Type 1 diabetes is an autoimmune condition in which the pancreas produces little or no insulin, requiring people to manage their glucose through external insulin for life. More than 121,000 Australians live with the condition, and roughly 3000 new cases are diagnosed each year. Over decades, persistently elevated glucose damages small blood vessels, driving complications such as retinopathy, neuropathy, and kidney disease. Recent Australian data estimate average annual direct healthcare costs of AUD 7084 per person with type 1 diabetes, more than two and a half times the AUD 2789 spent on people without diabetes, and the 2022 Australian National Diabetes Audit identified peripheral neuropathy and retinopathy as the most frequently reported complications, with rates climbing as the duration of disease lengthens.

Automated insulin delivery systems represent the newest generation of diabetes technology. These hybrid closed-loop devices link a continuous glucose monitor to an insulin pump via a control algorithm that adjusts insulin dosing every few minutes, mimicking some of the feedback function of a healthy pancreas. A 2024 consensus statement convened and endorsed by Diabetes Australia and numerous other societies declared automated insulin delivery the standard of care and recommended that all Australians with type 1 diabetes should have access to it. Yet fewer than 20 percent of people with the condition in Australia currently use such systems, largely because of cost and the limited funding arrangements available to adults outside top-tier private health insurance.

The new analysis focused on the MiniMed 780G advanced hybrid closed-loop system and compared it with the combination of multiple daily injections and continuous glucose monitoring. Clinical inputs came from the ADAPT randomized controlled trial, a European study in adults with type 1 diabetes and baseline HbA1c of at least 8.0 percent. In that trial, participants using the MiniMed 780G achieved a mean reduction in HbA1c of 1.54 percentage points from a baseline of 9.04 percent, while those on injections plus glucose monitoring achieved only a 0.20 percentage point reduction. The researchers assumed that intermittently scanned and real-time continuous glucose monitors had equivalent HbA1c efficacy, an assumption supported by published meta-analyses, though the modeling restricted treatment effects to HbA1c differences alone.

The simulation used the IQVIA Core Diabetes Model, an established platform for projecting long-term outcomes in diabetes. The model consists of multiple interdependent Markov sub-models covering both acute events and chronic complications, run through patient-level Monte Carlo simulations with tracker variables allowing the sub-models to interact. Physiological parameters such as HbA1c drive individual event risks and progress naturally over time based on long-term epidemiological studies. The simulated cohort reflected Australian registry data, with a mean baseline age of 43.3 years and a mean diabetes duration of 20.3 years. The base case analysis took a public healthcare payer perspective, captured only direct costs, ran over a 50-year horizon, and applied the 5 percent annual discount rate that Australian health economic guidance requires for future costs and outcomes.

The base case results showed that long-term use of the MiniMed 780G was projected to yield 13.32 quality-adjusted life years compared with 11.81 for injections plus monitoring, a difference of 1.51 quality-adjusted life years. Lifetime direct costs were higher with the technology, at AUD 367,341 versus AUD 317,819, driven mainly by device acquisition and consumable expenses. However, complication costs ran in the opposite direction, falling for users of the automated system because fewer people progressed to diabetes-related damage. The resulting incremental cost-effectiveness ratio was AUD 32,840 per quality-adjusted life year gained, well below the AUD 50,000 per quality-adjusted life year threshold that more than 60 percent of Australian cost-utility analyses published between 1992 and 2022 have assumed, even though Australia has no official willingness-to-pay threshold.

The projected clinical benefits extended across nearly every complication category. Simulated users of the automated system spent 1.15 years longer free of any diabetes-related complication, a 128 percent increase, with the largest gains seen in microvascular outcomes: 9.67 additional years free of macular edema, 8.20 years free of proliferative retinopathy, 7.58 years free of microalbuminuria, 6.25 years free of neuropathy, and 5.78 years free of severe vision loss. These delays translated into meaningful savings, including AUD 18,882 less in ulcer, amputation, and neuropathy costs, AUD 15,323 less in eye complication costs, AUD 8150 less in cardiovascular costs, and AUD 5428 less in renal costs over a lifetime.

Sensitivity analyses underscored how strongly the result depends on the long view. Over shorter horizons of 10 to 15 years, the incremental cost-effectiveness ratio exceeded AUD 50,000 per quality-adjusted life year, likely because many diabetes complications take decades to develop, so the savings and quality-of-life gains from preventing them only accumulate late. The analysis was also sensitive to the inclusion of hypoglycemic event rates and to the quality-of-life benefit associated with reduced fear of hypoglycemia, which the ADAPT trial had quantified. When that fear-related utility benefit of 0.0544 annually for the automated system was removed entirely, the ratio rose to AUD 55,976 per quality-adjusted life year, pushing the result above the conventional threshold. Conversely, adding severe hypoglycemic events and diabetic ketoacidosis lowered the ratio to AUD 19,234 per quality-adjusted life year, since preventing these acute crises avoids expensive emergency care.

A secondary analysis from the societal perspective, which incorporated indirect costs such as lost productivity from absenteeism using the human capital approach, made the case for the technology even stronger. Without hypoglycemia and ketoacidosis included, the societal-perspective ratio was AUD 23,378 per quality-adjusted life year; with those events included, it fell to just AUD 9446. Previous research has shown that even non-severe hypoglycemic episodes impair workplace productivity, and Australian users of advanced glucose technologies report in community surveys that the devices ease limitations on eating, sleeping, exercising, and socializing while improving emotional well-being. Severe hypoglycemic events causing loss of consciousness can also trigger temporary suspension of a person’s driving license in Australia, a consequence that compounds the personal burden beyond anything a cost model can capture.

The authors acknowledge limitations, including the reliance on a European trial, the assumption of equivalent HbA1c efficacy between glucose monitor types, the exclusion of complications not captured in the model such as oral, hearing, and autonomic effects, and the restriction to people with baseline HbA1c above 8.0 percent. Even so, real-world Australian studies, including a multicenter Queensland evaluation showing that starting automated insulin delivery improved time in range and HbA1c, particularly for those with higher baseline levels, support the generalizability of the findings. With the Australian Government’s Insulin Pump Program currently means-tested for people under 21, and Diabetes Australia and a 2024 parliamentary committee report both calling for stepwise expansion of access, the analysis concludes that broader funding of automated insulin delivery for adults with above-target HbA1c would improve long-term outcomes and likely represent an efficient use of healthcare resources.

Subject of Research: Long-term cost-effectiveness of advanced hybrid closed-loop automated insulin delivery versus multiple daily injections with continuous glucose monitoring in Australian adults with type 1 diabetes

Article Title: Long-Term Cost-Effectiveness of an Advanced Hybrid Closed-Loop Automated Insulin Delivery System Versus Multiple Daily Insulin Injections Plus Continuous Glucose Monitoring in Adults with Type 1 Diabetes in Australia

Article References: Glastras, S., Read, M., Ozdemir Saltik, A. Z., Huynh, M., Challis, G., Hill, M., Field, M., Pollock, R., & de Portu, S. (2026). Long-Term Cost-Effectiveness of an Advanced Hybrid Closed-Loop Automated Insulin Delivery System Versus Multiple Daily Insulin Injections Plus Continuous Glucose Monitoring in Adults with Type 1 Diabetes in Australia. Advances in Therapy. https://doi.org/10.1007/s12325-026-03752-8

Image Credits: AI Generated

DOI: 10.1007/s12325-026-03752-8

Keywords: type 1 diabetes, automated insulin delivery, hybrid closed-loop, MiniMed 780G, cost-effectiveness, continuous glucose monitoring, health economics, Australia, HbA1c, quality-adjusted life years, diabetes complications, IQVIA Core Diabetes Model

The protocol, published in Health Science Reports and registered with the international PROSPERO registry, describes a mixed-methods systematic review and meta-analysis that will pull together randomised trials, real-world cohort studies, qualitative interviews and implementation research on hybrid closed-loop systems such as CamAPS FX, Medtronic MiniMed platforms, Tandem Control-IQ and Omnipod 5. What makes the effort timely is the sheer speed of adoption. In England and Wales, the 2025 National Pregnancy in Diabetes Audit recorded that nearly 45 percent of women with type 1 diabetes at 28 weeks of pregnancy were using the CamAPS FX closed-loop system, with another 12 percent on a different hybrid platform. Yet that figure comes from a single publicly funded health system, and the researchers caution it cannot be extrapolated to countries where reimbursement, regulatory approval and clinical workforce differ wildly.

The stakes are high because type 1 diabetes in pregnancy remains stubbornly dangerous despite decades of specialist care. Women with the condition face elevated risks of pre-eclampsia, preterm birth, caesarean delivery, excessively large babies, neonatal hypoglycaemia and admission of their newborns to intensive care. The mothers themselves risk severe hypoglycaemia, diabetic ketoacidosis and progression of retinal and kidney damage. Large national audits continue to report worse outcomes than in the general obstetric population, even as technology use climbs. The core problem is physiological: insulin sensitivity shifts across gestation, glucose variability increases, and the relentless self-management burden falls on women, families and care teams already stretched thin.

Continuous glucose monitoring has transformed how clinicians measure that burden. The metric at the heart of the new review is time in range, specifically the pregnancy-specific target of 3.5 to 7.8 millimoles per litre, a window far tighter than the standard adult range. Landmark studies such as CONCEPTT and subsequent large cohort analyses established that more time in this range and less glucose variability translate directly into fewer large-for-gestational-age births, fewer hypoglycaemic newborns and fewer intensive care admissions. But even sensor-augmented pumps and monitored injection regimens still depend heavily on users and clinicians manually adjusting insulin doses, leaving dangerous highs and lows to persist precisely when insulin requirements are changing fastest.

Automated insulin delivery systems promise to close that gap. These platforms couple a glucose sensor, an insulin pump and an embedded algorithm that continuously modulates insulin delivery in response to sensor readings, adjusting basal rates and, in some systems, delivering automated correction doses. In non-pregnant populations, trials and meta-analyses show such systems increase time in range and reduce time spent dangerously low without clear increases in severe hypoglycaemia or ketoacidosis. Pregnancy, however, is not simply an extension of ordinary type 1 diabetes care. Many commercial algorithms were never designed around pregnancy-specific glucose targets, raising unresolved questions about device suitability, target settings, the best timing for initiation, clinician oversight, use during labour and adaptation to rapidly shifting insulin needs.

The clinical evidence so far is promising but fragmented. Randomised trials including AiDAPT, which tested CamAPS FX, and CRISTAL, which evaluated the MiniMed 780G, have reported improved glycaemic outcomes in selected settings, but they differ in device, comparator, algorithm targets, timing of initiation and support models. Observational cohorts and case series add real-world texture but vary enormously in size, baseline technology use and outcome definitions. The protocol’s authors argue that commercial automated insulin delivery in pregnancy must not be treated as a single homogeneous intervention; the meaning of any pooled result depends on which system was used, when it was started, how it was supported and which outcomes were measured.

That heterogeneity shapes the review’s unusually rigorous methodology. The team will not pool randomised and non-randomised studies together, and pairwise meta-analyses will be run separately for each comparator: closed-loop versus sensor-augmented pump therapy, versus monitored multiple daily injections, and versus clearly defined usual care. The primary outcome, pregnancy-specific time in range, will be harmonised by prioritising second-trimester estimates, with third- and first-trimester values used in descending order of preference. Network meta-analysis is deliberately not planned, because differences in devices, populations, timing and care models could violate the statistical assumptions such analyses require. Rare events such as diabetic ketoacidosis, severe hypoglycaemia and perinatal mortality will be handled with sparse-data methods or synthesised narratively when estimates are unstable.

What truly distinguishes the protocol is its breadth beyond glucose numbers. The review will capture maternal outcomes including hypertensive disorders, retinopathy progression and mode of delivery; neonatal outcomes from birthweight centiles to intensive care admission; and early postpartum outcomes grouped into the first 48 hours, the first two weeks after discharge and out to 12 weeks, a period when breastfeeding, sleep disruption and collapsing insulin requirements create new hazards. Psychosocial measures, including diabetes distress, fear of hypoglycaemia, sleep quality, anxiety and depression, will be extracted with full attention to instrument validation. Crucially, qualitative studies of women’s and clinicians’ lived experiences will be synthesised thematically and then integrated with the quantitative findings through structured side-by-side comparison, so that numbers can be explained, contextualised or challenged by human experience.

Implementation realities also take centre stage. The review will document device and consumable costs, reimbursement arrangements, out-of-pocket expenses, clinician workload, training demands, uptake and retention rates, and reported barriers to equitable access, always tied to the specific jurisdiction and regulatory context of each study. The authors are explicit that access, equity and workload will be treated as descriptive domains, with causal claims made only where study designs genuinely support them. Certainty in the quantitative evidence will be graded using the GRADE framework, while confidence in qualitative findings will be assessed with GRADE-CERQual, allowing the two evidence streams to be weighed on their own terms.

