Artificial intelligence in mobile health applications: A comprehensive review of its role in diabetes care

Abstract

This review explores the integration of artificial intelligence (AI) in mobile health applications for diabetes care. It focuses on key AI methodologies - machine learning, deep learning, and natural language processing - and their roles in glucose monitoring, personalized self-management, risk prediction, and clinical decision support. Drawing on recent literature (2018-2024), the study outlines the benefits of AI in improving accuracy, engagement, and precision in diabetes treatment. Challenges such as data privacy, algorithmic bias, and regulatory barriers are also examined. A new section discusses when AI technologies may become burdensome, especially in low-resource settings or for users with limited digital literacy. The review concludes with directions for enhancing model explainability and integrating AI with wearable and Internet of Things devices, emphasizing the need for ethical and equitable implementation in future diabetes management strategies.

Key Words: Artificial intelligence; Mobile health; Diabetes management; Predictive analytics; Clinical decision support; Personalized self-management

Core Tip: This review highlights the transformative role of artificial intelligence in mobile health applications for diabetes care. It synthesizes recent advances in machine learning, deep learning, and natural language processing, examining their use in glucose monitoring, personalized interventions, and clinical decision support. The review also discusses ethical challenges, data privacy, and situations where artificial intelligence may become burdensome. By bridging technology and practice, this study offers insights into building more equitable, efficient, and patient-centered diabetes management systems.

INTRODUCTION

Overview of mobile health and artificial intelligence in healthcare

Emerging mobile health (mHealth) technologies have significantly transformed healthcare delivery by improving patient access, facilitating remote monitoring, and enhancing the management of chronic diseases such as diabetes. According to Khan and Alotaibi[1], mHealth encompasses the use of mobile devices, wearable technologies, and wireless communication in medical and public health practice. The rapid advancements in artificial intelligence (AI), particularly in machine learning (ML) and deep learning (DL), have further expanded the capabilities of mHealth applications, contributing to improvements in disease prediction, diagnosis, treatment, and self-management. AI-driven algorithms enable the processing of large volumes of complex medical data, generating clinically relevant insights that support evidence-based decision-making and optimize patient outcomes[2].

Diabetes represents a major global health challenge, contributing substantially to morbidity and mortality. The increasing prevalence of diabetes places a significant burden on healthcare systems, necessitating more efficient and scalable management solutions. AI-powered mHealth applications have emerged as a potential tool for improving diabetes care, offering functionalities such as continuous glucose monitoring, early risk assessment, personalized lifestyle recommendations, and clinical decision support through automated data analysis[3,4]. These technological advancements signify a shift from traditional episodic healthcare toward real-time, patient-centered management, wherein intelligent systems play a crucial role in facilitating proactive and adaptive disease management strategies[5].

Role of AI in diabetes care: opportunities and challenges

AI has demonstrated great promise in improving diabetes care, including early detection, real-time monitoring, and customized therapeutic interventions. Indeed, AI-enabled predictive analytics have projected an individual’s risk of developing diabetes, considering the lifestyle, genetic, and clinical variables[6]. Furthermore, advanced ML algorithms can analyze the extensive datasets from continuous glucose monitoring (CGM) devices to forecast blood glucose levels with greater accuracy, enabling timely interventions to minimize complications associated with diabetes[7]. AI-based decision support systems also aid healthcare practitioners in optimizing the treatment regimen to reduce human errors[8].

However, despite all this, there are myriad of challenges that prevent the wide-scale adoption of AI in diabetes management. The use of mHealth apps and devices entails the collection and processing of extremely sensitive health information, so concerns over privacy and security remain a significant barrier to its widespread adoption[9]. It has to be aligned with various regulatory frameworks like the General Data Protection Regulation and the Health Insurance Portability and Accountability Act for the privacy and confidentiality of patients. Additionally, most AI models are challenged by issues of generalizability: Algorithms trained in particular populations do not perform well in diverse demographic groups[10]. AI-related bias in healthcare solutions can also lead to discriminatory practices in clinical diagnosis and treatment recommendations, requiring extensive validation and ethical considerations in deploying AI solutions.

Research objectives and scope

As the domains of AI and mHealth practically meld regarding diabetes management, the current paper outlines a state-of-the-art review of AI-based technologies in mHealth applications in relation to diabetes care. To that end, this paper evaluates: (1) The various AI methodologies, including ML, DL, and natural language processing, that are currently being applied in managing diabetes; (2) Evaluates the applications of AI in glucose monitoring, personalized self-management, early risk assessment, and clinical decision support; (3) AI-powered diabetes care solutions face several challenges, including data privacy and bias issues and regulatory compliance; and (4) Discusses future directions for research and technological development toward more effective and accessible solutions for AI diabetes management.

This literature review builds on previous research findings published in peer-reviewed literature and considers their implications in future diabetes care. Although this paper attempts to bridge the gap between innovation in technology and its actual application in clinical practice, discussions here also contribute to those on the role of AI within mHealth and how this may implicate decision-making processes toward more effective, ethical, and equitable solutions within diabetes management.

LITERATURE SEARCH STRATEGY

A literature search was performed according to a pre-planned strategy, which allowed a well-considered, organized approach to reviewing the role of AI in mHealth applications for diabetes management. The methodology was determined prior to the search and was strictly followed in the selection of studies, ensuring that only bodies of work with rigorously peer-reviewed content of appropriate quality were subjected to full-text review. This literature review covered the studies available to date in the following databases: PubMed, IEEE Xplore, ScienceDirect, and Google Scholar, using the terms: “Artificial Intelligence in Diabetes Care”, “machine learning in Diabetes Management”, “Deep Learning in mHealth”, “AI-based Glucose Monitoring”, and “Diabetes Clinical Decision Support Systems”[11].

To reduce the chances of selection bias in the articles to be included, the procedure was based on the search strategy that conforms to the PRISMA guidelines[12]. As per the methodology, the studies were included if they satisfied the following criteria: Published between the years 2018-2024, focused on applications of AI in diabetes management through mHealth, provided data or systematic reviews related to AI techniques, and discussed patients’ clinical outcome and consequences. On the contrary, those studies that failed to demonstrate robust validation were not peer-reviewed or dealt with applications of AI in the management of care unrelated to diabetes were excluded[13].

Next, we assessed their methodological quality based on the research design, data source, and methods for validating the AI models. For example, studies that used large datasets, real-world patient data, and robust ML algorithms were given special attention to establish the proposed solution’s reliability and generalizability. Following this, the selected articles were organized into the following sections, according to their contributions regarding different AI applications for diabetes care.

Classification of AI applications in diabetes care

Based on the earlier discussions, various applications of AI in diabetes care were first categorized under the broad objectives of AI toward better disease management, patient engagement, and clinical decision making. Earlier literature established several prior taxonomies related to the applications of AI in healthcare based on the functions they perform, the data they use, and the predictions they provide[14]. This literature review includes five major categories of AI applications: (1) Predictive analytics for diabetes onset and complications; (2) AI-driven CGM and glucose prediction; (3) AI-based personalized diabetes self-management; (4) AI-enabled risk stratification and early detection; and (5) decision support systems for healthcare professionals[10].

One of the most powerful applications of AI involves predictive analytics, wherein ML algorithms have been applied to risk factors among the patient population to predict disease progression[15]. There are now findings that AI models synthesized from large-scale datasets predict diabetes onset decades before its clinical manifestations with the potential of early interventions that are personalized for risk-reducing strategies[16]. This has seen a marked improvement in glucose control, fundamental in reducing complications and preventing hospitalization, through AI-based analysis of CGM data and prediction of hypoglycemic or hyperglycemic events in real time[17].

Another very relevant category is AI personalization in self-management, which individualizes diabetes care. AI-enabled mobile apps give real-time data and behavioral patterns of patients, making lifestyle and treatment recommendations on diet, physical activity, and medication adherence[12]. These apps create a more engaged and empowered patient role in disease management, which is expected to yield better results in the longer term from diagnosis[13].

