for real-world efficacy.”
Tackling the fear of nocturnal hypoglycaemia
One of the most challenging aspects of diabetes management is the risk of nocturnal hypoglycaemia, a condition that can lead to severe health consequences if undetected. Traditional glucose alarms have long been a staple in CGM systems, but their disruptive nature often leads to alarm fatigue or missed notifications. “We recognised that fear of nocturnal hypoglycaemia creates significant anxiety for many users,” Herrero says. “Our Night Low Predict feature is designed to alleviate this by providing alerts and actionable insights. Users can take preventive measures before bed, reducing the risk and improving sleep quality.”
The Night Low Predict functionality evaluates the hypoglycaemia risk over seven hours, offering mitigation recommendations. “The model considers multiple factors, including recent and past glucose readings (up to 28 days), insulin administration, and time of day, to calculate the likelihood of overnight hypoglycaemia,” Herrero explains. “If a high risk is detected, users are notified well in advance, allowing them to adjust their regimen or take preventive actions to avoid complications.
“By providing early warnings and actionable advice, the app reduces the need for intrusive alarms during the night. This is not just about technology for optimising glucose control; it is about improving the overall quality of life for people with diabetes. Sleep is critical, and anything that disrupts it can have cascading effects on health and well-being.”
Human-centred design for maximum impact
Developing an AI-driven tool is only part of the equation. Ensuring it integrates seamlessly into users’ lives requires profoundly understanding their needs and behaviours. Roche’s development process emphasised user-centred design, incorporating extensive usability studies to refine the app’s interface and features.
“We did not want to create a tool that simply impressed on a technical level,” Herrero continues. “It had to be something people would use, trust, and rely on daily. Through user feedback, we identified pain points, such as notification overload, and addressed them by offering customisable settings. Users can personalise thresholds, adjust notification timings, or pause alerts temporarily.”
The app also incorporates visual tools to make data interpretation more intuitive. Graphs and dashboards summarise past, present, and future glucose fluctuations, giving users a comprehensive view of their condition. “We focused on creating an interface that communicates complex information in a way that is easy to understand,” Herrero adds. “It is about balancing technical sophistication with simplicity of use.”
Bridging technological and regulatory challenges
While the potential of AI in diabetes care is immense, the path to implementation is not without obstacles. Regulatory compliance and system interoperability are significant hurdles that developers must navigate. “Meeting the highest medical device standards was non-negotiable,” Herrero says, highlighting Roche’s proactive approach. “Our app complies with rigorous certifications and regulations, including the CE marking, and adheres to data privacy laws such as GDPR.”
Interoperability was another critical consideration. The app seamlessly integrates with Roche’s Accu-Chek SmartGuide CGM sensor and Accu-Chek Care, Roche’s clinical platform for diabetes management. “We wanted to create a platform, not just a standalone product,” Herrero says. “The ability to share data across systems and with healthcare providers is essential for delivering holistic care.”
The cloud-based architecture supporting the app, which hosts the machine learning models, significantly enhances scalability and usability. This approach enables Roche to reach a wide range of mobile devices by eliminating computing power and energy consumption constraints. Furthermore, it allows for continuously refining existing predictive models based on performance monitoring and user feedback without requiring users to download new app versions manually.
The road ahead for AI in healthcare
Herrero is optimistic about AI’s broader implications in chronic disease management. “What we are doing with the Accu-Chek SmartGuide Predict app is just the beginning. The lessons we have learned here about data integration, machine learning, and user-centric design can be applied to a wide range of healthcare challenges,” he says.
However, he also acknowledges the need for collaboration across the healthcare ecosystem. “AI solutions do not exist in a vacuum. For these technologies to truly make an impact, there needs to be alignment between developers, regulators, clinicians, and patients. Only then can we overcome the barriers to adoption and unlock the full potential of AI.”
As for the future of diabetes care, Herrero envisions a shift towards even greater personalisation and autonomy. “The ultimate goal is to reduce the burden of diabetes management to the point where it becomes almost invisible in people’s daily lives,” he explains. “Technology should act as a silent partner, always working in the background to ensure optimal outcomes.”
A significant advancement in diabetes control
The Accu-Chek SmartGuide Predict app represents a paradigm shift in diabetes management. It combines cutting-edge AI with a human-centred approach to empower users. Its predictive capabilities, intuitive design, and seamless integration address longstanding challenges in diabetes care while setting a new standard for digital health solutions.
Herrero’s insights underscore the transformative potential of AI when applied thoughtfully and rigorously. He aptly concludes, “The future of healthcare lies in proactive, data-driven solutions that empower individuals. By anticipating needs rather than reacting to problems, we can create a world where managing chronic conditions is no longer a burden but a seamless part of life.” This convergence of technology, clinical expertise, and user-centric innovation offers a blueprint for how AI can redefine not just diabetes care but the entire landscape of chronic disease management.
