Glucose Monitoring Metrics in Individuals With Type 1 Diabetes Using Different Treatment Modalities: A Real-World Observational Study Free

Type 1 diabetes is a chronic condition characterized by the autoimmune destruction of β-cells in the pancreas, affecting an estimated 9 million people worldwide (1). Achieving optimal glycemic control is essential for effectively managing the disease and reducing the risk of long-term complications. Historically, glycemic level has been primarily assessed through HbA_1c, which reflects mean blood glucose over the last 2–3 months. Most studies that establish a significant link between diabetes complications and glucose control rely on HbA_1c as a key indicator (2–4). Despite advances in treatment, individuals with type 1 diabetes often struggle to achieve optimal metabolic control, as evidenced by the fact that, in many populations, fewer than 30% achieve the target HbA_1c of 7.0% (53 mmol/mol) (5,6).

The introduction of continuous glucose monitoring (CGM) 20 years ago has provided an alternative to HbA_1c, allowing for the assessment of glucose metrics such as time in range (TIR: 3.9–10.0 mmol/L), time below range (TBR: <3.9 mmol/L), time above range (TAR: >10.0 mmol/L), and glycemic variability. Expert consensus guidelines have been published on how CGM metrics should be reported in a standardized manner for research purposes (7,8), including relevant clinical targets to strive for in different groups of individuals with diabetes (9). For type 1 diabetes, the recommendations are more than 70% TIR (TIR_>70%), less than 4% TBR (TBR_<4%), and less than 25% TAR (TAR_<25%) (9). CGM is considered a standard of care in type 1 diabetes management and valued by persons living with diabetes (10,11). Different types of CGMs are on the market—not all with the ability to set glucose threshold alarms, although there is evidence of superiority of CGM with the alarm features compared with CGM without alarms (12,13). In general, however, access to CGM is not universal, and availability can vary by country and health care system.

Also, in terms of insulin treatment strategies for type 1 diabetes, there have been significant advancements in recent years. Overall, there are two treatment options: multiple daily injections (MDI) or insulin pump therapy—both of which can be used with or without CGM. The most advanced insulin pump systems—automated insulin delivery (AID) systems—enable continuous automatic adjustment of insulin delivery in response to changing CGM levels. The glycemic efficacy of the different AIDs is described in numerous randomized clinical trials (14–23) and observational/real-world/register-based studies (24–28), suggesting significant improvement in HbA_1c and TIR and a reduction in TBR and TAR. However, in general, the investigation of efficacy of sensor-based treatment is mostly based on studies using a single device, is rarely based on comparisons between different brands, and never includes data on all available treatment modalities in a population.

This study aims to address this gap by evaluating the association between CGM-derived glycemic outcomes and different insulin treatment modalities in individuals with type 1 diabetes attending Steno Diabetes Center Copenhagen (SDCC), the largest diabetes clinic in Northern Europe. Specifically, we hypothesize that individuals using AID achieve better glycemic outcomes compared with those using CGM with other treatment modalities.

This was a cross-sectional, observational, population-based study using data from recent CGM uploads and electronical medical records (EMRs). The study was approved by the Danish Data Protection Agency in the Capital Region of Denmark (R-22031406).

SDCC is a specialized diabetes outpatient clinic in Denmark, where the health care system is publicly financed and care is free for all. At the time of the study, more than 11,000 individuals attended SDCC for regular treatment of their diabetes. Among these, 6,314 were adults and children with type 1 diabetes. It is estimated that about 90% of the children and 70% of adults with type 1 are granted the use of CGM from either the hospital system (SDCC) or their community health care system, but exact numbers are not available. In Denmark, CGM can be granted to all children, but, currently, adults need to fulfill at least one of the following criteria: HbA_1c above 8.5% (70 mmol/mol), impaired hypoglycemia awareness, problems measuring capillary finger prick glucose during work, or other specific individual indications. For both children and adults, different CGM systems are used: intermittently scanned CGM (isCGM) with and without alarm features and real-time CGM with alarms. Insulin pump treatment can be prescribed to all children with a wish for pump treatment and to adults willing to use it who are not achieving their target HbA_1c of 7.0% (53 mmol/mol) despite utilizing other initiatives such as CGM, carbohydrate counting, and analog insulin. The choice of equipment depends on the individual’s preference for the pump system and brand as well as whether the warranty still applies to a previous insulin pump. AIDs have been started directly from MDI in a few cases, especially in the pediatric population, but only at diabetes onset in the very young, because of national guidelines. Thus, almost all have been on another pump with expired warranty before changing to AID. A first-generation AID has been available at SDCC since January 2019, and two second-generation AIDs with autocorrection feature have been available there since 2021. In the clinic, MDI-treated individuals are, in almost all cases, treated with degludec (once daily) or glargine (once or twice daily) and aspart or faster aspart (three to six times daily).