Besides personalization in self-management, AI has also been important in early detection and risk estimation using DL and natural language processing techniques to highlight those patients who can develop complications such as diabetic retinopathy and neuropathy. State-of-the-art AI applications trained on huge medical imaging datasets showed excellent accuracy in the identification of these complications, which supports early diagnosis with timely intervention[17].

Finally, AI-enabled decision support underpins optimized treatment regimens and reduces diagnostic errors. In this regard, AI models utilize information from integrated electronic health record (EHR), wearables, and laboratory results to suggest recommendations on treatment that would be personalized. A number of recent investigations have found such systems to improve clinical decision-making and enhance the workflow of diabetes care[15].

While in-depth discussion of each application follows in the next sections, this structured approach to systematically classifying AI applications in diabetes care allows one to synthesize the literature and map out literature gaps for future research. Each application is discussed in further detail with regard to benefits, challenges, and implications for diabetes care.

AI TECHNOLOGIES AND APPLICATIONS IN DIABETES CARE

ML, DL, and natural language processing

AI has combined ML, DL, and natural language processing (NLP) to revolutionize diabetes care. It could further help provide predictive capabilities, real-time decision-making, and personalize treatment recommendations for the disease[18]. Indeed, the accuracy of predicting diabetes, its progression, and probable complications are now expected to be very accurate by using ML algorithms on large databases of complex patient information[19]. Most recently, DL methods have successfully been applied to the analysis of medical images for the detection of diabetic retinopathy, making quite an important contribution to early diagnosis and intervention of this condition[20].

In fact, convolutional neural networks, among various DL techniques, have found their application in automated analysis of retinal images for the early diagnosis of diabetic retinopathy[21]. Relatively recent studies involving large-scale labeled datasets demonstrate the potential of AI-embedded models for detection of small changes in retinal vessels at a level of accuracy beyond that of the clinical standard as it exists today based on human judgment. AI-embedded models could also utilize recurrent neural networks and long short-term memory networks in time-series analysis of glucose data, which enable glucose level forecasts to be more accurate than those inferred from the previous patterns. Guo et al[22] shows how this predictive power is expressed in proactive diabetes management and is reflected in a reduced risk for such acute complications as diabetic ketoacidosis.

NLP has found various applications in the field of diabetes care, including AI-integrated virtual assistants that help patients self-manage their disease by providing real-time feedback on blood glucose levels. Furthermore, recent evidence shows the potential of state-of-the-art NLP models in eliciting meaningful information out of unstructured EHR data-sources, like progress notes, for identifying trends in patient behavior, predicting medication nonadherence, and even facilitating the construction of personalized care plans. This is particularly valuable in multilingual or poorly structured clinical documentation; with this approach, clinicians’ workloads are alleviated and the data’s utility is enhanced across various fragmented systems. According to Rashid et al[23], this ultimately led to the development of AI-powered, NLP-enabled chatbots capable of answering patient queries, advising dietary changes, and reminding patients about medication adherence. These applications make the patient process easier while reducing the burden on healthcare providers. According to Tarumi et al[21], NLP methods have been applied to extract relevant care information for diabetes management from unstructured clinical notes in the EHR. Data automatically extracted in this manner can be used to inform clinical decisions and improve clinician workflows.

AI-driven glucose monitoring and prediction

The benefits of CGM have exponentially increased through several AI advancements, allowing predictive models to forecast glycemic variations to limit hyperglycemia and hypoglycemia[24]. Historical glucose trends and further physiological data make the AI-augmented CGM systems provide personalized predictions of future glucose levels. That will allow real-time decision-making concerning insulin dosing, significantly improving the possible quality of glycemic control and complication rates[25].

Comparatively, CGM significantly reduces the occurrence of adverse glycemic events by 30% to 50%, as evidenced by various studies using conventional monitoring techniques[7]. A novel front integrating photo- and electrochemical sensors into an AI-based noninvasive glucose monitoring method is under development, which again aims to reduce the burden imposed on patients by repeated finger-prick tests[26]. The AI-enabled optical technology of Raman spectroscopy measures glucose concentration through the skin, noninvasively, without needing a blood sample[22]. Such novel technologies will ensure more regular glucose monitoring by patients and relate to a better prognosis.

