The implication of artificial intelligence in diabetic ketoacidosis

ATTD 2021 Invited Speaker Abstracts.

Artificial intelligence (AI) is increasingly being used to improve patient care and management. In this paper, we propose explainable AI (XAI) models for predicting severe hypoglycemia (SH) and diabetic ketoacidosis (DKA) episodes in adults with type 1 diabetes (T1D) and relapses in adults with relapsing-remitting multiple sclerosis (RRMS). We follow a three-step process in this study: (1) develop baseline machine learning (ML) models, (2) improve the models using ReliefF feature selection technique, and develop sex-stratified models, (3) explain the models and their results using SHapley Additive exPlanations (SHAP). We built six ML models (XGBoost, LightGBM, CatBoost, AdaBoost, random forest, and linear regression) for all scenarios. Applying the ReliefF feature selection led to improved model performance in predicting all outcomes compared to the baseline models. Additionally, sex-stratified models further improved the prediction of SH episodes and relapses. The F1 scores for predicting SH episodes in male and female patients were 84.07% and 84.95%, respectively, and the DKA prediction model achieved an F1 score of 78.67%. The proposed relapse prediction models outperformed existing models with F1 scores of 84.55% (males) and 76.11% (females), and ROCs of 70.26% (males) and 69.05% (females). Our results highlight the importance of considering sex differences, socioeconomic factors, and physical and mental health in medical outcome prediction. Boosting ML algorithms were found to be effective in detecting SH and DKA in T1D patients and relapses in RRMS patients compared to conventional tree-based ML and statistical models.

Diabetic ketoacidosis (DKA) is a severe metabolic complication of diabetes, posing significant challenges in clinical management due to its complexity and associated risks. Factors such as delayed diagnosis, poor treatment adherence, and comorbid conditions, including severe infections, exacerbate the clinical scenario. This article highlights the key challenges in managing DKA, emphasizing the importance of rigorous monitoring of acid-base balance, fluid replacement, glycemic control, and correction of electrolyte disturbances, such as hypokalemia. Emerging strategies include advances in protocols leveraging artificial intelligence, patient health education, and the development of ultra-rapid-acting insulins. However, barriers such as limited access to high-quality care, inadequate healthcare team training, and variations in clinical protocols hinder effective management. Future perspectives underscore the importance of integrated approaches, educational programs, and innovative technologies to improve outcomes and reduce mortality associated with DKA.

Objective: To compare clinical recommendations on type 1 diabetes management based on the ADA’s Standards of Medical Care - 2022 generated by two generative artificial intelligent (AI) platforms for internal consistency (vertically within a platform) and external consensus (horizontally between platforms and with clinicians’ clinical decisions). Methods: A complex clinical case with type 1 diabetes was selected; the patient has a history of severe hypoglycemia and diabetic ketoacidosis (DKA). Before entering the case into each AI platform for analysis, the clinical note - chief complaint, subjective, and objective information - was edited and organized to enhance clarity and to standardize the language. The case was analyzed by two AI platforms three times; all AI responses were reviewed and compared to the clinician’s assessment and plan, thus taking a qualitative data analysis approach. Results: Conclusions: Although this initial step only included one case, both AI platforms could generate clinical recommendations with high internal consistency, especially in the effectiveness measures. For external consensus on the safety measures, ChatGPT emphasized the specific safety recommendations that aligned closer with the standards of care. Further analyses are warranted to include more cases, ensuring this approach can potentially augment clinician’s care practices in type 1 diabetes. Disclosure C.F. Young: Other Relationship; Abbott. S. Wong: None.

There is an increasing trend of observing Ramadan for people with diabetes (DM) although they are exempt from fasting. DM itself is also a growing epidemic in muslim people especially in the Middle-East and North Africa (MENA) Region. Hypoglycaemia (hypo), significant hyperglycaemia (glucose level equal or over than 300 mg/dl) are the cut off numbers to break the fast according to the international authorities. Not only these two, but dehydration, thrombosis, diabetic ketoacidosis and hyperosmolar non-ketotic coma are the fearful complications during the observance. Glycaemic control can be obtained with careful attention to the glucose readings throughout the day while taking into consideration of fasting and postprandial periods (after sahoor, and iftaar). Time in Range (TIR) as well as glucose variability are important factors, -lately emphasized for diabetes management other than Hba1c (A1c)- so as to get rid of later complications -macrovascular and microvascular ones- provoked by advanced glycation end-products (AGEP) and oxidative stress. Anxiety of everyday finger pricking for glucose measurement can be replaced with artificial intelligence (AI) used equipments for those who can afford. They are either part of continuous glucose monitoring systems (CGMS) or insulin pump delivery machines giving rise to less glucose spikes, less use of total insulin and smart warnings for hypo.

