Automated Insulin Delivery Associated with Superior Glycemic Outcomes in Type 1 Diabetes: A Swedish National Registry Analysis

Modern insulin therapy provides many different treatment modalities for type 1 diabetes (T1D) management in terms of glucose monitoring technology, means of insulin administration, and insulin analogs. Multiple daily insulin injection (MDI) remains the most common form of therapy worldwide but is often insufficient for maintaining acceptable glucose control. Insulin pump therapy (continuous subcutaneous insulin infusion [CSII]) has proven to provide modest improvements in glycemic control when compared with MDI.^1–3 Replacing conventional glucose monitoring (self-monitoring of blood glucose [SMBG]) with continuous glucose monitoring (CGM) in combination with either MDI (MDI-CGM) or CSII (with either an auxiliary CGM system or integrated with the pump, so-called sensor-augmented pump [SAP]) improves glycemic control,^4,5 but it is unclear if insulin pumps retain their superiority.⁶ Despite improvements, the observed outcomes still fall short of bringing population-level HbA1c values within the internationally recognized therapeutic target of <53 mmol/mol.^4,7

Automated insulin delivery (AID) systems, augmenting SAP systems with algorithm-driven insulin dosing to improve glycemic control, have emerged as the next technological step to push the treatment envelope. Several generations of AID systems have now been evaluated in randomized controlled trials (RCTs).⁸ The earlier systems incorporate features to down-regulate the basal delivery on existing (low glucose suspend [LGS]) or expected hypoglycemia (predicted LGS [PLGS]), whereas the modern systems can increase basal insulin during rising glucose and decrease or suspend insulin during falling glucose, with some also delivering automatic correction boluses. Commercially available AID systems include Medtronic MiniMed™ 670G and 780G, Tandem t:slim X2 and Tandem Mobi with Control-IQ technology, Omnipod 5, Medtrum TouchCare System Nano Pump, and CamAPS (in myLife YpsoPump, Dana Diabecare RS, and DANA-I pumps). Pivotal multicenter trials of CamAPS,⁹ Medtronic MiniMed 670G and 780G,^10–12 Tandem Control-IQ technology,¹³ DBLG1,¹⁴ and Insulet Omnipod 5 systems¹⁵ consistently demonstrate superior glycemic outcomes compared with SAP or MDI with lower hemoglobin (Hb)A1c, higher time in range (TIR, 3.9–10.0 mmol/l), and reduced hypoglycemia (<3.9 mmol/l).

Beyond clinical trials, real-world data (RWD) are essential to evaluate system performance, safety, and sustainability over time. Several large-scale studies have demonstrated that glycemic benefits of AID persist in clinical practice settings. Examples include real-world evaluations of Tandem Control-IQ technology,^16,17 Medtronic MiniMed 780G,^18,19 Diabeloop DBLG1,²⁰ CamAPS FX,²¹ and Insulet Omnipod 5.²² Collectively, these studies report sustained improvements in HbA1c and increased TIR, without evidence of increased severe hypoglycemia or diabetic ketoacidosis, consistent with the results from RCTs.

These findings have been reinforced by registry data,²³ systematic reviews,^24–26 and meta-analyses,^27,28 also covering additional AID systems, confirming that AID systems outperform conventional therapy across multiple outcome domains, with effect sizes consistent across age groups and trial settings.

Based upon these encouraging results, several consensus guidelines now recommend AID to be considered for all populations with T1D, as it increases the likelihood of reaching recommended glycemic targets.^4,29,30 The British National Institute for Health and Care Excellence recommends use for children and adolescents in gestational diabetes and for adults with HbA1c over 58 mmol/mol.^31,32

Despite robust evidence supporting AID systems, several evidence gaps remain. Most RWD studies rely primarily on CGM-derived metrics such as TIR, with only a few reporting HbA1c, which remains the gold standard for assessing long-term risk and for health-economic modeling and allows for comparison to non-CGM-based therapies. Moreover, comparative analyses between treatment modalities or AID systems are scarce, and long-term follow-up beyond 1 year is uncommon. Addressing these gaps is essential to strengthen the evidence base for clinical guideline development and reimbursement decisions.

National registries provide a unique opportunity to evaluate the impact of different treatment modalities in a nationwide population. In Sweden, structured diabetes care is supported by the Swedish National Diabetes Register (NDR), covering >95% of individuals with T1D.³³ CGM adoption became widespread in the mid-2010s, and AID uptake has accelerated since 2017. This study uses NDR data to evaluate the association between insulin delivery/glucose monitoring combinations and HbA1c in Swedish adults with T1D from 2014 to 2024, focusing on AID systems. We hypothesized that adults using AID would achieve significantly better HbA1c compared with those treated with MDI and CGM, consistent with the findings from clinical trials and real-world studies described above.