The integration of AI algorithms into smart insulin pens and closed-loop systems is even further extended, commonly termed artificial pancreas technologies[23]. Using deep reinforcement learning models, these systems optimize insulin delivery in real time, adjusting insulin dosages dynamically based on glucose level changes, physical activity, and dietary intake[21]. Compared with traditional models of insulin, studies suggest that dosing with the assistance of AI leads to improved glycemic variability while reducing the manual burden of insulin management[25].

AI for personalized diabetes self-management and coaching

AI helps in personalizing diabetes management: Treatment recommendations are based on variable patient information. In this regard, AI-enabled mHealth apps will soon be able to suggest personalized dietary plans, exercise schedules, and medications based on real-time data collected from wearables and CGM[21]. AI models will analyze lifestyle behaviors and physiological responses to make changes in the regimen, thus improving patient independence and adherence to treatment[22].

Recent advancements in reinforcement learning have led to AI-based decision-making frameworks that change diabetes treatment dynamically based on day-to-day glucose and physical activity fluctuations[27]. AI-empowered digital therapeutics could intervene to attain a better glycemic level in type 2 diabetes patients by delivering interactive behavioral coaching via a smartphone application[28]. In this regard, behavioral AI algorithms adjust the intervention strategy based on patient engagement data in order to maintain high adherence to treatment regimens[29].

The major strength of AI-based self-management tools lies in the multi-dimensional approach that allows the synthesis of various health parameters such as heart rate variability, sleep, and stress for comprehensive diabetes management[11]. Such a multi-dimensional perspective gives way to predicting lifestyle-induced fluctuations in glucose with the help of AI-driven systems which may then suggest measures to maintain optimal glycemic control[23]. Recently, AI-powered voice recognition technology has been used to develop hands-free interaction with applications for diabetes management, thus enhancing usability even for those patients who may have physical challenges[22].

AI in early detection and risk assessment

AI outplays traditional predictive diabetes models by integrating comprehensive data across various domains - such as EHRs, genetic and imaging data, and information from wearable sensors - to offer detailed early detection and risk assessment of diabetes and its complications. Oftentimes, the above models utilize single or few modalities, focusing on standard clinical metrics including, but not limited to, the measures of fasting blood glucose, the hemoglobin A1c level, and body mass index. In comparison, the AI model integrates several multidimensional data sources to search for subtle signals resulting in an augmented risk for developing diabetes even before overt clinical symptoms present[23].

Perhaps the most significant contribution of AI to the development of early prediction has taken the form of ML algorithms used in analyzing patient information records to predict the onset of diabetes. Other supervised learning techniques, such as support vector machines, random forests, and gradient boosting algorithms, provide useful alternatives towards highly accurate ways of stratifying a population based on the risks attributed to each individual due to various factors that predispose them to diabetes[25]. For instance, when applied to large-scale epidemiological datasets, AI models trained thus could estimate the risk of developing diabetes in the next five years, thus giving room for early intervention through preventive measures[21].

Apart from clinical records, AI-powered image analysis has spread its wings in revolutionizing the screening for complications due to diabetes. For instance, recently published DL algorithms applied to retinal imaging show excellent performance in detecting diabetic retinopathy, one of the leading causes of blindness worldwide[30]. Convolutional neural networks trained on extensive and accurately annotated datasets assess retinal images for microaneurysms, hemorrhages, and neovascularization with performance levels compared to ophthalmologists[22]. Several AI-based systems screening for retinopathy have already been approved by the United States Food and Drug Administration and subsequently incorporated into clinical practice for fast, automated diagnoses with no eye specialist necessary.