The anticipated limitations are candidly acknowledged: likely few randomised trials, inconsistent definitions for neonatal hypoglycaemia and device-related adverse events, off-label use of systems with non-pregnancy-specific settings, and qualitative evidence skewed toward specialist early-adopter centres. Yet the strengths are substantial, including prospective registration, alignment with PRISMA-P reporting standards, no language restrictions on searches extending back to 2010, and dual independent screening and appraisal at every stage. When the completed review arrives, it should give clinicians, guideline developers and health services something the field currently lacks: a clear, device-specific, globally honest account of whether automated insulin delivery improves diabetic pregnancies, and what it takes, technically, financially and organisationally, to deliver that benefit safely and equitably.

Subject of Research: A mixed-methods systematic review protocol evaluating commercial automated insulin delivery systems in type 1 diabetes pregnancy

Article Title: Commercial Automated Insulin Delivery Systems in Type 1 Diabetes Pregnancy: Protocol for a Mixed‐Methods Systematic Review and Meta‐Analysis of Glycaemic, Perinatal, Psychosocial and Implementation Outcomes

Article References: Li, J., Sheklabadi, E., Goldstein, R. F., Ng, A. H., Teede, H., & Naderpoor, N. (2026). Commercial Automated Insulin Delivery Systems in Type 1 Diabetes Pregnancy: Protocol for a Mixed‐Methods Systematic Review and Meta‐Analysis of Glycaemic, Perinatal, Psychosocial and Implementation Outcomes. Endocrinology, Diabetes & Metabolism, 9(5), Article e70309. https://doi.org/10.1002/edm2.70309

Image Credits: AI Generated

DOI: 10.1002/edm2.70309

Keywords: type 1 diabetes, pregnancy, automated insulin delivery, hybrid closed-loop, continuous glucose monitoring, time in range, perinatal outcomes, CamAPS FX, systematic review, meta-analysis, health technology, implementation science

To understand why this result is attracting attention, it helps to look at the regulatory asymmetry between the two sides of the Atlantic. In 2018, the United States Food and Drug Administration created the “integrated” CGM, or iCGM, designation, a performance-based regulatory pathway that imposes stringent accuracy and interoperability standards on manufacturers. Devices cleared under this pathway can reliably “integrate” with other digitally connected health technologies, including insulin pumps and smart pens, because regulators have verified their performance across the full measuring range. In Europe, by contrast, CGM systems have traditionally entered the market through CE marking, a process that assesses general safety and performance but has been criticized in the clinical literature as less stringent, less device-specific, and less transparent than the FDA approach. For patients and clinicians, this has meant that two sensors carrying the same CE mark could, in principle, perform very differently.

The eCGM proposal was formulated to address precisely this concern. Largely mirroring the FDA’s iCGM requirements, the eCGM criteria specify accuracy expectations across three glycemic strata, define limits for paired glucose comparisons against laboratory reference measurements, set confidence interval thresholds for proportional agreement, and impose demanding requirements on how the supporting clinical study must be designed. Among these design requirements are a minimum of 100 participants, a majority with type 1 diabetes, sensors drawn from at least three manufacturing lots, the use of laboratory-grade reference analyzers, and crucially a distribution of comparator glucose values in which at least 8 percent fall below 70 mg/dL and at least 5 percent exceed 300 mg/dL. That last requirement is far from trivial: recruiting enough data points at dangerously low and very high glucose levels, while keeping participants safe, is one of the hardest parts of CGM performance testing.

The new analysis, led by Nina Jendrike and Guido Freckmann of the Institut für Diabetes-Technologie Ulm GmbH together with Korean collaborators, assembled a pooled dataset from three clinical investigations to meet these stringent conditions. The first pivotal study, conducted in the Republic of Korea between February and August 2022 with 84 participants, supported national regulatory approval in 2023. A second pivotal study, run in Germany between May and October 2022 with 50 participants, contributed to the CE marking granted in early 2024. Because the two pivotal trials together did not quite reach the glycemic distribution demanded by the eCGM proposal, the team added a post-market clinical follow-up study between October and December 2025, enrolling 30 additional participants specifically to generate supervised hypoglycemic and hyperglycemic excursions, with venous blood sampled as frequently as every five minutes during those episodes.

The resulting pooled dataset comprised 164 participants, of whom 86 percent had type 1 diabetes, half were female, and two-thirds managed their diabetes with multiple daily injections. Across the three studies, 168 primary sensors were worn, and 163 of them yielded 17,655 paired CGM-comparator measurements suitable for the accuracy analysis. Of these pairs, 10.9 percent had reference plasma glucose values below 70 mg/dL and 5.7 percent exceeded 300 mg/dL, comfortably satisfying the eCGM distribution requirements. All CGM readings were retrospectively generated by the manufacturer using the currently marketed optional-calibration algorithm applied to the recorded raw sensor signals, while the statistical analysis itself was carried out independently by the investigators.

The headline numbers are striking. In the hypoglycemic range, where CGM accuracy matters most for safety, approximately 91.0 percent of sensor readings below 70 mg/dL fell within ±15 mg/dL of the paired laboratory value, placing the device at the upper end of the 85.7 to 93.2 percent range reported for FDA-cleared iCGM systems. In the euglycemic range of 70 to 180 mg/dL, 78.6 percent of readings agreed with the comparator within ±15 percent, rising to 84.4 percent in the hyperglycemic range above 180 mg/dL, where competing iCGM devices have reported 85.5 to 92.6 percent. Overall, 90.3 percent of all readings landed within ±20 percent of the reference value across the measuring range, and the pooled mean absolute relative difference, a widely cited accuracy metric known as MARD, came out at 9.5 percent. Consensus Error Grid analysis reinforced the clinical picture: 94.6 percent of readings fell into zone A, meaning no effect on clinical action, 5.4 percent into zone B with little or no effect on outcome, and essentially none into the zones representing meaningful clinical risk.

Beyond raw accuracy, the analysis probed how performance holds up over the sensor’s 15-day wear period, a question of real practical importance since CGM accuracy often drifts at the beginning and end of sensor life. Accuracy was slightly lower during the first three days of wear but then stabilized and remained consistent through day 15, a pattern the authors describe as in line with modern CGM technologies. Kaplan-Meier survival analysis, which included all 168 primary sensors, estimated a 90.2 percent probability that a sensor would survive to the end of its expected 15-day-and-30-minute lifetime, with a mean survival time of 14.4 days. Data availability averaged 99.7 percent, and of the 267 data gaps identified, over 91 percent lasted less than 15 minutes; only two gaps exceeded one hour, both traced to smartphone-related user issues rather than sensor malfunction. Safety findings were similarly reassuring: eight adverse device effects were recorded, all mild and non-serious, consisting of erythema, skin pressure marks, and minor bleeding at insertion sites, all recognized and expected consequences of CGM use.

The authors are careful to situate these results within their limits. The analysis was conducted under controlled clinical conditions, so real-world performance may vary with physiological and environmental factors, and the study population excluded children and pregnant women, meaning results cannot simply be generalized to those vulnerable groups. The pooled dataset, while heterogeneous in study design and population, was also funded by the device manufacturer, which provided the systems and retrospectively generated the CGM readings, although the investigators retained independent control over the statistical analysis and interpretation. The eCGM framework itself remains a proposal rather than a formally established European regulatory requirement, and the authors note that further consensus among clinicians, regulators, and manufacturers will be needed before it can be embedded in official guidance. In parallel, the IFCC Working Group on CGM has developed its own comprehensive guideline aimed at a formal ISO standard, signaling that the regulatory landscape for glucose sensors is consolidating rapidly.

Even with those caveats, the significance of the result is hard to overstate. This is the first time that explicit compliance with the eCGM accuracy criteria has been demonstrated for any CGM system, and notably it was achieved by a device that has not itself received FDA iCGM clearance, a status that would otherwise have implied such compliance. For the roughly millions of Europeans who rely on CGM to steer insulin therapy, the study offers something previously absent from the CE-marking landscape: a transparent, quantified, internationally benchmarked demonstration that a sensor can read glucose accurately enough to dose insulin on, detect hypoglycemia dependably, and feed data continuously to an automated insulin delivery algorithm. As CGM technology becomes the backbone of closed-loop diabetes care, independent verification of the numbers behind those algorithms is not a technicality. It is the foundation on which patient safety, and the next generation of automated diabetes therapy, will be built.

Subject of Research: Clinical performance of the CareSens Air CGM system against the proposed European eCGM accuracy criteria

Article Title: Performance of the i-SENS CareSens Air CGM System in Compliance with the European CGM (eCGM) Clinical Performance Criteria: A Pooled Analysis

Article References: Jendrike, N., Kim, K.-S., Lee, S.-H., Yoo, W. S., Park, C.-Y., Öter, S., Morent, L., Eichenlaub, M., & Freckmann, G. (2026). Performance of the i-SENS CareSens Air CGM System in Compliance with the European CGM (eCGM) Clinical Performance Criteria: A Pooled Analysis. Diabetes Therapy. https://doi.org/10.1007/s13300-026-01917-w

Image Credits: AI Generated

DOI: 10.1007/s13300-026-01917-w

Keywords: continuous glucose monitoring, eCGM, CareSens Air, diabetes technology, sensor accuracy, MARD, insulin dosing, automated insulin delivery, hypoglycemia, CE marking, clinical performance, i-SENS

The review, conducted by researchers working at the intersection of biomedical engineering and endocrinology, synthesizes evidence from physiology, clinical studies, and computational modeling to answer a deceptively simple question: what hormonal processes underlie the sex-specific impact on glucose regulation, and can modern diabetes care continue to overlook physiological differences between the sexes without compromising precision and equity? The authors argue that the answer to the second question is increasingly no, particularly as automated insulin delivery systems become the standard of care and their algorithms remain calibrated to a sex-neutral, hormonally static patient.

The physiological backdrop is well established. Across an approximately 28-day cycle, estradiol and progesterone oscillate in patterns that modulate thermoregulation, energy expenditure, substrate utilization, and insulin sensitivity. Studies in healthy women show that these endocrine fluctuations influence glucose kinetics, gastric emptying, incretin secretion, and even exercise performance in a phase-dependent manner. Estradiol tends to enhance insulin-mediated glucose uptake and lipid oxidation, whereas progesterone induces relative insulin resistance and increases hepatic glucose output. The cyclic interplay between these hormones generates measurable variation in glycemia and metabolic efficiency throughout the cycle.

In women with type 1 diabetes, even modest hormonal fluctuations can produce clinically meaningful changes in insulin sensitivity and glycemic control. Yet the clinical literature is inconsistent. Some studies report increased insulin resistance or higher glucose levels during the luteal phase, when progesterone dominates; others observe minimal or no systematic phase-related effects; and interindividual variability frequently exceeds the average effect size. The review attributes these discrepancies to methodological heterogeneity, differing definitions of cycle phases, and the confounding influence of physical activity, stress, diet, and sleep, factors that are difficult to control in real-world settings but that interact directly with hormonal and metabolic pathways.

The complexity deepens considerably in the context of polycystic ovary-like metabolic dysfunction, which the review reports reaches a pooled prevalence of nearly 25 percent among women with type 1 diabetes. Here the pathophysiology is driven primarily by chronic exposure to supraphysiological peripheral insulin levels rather than classical insulin resistance. Exogenous hyperinsulinemia stimulates ovarian theca cell steroidogenesis and suppresses hepatic production of sex hormone–binding globulin, raising circulating androgens. Elevated androgens then impair insulin sensitivity through tissue-specific mechanisms: in skeletal muscle they disrupt post-receptor signaling, including IRS-1/PI3K/Akt activation and GLUT4 translocation; in adipose tissue they promote visceral fat accumulation, enhanced lipolysis, and increased free fatty acid flux. The result is a vicious cycle in which exogenous insulin excess drives ovarian hyperandrogenism, which further deteriorates insulin sensitivity and amplifies long-term cardiometabolic risk.

Against this physiological landscape, the review evaluates the state of mathematical modeling. Physiological models—computational representations built on differential equations describing hormone secretion, glucose–insulin kinetics, and metabolic fluxes—are the foundation of in silico simulators used to test insulin dosing algorithms and closed-loop control strategies before clinical deployment. The field’s canonical frameworks, from the Bergman Minimal Model of 1981 and Sorensen’s comprehensive physiological model of 1985 through the Hovorka and Dalla Man models and the widely adopted UVA/Padova simulator, have achieved remarkable methodological maturity and clinical validation. But the review finds that virtually all of these frameworks rely on sex-neutral assumptions and do not incorporate the cyclical effects of estradiol and progesterone on insulin sensitivity, glucose uptake, or hepatic glucose production.

The authors systematically classified 25 unique studies into categories spanning sex-specific physiological models of energy metabolism, machine learning approaches to menstrual phase detection from wearable data, deep learning frameworks for glucose forecasting in type 1 diabetes, and mechanistic models of the hypothalamic–pituitary–ovarian axis. Notable contributions include Fischer and Röblitz’s mechanistic model of the ovarian cycle describing estradiol, luteinizing hormone, follicle-stimulating hormone, and progesterone dynamics, validated in women undergoing in vitro fertilization, and machine learning frameworks that identify menstrual phases from heart rate, temperature, and sleep metrics with high accuracy. Yet only a handful of studies attempt to couple ovarian hormone dynamics directly to glucose regulation, and fewer still in a diabetes-specific context.

Three recent frameworks emerge as the most significant attempts to bridge the divide, and the review provides a detailed comparative analysis of their abstraction levels. Manrique-Córdoba and colleagues modify a single parameter governing peripheral insulin action within the oral glucose minimal model, representing the cycle implicitly through cycle day; the approach preserves interpretability but cannot distinguish the mechanistic contributions of individual hormones. Díaz and colleagues take a control-oriented approach, discretizing the cycle into follicular and luteal phases and translating insulin sensitivity variability, derived from euglycemic clamp data, into phase-dependent adjustments of basal rate, carbohydrate ratio, and correction factor—an approach directly relevant to automated insulin delivery but agnostic to underlying endocrine mechanisms. Ramírez offers the most mechanistically integrated strategy, coupling a phenomenological ovulatory cycle model with a minimal glucose–insulin–beta-cell system in which estradiol and progesterone are dynamic state variables modulating insulin sensitivity, insulin secretion, and beta-cell mass through saturating nonlinear functions, though the model remains exploratory and unvalidated in diabetic populations.

The review’s central conclusion is that no existing framework simultaneously provides hormonal explicitness, clinical validation in type 1 diabetes populations, and direct applicability to closed-loop insulin delivery. Current automated insulin delivery algorithms do not incorporate explicit hormonal modeling, and phase-adaptive control has not been clinically validated as a strategy to improve glycemic outcomes across the menstrual cycle. Clinical studies of automated insulin delivery across cycle phases have yielded mixed results, with some reporting no statistically significant differences in overall glycemic outcomes and others noting contrasts in time in range, underscoring the need for larger, better-controlled investigations.

The authors emphasize that hormonal fluctuations alone cannot fully explain the variability observed in glycemic responses across menstrual phases. Lifestyle factors, behavioral routines, psychosocial context, ethnicity, geographical environment, and emotional state all contribute and warrant further investigation. Nevertheless, the path forward is clear: clinically validated, control-oriented frameworks capable of identifying, tracking, and integrating ovarian hormonal dynamics into insulin delivery algorithms are needed to advance truly personalized diabetes care for women. By mapping the field’s achievements and its gaps, this review provides both a foundation and a roadmap for the next generation of sex-specific, hormone-aware diabetes technology.

Subject of Research: Mathematical modeling of ovarian hormone effects on glucose–insulin regulation in type 1 diabetes

Article Title: Modeling ovarian hormone effects on glucose–insulin control in type 1 diabetes

Article References: Albaladejo‐Carrasco, N., Furió‐Novejarque, C., Nattero‐Chávez, L., Bondia, J., & Díez, J.-L. (2026). Modeling ovarian hormone effects on glucose–insulin control in type 1 diabetes. Bioengineering & Translational Medicine, Article e70168. https://doi.org/10.1002/btm2.70168

Image Credits: AI Generated

DOI: 10.1002/btm2.70168

Keywords: type 1 diabetes, menstrual cycle, ovarian hormones, estradiol, progesterone, insulin sensitivity, glucose regulation, mathematical modeling, automated insulin delivery, closed-loop systems, hyperandrogenism, sex-specific medicine

The trial builds on a 13-week multicenter randomized controlled trial conducted in the United States and France, in which 194 adults aged 18 to 70 were assigned either to the Omnipod 5 System or to continue their existing non-automated insulin pump therapy with continuous glucose monitoring. In that original phase, the automated system clearly outperformed conventional pumps: participants spent 61.2 percent of their time in the target glucose range of 70 to 180 milligrams per deciliter, compared with 43.8 percent for controls, a difference of more than four hours per day spent in the healthy range. The extension study followed 75 of the 76 French participants who agreed to continue, including 23 who had originally served as controls and now transitioned to the automated system.

The technology itself represents a notable engineering achievement. Omnipod 5 is the first tubeless automated insulin delivery system, combining a disposable, wearable insulin pump called a Pod with a built-in control algorithm, a wirelessly connected smartphone application, and the Dexcom G6 continuous glucose monitor. Operating in Automated Mode, the system delivers tiny micro-boluses of insulin every five minutes, continuously nudging glucose toward a user-selected target that can be customized in 10 milligram per deciliter increments between 110 and 150 milligrams per deciliter. This closed-loop design replaces much of the guesswork of manual pump therapy with an algorithm that responds to every fluctuation the sensor detects.

The results after one year were striking. Participants spent a median of 99.7 percent of the study period wearing and using the system, and a median of 95.6 percent of the time in Automated Mode rather than reverting to manual control, a strong signal of satisfaction and trust in the algorithm. Most of the time, 63.2 percent, users chose the most ambitious 110 milligram per deciliter target. Compared with their pre-trial baseline, participants increased their time in range by an average of 17.9 percentage points, equivalent to 4.3 additional hours per day in the healthy glucose zone, while their mean glucose fell by nearly 28 milligrams per deciliter. Time spent above range dropped by 17.7 percentage points, and critically, time spent in hypoglycemia did not increase at all.

Perhaps the most dramatic shift appeared in a standard laboratory measure of long-term glucose control. The average hemoglobin A1c of the cohort fell from 8.33 percent at baseline to 7.18 percent after 12 months, an improvement of 1.14 percentage points that remained statistically robust at every quarterly checkpoint. At the start of the study, not a single participant met the internationally recommended target of an A1c below 7 percent; by the end, 30 participants, or 40.5 percent of the cohort, had crossed that threshold. The proportion meeting the stricter goal of below 8 percent rose from roughly one in four to more than 90 percent of all participants. Notably, participants with the highest starting A1c levels experienced the greatest improvements, suggesting the system holds particular value for those struggling most with conventional therapy.

Safety data were equally encouraging. Over the full year, researchers recorded just nine adverse events, a rate of 12 per 100 person-years, including a single episode of prolonged hyperglycemia. There were no cases of severe hypoglycemia, diabetic ketoacidosis, or other serious glycemic emergencies, rates well below population-level figures reported across Europe. Seventeen device deficiencies were reported among 13 participants, most involving the handheld controller, but none resulted in serious harm. Body mass index rose slightly, by 0.7 kilograms per square meter, an increase the authors judged unlikely to be clinically meaningful.

The durability of these gains matters because time in range is increasingly recognized as a predictor of long-term outcomes. A growing body of evidence links greater time within the target glucose range with reduced risks of the microvascular complications of diabetes, including retinopathy, nephropathy, and neuropathy. The proportion of participants in this study meeting international consensus targets of at least 70 percent time in range together with less than 4 percent time below range rose more than fourteen-fold, from 1.4 percent at baseline to 20 percent at one year. Although the study was not designed to measure complication rates directly, the authors argue that sustained glycemic improvements of this magnitude may translate into clinically meaningful long-term protection.

Patient-reported measures reinforced the clinical picture. Scores on the Diabetes Quality of Life questionnaire and the Hypoglycemia Confidence Scale improved during the randomized phase and were maintained at both six and twelve months of the extension, indicating that users did not merely tolerate the technology over time but continued to feel confident managing hypoglycemia and less burdened by intensive treatment. The near-universal uptake is itself telling: all but one of the 76 eligible participants opted to continue into the extension, and only one dropped out, because of a move abroad.

The study does have limits. All participants were French adults who were already experienced insulin pump users with baseline A1c values between 7 and 11 percent, so the findings may not generalize to people using multiple daily injections or those with very different starting glucose control. Outcomes were exploratory, without pre-specified hypotheses or formal sample-size considerations, and a slight dip in time in range from the end of the randomized phase to the end of the extension may reflect less intensive clinical monitoring, underscoring that ongoing professional support still matters. Even so, the authors conclude that the short-term glycemic, safety, and quality-of-life benefits observed in the controlled trial were faithfully maintained over a full year of routine use, adding to a mounting body of evidence that automated insulin delivery deserves a place as a first-line therapy for people with type 1 diabetes.

Subject of Research: Twelve-month extended use of the tubeless Omnipod 5 automated insulin delivery system in adults with type 1 diabetes

Article Title: Extended Use of the Omnipod 5 Automated Insulin Delivery System in Adults With Type 1 Diabetes: 12‐Month Extension of a Randomized Controlled Trial

Article References: Renard, E., Weinstock, R. S., Penfornis, A., Riveline, J.-P., Thivolet, C., Ly, T. T., & for the OP5‐003 Research Group (2026). Extended Use of the Omnipod 5 Automated Insulin Delivery System in Adults With Type 1 Diabetes: 12‐Month Extension of a Randomized Controlled Trial. Endocrinology, Diabetes & Metabolism, 9(5), Article e70321. https://doi.org/10.1002/edm2.70321

Image Credits: AI Generated

DOI: 10.1002/edm2.70321

Keywords: type 1 diabetes, automated insulin delivery, Omnipod 5, continuous glucose monitoring, time in range, HbA1c, closed-loop insulin pump, hypoglycemia, diabetic ketoacidosis, patient-reported outcomes, randomized controlled trial, diabetes technology