From the beginning of November, General Practitioner (GP)s are able to offer people with common mental health problems, such as stress, anxiety and mild-to-moderate depression, a digital tool to support them while they await referral to mental health services such as Improving Access to Psychological Therapies (IAPT) and Child and Adolescent Mental Health Services (CAMHS). Wysa Premium is an Artificial Intelligence (AI)-enabled mental health app which has been approved by the National Health Service (NHS). The app is clinically safe, adhering to strict Digital Technology Assessment Criteria (DTAC), ensuring that it meets NHS and social care clinical safety, data protection, technical security, interoperability and usability and accessibility standards. Access to the app is via a mobile phone or tablet. The app provides an ‘emotionally intelligent’ service which responds to the emotions you express and uses evidence-based cognitive-behavioural techniques such as meditation, breathing, yoga, motivational interviewing and micro-actions to help you build mental resilience skills and feel better. It was developed by a team of 15 people, consisting of psychologists, designers, developers, and over 500,000 users to understand how AI chat can help us learn skills to build emotional resilience. Wysa has already been rolled out across a number of NHS IAPT and CAMHS talking therapy services. Following a diagnosis of pancreatic cancer, median survival is just 9 months, with chances of 5-year survival being less than 10%, and only 1% surviving for 10 years. Almost 90% of people are diagnosed at advanced stages of the disease when it is too late for effective treatment. Because it is relatively rare and treatment outcomes are poor, there is no national screening program for pancreatic cancer. At present, the National Institute for Health and Care Excellence (NICE) only recommends screening those with a strong family history of pancreatic cancer, hereditary pancreatitis or other hereditary syndromes that increase the risk. However, a new study suggests that pancreatic cancer could be detected earlier by looking for simple signs such as dramatic weight loss or new onset diabetes.1 The team analysed data from 8777 patients diagnosed with pancreatic cancer compared with a control group of 34,979. They found that dramatic weight loss could be detected 2 years before the patients received an official diagnosis of pancreatic cancer, with mean BMI being 3 units lower in those with pancreatic cancer. Average rise in HbA1c of 6.5 mmol/mol (0.6%) was detectable even earlier, from 3 years before the diagnosis. They also found that in people with existing diabetes, dramatic weight loss was associated with a higher risk of pancreatic cancer than in those without the condition. However, in people without diabetes, raised HbA1c was associated with a higher risk than for people with diabetes. Weight loss and increasing blood glucose levels are known early indicators of pancreatic cancer, but NICE guidelines on diagnosing pancreatic cancer do not include them as diagnostic features. The researchers suggest that BMI and HbA1c assessments should take place regularly for people both with and without diabetes to help identify possible cases of pancreatic cancer. In those with pre-existing diabetes, dramatic changes in weight, particularly if the individual has not been trying to lose weight, should be a warning sign. This might allow for earlier surgical intervention before the cancer has become too advanced. Another paper suggests that some specific biomarkers might be able to differentiate between newly diagnosed type 2 diabetes and type 3c, 10% of whom may have underlying pancreatic cancer. The presence of adiponectin and interleukin IL-1Ra showed the strongest diagnostic potential and the authors suggest that further studies are carried out.2 Having analysed European data, the charity Asthma + Lung UK have announced that more people die from pneumonia each year in the UK than anywhere else in Europe.3 Pneumonia is an inflammation of the lung tissue affecting one or both lungs that occurs as a result of an infection. Most cases of pneumonia are caused by a bacteria called streptococcus pneumoniae, known as pneumococcus. It can be spread by aerosol, droplets or direct contact with respiratory secretions of someone carrying the organism. Symptoms of bacterial pneumonia include shivering fits, fever, pains in the chest and coughing. The cough may be dry or may produce thick mucus which can be yellow, blood-stained or rust-coloured. Breathing becomes fast and shallow with difficulty or pain on breathing deeply or coughing. The charity's analysis shows that each year in the UK almost 42,000 people die from pneumonia and other lower respiratory tract infections, although pneumonia accounts for over 80% of deaths. This makes the UK the worst in Europe above Russia in second place and Germany in third. The UK also has the third highest 5-year average mortality from pneumonia and other lower respiratory tract infections at 60.9 deaths per 100,000 people. Monaco has 66.9 with Portugal, at 74.9, being the worst in Europe. The charity is encouraging people at higher risk of infection or death from pneumonia to come forward for vaccination. People living with diabetes are one such group and everyone living with diabetes should be vaccinated. Unlike the flu vaccine which needs to be given annually, most people with diabetes only need to have the pneumococcal pneumonia vaccine once, although some people may need a booster every 5 years, particularly if they have a long-term kidney or spleen problem. The vaccine can reduce the likelihood of getting pneumonia by 50%–70%, dramatically reducing the risk of serious illness and death. Animal studies have shown that exposure to light at night may interfere with circadian rhythms and affect glucose homeostasis. Other research has also shown that ongoing exposure to moderate indoor light during sleep increased the prevalence of diabetes in older adults compared to those sleeping in a darker room. A new national, cross-sectional study from China has now shown that higher levels of exposure to outdoor artificial light at night are also significantly linked with markers of diabetes and impaired glucose homeostasis.4 The researchers assessed 98,658 participants from the China Noncommunicable Disease Surveillance Study across 162 sites. The mean age of participants was 42.7 years and just under half were female. Diabetes was defined based on American Diabetes Association (ADA) criteria. Light exposure was determined from satellite images. The results showed that exposure levels to outdoor light at night were positively linked with 2 h and fasting glucose concentrations, HbA1c, and insulin resistance. Exposure to light was divided into quintiles, and for each higher quintile of exposure there was a 7% significant increase in diabetes prevalence. People living in areas with the most exposure to light at night had a 28% higher prevalence of diabetes than those living in places with the lowest exposure. The underlying belief is that our internal body clock regulates a variety of processes such as metabolism and hormone synthesis. It can also affect sleep patterns by interfering with synthesis of the hormone melatonin, which is essential for sound sleep. The authors accept that this study cannot prove a direct causal link. Areas with the most light pollution are more likely to be urban and it is well known that living in an urbanised area increases your risk of obesity through increased access to higher-fat food and less physical activity. Also, the exact degree of an individual's exposure to night-time light cannot be proven. More research is needed to prove a direct link, ideally through studies that involve direct measurement of individual exposure to light at night. Teplizumab, a monoclonal antibody that suppresses the body's immune response, delaying the onset of type 1 diabetes, has been approved by the FDA for use in people aged eight and over, who have two or more diabetes autoantibodies and changes in blood sugar stability. The company believe that there are currently about 30,000 people in the US who would match the prescribing criteria. The company that produces it, Provention Bio, are hoping to fully launch the drug, under the name Tzield, from January 2023, working alongside Sanofi, who are in a co-promotion agreement with Provention Bio. Although initially licensed only for this particular indication, further trials are ongoing in its use in those newly diagnosed with type 1 diabetes. However, the company have chosen to price the drug at $13,850 (£11,500) per vial. This means that the 14-day course will cost almost $194,000 (£161,000). The company argue that the cost is justified as this is the first drug that can delay the onset of type 1 diabetes. However, some specialists think that this may make it harder for people to access the treatment if insurance companies are not prepared to pay. Although the drug can delay the onset of type 1 diabetes in those at highest risk by up to 3 years, the cost far outweighs the cost of treating them for diabetes over that time. In 2017, the ADA calculated that diabetes cost an individual about $16,750 (£13,950) per year for care and treatment.5 Delaying the onset of diabetes for 3 years would therefore save an individual about $50,250 (£41,830), far less that the cost of Tzield. Of course, this does not take into account the psychological impact of being diagnosed with type 1 diabetes, not only for the child or young person, but also for their family. Teplizumab was awarded an Innovation Passport by the MHRA last summer which means that it could be fast-tracked to approval in the UK. For a drug like templizumab to have the greatest impact, it is vital that you can find those people who are most at risk of developing type 1 diabetes. That is why the launch of a new trial screening programme for type 1 diabetes, on World Diabetes Day, is so timely. ELSA (EarLy Surveillance for Autoimmune diabetes) will aim to recruit 20,000 children, aged three to 13 years, to assess their risk of developing type 1 diabetes. This hopefully will lay the foundation for a national screening program when treatments, such as teplizumab, are available in the UK. The program, which is funded by Diabetes UK and JDRF, will provide vital insights into practical and effective ways to screen large numbers of children. It will also examine how best to support those at high risk and their families to ensure the earliest, safest diagnosis possible. Those who are identified as high risk will also be directed to clinical trials testing the potential innovations which could prevent or delay the condition. The researchers will assess the risk of type 1 diabetes through a combination of finger prick and venous blood tests, looking for autoantibodies associated with the development of type 1 diabetes. These can be found years, or sometimes decades, before people begin to experience any symptoms. People with two or more autoantibodies have an 85% chance of developing type 1 diabetes within 15 years. As no treatment is currently available in the UK, children and their families will be offered support and education to help prepare them for the diagnosis of type 1 diabetes. This will include awareness of the signs and symptoms of diabetes, so that children can be diagnosed sooner. Currently, one in four children ends up in diabetic ketoacidosis at the time of diagnosis, as symptoms have been missed. Families of at-risk children and healthcare professionals will also be interviewed to understand the most effective way to deliver a future screening programme. This will help make the case for a national routine screening program for type 1 diabetes in the UK. Up to now, most studies looking at the development of type 1 diabetes have relied on research with family members of people already diagnosed with the condition. However, as most people who develop type 1 have no family connection to the condition, this study wants to test a much wider group of children and is therefore being promoted to anyone who wants to take part. Those who choose to take part will be sent a finger prick blood test to complete at home. Some schools and general practices will also be offering screening. 99% of those tests are expected to be negative but for the 1% who test positive, this will be followed up with a venous blood test. For those who are found to have one antibody, the family will be invited to an education session to explain what this means. For those who test positive for two or more antibodies, a glucose tolerance test will be carried out to see if insulin needs to be started straight away. Education sessions will then be provided to explain the signs and symptoms of type 1 and to talk about other steps the family might choose to take, such as getting involved in further research. A new large-scale British trial led by the University of Exeter, in collaboration with many other units across the UK, suggests that allowing people with type 2 diabetes to choose their own medication, after trying several, could be a good way of deciding the best treatment for an individual.6 The TriMaster study followed 448 people with type 2 diabetes, who had each been given three different commonly prescribed second or third line once-daily glucose-lowering drugs in turn for a 16-week period: pioglitazone (a TZD), sitagliptin (a DPP4 inhibitor) and canagliflozin (an SGLT2 inhibitor). The trial was a randomized double-blind, three-way crossover trial, so the participants were not sure which drug they were taking at which point during the trial. Researchers monitored the effects on weight and blood sugar levels and recorded any reported side effects. Overall, the three drugs showed similar levels of glycaemic control although the participants were not aware of these results until the end of the trial. After trying all three drugs, the participants were allowed to choose to continue with the one they personally felt had worked best for them. 25% preferred pioglitazone, 35% sitagliptin and 38% canagliflozin. Although they were not aware of this at the time of choosing, participants tended to choose the drug which had lowered their glucose most effectively. Less surprisingly, they also chose the drug which had given them fewest side effects. Allowing the individual to continue with their drug of choice saw more people achieve the lowest HbA1c for them (70% vs. 30%) and the fewest side effects (67% vs. 50%). The study suggests that putting people with type 2 diabetes in charge of their own medication could be a more effective approach to finding the best treatment for them—and to find a treatment that they may be more likely to adhere to. Perhaps it is time to really give a personalised and patient-centred approach to diabetes treatment decisions. The NHS Health Check is designed to look for some of the most common health conditions that affect people as they age, including early signs and symptoms of stroke, kidney disease, heart disease, some forms of dementia and type 2 diabetes. It is offered to all 40–74-year olds who have not already been diagnosed with any of these conditions. Up until now, this has been delivered in a face-to-face way but, in light of many services going digital as a result of the pandemic, a digital version of the NHS Health Check is being piloted as a way of reducing pressure on GP services. The pilot is taking place across three GP practices in Cornwall and will see around 2000 people being invited to take an online questionnaire, as well as being given a kit for taking a blood sample at home. Blood pressure will be checked either at a local pharmacy or in the GP waiting room. Anybody whose results suggest an underlying health condition will be followed up within the GP practice setting. Learning from the pilot will help the NHS understand what a new digital NHS Health Check could look like in the future and how it could help reduce pressure on frontline services. Some concerns have already been raised, one being around how this will impact people who are digitally excluded and whether this could widen inequalities in health unless face-to-face access is also available. The pilot also needs to evaluate whether this leads to greater distress in people who may not know how to interpret the findings but have no-one on hand to explain the results. Research, using data from the Global Burden of Disease study, has found that the number of people, aged between 15 and 39, who have been diagnosed with type 2 diabetes has increased almost fourfold between 1990 and 2019.7 They reviewed data from 204 countries and territories and found that the rise in the diagnoses of younger people has been greater in the UK than anywhere else in the world. In the UK, over the 30 years reviewed, prevalence went from 94 cases per 100,000 young people in 1990 to 347 by 2019, the fastest rate of growth of any country studied. Across the world as a whole, the number of young people being diagnosed increased from 117 per 100,000 in 1990 to 183 per 100,000 in 2019 with the main attributable risk factor being a high BMI, although air pollution and smoking also appeared to have an impact. In response to these findings, charities, including Diabetes UK, have called on the Government to do more to tackle obesity and the serious health conditions it is associated with, such as early-onset type 2 diabetes.

