Artificial intelligence shows promise in diabetes self-care, but trust is key

AI technologies hold immense potential to transform diabetes self-care by improving monitoring, enabling personalized interventions, and alleviating pressure on healthcare providers. District nurses, in particular, could benefit from AI-driven alerts and recommendations, freeing up time for person-centered care and reducing administrative load.

Artificial intelligence (AI) is transforming the landscape of self-care for patients with type 1 and type 2 diabetes, offering predictive accuracy, decision support, and real-time monitoring tools that can significantly enhance quality of life. A study, titled “Artificial Intelligence as a Tool for Self-Care in Patients with Type 1 and Type 2 Diabetes—An Integrative Literature Review,” published in Healthcare provides a comprehensive examination of how machine learning technologies like XGBoost, MLP, RF, and ChatGPT-4 are being integrated into diabetes care across multiple countries.

The researchers, based in Sweden, applied a rigorous methodology to analyze empirical studies from China, the United States, the United Kingdom, and other nations. Their goal was to evaluate how AI technologies support self-care in diabetes by helping monitor glucose levels, predict complications, assist in wound management, and address patients’ informational needs. Their findings suggest that while AI tools hold significant promise for empowering patients, concerns related to education, trust, and data security remain critical barriers to widespread adoption.

How does artificial intelligence improve blood sugar monitoring and prediction?

One of the core contributions of AI in diabetes care is its ability to support blood glucose monitoring and prediction. The review highlights several machine learning algorithms such as Extreme Gradient Boosting (XGBoost), multilayer perception (MLP), and Random Forest (RF) as highly effective in forecasting hypoglycemic and hyperglycemic episodes. These tools process vast amounts of patient data, including insulin use, HbA1c levels, and previous health episodes, to deliver accurate, real-time risk assessments.

For instance, XGBoost consistently outperformed other models in predicting glucose variability, especially during high-risk periods such as fasting during Ramadan or during nocturnal hours when hypoglycemia is harder to detect. Similarly, MLP models showed high performance in detecting nighttime blood sugar drops in patients with type 1 diabetes, offering critical support for avoiding life-threatening events. Random Forest models also demonstrated superior accuracy in predicting hypoglycemia during physical activity, a common trigger for sudden glucose dips in diabetic patients.

The integration of AI with continuous glucose monitoring (CGM) systems further enhances predictive capability. These systems continuously collect glucose data, which AI algorithms use to anticipate dangerous fluctuations before they occur. The predictive power of AI not only aids in glucose management but also reduces anxiety for patients and caregivers, who often struggle with the uncertainty of managing chronic conditions in real time.

Can AI be trusted to support decisions on diabetic complications like foot ulcers?

Another major area explored by the study is the use of AI as a decision support tool in managing complications such as diabetic wounds and amputations. Diabetic foot ulcers, which can lead to severe infections and amputations if untreated, represent a high-risk area where early detection is critical. AI-powered applications such as the CARES4WOUNDS (C4W) system were shown to accurately assess wound size and severity, performing comparably or better than traditional methods.

In one case, the AI tool ChatGPT-4 was tested for its ability to recommend levels of amputation in severe foot ulcers. The AI matched clinician recommendations in over 83% of cases, demonstrating not just computational accuracy but potential for real-world application in clinical decision-making. Moreover, machine learning models like XGBoost and RF were again validated as strong predictors for the risk of amputation in diabetic foot ulcer grade 3 patients.

These tools offer clinicians and patients timely insights that can shape interventions and treatment strategies. However, their success also depends on the reliability of underlying datasets and algorithm transparency. Overreliance on AI without clinical judgment, or lack of understanding of model limitations, can lead to overdiagnosis or inappropriate treatment paths, highlighting the need for careful integration and clinician oversight.

What do patients want from AI tools in managing diabetes?

Perhaps the most human-centric finding in the review is that patients with diabetes are not just passive recipients of AI technology—they have clear demands and expectations. Across the studies, patients expressed a strong desire for education, usability, and transparency in AI systems. Many requested brochures, digital guides, and access to peer experiences to better understand how to use these tools effectively and safely.

One example cited involved young adults with type 1 diabetes transitioning to university life. AI-enabled sensors and apps allowed for remote blood glucose monitoring, not only benefiting patients directly but also involving roommates and friends through alert systems. This feature increased the patient’s sense of independence while also building a safety net of support.

Despite these benefits, patients also voiced concerns around data privacy, safety, and trust. The study emphasizes that education must accompany AI deployment. Without proper onboarding and explanation of how these tools function, and how patients can interpret or override AI decisions, there is a risk of alienation or misuse. Patients need to feel empowered, not overwhelmed, by digital health innovations.

From a policy standpoint, this points to the need for healthcare systems to invest in digital health literacy, ensuring patients understand their AI tools as much as they understand their medications. It also highlights the ongoing ethical debate around data governance, consent, and AI accountability, especially as predictive tools become more integrated into daily care routines.

What does this mean for the future of AI in diabetes care?

AI technologies hold immense potential to transform diabetes self-care by improving monitoring, enabling personalized interventions, and alleviating pressure on healthcare providers. District nurses, in particular, could benefit from AI-driven alerts and recommendations, freeing up time for person-centered care and reducing administrative load.

Yet the path forward is not without challenges. Overreliance on AI may reduce patient autonomy, and the potential for algorithmic error poses risks that require human oversight. Additionally, the study notes that while AI can significantly assist in routine monitoring and prediction, it cannot replace the nuanced judgment, empathy, and communication offered by trained healthcare professionals.

To ensure safe and effective integration of AI into diabetes care, a multi-stakeholder approach is essential. Policymakers must enforce data protection and regulatory standards. Clinicians must be trained in interpreting and validating AI outputs. Patients must be given tools and education to make informed choices. Only then can AI be trusted to deliver not just smarter healthcare, but more human-centered care.