On the other hand, there is growing evidence suggesting that this AI could lead to earlier recognition of diabetic neuropathy. The classical method for diagnosis through nerve conduction studies is an invasive and time-consuming approach. In this context, new AI models based on data collected by wearable sensors, gait analysis, and electromyography have come up with new early indicators of developing neuropathic damage markers, which would allow timely therapeutic intervention[8]. This certainly extends AI-driven risk estimations of diabetic complications to neuropathy. Overall, DL models estimate electrocardiograms and echocardiograms efficiently and accurately and detect early signs of diabetic cardiomyopathy for medical treatment[20].

Genomic analysis for estimating diabetes risk with the support of AI is another emerging field. Genome-wide association studies have aligned several different genetic variants with the risk of developing diabetes. The integration of genetic data into AI models, along with information on environmental and lifestyle factors, may improve diabetes risk prediction beyond current methods[23]. In conjunction with federated learning techniques, this allows large-scale, multi-institutional collaborative AI model training without compromising patient data privacy and would further strengthen and enhance the generalizability of the developed models for predicting diabetes risk[22].

These complex data analyzed with the help of AI, coupled with personalized diabetes prevention approaches, provide an entirely new perspective. AI-based models remove the one-size-fits-all concept from diabetes prevention, offering tailored advice on diet, lifestyle change, and pharmacological intervention with real-time risk assessment leads. It provides an offer that should be placed on the table for integrating AI in workflows of early detection and risk assessment of diabetes and its complications to reduce burdens on diabetes and its complications worldwide.

AI-enabled decision support for healthcare providers

The use of AI-driven decision support systems in diabetes management has revolutionized not only clinical decision-making regarding the choice of treatment but also improved outcomes in patients with diabetes. Diabetes management suggestions are based on the integration of patient-related glucose monitoring trend data, laboratory test results, medication adherence, and lifestyle information, therefore, reflecting the evidence from the literature tasks that such a system is likely to allow clinicians to build a data-driven and less variable treatment plan that adheres to clinical guidelines for diabetes management[22,23].

Insulin dose optimization is one of the most important applications of AI in decision support. Conventional insulin therapy necessitates a rigid scheme of blood glucose monitoring and dose adjustment, which is conducted manually, making it cumbersome for both patient and clinician alike. Several recently developed computerized insulin dosing algorithms use data from CGM and other continuous monitoring devices to adjust insulin delivery dynamically and in real time[31].

With regard to hybrid closed-loop insulin pumps, such systems have improved glycemic control, reduced hypoglycemia risk, and improved quality of life for the patients using them[32]. AI-based decision support systems also assist in various medications used in treating type 2 diabetes mellitus[23]. In this way, AI could analyze patient responses to several antihyperglycemic agents, giving personalized recommendations for achieving the desired level of glycemic control without increasing unwanted side effects. Reinforcement learning models update treatment recommendations continuously based on long-term observation of the patient, which ensures that the therapy regimen remains relevant in the long term. According to Khodve and Banerjee[8], such tools identify patients at risk of early treatment failure, making possible prompt therapeutic changes.

Other important areas where AI facilitates support in managing diabetes are its complications. One commonly associated complication of diabetes is cardiovascular disease, followed by renal impairment and metabolic syndrome, which assert a considerably multidisciplinary approach to care. In that respect, risk models using AI incorporate data from EHR, imaging results, and biomarker studies to yield integrated risk estimates for nephropathy and other complications of diabetes, including cardiovascular disease[5]. The resulting predictions allow for intervening much earlier than would otherwise be possible; thus, decreasing the rates of hospitalization and healthcare costs[21].

This has led to the concept of support extending from individual patient care to population-based management of diabetes. AI could offer predictive analytics tools to help healthcare providers identify a population at risk and, therefore, design prevention programs targeted at the right population and health interventions personalized to the individual. In that respect, AI not only plays a very important role in facilitating early screening and community-based interventions but also contributes to a significant reduction in the incidence and severity of diabetes among high-risk populations[20,22].

While AI clinical decision support shows great promise, several challenges remain. The central concerns with AI models are interpretability and explainability, given that any healthcare professional must first trust and understand the model’s recommendation to use it in patient care[18]. Besides these, regulatory frameworks and ethics that ensure the safe and effective adoption of AI DSS in clinical practice are of significant importance[8]. Lastly, the transferability of AI models needs to continue across population groups to adequately reduce the risk of bias in treatment recommendations for diabetes, as well as the transferability of models across diverse populations[23].

AI-enabled decision support systems are revolutionary in diabetes management, giving precision-driven treatment recommendations, alleviating the burden of care, and improving outcomes in patients. Therefore, adding AI technologies in the future will develop the routine clinical workflow in diabetes management into a more personalized, proactive, and data-rich approach for all patients living with diabetes.

Summary table: AI applications in diabetes care

Table 1 summarizes the 5 major AI application areas in diabetes care discussed in this review. Each of these is characterized by its dominant AI techniques and contribution to diabetes management. Predictive analytics uses ML to project disease onset and complications, whereas real-time glucose monitoring involves CGM using DL methods, such as long short-term memory, for real-time glucose prediction. Personalized self-management involves lifestyle intervention. This table not only illustrates the breadth of AI integration in diabetes management but also emphasizes the distinct technological methodologies underpinning each function. Such classification enables a clearer identification of strengths and research gaps, aiding in the future optimization of AI-based mHealth strategies.

Table 1 Comparison of artificial intelligence applications in diabetes care.

AI application area	Primary AI techniques used	Key benefits
Predictive analytics	Machine learning, ensemble models	Early diabetes prediction and prevention
Continuous glucose monitoring	Recurrent neural networks, LSTM	Real-time glycemic control and hypoglycemia prevention
Personalized self-management	Reinforcement learning, behavioral AI	Tailored interventions based on patient data
Early risk assessment	Deep learning, NLP	Such as diabetic retinopathy and neuropathy
Clinical decision support	Rule-based systems, reinforcement learning	Optimized treatment plans and reduced clinician burden

AI: Artificial intelligence; LSTM: Long short-term memory; NLP: Natural language processing.

BENEFITS, CHALLENGES, AND LIMITATIONS

Improved patient engagement and self-management

The rapid evolution of AI in contemporary medicine has fostered a sea change in patient involvement and self-management within diabetic care through personalized advice, real-time monitoring, and predictive analytics. AI-imbued mHealth apps empower patients through the provision of CGM data, automated insulin dosing suggestions, and lifestyle changes recommendations, all underpinned by real-time data analysis[33]. AI-enabled virtual health assistants have been shown to enhance adherence to treatment regimens by sending reminders related to medication intake, exercise, and dietary modifications[34]. This invariably reduces the cognitive burden on patients, enabling them to focus on sustaining optimal glucose with integrated minimal effort.

Self-care applicators have demonstrated the potential to influence better glycemic control and reduce admissions for diabetes in a number of significant studies[35]. For example, reinforcement learning algorithms could adjust dynamic insulin dosing; patients could dynamically achieve the optimal therapeutic recommendation based on a unique physiologic response to treatment[36]. Furthermore, AI integrated into wearable devices enables continuous feedback on glucose fluctuations, which allows timely notifications on preventive actions to be taken by a patient before the onset of hyperglycemia or hypoglycemia[37]. Of course, while all of these developments enhance engagement among patients, there are also grave concerns about the increasing divide in digital literacy. The very tools through which care is delivered do not become available equitably in all instances, and disparity will thus also follow in diabetes management. AI-based technologies can be daunting for older and lower-income individuals, who become targets for specific educational and access interventions[38].

AI’s role in precision medicine and healthcare accessibility

AI has revolutionized precision medicine by developing tailored treatment approaches from patients’ data. The AI modalities utilize information from the genetic, behavioral, and clinical data sets to optimize medication regimens, forecast disease progression, and adjust treatment plans in their course[23]. However, it is simple to use, and the AI-supported clinical decision support system has greatly empowered health providers in personalizing diabetes treatment, with the end results of improved glycemic outcomes and reduced complications[39]. Through AI, clinicians can prescribe meal plans customized to the patient’s metabolic response, suggest exercise regimens, and optimize pharmacological interventions.

One of the most significant impacts of AI on diabetes management is its potential to improve access to care in poorly resourced settings. For example, AI-enabled diagnostic tools screen remote consultation for diabetic retinopathy and neuropathy, thus permitting early diagnosis and treatment interventions[5]. This is particularly important in poor resource settings where access to endocrinologists and other diabetes specialists is very limited. AI-enabled telemedicine platforms provide virtual consultations and remote monitoring, reducing the healthcare infrastructure burden and increasing access to diabetes care in remote areas[40]. Moreover, AI chatbots and virtual assistants provide 24/7 support, ensuring that patients will be counseled regarding their conditions, even in instances when healthcare providers may not be available.

However, despite this progress, many high implementation costs and technological barriers remain to applying AI in resource-poor settings[38]. Indeed, many AI models are created with complex computational requirements and large-scale training datasets, which imply problematic ambitions for applying them at rural healthcare levels. Besides, variable connectivity and infrastructural disparity inhibit the adoption of AI in certain areas. All these require public-private partnerships, investments in digital health infrastructure, and AI solutions adapted to resource-constrained environments.

Data privacy, security, and ethical considerations

The role of integrating AI in diabetes care concerns primarily revolves around data privacy, security, and ethics related to healthcare applications based on AI. These applications depend on massive databases of patients’ information, and such needs impose heavy responsibilities relating to data protection against unauthorized access or data leaks. According to Sridhar and Lakshmi[41], the details found in health records are very sensitive, such that any data leak may cause identity theft, insurance fraud, or some types of third-party exposure. Several regulations are intended to protect patients’ information, such as the General Data Protection Regulation and the Health Insurance Portability and Accountability Act; however, they sometimes prove quite complicated to comply with when any cross-border deployment of AI is involved[42].

One of the most serious ethical issues is that AI healthcare systems might reduce human intervention in medical decisions. Although AI proves invaluable in increasing the accuracy of diagnoses, it will never replace the human touch in medical recommendation decisions[43]. Some other AI applications in medicine have earned the well-deserved reputation of “black boxes”. These DL networks raise serious issues related to transparency and explainability in making clinical decisions. Patient safety often finds itself in the crosshairs when clinical judgment artificially competes with AI-derived insight. One of the most serious issues of AI today is algorithmic bias, as AI models trained on homogenous data sets make invalid predictions on heterogeneous populations, thereby producing inequality in diabetes outcomes[44].

However, much is being done. Heading off these concerns, the work is now developing an area of explainable AI that offers interpretable insights explaining to clinicians how AI models generate their recommendations. An even more active area of research is the possible applications of federated learning, where AI models are trained on distributed data sets while keeping patients’ information private[45]. Such enhancements in ethics and AI privacy will indeed go a long way toward building trust in the equitable application of AI in diabetes care.

AI model generalizability, bias, and regulatory barriers

AI models in diabetes care have issues with generalizability as they have been trained on certain datasets, and consequently, the demographics are well represented. ML algorithms require an extensive validation process to be carried out on several different populations so that they are reliable and accurate[45]. Some authors propose federated learning, a decentralized approach to training AI, as a means through which the generalization of AI models could be enhanced without compromising patients’ confidentiality[46]. On the other hand, federated learning has so far remained quite impractical, given its demands in terms of synchronization between health institutions and data-sharing protocols.

The other AI challenges translate into the regulatory agency issues. AI-enabled medical devices undergo a complicated approval process that thoroughly tests clinical validation processes in compliance with the relevant legal requirements[47]. Accordingly, any model of AI for diabetes prediction and management has to go through clinical trials to test its safety, efficacy, and reliability before recommending it as a standard medical practice.

The area still hotly debated includes liability for medical decisions generated by AI. When the diagnosis or treatment recommendation produced by an AI system is wrong, proving who is responsible for the ultimately flawed decision will not be easy. The legal structures must adapt to liability issues without losing the basic premise that the role of AI is to assist, not supplant, clinical decision-making[11]. It is with respect to such regulatory challenges that AI in diabetes management is secure and effective. Through these outlined hurdles, AI will continually forge ahead in changing diabetes management, contributing to ethical, just, and secure health care. The next sections discuss the possible future and paths toward enhancement strategies in order that diabetes care can realize its full potential with the help of AI.