Introduction and Objective: The current study aimed to assess the applicability of different clustering methods in an extensive range of Chinese inpatients with diabetes Methods: We performed hard and soft clustering diagnosis and prediction of complications in two follow-up visits among Chinese inpatients with diabetes in Heilongjiang. Furthermore, the hard and soft clustering model was validated using data from Chinese inpatients with diabetes in Beijing. Results: The results between the Heilongjiang cohort and the Beijing Hospital cohort were consistent, with the highest proportion of participants in the mild obesity-related diabetes (MOD) subgroup, followed by severe insulin-resistant diabetes (SIRD). In the K-means model of two follow-up visits with Heilongjiang data, the mild age-related diabetes (MARD) subgroup had the highest prevalence of distal symmetric polyneuropathy (DSPN) and coronary heart disease (CHD), the severe insulin-deficient diabetes (SIDD) subgroup had the highest prevalence of diabetic ketosis (DK) and the highest level of TG and eGFR (all P value <0.001), and the MOD subgroup had the highest prevalence of hypertension (P value<0.001). In the Gaussian mixture model of two visits with Heilongjiang data, the MOD subgroup had the highest prevalence of hypertension (P value<0.001), severe insulin-deficient diabetes (SIDD) subgroup had the highest value of LDL-C (P value <0.05), and the MIX subgroup had the highest the percentage of alcohol drinking (P value <0.05) Conclusion: Both hard and soft clustering methods were applicable to a comprehensive range of Chinese inpatients with diabetes Disclosure J. Zhang: None. W. Wang: None. L. Guo: Research Support; Abbott, AstraZeneca, Bayer Pharmaceuticals, Inc, Eli Lilly and Company, Innovent Biologics, Merck & Co., Inc, MSD Life Science Foundation, Novo Nordisk A/S, Sanofi, Jiangsu Hengrui Pharmaceuticals Co., Ltd, Tonghua Dongbao. Q. Pan: None. Funding Capital's Funds for Health Improvement and Research (2024-1-4053); National High Level Hospital Clinical Research Funding (BJ-2024-144)

Ketone bodies detection: Wearable and mobile sensors for personalized medicine and nutrition

Last year Medtronic and IBM Watson Health launched their artificial intelligence (AI)-powered Sugar.IQ diabetes management app in the USA. This is linked to Medtronic's Guardian Connect system and is only available for users of this system in the USA on iOS mobile phones. The app responds to continuous glucose monitoring (CGM) data and evaluates how blood glucose is affected by food intake, insulin doses from a pen and other daily routines, enabling it to make suggestions to improve blood glucose control. Initial findings suggested that people using the app spent 36 min more per day in time in glucose target range 1. The original app was designed to give users predictive alerts up to 60 min before they experience hyper- or hypoglycaemia; however, in January, Medtronic and IBM Watson announced a new feature, IQcast. This uses AI technology to predict the likelihood of a user experiencing a hypoglycaemic event (‘hypo’) within the next 1 to 4 h. This long-term forecast has been an aim since the beginning, and follows on from a pilot study back in 2016 in which cognitive analytics were applied to data from 600 people's insulin pumps and CGM devices to predict ‘hypos’. The IQcast feature analyses multiple signals to assess whether someone with diabetes has a low, medium or high likelihood of becoming hypoglycaemic over the next 1 to 4 h. The degree of predictive accuracy increases as a ‘hypo’ becomes more imminent. By using AI, the system can reveal patterns that the individual may find hard to see and offer insights into how lifestyle choices can impact diabetes management and time in range. Although there are no plans to launch the product outside the USA at present, this use of AI to give greater insights into day-to-day diabetes management is an exciting breakthrough which could, in the future, help minimize the risk of hypoglycaemia. After much lobbying and lengthy discussions, the Driver and Vehicle Licensing Agency (DVLA) has implemented the recommendations of the Secretary of State for Transport's Honorary Medical Advisory Panel on driving and diabetes to allow the use of CGM or flash monitoring instead of finger-prick blood tests when driving from February 2019 2. Drivers can still choose to use blood tests if they prefer and should still test their blood if their device says that their blood glucose is below 4 mmol/l or if they experience symptoms of hypoglycaemia. As always with these devices, a blood test is recommended if they give a reading that is not consistent with an individual's symptoms. Drivers who use CGM or flash glucose monitoring must still carry a blood glucose testing kit with them when driving to demonstrate that they could undertake a blood test if necessary. The rules on testing before driving and every 2 h during a journey still apply, so those using the Libre will need to ‘swipe’ the sensor before driving and at least every 2 h afterwards to provide evidence that they have tested at these times. This is less easy for those using CGM systems and so it may be prudent to set ‘hypo’ alarms slightly higher while driving, to ensure adequate warning before such an event occurs. It may also be sensible to keep a note of when CGM has been consulted to comply with this regulation. These new rules apply only to drivers of cars or motorcycles. Anyone driving with a Group 2 licence (larger vehicles and public service vehicles) must still test with a finger-prick blood test at least twice daily, including on days when not driving, and no more than 2 h before the start of the first journey and every 2 h after driving has started. They must still have a meter which can store testing data over a 3-month period. At the end of March the European Medicines Agency (EMA) gave approval for the use of dapagliflozin in people with Type 1 diabetes and a BMI > 27 kg/m2 as an adjunct to insulin, if the insulin alone is not providing adequate glycaemic control. The phase III DEPICT trial showed that 5 mg dapagliflozin once daily, as an adjunct to insulin, led to significant and clinically meaningful reductions in HbA1c, weight and total daily insulin dose at both 24 and 52 weeks. The main serious side effect was an increase in the number of diabetic ketoacidosis events. Dapagliflozin should not be used in people who have low insulin requirements, and insulin therapy should be optimized to prevent ketosis and ketoacidosis while minimizing the risk of hypoglycaemia. Use in Type 1 diabetes should only be initiated and supervised by specialist doctors, and individuals should be educated about the risk factors for and signs and symptoms of diabetic ketoacidosis. At the same time, the EMA's drugs advisory panel recommended the use of sotagliflozin, a sodium-glucose co-transporter (SGLT)1/2 inhibitor, as an adjunct to insulin for certain people with Type 1 diabetes with similar restrictions as for dapagliflozin. The EMA is expected to ratify this recommendation in April. Interestingly, the US Food and Drug Administration (FDA) did not approve sotagliflozin because of the increased risk of diabetic ketoacidosis. Whether the FDA will ask for further trial data is not clear. An international consensus guideline for the use of SGLT2 inhibitors in people with Type 1 diabetes has also been published 3. This recommends several steps people with Type 1 diabetes should take to minimize the risk of ketoacidosis while using these drugs. Advice includes following a regular diet, consuming only a small amount of alcohol, and following prescribed treatment protocols. Clinicians are recommended to prescribe the lowest dose possible when blood ketone levels are < 0.6 mmol/l and to adjust and monitor insulin doses based on the individual's level. Research on hybrid closed-loop insulin delivery systems (the artificial pancreas) have shown excellent results, with people spending more time in target range with less hypoglycaemia. However, a new paper examining the real-world experience of young people using such systems (in this case the Medtronic MiniMed 670G, the only such system to have been approved and marketed) found that nearly 40% of users stopped using the system after a few months 4. The study involved 93 children and young people with Type 1 diabetes from one clinic population who had freely chosen to use the system. The commonest reason for stopping was the difficulty of staying in ‘automode’, whereby the system is doing all the work for you. Other reasons included technical difficulties such as the frequency of alarms, early sensor failure, the need to calibrate the system regularly, skin reactions to the adhesives used, and sensor supply issues. Why would people choose not to use automode? First, the system has to learn about your diabetes for several weeks and some people find that frustrating, especially if it leads to both highs and lows. The system is programmed to keep the user at a set level which cannot be changed and so some people switch off automode if they want to run higher or lower. The system also relies on accurate carbohydrate counting to learn how much insulin is needed and on accurate and regular calibration. If this is not done correctly, the system may under- or overdose insulin, and users may choose to switch automode off. The 62% of people who continued to use the system also did not always use automode, ranging from 10% to 90% of the time; however, those who did use automode saw a significant decrease in HbA1c at 6 months, but by 12–24 months, this was almost back to baseline levels. The more time people spent in automode, the better the improvement in HbA1c. The main message seems to be that, although this technology has great potential, it is not simple to use and people have to put a lot of effort in to making the most out of it. The goal of oral insulin came a step closer recently with the announcement of the first human study investigating a capsule which is designed to replace injections of biological drugs. Insulin is broken down and digested in the stomach and various methods to protect the insulin are being investigated. The RaniPill has been tested in more than 100 animal studies and has shown 100% equivalence with injections. It employs a special enteric coating that protects the capsule in the acidic environment of the stomach. Once the pill moves into the intestine and pH levels rise, the enteric coating dissolves and a chemical reaction takes place to inflate a micro-balloon. The pressure in the balloon then pushes a dissolvable microneedle, filled with the drug, into the intestinal wall, where it is rapidly absorbed. The capsule remnant passes through the gut. At present the technology can deliver doses up to 3 mg (equivalent to about 80 units of insulin) and bioavailability appears to be equal to or better than subcutaneous injections. Initial studies in humans demonstrate the capsule is well tolerated with no adverse events and the absorption is not affected by whether the participants have eaten or not 5. Other research has led to the development of pea-sized capsules that settle on the floor of the stomach because of their acorn shape. The capsule then pushes out a small needle, made almost entirely of insulin, which penetrates the mucous membrane in the stomach and is dissolved, releasing insulin into the blood stream. The device then passes through the gut and is expelled. Trials have been carried out in pigs and show the device is as effective as insulin injection, with no apparent damage to the stomach 6. Obviously, from a diabetes point of view, there are still questions to be answered, particularly in terms of variable dosing of short-acting insulin; however, these devices could provide an oral solution to taking set doses of basal insulin in people with diabetes in the future. A few years ago, Rachel Humphrey was travelling with her teenage son when he was prevented from boarding a flight because airport security wanted to X-ray his insulin pump. Most pump and CGM manufacturers advise against putting the devices through X-ray machines as a precaution, in case they are affected. After her experience, Rachel has campaigned to raise awareness of this issue and to ensure other people do not have the same negative experience. This has resulted in a new Medical Device Awareness Card for people with Type 1 diabetes to use when travelling abroad, launched in partnership with the UK Civil Aviation Authority and Airport Operators Association 7. The card provides information for both the Security Officer and the passenger, and highlights the fact that regulations allow passengers with these devices to ask for an alternative security screening process. It also stresses that passengers should not be asked to remove these medical devices from their body. People with Type 1 diabetes are advised to carry confirmation that they need to wear these devices, such as a letter from their diabetes team, and are advised to check with the airports to ensure a problem-free flight. The pump company Cellnovo has ceased production and marketing of its insulin micropump with immediate effect. The company had been marketing in the UK, six other European countries, Israel, New Zealand and Australia and had over 1500 users in 2018. It was awaiting FDA approval to expand its reach into the USA. The decision appears to be mainly financial, with the CEO stating that ‘unfortunately, we were unable to withstand the pressures of the competitive environment and the challenges of rolling out a breakthrough system at a sustainable cost’. No new customers will be started on the system, and the company is now working ‘to ensure as smooth a transition as possible for patients and their healthcare teams’. The appointed administrators have concluded an initial review and have put a restructuring plan in place to focus all activities on identifying new strategic partners and financial investors. This suggests that the company hopes that they may become commercially viable again in the future. An abstract presented at the Diabetes UK Professional Conference suggests that C-peptide testing a few years after diagnosis of Type 1 diabetes should become routine practice as it can help identify people who have been misdiagnosed and actually have Type 2 diabetes or a monogenic form 8. The team from the Western General Hospital, Edinburgh, used the C-peptide test in people who had been diagnosed with Type 1 diabetes for ~3 years. Of the 757 people tested, 103 (13.7%) had a C-peptide level > 200 pmol/l. Eight people were rediagnosed as having monogenic diabetes and 27 have been diagnosed with Type 2 diabetes. Twelve people stopped insulin as a result of testing. Nine people had C-peptide levels between 600 and 900 pmol/l, but also had evidence of autoimmunity confirming the diagnosis of Type 1 diabetes, even though they were still producing some insulin. The cost of C-peptide testing has also fallen and is now between £6 and £10 per test. This is potentially very cost-effective if enough people can stop taking insulin. The team have therefore submitted a successful business case to NHS Scotland to roll out C-peptide testing on a national basis. NHS England have announced pilot projects later in 2019 in London and on the South Coast for the treatment of people with diabetes who live with diabulimia, a condition whereby people with Type 1 diabetes restrict their insulin intake in order to lose weight, leading to erratic blood glucose control and an increased risk of complications 9. The proposed intervention is based on lifestyle coaching to address issues related to unrealistic body image, particularly amid growing concerns about the potential damage social media can have on young people's mental health. The pilots will be provided by a wide range of healthcare staff, including mental health therapists and specialist diabetes nurses, to address physical and mental health needs. There will also be an online learning element for carers and families to help them better understand the condition and offer support. NHS England say that the services will provide: If successful, the plan is to roll out these services across the country, although Claire Murdoch, National Director for Mental Health at NHS England, stressed that the NHS cannot do this alone and that ‘wider society needs to take a long hard look what more we can do together to protect young people's well-being’. Recent Medical Technology Guidance from the National Institute of Health and Care Excellence (NICE) has recommended the use of UrgoStart dressings for people with diabetic foot ulcers, once other modifiable factors such as infection have been treated 10. The interactive dressings include a special matrix, called TLC-NOSF, which contains a particular potassium salt. This matrix acts on two key factors that impair wound healing and helps deliver oxygen and nutrients to the wound. Research studies have shown faster wound healing and greater reduction in wound size compared with standard non-interactive dressings. They also reduce pain and discomfort. The decision by NICE was also based on evidence that using UrgoStart dressings is associated with a cost saving of £342 per person after 1 year. As over 90% of all diabetes-related amputations are preceded by a foot ulcer, the use of this technology could help prevent amputations. A recent review of data from the Health Survey for England has shown an increasing gap between actual weight and people's perceptions of their own weight and suggests that there has been a normalization of overweight and obesity in England 11. A total of 23 459 people with a BMI of ≥25 kg/m2 answered a question on weight perception. Respondents of the survey were asked, ‘Given your age and height, would you say that you are…’ and were provided with the following four options: (1) about the right weight, (2) too heavy, (3) too light, and (4) not sure. The survey also asked whether people were trying to lose weight. Among individuals with overweight or obesity, 38.5% of men and 17.2% of women, perceived their weight as about the right weight. Among people with obesity, the proportion of men perceiving their weight as about right weight doubled from 6.6% in 1997 to 12.0% in 2015. The number of people under-estimating their weight increased between 1997 and 2015, from 48.4% to 57.9% in men and 24.5% to 30.6% in women. Generally, men were more likely than women to underestimate their overweight or obesity status (38.8% vs 16.8%) and consequently were less likely to try to lose weight (48.1% vs 71.1%). Those who were overweight were more likely to underestimate their weight status than those with obesity (40.8% vs 8.4%). Only about half of overweight individuals were trying to lose weight compared with over two-thirds of those with obesity. With two-thirds of the population of England being overweight and obese, it is not surprising that this is becoming the new normal and therefore people's perception of their own weight status may not be as clear as it once was. This study has potentially identified vulnerable subgroups of populations who are most likely to misperceive their weight status and may help in designing obesity prevention strategies. Researchers in Australia think that they may have found a positive link between the rotavirus vaccine and the development of Type 1 diabetes in children 12. Type 1 diabetes in those aged 0–4 years had been on the rise in Australia since the 1980s but has declined since 2007, the year that the rotavirus vaccine was introduced as a routine vaccination in childhood. While this does not conclusively link the rotavirus vaccine with protection against Type 1 diabetes, the discovery builds on earlier research suggesting that natural rotavirus infection may be a risk factor for Type 1 diabetes. The rates of Type 1 diabetes in older children, who had never been vaccinated against rotavirus, went unchanged, suggesting that younger children may have been protected by the vaccination. The team plan to continue the research as to whether the rates of Type 1 diabetes remain lower as these children grow older. This work was funded by Diabetes UK. None declared.

The 18th International Conference on Advanced Technologies and Treatments for Diabetes (ATTD 2025) highlighted groundbreaking advancements – including closed-loop insulin delivery systems, artificial intelligence (AI)-integrated monitoring platforms, continuous glucose and ketone monitoring (CGM/CKM), and regenerative cellular therapies – that are redefining global diabetes care. While these innovations significantly enhance glycemic control, patient autonomy, and clinical decision-making, equitable access remains a critical challenge, particularly in low- and middle-income countries (LMICs). The objective of the study was to synthesize key clinical and technological insights from ATTD 2025, with a focus on their applicability in LMICs. This review evaluates the efficacy and adoption potential of automated insulin delivery (AID) systems, AI-driven decision tools, next-generation CGM/CKM devices, and beta-cell replacement strategies, while also examining regulatory, infrastructural, and ethical considerations – using India as a case study for scalable implementation. Early adoption of AID systems in pediatric populations demonstrates substantial improvements in time-in-range (TIR) and psychosocial outcomes. CKM technologies offer earlier detection of diabetic ketoacidosis, and AI-powered platforms are driving personalized, real-time diabetes management. Despite these benefits, widespread adoption in LMICs is constrained by fragmented regulatory frameworks, affordability issues, limited reimbursement, and health system disparities. India’s emerging regulatory reforms, AI sandbox programs, and expanding digital health infrastructure suggest a replicable model for contextualizing global innovations. Bridging the diabetes technology divide in LMICs requires a systems-level approach that aligns innovation with affordability, digital readiness, and policy support. Multisectoral collaboration – including regulatory adaptation, public–private partnerships, and patient-centered design – is essential to achieving inclusive, sustainable access to next-generation diabetes care.

