Health AI Chronicle - Edition 257
Practical AI in healthcare this week comes from the research pipeline rather than the deployment floor: a Kaunas University of Technology team in Lithuania has built a graph neural network that forecasts blood-sugar swings and hypoglycemia risk up to an hour in advance for people with type 1 diabetes.
Section A Overview - Practical AI in Healthcare
Practical AI in healthcare this week comes from the research pipeline rather than the deployment floor: a Kaunas University of Technology team in Lithuania has built a graph neural network that forecasts blood-sugar swings and hypoglycemia risk up to an hour in advance for people with type 1 diabetes. The model performed strongly even on patients it had never seen before, but the researchers themselves caution that real clinical validation - not just historical data testing - still lies ahead. It's a reminder that Europe's practical-AI pipeline runs through university labs as much as hospital wards, with genuine clinical benefit still a few validation steps away.
Section B Overview - Reducing Administrative Burden
Reducing administrative burden this week comes down to one large funding round with genuinely European ambitions: Stockholm's Tandem Health raised €86.5 million to scale an AI-native clinic operating system now live in 14 European markets and more than 10,000 care organizations. Clinicians using the platform report a 29% cut in administrative time and a 30% drop in administrative stress, with NHS audits reportedly putting its note accuracy at 97%. The raise, and Tandem's parallel acquisition of Dutch rival Juvoly, point to a European ambient-AI documentation market that's consolidating around a homegrown champion rather than ceding the field entirely to US platforms.
Section C Overview - How Patients Use AI
How patients use AI this week is anchored by a sobering scoping review: cancer patients, survivors and caregivers like what generative AI already offers - clearer health information, simplified clinical reports - but have had almost no voice in designing the tools built for them. Of 32 studies reviewed, only three even examined who's accountable when AI in oncology gets something wrong, with safety, transparency and equity left largely unexamined. For #patientsuseai, it's a reminder that patient enthusiasm for AI is running well ahead of patient involvement in shaping it.
Summary Section A: Practical AI in Healthcare
Summary: ### Lithuanian Researchers Build AI Model That Forecasts Glucose Swings Up to an Hour Before They Happen
A team at Kaunas University of Technology (KTU) in Lithuania, led by Professor Rytis Maskeliūnas and PhD student Muhammad Abdullah Sarwar, has developed an AI model that combines glucose readings with insulin dosing, carbohydrate intake and physical-activity data to forecast blood-sugar fluctuations and hypoglycemia risk 30 and 60 minutes in advance. The system uses graph neural networks and attention mechanisms to identify which historical data points matter most for a given patient, and it performed well even on people whose data it had never encountered before, based on tests against international type 1 diabetes datasets. Accuracy improved further when the model was trained simultaneously on both glucose forecasting and risk assessment. The researchers are careful to note that further clinical studies are essential before the tool could move from historical data analysis into real-world bedside use. For a field crowded with diabetes-tech promises, it's a reminder that predictive accuracy on paper is only the first step toward something patients can actually rely on.
Hashtags: #PracticalAI #DiabetesTech #Lithuania #DigitalHealthEurope
Source: News-Medical.Net: New AI Model Forecasts Glucose Changes in Type 1 Diabetes Patients
Summary Section B: Reducing Administrative Burden
Summary: ### Sweden's Tandem Health Raises €86.5M to Build an "AI-Native Operating System" for European Healthcare
Stockholm-based Tandem Health has closed an €86.5 million ($100 million) Series B led by the EQT-managed Scaleup Europe Fund, bringing its total funding to roughly €138 million, to expand an AI medical assistant that handles documentation, coding and clinical decision support throughout a patient visit. The platform is now used by more than 10,000 care organizations across 14 European markets, including Ramsay Santé, Humanitas and parts of the NHS, with each AI component independently certified as a Class IIa medical device under EU rules. Clinicians using Tandem reported a 29% reduction in time spent on administrative work, a 30% improvement in job satisfaction and a 30% drop in administrative stress, and NHS audits reportedly found the platform's generated notes to be 97% clinically accurate. Industry analysis frames the raise as a consolidation moment for Europe's ambient-AI documentation market, with Tandem - having already acquired Dutch rival Juvoly - positioned as a homegrown challenger to US players like Abridge and Microsoft's Dragon Copilot. For hospitals weighing where administrative relief will actually come from, it's a sign that European-built, EU-regulated platforms are starting to scale rather than just pilot.
Hashtags: #AdminBurden #Sweden #DigitalHealthEurope #AIFunding
Summary Section C: How Patients Use AI (#patientsuseai)
Summary: ### Scoping Review Finds Cancer Patients Have Had Almost No Say in Designing the Generative AI Tools Built for Them
A scoping review led by researchers at Flinders University in Australia, published in the European Journal of Cancer, screened 2,441 records and analyzed 32 studies on how people affected by cancer - patients, survivors and caregivers - actually feel about generative AI in oncology. Patients responded positively when AI made health information easier to access and simplified dense clinical reports, and their trust hinged on whether responses felt personalized, accurate and emotionally appropriate. But the review's central finding is a gap: as lead author Dr. Bradley Menz put it, "designing AI for patients is not the same as designing it with them," and of the 32 studies reviewed, only three even examined questions of accountability, with safety, transparency, equity and human autonomy barely assessed at all. Co-author Associate Professor Ashley Hopkins called for future research to assess patient perspectives "with greater conceptual breadth and depth." For #patientsuseai, it's a concrete reminder that patient enthusiasm for AI in cancer care is outpacing patients' actual involvement in building it.
Hashtags: #patientsuseai #Oncology #PatientVoice #GenerativeAI
Daily Health AI Chronicle • Edition 257 • September 16, 2026 Practical AI in healthcare news from Europe, Canada, and beyond - focused on clinical deployment, patient impact, and administrative burden reduction. Sources: News-Medical.Net, EU-Startups, EurekAlert! (Flinders University)
Book Lucien for your next event
Keynotes, masterclasses, panels and board-room sessions.