Individuals were eligible for inclusion in the study if they used any kind of CGM and had at least one upload of their CGM device in the period from 1 September 2021 to 31 August 2022 to one of the following three software applications: 1) Stenopool, which is SDCC’s own uploading system containing direct uploads from devices and data uploaded to LibreView (Abbott, Alameda, CA), and, afterward, transferred to Stenopool; 2) Glooko, (Mountain View, CA); or 3) Carelink (Medtronic, Northridge, CA). For each individual, the most recent 4-week CGM data available were extracted from the relevant databases. Exclusion criteria included use of DIY looping (do-it-yourself looping with noncommercial closed-loop systems), less than 20% active CGM time during a 4-week consecutive period (<5.6 days), and inability to determine, from the EMR, whether an individual’s isCGM had alarm features. The 20% data availability criterion was set arbitrarily, but was chosen to ensure that the included population reflected the real-world population. A data availability criterion of 70% would have excluded more than 400 individuals with inferior glycemic control as shown in Supplementary Table 1, which would have minimized the generalizability of the study.

Information on date of the extracted CGM data, date of birth, sex, date of diabetes onset, insulin treatment regimen, CGM type including brand name and whether alarms were available for the isCGM, insulin pump type, and HbA_1c levels obtained as part of usual care were collected from the EMR (EPIC Systems, Verona, WI). For HbA_1c values, the data point closest to the date of the extracted CGM data was used. Only HbA_1c measured within a range of ±3 months from the obtained CGM data were used for analysis of associations with CGM outcomes.

Based on our prespecified subdivision of insulin treatment regimen at the time of CGM upload, the participants were classified into four main groups: 1) CGM and multiple daily insulin injections (MDI), 2) CGM and insulin pump but without sensor integration (sensor unintegrated insulin pump [SUP]), 3) CGM and insulin pump with low-level integration (sensor-augmented pump [SAP]), and 4) CGM and insulin pump with high-level integration (AID). The MDI and SUP groups were further subdivided into whether their CGM had available alarm features (+A/−A). The categorization and number of users of the different devices by brand names appear in Supplementary Table 2.

Clinical characteristics of the four groups are presented. Continuous outcomes are presented as median with interquartile range (IQR), and binary outcomes are presented as numbers with percentages. Overall and for each treatment group, continuous outcomes are presented as median with IQR, and binary outcomes are presented as numbers with percentages. The distributions of TIR, TAR, and TBR were analyzed by quantile regression for all percentiles 1–99. The probabilities of achieving TIR_>70%, TAR_<25%, and TBR_<4% were all analyzed by logistic regression. The associations with treatment modality were reported as odds ratios relative to the MDI without alarm. HbA_1c was analyzed by linear regression. In all analyses, models only including treatment modality were used, as well as models with sex-specific effects of age and diabetes duration.

All statistical analyses and data management steps were done in R version 4.3.0 (29).

A CONSORT (Consolidated Standards of Reporting Trials) flow diagram for inclusion of individuals is provided in Fig. 1. The final study cohort comprised 3,041 individuals with type 1 diabetes. Among all the included CGM users, 1,622 (53.3%) used MDI, 504 (16.6%) used SUP, 354 (11.6%) used SAP, and 561 (18.4%) used AID. Among the MDI users, the CGM was, in 943 cases (58.1%), without alarm function (−A MDI); for the SUP users, 241 (47.8%) were without alarms (−A SUP), whereas, by definition, none in the SAP and AID groups were without alarm function (i.e., they did not use the isCGM Libre 1) (Supplementary Table 2). The study population (49.7% women) had a median (IQR) age of 41.5 (24.5–58.6) years and a diabetes duration of 18.6 (8.5–32.5) years. On average, AID users were slightly younger with a shorter diabetes duration compared with the rest of the cohort. A comparison of the included individuals with those excluded because of less than 20% CGM use (N = 89) is presented in detail in Supplementary Table 1. Most importantly, the excluded participants differed from the included (N = 3,041) by having a significantly higher HbA_1c (8.1 vs. 7.5%) ([65 vs. 58 mmol/mol], P < 0.001).

Table 1 lists CGM-derived glycemic metrics for the four main treatment groups. For almost all parameters, a difference between the groups was detected in favor of more advanced technology. The overall distribution of TIR, TAR, and TBR in the six subtreatment groups is shown in Fig. 2. The recommended TIR_>70% was obtained by 21.1% of −A MDI users and 25.9% of +A MDI users, 21.6% of −A SUP users and 17.9% of +A SUP users, 31.4% of SAP users, and 58.5% of AID users. For the same six subgroups, TAR_<25% was obtained by 21.2% and 24.4% in the MDI groups, 21.6% and 18.3% in the SUP groups, 25.7% of SAP users, and 48.1% of AID users. A TBR_<4% was obtained in 66.7% of −A MDI users but 71.7% of +A MDI users, in 54.8% of −A SUP users and 62.7% +A SUP users, in 81.4% of SAP users, and in 85.0% of AID users. Thus, it appears that alarm features for MDI and SUP groups only had an impact on reducing TBR_<4%. In Fig. 3, the estimated odds ratios for obtaining the recommended targets (TIR_>70%, TAR_<25%, TBR_<4%, or all three) demonstrate that, compared with our reference group −A MDI, SUP users are not, in general, doing better than MDI users for any targets. In comparison with the −A MDI group, the odds ratios for obtaining all targets (TIR_>70%, TAR_<25%, TBR_<4%) were 1.5 (95% CI 1.1–2.1) for the +A MDI group, 0.9 (95% CI 0.5–1.5) for the −A SUP group, 0.8 (95% CI 0.5–1.4) for the +A SUP, 2.4 (95% CI 1.6–3.5) for the SAP group, and 9.4 (95% CI 6.7–13.0) for the AID group.

In the adjusted quartile regression model, age and diabetes duration had an effect on TIR for the AID by a 1.6% increase per 10 years of age and 2.0% increase per 10 years of duration. The effects of age and diabetes duration for the other treatment modalities are depicted in Supplementary Figs. 1 and 2. The age effect was predominantly seen in SUP and AID, while the duration effect was seen in all treatment groups.

Evaluation of the recently developed Glycemic Risk Index, a single-number summary of the quality of glycemia, was markedly lower and better in the AID group compared with the other treatment modalities (30).

HbA_1c data within our defined timeframe were available for 2,394 (79%) of the 3,041 participants. The overall medians of latest measured HbA_1c for the different treatment groups are shown in Table 1. Compared with the −A MDI group, the SAP and AID groups had an average of 0.6 (95% CI 0.4–0.8)% [7 (95% CI 4–9) mmol/mol] and 0.9 (95% CI 0.7–1.0)% [10 (95% CI 8–11) mmol/mol] lower HbA_1c, respectively (adjusted for sex, age, and diabetes duration).

In this population-based, cross-sectional study comparing CGM-derived glycemic outcomes across individuals with type 1 diabetes using different treatment modalities, an almost 10-fold higher odds ratio for achieving all recommended sensor-based treatment targets was demonstrated with AIDs as compared with injection therapy supported by a CGM without alarm features. Notably, usually, insulin pump therapy without CGM integration did not generally yield better outcomes than MDI treatment. Furthermore, CGM alarm features seemed to have a positive impact on achieving glucose targets in the MDI group, but not in the SUP group. In our cohort, AID users were younger and had a shorter diabetes duration compared with the other treatment modality groups, indicating that prescription of AID as the first insulin pump was adopted faster in the younger part of the population. As visualized in Supplementary Fig. 1, lower age was associated with a marginally lower achievement of TIR, and, in Supplementary Fig. 2, shorter diabetes duration seems to be associated with better TIR across all treatment modalities. Both findings are consistent with existing literature. In our analysis, odds ratios are adjusted for sex, age, and diabetes duration and remain highly significant after this adjustment. Nevertheless, it should be acknowledged that a difference in baseline characteristics exits between groups.

To our knowledge, no similar real-world studies, involving both adults and children with type 1 diabetes, have examined CGM metrics across all different treatment modalities. However, a recent multinational, register-based study by Dovc et al. (31) investigated similar glycemic differences in children and youths using MDI or insulin pumps with either real-time CGM or isCGM. Similar to the current study, the authors found that users of insulin pumps with real-time CGM performed better than those who used insulin pumps with isCGM or MDI users with CGM or isCGM. They did not report specifically on AID treatment, because of limited numbers. However, it is still debated whether the difference seen in studies comparing real-time CGM and isCGM are due to the presence of alarm features—always present in real-time CGM but not in older versions of isCGM—or the need for scanning the sensor when using isCGM (32).

The data collection method used in the current study is truly unique, as it involves gathering data from three distinct uploading platforms, requiring some manual effort. Although our clinic is currently transitioning all device uploads to our proprietary Stenopool platform (33), only two-thirds of the devices were integrated with this platform during the current study.

This study boasts several notable strengths. Firstly, our large single-center study population enables us to describe glycemic outcomes in a cross-sectional manner, from a setting where allocation of devices largely follows fixed rules as described in Research Design and Methods and AID allocation is based on pump failure and lost warranty of previous pumps. Secondly, we were able to present CGM data in greater detail than is traditionally done. This included displaying the distribution of participants achieving different times in glucose ranges; presenting odds ratios for achieving recommended targets; and analyzing the effects of age, sex, and diabetes duration. These expanded analyses provide a more comprehensive understanding of the glycemic outcomes associated with different devices in our study population. We have presented differences across devices for new parameters as time in tight range (3.9–7.8 mmol/L) and Glycemic Risk Index, which both, most probably, will be increasingly used in clinical practice. Finally, all outcomes were calculated from individual raw CGM data files, allowing us to conduct in-depth analysis not otherwise possible as, for example, the Glycemic Risk Index in different treatment groups.

Our study has some limitations to consider. Firstly, it is important to acknowledge that the study design is cross-sectional, limiting our ability to establish causality regarding the glycemic effects of the devices. Furthermore, allocation of devices in real-world settings was not randomly assigned, as individuals may have varying needs and preferences for device choices. For example, some individuals prefer a patch pump without an available CGM integration over an AID with tubing. Additionally, we had no measures of person-related outcomes or socioeconomic factors which might influence choice of devices and impact glycemic outcome (34). Nevertheless, individuals treated with SUP, SAP, and AID are fulfilling exactly the same indications and preferences for treatment, and the differences between them are based only on when their pumps should be replaced because of expired warranty. Therefore, also based on previous studies in this field, we are confident that the differences in glycemic control are mainly due to efficacy of treatment modality.

Secondly, we lacked uploaded data from up to an estimated 1,000 individuals in our clinic who we believe are CGM users. This may have been because of various reasons, such as incomplete patient identification information on the LibreView account, failure to download device data during or before consultations, or, simply, long intervals between uploads and consultations during the COVID pandemic. We have no indication of this group being different in device use or metrics than the attending population. Thirdly, we excluded approximately 90 people because they did not meet the requirement of having 20% of CGM data within 4 weeks. Compared with the included individuals, they had a significantly higher HbA_1c, but did not differ on sex, age, diabetes duration, or devices used. In our analysis, we included data for all individuals with >20% active sensor time, to reflect the full range of glycemic control observed across the population with varying sensor usage. Limiting the analysis to only including data for individuals with >70% active sensor time would have excluded ∼500 individuals and reduced the generalizability of the findings of the study. Finally, it is worth noting that the four subgroups used different CGM types with slightly different accuracies, as the isCGM systems are only used among the MDI and SUP groups. To the best of our knowledge, this aspect is not considered to have a major impact on our results.

Our study examined a diverse population of individuals with type 1 diabetes across different age groups, and the results demonstrated that those who used AIDs had significantly better glycemic outcomes compared with all other treatment modalities. Our findings suggest that AID should be considered as a first choice when initiating pump therapy in CGM users and be available for all interested individuals with type 1 diabetes. However, it should be acknowledged that funding and availability of AIDs may limit accessibility.

Acknowledgments. The authors thank Liv Boelskifte Skovhus, Steno Diabetes Center Copenhagen, for performing data analysis of the raw data from the CGM files, and Antonius Manders, Steno Diabetes Center Copenhagen, for collecting clinical data from EMR.

Funding. The authors received no financial support for the study, authorship, or publication of this article. SDCC is a public hospital and research institution under the Capital Region of Denmark, which is partly funded by a grant from the Novo Nordisk Foundation.

Duality of Interest. K.N. holds shares in Novo Nordisk; has been a paid consultant for Novo Nordisk and Medtronic; and has received speaker honorarium and advisory board honorarium to her institution from Medtronic, Novo Nordisk, and Convatec; her institution has received research funding from Zealand Pharma, Novo Nordisk, Medtronic, and Dexcom. C.L. was employed by SDCC during the conduct of the study, but, as of 1 November 2022, is employed by, and owns shares in, Novo Nordisk. H.U.A. holds shares in Novo Nordisk and is on advisory boards for Abbott and Sanofi. J.S. holds shares in Novo Nordisk; has been a paid consultant for Medtronic; and has received speaker honorarium from Medtronic, Novo Nordisk, Rubin Medical and Sanofi Aventis; her institution has received research funding from Novo Nordisk, Convatec, and Medtronic. D.V. has received research grants from Bayer A/S, Sanofi, Novo Nordisk A/S and Boehringer Ingelheim. D.V. holds shares in Novo Nordisk A/S. B.C. holds shares in Novo Nordisk A/S. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. K.N. and C.L. designed the study; K.N., C.L., K.G.T., C.S., and H.U.A. researched data; A.G.R. and B.C. did the statistical analysis; and K.N. wrote the manuscript, which all authors critically reviewed. K.N. and A.G.R. are the guarantors of this work and, as such, have full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

Prior Presentation. The data were presented as a late-breaking abstract at the Scientific Sessions of the American Diabetes Association, San Diego, CA, 23–26 June 2023.