ChatGPT, a popular large language model developed by OpenAI, has the potential to transform the management of diabetes mellitus. It is a conversational artificial intelligence model trained on extensive datasets, although not specifically health-related. The development and core components of ChatGPT include neural networks and machine learning. Since the current model is not yet developed on diabetes-related datasets, it has limitations such as the risk of inaccuracies and the need for human supervision. Nevertheless, it has the potential to aid in patient engagement, medical education, and clinical decision support. In diabetes management, it can contribute to patient education, personalized dietary guidelines, and providing emotional support. Specifically, it is being tested in clinical scenarios such as assessment of obesity, screening for diabetic retinopathy, and provision of guidelines for the management of diabetic ketoacidosis. Ethical and legal considerations are essential before ChatGPT can be integrated into healthcare. Potential concerns relate to data privacy, accuracy of responses, and maintenance of the patient-doctor relationship. Ultimately, while ChatGPT and large language models hold immense potential to revolutionize diabetes care, one needs to weigh their limitations, ethical implications, and the need for human supervision. The integration promises a future of proactive, personalized, and patient-centric care in diabetes management.

The T1D Exchange Quality Improvement Collaborative (T1DX-QI) continues to be a leader in driving innovation advancements in type 1 diabetes (T1D) care.1 With over 50 diabetes centers in its consortium across the United States, including 32 pediatric and 18 adult centers, T1DX-QI has been able to capture T1D electronic medical record data on over 55 000 people with T1D. Through patient and parent partners, an engaged group of multidisciplinary healthcare providers, and an advisory board specifically focused on racial-ethnic equity, T1DX-QI has been able to glean vital insights about real-world diabetes care and outcomes, especially among those underserved and traditionally excluded from research.2 This year′s quality improvement conference covers some of the most pressing issues in diabetes care, with a focus on using technology to improve outcomes in high-risk underserved populations. Several abstracts, contributed by Noland, Lockee, Kaplin, and Izquierdo, describe use of big data and artificial intelligence algorithms to identify patients at high risk for a variety of complications, including poor glycemic outcomes, hospitalizations for diabetic ketoacidosis, and long-term complications.3-6 One abstract by Vandervelden details how the development of a T1D dashboard enabled better population health management through identification and systematic tracking of high-risk patients.7 A new QI portal dedicated to improving healthcare clinic self-monitoring and facilitating sharing of ideas has been described. This may be a model for other learning health networks to build cross-collaboration.8 These ways of harnessing real-world data have the potential to identify new at-risk populations and drive change in clinical care using QI methodologies.9, 10 Another theme of the abstracts has been a hot topic in much of the literature: continuous glucose monitoring and insulin pump equity in T1D were explored by Adams, Gandhi, Wong, and Mathias.11-14 Inequity in diabetes technology has been demonstrated in numerous studies from the T1D Exchange network based on the US population15 and in comparison to a population from the German/Austrian Diabetes-Patienten-Verlaufsdokumentation registry.16 These and other papers underscore how critical and pervasive the inequity remains. Although acknowledgement of these issues continues to be vitally important, what is exciting about the abstracts this year is that many diabetes centers have started to develop and test solutions to improve technology use among underserved populations.11-14 Interventions ranging from better outreach and tracking of patients who are eligible but may not be using technology16 to race-targeted approaches that change workflows or lower barriers for technology eligibility17, 18 and modification of clinical pathways to prescribe, authorize, and support underserved patients while using technology are all explored by Wong, Virani, and Byer-Mendoza.13, 19, 20 Future work will continue to enable cross-pollination of interventions such that a suite of possible solutions to form a change package may eventually be developed to be employed by clinics outside of the network. The last major focus of abstracts is the important topic of screening for social determinants of health in diabetes care. Multiple clinics have explored how to incorporate social determinants of health screening into current workflows,10, 11, 19, 21, 22, some using manual processes, others using electronic medical record systems and leveraging institution-wide initiatives in primary care or pediatrics to achieve specialty-care level reach.10 Overall, these abstracts show that screening for social determinants of health is feasible; valued by providers, patients, and institutions; and has potential for significant impact on addressing the unmet needs of vulnerable populations who are often at highest risk for short- and long-term complications. This year has been an exciting time of major strides for the T1D Exchange Collaborative, keeping current issues at the forefront of clinical real-world care. T1DX will continue to search for new ways to drive change for patients and their families, using novel QI methodology, stakeholder engagement, and data-driven approaches. Future directions may include diabetes technology data integration in the electronic medical record, new psychosocial care models that address highly prevalent psychological issues in people with T1D, and shared decision aids that promote patient-centered care in diabetes. OE conceptualized the manuscript. SA wrote the manuscript. SM, NR, and RR reviewed/edited and approved final versions of the manuscript. OE, NR are the guarantors of this work. All members of the T1DX-QI Collaborative, patients, and partners. The Helmsley Charitable Trust funds the T1DX-QI Collaborative. SA is a healthcare disparities advisor for Beta Bionics and Medtronic. NR has no disclosures. SM has no disclosures. OE is a member of the Medtronic Diabetes Health Equity Advisory Board; He is the Principal Investigator for research projects funded by Eli Lilly &Co, Medtronic Diabetes, Abbott Laboratories, and Dexcom Inc. All the funds for these industry-funded projects and board roles are paid directly through his organization, T1D Exchange. RR is Associate Editor for Journal of Diabetes.

News and Views: A Roundup of some new developments in diabetes.

Comparative analysis of artificial intelligence chatbots in health care: Managing diabetic ketoacidosis during the COVID-19 era

Sensing biomarkers in exhaled breath offers a potentially portable, cost-effective, and noninvasive strategy for disease diagnosis screening and monitoring, while high sensitivity, wide sensing range, and target specificity are critical challenges. We demonstrate a deep learning-assisted plasmonic sensing platform that can detect and quantify gas-phase biomarkers in breath-related backgrounds of varying complexity. The sensing interface consisted of Au/SiO2 nanopillars covered with a 15 nm metal-organic framework. A small camera was utilized to capture the plasmonic sensing responses as images, which were subjected to deep learning signal processing. The approach has been demonstrated at a classification accuracy of 95 to 98% for the diabetic ketosis marker acetone within a concentration range of 0.5-80 μmol/mol. The reported work provides a thorough exploration of single-sensor capabilities and sets the basis for more advanced utilization of artificial intelligence in sensing applications.