This was a retrospective, registry-based cohort study using data from the Swedish NDR between January 1, 2014, and December 31, 2024. The NDR is one of the world’s largest national quality registries for diabetes, with >95% coverage of individuals living with T1D in Sweden. It includes comprehensive data on demographics, treatments, biomarkers, comorbidities, and outcomes. All endocrinology clinics, pediatric diabetes units, and primary care centers treating people with diabetes in Sweden contribute to the registry. The NDR is maintained by the Swedish Society for Diabetology on behalf of the Swedish Association of Local Authorities and Regions.³³

The study period starts about when CGM adoption became common in Sweden and extended through the most recent available year. The analysis was restricted to adults (≥18 years) with a confirmed diagnosis of T1D and at least one recorded HbA1c value per calendar year. In 2024, a total of 52,570 adults across 90 clinics met the inclusion criteria. Clinics involved ranged from small regional hospital units to large-scale university centers.

The primary outcome was glycated HbA1c (mmol/mol). For each eligible individual, all HbA1c measurements recorded within a calendar year were averaged to generate a single annual value per patient. The analysis was performed on clinic-level aggregated data, meaning that for every clinic, calendar year, and treatment combination, one aggregated data cell was provided by averaging over the annual values calculated for each patient using that treatment combination. To prevent any possibility of reidentification in small strata, standard statistical disclosure control rules were applied prior to data release from the registry: cells with <5 individuals were excluded, and those with <15 individuals were recoded as containing 10 participants.

A treatment combination was defined by the insulin delivery technology and glucose monitoring method registered in the NDR (Table 1). The specific brand or model of the pump or CGM sensor was not always specified, and best efforts have been made to find as descriptive categories as possible. Treatment combinations with fewer than 100 users per year were excluded. Categories were mutually exclusive except for the common AID and CSII-CGM categories.

Base characteristics regarding age, gender distribution, duration since diabetes diagnosis, smoking, and physical activity (PA) habits were extracted from the registry (Table 2). PA was measured using patient-reported information collected during clinic visits. Patients were asked to self-estimate their regular PA level. In addition, participants reported the frequency with which they engaged in these activities. Physically inactive corresponds to patients responding that they have less than one instance of 30 min PA per week (session bouts of 10 min may be added together). Reporting was generally high (>90%) for all variables, and missing data were imputed using multivariate imputation by chained equations.³⁴

The dataset consists of over 600,000 patient years (Table 1). More than a third of those correspond to the use of MDI-CGM (n = 219,245). Due to the lack of device registration, especially in the first years of the dataset, more than 130,000 patient years are attributed to the Unknown combination. Over 100,000 patient years relate to pump use, of which about 30,000 correspond to AID systems.

The covariates were in general not correlated, except age and duration (ρ = 0.35) and age and proportion male (ρ = 0.54). The population has positive health trends, with fewer patients being inactive (P < 0.001) and fewer daily smokers (P < 0.03) every year, in line with the trends in the general Swedish population during this time period.³⁷ The patients are also getting older (P < 0.03) and have had diabetes for a longer time (P < 0.001). The body mass index (BMI) increased from 26.0 to 26.9 over the time period (P < 0.001).

The MDI users were substantially older than the CSII-CGM and AID users (50 vs. 41 years, P < 0.001). Within the AID group, the Medtronic 780G users were slightly older than the Medtronic 670G and the Tandem Control-IQ users (43.3 vs. 41.8 and 40.1 years, P < 0.001).

There were small but significant differences in duration of diabetes between almost every treatment combination, with the longest duration for Medtronic 780 (26.8 years) and the shortest for Unknown (23.6 years) users.

In terms of BMI, the Medtronic 780G and 670G users had significantly higher BMI (27.5 kg/m²) than all other treatment combinations, and the AID systems, including Tandem Control-IQ, had significantly higher BMI (27.2 kg/m²) than the CSII-CGM and MDI groups (both about 26.5 kg/m², P < 0.001).

The level of physical inactivity was highest for the MDI and Unknown groups (15%–19%) and significantly lower for the CSII-CGM group (14%) and yet lower for the AID combinations (11%). Within the AID group, there was a significant difference between the Tandem Control-IQ and Medtronic 670G users on one hand and the Medtronic 780G users (11, 10 vs. 13%, P < 0.001).

The gender distribution also differs between the treatment groups, with a male-dominated population for the MDI (58% male), whereas the CSII-CGM users are leaning more female (47% male), with Tandem t:slim X2, Unknown, and OmniPod having relatively most females (45% male). Wedged in-between are the AID systems with close to equal proportions (49% male).

Finally, daily smoking was twice as common in the MDI group (8%) compared with the AID users (4%, P < 0.001). The CSII-CGM group lands in-between (6%). Fewest smokers were found in the Medtronic 670G group (3%), which had a significantly smaller proportion than the other AID systems (P < 0.001).

The treatment effect of the AID systems as a group was consistently associated with a larger reduction of HbA1c compared with the reference MDI-CGM (−4.0, P < 0.001). The largest effect within this group was attained by Tandem Control-IQ technology (−4.8 mmol/mol, P < 0.001), followed by Medtronic 780G (−3.1 mmol/mol, P < 0.001) and Medtronic 670G (−2.9 mmol/mol, P < 0.001). The differences between these systems in turn were also significant (Table 3).

The treatment effect in the CSII-CGM group was larger than the MDI-CGM reference but lower than the AID group. CSII had similar effect as the reference. MDI-BG was the only option with inferior effect compared with MDI-CGM.

In this nationwide, registry-based analysis of more than 600,000 annualized HbA1c measurements from adults with T1D in Sweden between 2014 and 2024, we found a clear hierarchy in terms of HbA1c reduction effect (Δ, mmol/mol) with Δ_AID>Δ_PLGS≥Δ_CSII-CGM>Δ_MDI-CGM≥Δ_CSII-BG≥Δ_MDI-BG^a among the evaluated treatment modalities.

Direct comparisons between Medtronic 670G and 780G in the FLAIR crossover trial demonstrated a mean HbA1c reduction of –5 mmol/mol for the 780G users versus –3 mmol/mol with 670G.⁴⁷ Real-world evidence from a pediatric cohort corroborates these findings with 6.5 mmol/L lower HbA1c 6 months after AID onboarding for the users of the newer system than corresponding follow-up data for the 670G system.⁵⁰ However, patients already using 670G and transferring at onset to 780G saw no improvement. No significant difference in terms of HbA1c could be found in our results.

Our study reports superior HbA1c outcomes for Control-IQ technology users compared with Medtronic 780G users. In contrast, some real-world studies have also reported higher TIR outcomes with Medtronic 780G compared with Control-IQ technology.^51–54 We selected HbA1c as the outcome metric due to the ability to compare outcomes across individuals using different proprietary technologies. Concerns about using TIR to compare the efficacy of AID systems utilizing different sensor systems have been raised,⁵⁵ with head-to-head comparisons between sensor systems indicating systematic differences,^56,57 and international consensus guidelines caution against comparison of CGM metrics across sensor types.⁵⁸ Results with HbA1c as outcome are mixed, with two studies comparing Control-IQ technology and 780G reporting similar HbA1c improvements from baseline (−6 to −8 mmol/mol),^53,59 whereas other studies confirm similar trends as this study with lower HbA1c with Control-IQ technology.^49,54,60 Given that the present analysis is not a head-to-head comparison and prior evidence remains inconsistent, the observed superiority of Control-IQ in this study should be interpreted with appropriate caution.

A possible explanation for the overall superior results for the AID systems compared with other modalities could be a selection bias where patients with better glycemic control are more likely to transition to AID therapy. In 2018, when AID was introduced into clinical practice, this group had a lower HbA1c than the alternatives (Fig. 3). However, it should be recalled that the values represent averages over the year and that several studies have shown that the glycemic improvement after switching to AID is substantial already after 3 months with both Tandem and Medtronic AID systems.^61,62 Further studies are needed to investigate the effect of treatment switching and if the improvement effects are homogeneous over the population.

This nationwide, registry-based analysis provides real-world evidence on glycemic outcomes associated with different insulin delivery and glucose monitoring technologies in Swedish adults with T1D over a 10-year period.

There has been a substantial improvement in average HbA1c at the national level, with an approximately 8 mmol/mol reduction between 2014 and 2024 paired with a dramatic increase in CGM usage from virtually zero to over 90% of the population.

Use of AID technology was consistently associated with lower HbA1c compared with both MDI-CGM and CSII-CGM, with the most treatment impact from Control-IQ technology. The AID group results corroborate previous results from RCTs and RWD studies that AID systems are associated with significantly better HbA1c outcomes than non-AID modalities, supporting the broader adoption of AID in routine care, and underscore the value of registry-based surveillance in monitoring long-term outcomes.

Further analyses are necessary to assess potential differences among subpopulations, evaluate the dynamics of glycemic improvement from onset onward to determine its stability, identify potential confounders and transition effects associated with switching treatment modalities, and delineate causal factors related to settings, algorithms, and training. Such investigations are essential to optimize interventions and enhance glycemic control within the diabetes community.

F.S.: Conceptualization, data curation, formal analysis, methodology, and writing—original draft. All authors: writing—review and editing.

The authors thank the Swedish National Diabetes Registry for data provision and all participating clinics for continuous reporting.