The case for AI-enabled diabetes care in primary care

Professor Samuel Seidu, a GP and professor at the University of Leicester, looks at the potential benefits of AI in treating diabetes in primary care.

This is part of the Pulse Partners series. This article has been paid for by Roche Diagnostics, with editorial input by Pulse. The opinions in this article do not necessarily reflect the views of Pulse.

Caring for people with diabetes in general practice is becoming increasingly challenging. Rising prevalence, more patients with multiple long-term conditions, workforce pressures and limited appointment capacity mean GPs are managing complex needs with little room to pause.

According to Diabetes UK,¹ diabetes now affects more than 5.8 million people in the UK, with higher rates in deprived communities and certain ethnic groups, where wider social pressures often compound the burden. For many practices, this is a daily reality.

Much care still relies on intermittent data – HbA1c measurements, structured reviews and patient recall. While these tools reflect what has traditionally been available, they can make early changes hard to spot and put timely interventions at risk.

At the same time, the NHS 10-Year Health Plan calls for a shift from analogue to digital, from reactive to preventative care, and from hospital-centred models to care closer to home.² For GPs under sustained pressure, the question is not whether these goals are right, but how they can be achieved in a way that genuinely supports day-to-day practice.

This is where the growing use of artificial intelligence (AI) may offer practical support.

The growing role of AI in continuous glucose monitoring

Increasingly, AI is underpinning the data generated by continuous glucose monitors (CGM). This helps identify patterns, flag emerging risks, and generate predictive alerts, all designed to surface clinically meaningful information without overwhelming clinicians or patients with raw data. In practice, this means patients have clearer, more actionable information at their fingertips.

For GPs, these advances provide a far more nuanced picture of glycaemic variability across time of day, eating habits, physical activity, and medication. This goes well beyond what intermittent testing alone can show. Access to this richer data can support more confident, individualised decisions: identifying patterns of nocturnal hypoglycaemia, spotting post-meal spikes, informing medication titration, and shaping more focused conversations about lifestyle changes.

Crucially, predictive and trend-based features may also support better prioritisation. By helping to identify patients who are drifting out of range or showing early signs of deterioration, the AI-driven insights from CGM can enable earlier review, more targeted medication or lifestyle optimisation, and more focused use of remote consultations. In a setting where time and resources are limited, this ability to direct attention where it is most needed is particularly valuable.

This does not reduce the importance of the GP’s role. Patients still need interpretation, reassurance and clinical context – particularly those living with multiple conditions, health anxiety or limited confidence with digital tools. Technology can support these conversations, but it cannot replace the relationship at their centre.

Multidisciplinary working has long been central to effective diabetes care, with nurses, pharmacists and the wider team building continuity and trust around the patient. The difficulty is ensuring the right expertise is brought in at the right time, rather than once a problem has already escalated.

AI-enabled CGM can support a more coordinated approach by identifying clinically meaningful patterns earlier. For example, a patient experiencing intermittent nocturnal hypoglycaemia may previously have presented reporting they are ‘feeling off’, waking tired or with headaches, but with limited objective data to guide changes. The issue might only surface at routine review, or worse, after a more significant hypoglycaemic episode.

With AI-enabled CGM, recurring overnight trends can be detected and flagged proactively. This could trigger a targeted medication review with a practice pharmacist to adjust insulin timing or dose, or input from a diabetes nurse specialist to review injection technique or evening carbohydrate intake, all before escalation to urgent care.

In this way, technology strengthens rather than disrupts established team-based care. By highlighting which patients need timely intervention and directing them to the most appropriate professional, practices can prioritise more effectively, protect GP time for complex decision-making, and deliver earlier, joined-up support within the community.

There is also a real opportunity to address health inequalities, but only if digital innovation is introduced thoughtfully. When inclusion is considered from the outset, technology can narrow gaps in care rather than widen them. In practice, this means using flexible models of deployment. For example, some practices may offer CGM devices with remote monitoring for patients comfortable with apps and home data sharing, while others provide in-clinic review sessions or nurse-supported set-ups for those less confident with technology. Approached in this way, innovation can help strengthen access to and continuity of care, particularly in more deprived communities, rather than becoming another barrier to care.

Early diagnosis and risk prediction

While CGM provides a clear and practical example of how AI can support diabetes care, it also reflects a wider shift already underway. AI is increasingly being used in diabetic eye screening to analyse retinal photographs more quickly and consistently.³ Thousands of images can be reviewed in minutes, helping to identify early signs of damage – including diabetic retinopathy, a leading cause of blindness – sometimes before changes are obvious to the human eye.

Beyond complication management, AI is also being explored in early diagnosis and risk prediction. By analysing routine clinical data, emerging systems may help identify patterns that signal increased diabetes risk before it becomes clinically obvious. Researchers are currently examining whether combinations of features within children’s GP records – such as urinary infections, bedwetting and family history – could help flag those at higher likelihood of developing type 1 diabetes, enabling earlier testing and safety-netting to reduce the risk of diabetic ketoacidosis.⁴

Similarly, ongoing work is investigating whether ECG data might reveal patterns associated with future type 2 diabetes risk, potentially years before diagnosis. While still developing, these approaches point towards a more preventative model of care – supporting earlier conversations, closer monitoring of higher-risk patients and more targeted intervention within primary care.

AI-enabled diabetes care is not a solution to every problem, nor should it be presented as one. But it does offer a way to better align policy ambition with clinical reality – supporting prevention, prioritisation and multidisciplinary working while respecting the pressures facing primary care.

As general practice continues to evolve, the opportunity lies in shaping how these tools are used, ensuring they support those delivering care as much as those receiving it.

References: