Closed-Loop and Artificial Intelligence–Based Decision Support Systems

Comments

The use of a CGM-based decision support system (DSS) for real-time insulin dosing for people with T1D using MDI therapy was demonstrated to be feasible and safe. Both study groups showed improvements, but no significant differences were observed in glycemic control and patient reported outcomes (PROs) between the group that used the DSS and the control group.

This was a well-designed randomized controlled study, yet it is hard to appreciate the actual additional benefit of the DSS. The use of the DSS was accompanied by significant therapy change in both study groups. This included mainly the introduction of CGM and a switch to insulin analogues with a more stable insulin degludec. These two changes had a significant impact on the TIR in both groups, emphasizing the efficacy of CGM for MDI users. In addition, both groups had frequent contact with the study team (every 2 weeks), which might also have impacted the study outcomes. Furthermore, no insulin data were provided, a factor that limits the ability to evaluate changes in insulin administration between the groups.

Nevertheless, the authors demonstrated greater glycemic benefit (reduction in hypoglycemia) in a subgroup of participants who used the DSS recommendations (active users). These findings are in line with the well-known observation that technology helps if you use it. The more intriguing question is why only a third of the participants in the DSS group used the system and not all of them. One of the answers could be the device usability problems that occurred during the study (such as connectivity issues) as mentioned by the authors. It might be that a seamless-use device may increase the number of active users. The device's capabilities include a wide range of interventions, including a bolus calculator, hypoglycemia prediction, exercise and bedtime advice to assess the risk of hypoglycemia, and retrospective insulin dose titration every 2 weeks. It would have been useful to assess which components contributed to the reduction in hypoglycemia observed in the active subgroup who used the DSS.

The study demonstrates that DSS is a feasible option for people using MDIs. People who do not want or cannot use pump therapy should have technological tools to support insulin dosing decisions. Evaluating the DSS among different populations for a longer time compared to a regular care group might have provided different outcomes; this is left to be seen.

Comments

In the present study, the authors have used an open-source algorithm to prioritize the team review of patients' CGM data. In the manuscript, they describe in detail all the stages of the algorithm development and the way they introduced it to their clinic in a research program. Despite the fact the study had relatively small number of participants with a high percentage of newly diagnosed participants with T1D, it is an important step in the right direction.

The emerging artificial intelligence–based (AI) technologies and the new generation of sensors with the ability to passively transfer the CGM data to the Cloud, open a new horizon of opportunities for people with diabetes and their health-care providers (HCPs). It is obvious that in a busy T1D or T2D clinic there are people with diabetes (PwD) who need more attention at a certain point of time than others. In most places in the world, the HCPs meet their patients only a few times a year and do not regularly follow them in between visits. The CGM technology and the passive transfer of the data to the Cloud, which make those data accessible to HCPs, can theoretically change that reality. However, it is time consuming to review CGM data in between visits, analyze the data, prioritize the patients, and think of what needs to be changed in a patient's care and what kind of advice should be given to that patient. Handling this level of work is not feasible in most diabetes clinics around the world and especially not in a busy primary care clinic. Therefore, the future of diabetes follow-up visits depends on new technologies that will be able not only to prioritize which patients' data should be reviewed but also to interpret the data and to deliver advice on how to titrate the insulin regimens of those patients while those patients are using a pump, syringe, or a pen and to suggest behavioral changes in the aim to improve the metabolic control. Such an approach will be a game changer in the way we practice medicine globally.

Comments

In the present pilot feasibility study, the authors conducted a multicenter trial to investigate the ability of a self-administered version of glucose excursion minimization (GEM) program augmented with CGM to improve metabolic control (HbA1C) in adults who were recently diagnosed with T2DM. The GEM method was developed to empower people with diabetes to better understand the impact of food and exercise on their blood glucose levels. GEM has been administrated as a face-to-face intervention in adults diagnosed with T2DM during the past 10 years and was published in the literature as superior to routine care. The present study was the first attempt to run GEM on newly diagnosed people with T2D in a self-administered format. The authors hoped to achieve a reduction in carbohydrate intake and increase in physical activity and to diminish CGM glucose excursion. Altogether, a small group of patients were enrolled into the study and only 3 months of follow-up data were presented. Mean HbA1c levels were reduced by 1.8% by all participants; there were decreases in diabetes distress, depression symptoms, and BMI. In addition, patients felt more empowered with respect to their diabetes care. All improvements were achieved by the participants themselves, with the use of a CGM and a booklet with instruction and education.

It is a pity that the study did not have a control group, had only a small number of participants, and was conducted over a short period of time. Perhaps a decision support system based on artificial intelligence and designed to interact directly with people newly diagnosed with T2D could be helpful. Regardless, before conclusions can be drawn, a prospective randomized control study needs to be conducted with a sufficient number of newly diagnosed people with T2D and for a longer period of time.

Comments

This study, published in one of the highest-ranking medical journals, is the largest-to-date randomized clinical trial of very young children with type 1 diabetes ages 2 to 7 years old. The researchers were able to demonstrate that the CamAPS FX application running on a smartphone significantly improved children's glycemic control without increasing their time spent in hypoglycemia. Extensive comments following the original paper discussed various aspects of the study, such as health-care providers having substantially more unscheduled contacts with the participants while they were receiving the closed-loop (CL) treatment (40,41). Expanding the use of CL in this age group, known for volatile glycemic control, is certainly of importance to the progress of the clinical practice of diabetes. Despite the evident merits of the study, we should note that the first sentence in the abstract is somewhat misleading: CL performance in very young children with type 1 diabetes was not entirely “unclear” prior to this publication. The MiniMed 670G/770G system has been approved by the FDA for children ages 2 and up since 2020, based on clinical trial NCT02660827. Recently, the Omnipod 5 system was approved by the FDA for ages 2 years and above. While the publication of this trial was delayed until January 2022 (42), the results were first posted in January 2019, and the glycemic improvements were very similar to those presented in this paper; for example, both studies achieved time-in-range (TIR) improvement of 8–9 percentage points. Further, although the study presented here had a recruitment target of 1 to 7 years old, the actual age range of the participants was 2.3 to 7.9 years, with approximately one-third (36%) of the children under the age of 5 years old. Thus, the age range was very similar to the ages of the participants in the aforementioned 670G trial (42), two pilot trials with Control-IQ (43), an Omnipod 5 trial (44), and a real-life observation of children using MiniMed 670G (45). We can therefore conclude that this study follows a sequence of clinical investigations of the efficacy of CL in young children, which used multiple CL systems in pilot trials and for the purpose of regulatory approval. More results on this important topic are upcoming; for example, we can expect to soon see the results of the Pediatric Artificial Pancreas (PEDAP) trial of Control-IQ technology in very young children in type 1 diabetes (NCT04796779), which recruited 109 children ages 2 to 6 years old (submitted for publication).

Comments

This study presents both glycemic-control and patient-reported outcomes (PROs) associated with the use of a hybrid CL system by children and adolescents with type 1 diabetes. As observed in previous studies of this magnitude (1 –6), CL, compared to conventional therapy, improved TIR (in this case by 6.7 percentage points), primarily by reducing exposure to hyperglycemia above 180 mg/dL. The conventional treatment group was much broader than the control groups used by other studies (typically sensor-augmented pump) and included continuous subcutaneous insulin infusion or multiple daily insulin injections with or without CGM. Thus, unlike other studies, the focus of the outcome was on CL as compared to any other therapy for type 1 diabetes, not specifically on the improvement due to the control algorithm embedded in a CL system. This is an advantage and not necessarily a limitation of the study—the MiniMed 670G system used here has been extensively tested in clinical trials and in real life, and its capabilities for improving glycemic markers are well established (1,4,8). Further, it is probably a missed opportunity that the study excluded participants with higher glycated hemoglobin levels baseline HbA1c (above 10.5%), given that these people may benefit most from CL treatment (21). An intriguing element is the focus on PROs and quality of life improvement due to CL, which have been underinvestigated. We should note, however, that the statement “the efficacy of HCL on glycemic and psychosocial outcomes has not yet been established in a long-term randomized clinical trial” is not entirely accurate. A long-term randomized CL trial assessing health-related quality of life and treatment satisfaction in parents and children with type 1 diabetes was published in June 2021 (46), approximately 6 months before this paper. Nevertheless, this manuscript is of high interest due to the heterogeneity of its population and its emphasis on both glycemic and psychosocial outcomes.