When AI becomes a burden: use scenarios and limitations

While AI has numerous potentials related to diabetes management, it can also become burdensome when implemented without proper contextuality and infrastructural support. For instance, AI-based tools often need high computational power or even robust internet connectivity-unavailable in most low-resource settings. In addition, interaction with AI-driven applications or chatbots may not be viable for patients possessing low digital literacy or with most older adults, which may in itself widen health inequities. Equally significant is the potential overreliance on automated decision systems in day-to-day clinical practice to undermine the autonomy of health professionals or delay crucial interventions in those scenarios where human intuition is indispensable. This would mean AI systems must be selected relative to the considerations of the context, population, and resources within AI enablers and not embedded as barriers.

CONCLUSION

Enhancing AI accuracy, explainability, and generalizability

While AI technologies continue to mature, several questions remain around their accuracy, explicability, and generalizability for diabetes care. Training AI models on diverse datasets is paramount to their generalizability across different populations, minimizing biases and augmenting clinical applicability[16,48]. Regarding this, federated learning allows the training of AI models on decentralized data without compromising patient confidentiality, thus enhancing generalization and robustness[49]. Secondly, improving AI interpretability is another important issue as healthcare professionals need to be confident in the recommendations coming from AI. Various AI model methods, ranging from decision trees to attention-based neural networks, develop insight into how AI reaches a decision, hence increasing transparency in the clinical setting[50].

Integration with wearable devices and Internet of Things

Another promising future line in diabetes management revolves around integrating AI in wearables and the Internet of Things. Devices, including CGM, smart insulin pens, and fitness trackers, yield enormous amounts of real-time health information, which AI would utilize to generate personalized feedback and predictions[51]. With Internet of Things connectivity, diabetes management systems embedded with AI can pull together patient information in an unprecedented way from all sources - making recommendations on insulin dosing and lifestyle advice even more precise[52]. For example, AI-augmented remote monitoring makes it possible for the health service to reach its timely proactive interventions before complications arise even in rural and low-resource settings[42].

Ethical, policy, and regulatory considerations

While there remains the promise of AI in the management of diabetes, many ethical and regulatory challenges prove crucial in making such technologies safe and just for implementation. AI-augmented diagnostics and treatment recommendations must be rigorously regulated by authoritative bodies such as the United States Food and Drug Administration and the European Medicines Agency to ensure consistent clinical reliability and patient safety[44]. Notably, AI must be free from biases that will affect specific demographic groups disproportionately. Sheng et al[20] stated that there must be comprehensive policymaking on healthcare solutions driven by AI, balancing innovation and ethical considerations of autonomy and privacy of data. Standardization of AI governance and compliance frameworks across dissimilar jurisdictions will be paramount in further guiding the integration of AI into diabetes care[53].

Summary of key insights and future outlook

AI has already shown significant potential in revolutionizing diabetes care by improving early detection, personalizing treatment, and enhancing patient self-management. It will require further improvements in AI’s accuracy, explainability, and integration with wearable technologies to enjoy all the potential benefits that this technology has to offer[33]. Moreover, refining the ethical and regulatory frameworks will ensure that AI remains safe, effective, and accessible to all patients with diabetes[54].

Future research will be oriented toward optimizing AI algorithms for predictive improvement, dataset expansion for model generalizability, and workflow-congruent AI-driven DSS[50]. The future of AI in the management of diabetes is likely to be shaped by the work and collaboration of healthcare professionals, researchers, policymakers, and groups representing patients. Further evolution in AI technologies will probably see the further development and embedding of AI in the management of diabetes, new openings for improved patient outcomes, and effective healthcare delivery[16,48].

AI in diabetes care thus holds revolutionary promise, but its general implementation is hedged in by several current challenges linked to accuracy, integration, ethics, and regulation. With further research, innovation, and regulatory development, AI-driven solutions gradually gain relevance toward improving diabetes management and the quality of life for millions worldwide.