Comments

As noted in the introduction to this article, the first generation of commercial CL systems—MiniMed 670G/770G (1), Tandem's Control-IQ technology (2,3) and, more recently, MiniMed 780G (4) and Omnipod 5 (7)—enjoy a good reception with hundreds of thousands of users around the world. For these and other systems now entering the clinical practice (e.g., CamAPS FX [5] and Diabeloop's DBLG1 [6]), the studies shifted from controlled clinical trials to real-life observational investigations or records from the clinical practice. This manuscript is an example of the latter, presenting real-world outcomes for youth (N=191) using the Control-IQ technology for 6 months at a large pediatric clinic, the Barbara Davis Center for Diabetes at the University of Colorado. One of the most interesting outcomes of this observation is that the effect of this CL system on TIR was the same as in its two pivotal trials (2,3) and other published real-life data (9 –12): TIR increased by 11 percentage points with respect to the TIR before system initiation. This effect was sustained over time and was most pronounced for those who had the highest baseline HbA1c. Similarly, the glucose management indicator (GMI) used as a proxy to HbA1c was reduced by 0.4%, and the rate of CGM readings below 70 mg/dL was low, under 2%, matching all previously published data. It is therefore affirmed that in a distinct age group of children and adolescents with type 1 diabetes, guided initiation of CL during routine clinical practice, the results of CL use are sustained over long periods of time and very well matched the results obtained in other populations and in randomized controlled clinical trials.

Comments

While the outcomes related to average glycemia (e.g., improvement in HbA1c by 0.32% and improvement in TIR by 6.7%) are to be expected and are consistent with other trials of CL systems, the frequency of hypoglycemia (over 6% with CL throughout the study) and the occurrence of severe hypoglycemic episodes (a total of seven, four of which were in the CL group) were notably high. Typically, in studies with similar population, size, and duration, these numbers would be substantially lower (CGM time ≤70 mg/dL would be in the order of 2%) and there would be little or no severe hypoglycemia or diabetic ketoacidosis related to CL use (2,3,24,41).

The explanation of the high GGM time ≤70 mg/dL given by the authors is very informative and raises an issue that is insufficiently discussed: a post hoc comparison of time in hypoglycemia recorded with Dexcom G6 with hypoglycemia recorded with Libre Pro in the CamAPS FX group showed 2.8% GGM time ≤70 mg/dL based on Dexcom G6 readings but 11.3% GGM time ≤70 mg/dL based on Libre Pro readings. Thus, a discrepancy between sensors (e.g., one sensor reading systematically low) can dramatically bias the results of a clinical trial. This, of course, does not explain the multiple observed episodes of symptomatic severe hypoglycemia, but it is an issue that should continue to be discussed until standard between-sensor corrections are accepted to equalize the results between different CGM devices.

Another informative outcome of this paper is the emphasis on the importance of user experience (UX). FlorenceM and CamAPS FX use the same control algorithm, but have very different hardware configurations, one of which was notably more user-friendly: CamAPS FX had a more advanced sensor (one that did not require calibration fingersticks), better device connectivity, and a newer smartphone used as a system hub. This had a significant effect on system use (only 40% adherence with FlorenceM vs 93% with CamAPS FX) and thereby on the glycemic control achieved by these two systems (e.g., there was no HbA1c improvement in the FlorenceM group). The authors concluded that usability (i.e., reliability of system components as well as ease of use) plays an essential role in determining long-term adherence and efficacy, particularly in the adolescent age group, and this is the most important conclusion of this manuscript.

Comments

Several years ago, the EASE (Empagliflozin as Adjunctive to inSulin thErapy) program evaluated empagliflozin 10 mg and 25 mg daily doses (as approved in treatment of type 2 diabetes), and additionally a subtherapeutic 2.5 mg daily dose, on 26-week change in HbA1c (primary endpoint) and weight, TIR, insulin dose, blood pressure, and hypoglycemia (47). These trials included 1707 individuals with type 1 diabetes and concluded that “Empagliflozin improved glycemic control and weight in T1D without increasing hypoglycemia. Ketoacidosis rate was comparable between empagliflozin 2.5 mg and placebo but increased with 10 mg and 25 mg.” The latter, together with other observations, was particularly important and, perhaps, taken to an extreme, because SGLT2 inhibitors were abandoned from the treatment of type 1 diabetes.

This paper tries to reverse this unfortunate trend in the context of the emerging CL therapies. The logic behind this study is straightforward: CL algorithms are the best for controlling blood glucose levels overnight but still cannot prevent postprandial glucose excursions to a great extent. SGLT2 inhibitors are the best option for controlling hyperglycemia and have less pronounced effect on lower glucose levels. Thus, the actions of the CL and the drug are largely complementary. The difficulty, as pointed out by the EASE program (47), is the risk of euglycemic ketoacidosis, which this trial attempts to mitigate by using a subtherapeutic dose of empagliflozin (5 mg/day). The result is a rather dramatic improvement in TIR, with the combination of Basal-IQ (a predictive low glucose suspend system) and Control-IQ Technology, a hybrid CL, resulting in TIR≥80%. An advantage of this study, with respect to the few similar trials using adjuvant SGLT inhibitor with CL, is that the systems used, Basal-IQ and Control-IQ, are both commercially available, which means that this device-drug combination can be readily available to patients should the clinical paradigm change and appropriate precautions and recommendations be put in place. There are new-generation control algorithms that mitigate the risk of diabetic ketoacidosis by regulating the minimum amount of insulin injected. From an engineering point of view, simply using an adjuvant daily pill with these new algorithm-based systems could help prevent ketoacidosis and result in a dramatic 10 to 15 percentage point increase in TIR.

Comments

This paper was published shortly after the previously reviewed manuscript by Garcia-Tirado et al. (27) and presents a clinical trial on the same topic: adding SGLT2 inhibitor to a CL system to investigate whether this adjuvant therapy will improve the action of the control algorithm. Nevertheless, there are two substantial differences between these manuscripts. First, the CL system in this study was experimental and was compared to sensor-augmented pump therapy (SAP), whereas in the previous manuscript, two commercial off-the shelf systems were tested: the Basal-IQ predictive low glucose suspend system (PLGS) and Control-IQ Technology. Second, the dose of SGLT2 inhibitor, which was empagliflozin in both studies, was 25 mg/day in this study but 5 mg/day in the previous trial (27,28). Nevertheless, the findings were similar: in this study, empagliflozin improved TIR by 11.4% with an SAP system and by 7.2% with a CL system, whereas in the previous trial, empagliflozin improved TIR by 16.5% with a PLGS system and by 9.9% with a CL system. This brings about an interesting observation: a subtherapeutic SGLT2 inhibitor dose (e.g., 5 mg) yields similar, perhaps even better, improvement in TIR compared to a standard 25 mg dose. While it can be argued that the CL systems of the two studies are very different, experimental vs commercial, this comparison is between the changes in improvement within the same patient groups and conditions. Thus, we can speculate that a subtherapeutic dose of SGLT2 inhibitor brings improvements to the work of a CL system that are similar or even superior to the full dose, with presumably fewer and less intense side effects, such as ketosis. Indeed, this observation has been confirmed to some extent by the large-scale EASE studies, in which the use of 2.5 mg empagliflozin had adverse effects indistinguishable from placebo (42).

Comments

As new technology emerges and is integrated into daily care, there is a need to characterize the variables that contribute the most to its success and failure. Such an understanding will help to plan strategies for better acceptance and implementation and eventually enable us to get the best out of the technology. In this study, the authors concluded that user engagement is one of the main variables for the success of AID use. This conclusion is not surprising and is in line with data obtained from other studies when implementing new technologies such as pumps and sensors. This is also not surprising, as the current AID systems are hybrid and ask for a range of user interactions. It is important to understand that CL systems can fully automate some of user roles but only partially automate others, which is why some boluses were missed. Therefore, before initiating AID, it is highly important to address user expectations as well as health-care provider expectations.

In this subanalysis of a randomized controlled trial, individuals who started AID with the lowest TIR gained the greatest improvement in TIR, although they did not reach the TIR achieved by those who started AID with the highest TIR. This gap in TIR achievement should be further investigated and probably relates to user engagement. In a similar study methodology, Ekhlaspour et al. showed that people with higher initial HbA1c gained the greatest improvement switching to AID (48). Recently, data from real-world use studies also support these findings, among a large group of Medtronic 780G (49) and Contro-IQ initiators (11,12). Individuals with GMI ≥9% improved TIR by 27% (21), whereas the average increase for the general population is 9%–16% for most systems (50). These data change the paradigm shared by some health-care providers, namely that people who do not comply with all diabetes treatment recommendations or have difficulty using technology should not use AID. On the contrary, the data emphasize that people and adolescents who have the most difficulty adhering to current diabetes treatment recommendations will have the greatest improvements in glycemic control by using AID technology. Still, it will be helpful to have further data regarding the efficacy and safety of AID use among the population with poor control and low system engagement. However, recently published data presented at the 82nd scientific sessions of the ADA showed that reported diabetic ketoacidosis (DKA) occurrence was similar between AID, MDI, and pump users in the T1D Exchange Registry. These data imply that AID use is not associated with higher rates of DKA (51).

Comments

Preventing postprandial glucose excursions remains the main challenge for diabetes treatment and for establishing fully automated insulin delivery. People with diabetes, for simplicity, use only carbohydrate counting to dose premeal insulin, although it has long been known that protein and fat also contribute to postprandial hyperglycemia. Protein and, to a lesser extent, fat increase the production of glucose, fat delays gastric emptying, and both macronutrients increase insulin resistance and stimulate glucagon secretion, which further contributes to postprandial persistent increase in blood glucose. Ventrani and colleagues analyzed the glucose and insulin delivery response to the three meals of the day in relation to the reported meal composition of well-controlled adults using MiniMed 670G. The postprandial glucose profile differed between the three meals of the day and was shown to be dependent on the nutrient composition and meal size. Early postprandial glucose rise is observed with simple sugars and delayed persistent glucose rise is observed with fat and protein. Thus, food containing fat and protein also increases insulin requirements and may need to be considered to optimize glucose control. In an open loop, delayed nutrient absorption is dealt with by using an extended or dual bolus. In AID, the ability to adjust basal insulin was assumed to overcome delayed nutrient absorption, especially of fat. The present study showed that this assumption did not work fully for the 670G system, which modulates only basal insulin delivery. Thus, for those using systems in which an extended bolus is no longer available, an additional user-initiated correction bolus may become necessary after meals that are rich in protein and fat if the algorithm is unable to compensate for persistent hyperglycemia through basal modulation. It is likely that a system that also delivers automated insulin corrections such as MiniMed 780G will perform better, although more research is needed. Some systems (CamAPS) allow users to specify “slowly absorbed meal” to manage such meals, and the Control IQ enables a fixed extended insulin bolus of 50/50 for 2 hours. Still, there is a need to establish which strategy is better for use during AID, simple correction bolus or extended bolus.

The study emphasizes that food composition and choices still matter during AID use. Nevertheless, the outcomes of each meal (breakfast, lunch, and dinner) might have been different if they were tested in other populations with different age groups, lifestyles, and eating habits. The management of the composition of macronutrients will be important for consideration in the development of future generation systems. It will need to incorporate a prediction model for individual postprandial glucose responses, as people have widely different responses to the same food (52) and have different diurnal patterns of postprandial insulin sensitivity over meals (53).

Comments

One of the barriers to physical activity is the challenges that it possesses for people with type 1 diabetes (preventon of exercise-associated hypoglycemia, insulin corrections for postexercise hyperglycemia, and the management of postexercise, late-onset hypoglycemia). As AID systems use is increasing among people with type 1 diabetes, less is known about the ability of the system to cope with the rapid changes in insulin requirements posed by different types of physical activity. In this well-structured supervised study, Paldus and colleagues examined the influence of HIE and RE (anaerobic-based activities) as well as MIE (aerobic-based activity) on glucose levels and insulin delivery among well-controlled adults while using the 670G Medtronic system. Following preexercise adjustments, glycemic control was kept within the recommended glucose target ranges for 14 hours after exercise for all exercise types. We should note, however, that the study was done in a controlled hospital environment, including 4 hours of fasting before exercise; hence, there was no insulin bolus prior to the time of exercise, reducing the probability of upcoming hypoglycemia. Therefore, the study might not completely reflect common behavior in real life. Yet, the results demonstrate the ability of the system to prevent hypoglycemia posed by MIE (aerobic-based activity) and cope with hyperglycemia posed by HIE and RE (anaerobic-based activities). If the study had included a control arm, it would have been possible to assess the extent of the system's contribution to postexercise glycemic control.

The same recommendation for preexercise adjustment was implemented for all three physical activities. However, HIE and RE might need no adjustment before exercise, as these activities may cause a rise in glucose levels (54). The preexercise adjustment might also be the reason for the post-dinner hyperglycemia observed in the study after these two activities.

The study showed that physical activity could be made safe through a simple strategy of reducing insulin delivery 2 h beforehand, having no bolus insulin on board, and having a higher glucose level at the start of exercise, along with using the 670G. The outcomes of the study might not be generalized to other populations, such as children and adolescents (glycogen reserve and fat utility are lower compared to adults), those with different degrees of glycemic control, or those using other AID systems.

Comments

The use of Medtronic 670G increased TIR by roughly 10% with no change in hypoglycemia among a small group of people with T1D in Argentina. AID education, initiation, and follow-up were done virtually because of COVID-19 lockdown restrictions. The successful virtual transition to AID use supports previous studies that used this methodology to start AID with either pump or MDI users (55,56).

As with previous AID studies, the improvement in TIR was demonstrated in the first week of AID use and was sustained almost constantly through the 6-month follow-up; this observation is worth further investigation to understand the reason why no further improvement was seen after the first week, even though the initial improvement was sustained through the remainder of the study.

The excellent outcomes in this study are likely related to the high percentage (90%) of AID use throughout the 6 months of the study, implying a high level of user engagement. This reported rate of use is much higher than the reported rate of use in other clinical and real-life studies of the 670G. The high rate of auto-mode use in this study might also be related to the COVID-19 lockdown, which increases user engagement. Thus, as with any other technology, minimizing device discontinuation and maximizing efficient technology use leads to better glycemic control for a broad range of populations.

Nevertheless, no data are given for the user's effort, such as the percentage of user-initiated boluses. In addition, the author's statement about the “nonselected” population is somewhat misleading. The studied population is composed of individuals with reasonable level of glucose control to begin with, most of whom have already used advanced technologies such as a pump with PLGS. It will be interesting to see data from less controlled and less technology-savvy populations in Latin America.

Comments

There is no validated structured protocol that people using MDI should follow when switching to AID therapy. Most of the available data on transition come from AID studies, in which most of the participants already use an insulin pump with or without CGM. The run-in period, usually between several days and 4 weeks, is used to allow the individual to adapt to the new devices and to optimize insulin treatment prior to AID initiation.

Several AID studies included a transition from MDI to AID, with or without previous CGM use. In the FLAIR study, MDI users started with Medtronic pump therapy for 2 weeks, then went on to 2 weeks of CGM use before starting AID (4). In another study, adult MDI and SMBG (self monitoring of blood glucose) users were first trained for carbohydrate counting using a bolus calculator meter, followed by 2 weeks of pump use and finally 4 days to 2 weeks of SAP therapy before starting AID (57,58). In the Control IQ pivotal trial, all participants used the study CGM for 2 weeks, and then pump-naïve participants started 2 weeks of study pump use before starting AID (2).

The presented study describes a short 10-day protocol transition from MDI therapy to Medtronic 780G in children and adolescents, using optimal AID settings from the beginning of use. This protocol is similar to a previously published protocol by Petrovski et al, which consists of 7 days of CGM use with MDI, during which the participants attended daily AID group training sessions, followed by 3 days of manual SAP mode before initiating Medtronic 670G (59). Although this study included a small group of participants and the study duration was short, it is an important study to look at. This is because the study raises the question of whether a fast transition to AID is better and safer than a longer transition. The arguments for a longer transition include the option to optimize therapy and providing enough time to adapt to technology and learn the basic concepts and skills of SAP therapy. On the other hand, a short transition enables to improve glycemic control earlier, and users may comply better with AID principles of use. However, the study does not provide a complete answer, mainly because it does not include a control arm with a longer transition duration and follow-up. While the transition duration in the present study was only 10 days, it included a well-structured, intensive education and training. Follow-up after AID initiation included close contact (five phone and clinical visits during the 12 weeks following AID initiation) and available technical and clinical support at any time.

The authors' conclusion of a safe transition without DKA occurrence should be taken with caution. First, the included population was selected based on diabetes management adherence (such as frequency of blood glucose measurements and no DKA in the 6 months prior to AID use), so the ability to generalize the findings to other populations was limited. Second, the study was of short duration, with continuous support limiting the ability to evaluate skills of self-management using the new devices. It is also important to mention that transition should be tailored individually to the intended user and family.

An impressive improvement in glycemic control was seen at the 3-month follow-up: TIR increased by 37%, and HbA1c decreased by 2.1%. Probably the contributing factor to this improvement is the optimal AHCL settings of 100 mg/dL and 2 hours of active insulin time initiated from the beginning of the AID use. The study showed that it is safe to start with the 780G optimal settings in this suboptimally controlled population. Total insulin dose significantly increased from pre-AID use. This might be related to suboptimal treatment prior to initiating AID and may explain the need for a 40% increase in meal bolus. The study showed the beneficial gly-cemic effects associated with the AHCL system use among suboptimally controlled children and adolescents naïve to pump therapy in Qatar.