Real-Life Use of Automated Insulin Delivery in Individuals With Type 2 Diabetes

Abstract

Background:

The objective of this work is to document performance of automated insulin delivery (AID) during real-life use in type 2 diabetes (T2D).

Methods:

A retrospective analysis was performed of continuous glucose monitoring and insulin delivery data from 796 individuals with T2D, who transitioned from 1-month predictive low-glucose suspend (PLGS) use to 3-month AID use, in real-life settings. Primary outcome was change of time in range (TIR = 70-180 mg/dL) from PLGS to AID. Secondary outcomes included time above/below range (TAR/TBR) and total daily insulin (TDI).

Results:

Compared with PLGS, AID increased TIR on average from 63.2% to 72.6%, decreased TAR from 36.2% to 26.8%, and increased TDI from 70.2 to 76.3 U (all P < .001), without significant change to TBR. Glycemic improvements were more pronounced in those with worse glycemic control during PLGS use (P < .001).

Conclusions:

Real-life use of AID led to a rapid and sustained improvement of glycemic control in individuals with T2D.

Introduction

Type 2 diabetes (T2D) is a progressive metabolic condition characterized by chronic hyperglycemia, derived from the impairment of the action and secretion of insulin. ¹ Therapy for T2D is evaluated based on patient-centered treatment factors, and it generally includes lifestyle modification, non-insulin pharmacological treatment, and insulin replacement therapy when disease progression overcomes the effect of other anti-hyperglycemic agents. ² At initiation of insulin therapy, long-acting basal insulin is the insulin regimen of choice, followed by the addition of rapid-acting prandial insulin if glycemic goals are not met. ² Clinically available options for basal/bolus insulin therapy in T2D do not yet include automated insulin delivery (AID)—an established clinical reality, so far limited to the management of type 1 diabetes (T1D).^3-7 Although in T1D, AID has shown real-life performance comparable to what had been observed in clinical trials, ⁸ AID use in T2D has been evaluated in smaller inpatient and outpatient studies.^9-18 In a recent meta-analysis of randomized controlled trials of fully automated AID in T2D, ¹⁹ data from seven studies comprising 390 individuals showed an increased proportion of time spent within the target glucose range with AID versus conventional insulin therapy, with reduction of the overall time spent in hyperglycemia and no significant difference in terms of time spent in hypoglycemia. Comparable outcomes were obtained in a single-arm prospective study of 30 individuals with T2D ¹⁷ and in an observational study of seven individuals providing real-world data from the use of a community-derived open-source AID system. ¹⁸ This technology report presents results from the retrospective analysis of data from a large cohort of individuals with T2D who used an AID system off-label, in real-life settings.

Methods

The analysis included real-life data from individuals with T2D based in the United States, who were using a predictive low-glucose suspend (PLGS) system (Basal-IQ Technology, Tandem Diabetes Care, San Diego, California) and then updated their insulin pump software to AID (Control-IQ Technology, Tandem Diabetes Care). Individuals contributing data had baseline information available in the Tandem Diabetes Care Customer Relations Management Database and had uploaded their continuous glucose monitoring (CGM) and insulin delivery data—either through the Tandem Diabetes Care t:connect Uploader or through the mobile app—to the t:connect web application. Consent to the use of their data for research purposes was given by each individual as part of their onboarding to Tandem Diabetes Care, in initiating a t:connect account. The data extracted for analysis were de-identified and therefore not protected by the Health Insurance Portability and Accountability Act (HIPAA) Privacy Rule (https://privacyruleandresearch.nih.gov/pr_08.asp#8a); consequently, no institutional review board approval was sought for this retrospective analysis.

For all individuals, glycemic and insulin delivery metrics were considered in daily aggregated form (ie, one value per day) and were available for up to 30 days of PLGS use before system transition and up to 90 days of AID use after system transition. These metrics included percent time spent in 70 to 180 mg/dL (time in range [TIR]); mean of CGM readings; percent time spent >180 and >250 mg/dL; percent time spent <70 and <54 mg/dL; coefficient of variation of CGM readings; and total daily insulin (TDI). To be included in the analysis, an individual had to have at least 21 days contributing glycemic metrics and TDI during the month of PLGS use and each month of AID use ²⁰ ; for a day to contribute metrics, >50% of CGM readings were needed (ie, >144 readings). The day of system transition was excluded from the computations. Available baseline information was sex and age.

Primary outcome was TIR, which was compared between PLGS and AID using a general linear model with one group and four repeated measures, each reflecting one month of data. Secondary analyses were performed using the same statistical approach, considering the other glycemic metrics and TDI. Monthly metrics used in the statistical analyses were obtained by averaging daily metrics within each month; only days contributing metrics as defined above were considered in computing the monthly averages. To explore the impact of baseline (ie, pre-AID) glycemic control on the effect of transition to AID, data were stratified by glucose management indicator (GMI) ²¹ computed from CGM data during PLGS use (GMI_PLGS), defining the following four groups: GMI_PLGS ≤6.9%, 6.9%< GMI_PLGS ≤7.4%, 7.4%< GMI_PLGS ≤8.0%, and GMI_PLGS >8.0%. The impact of sex (female vs male) and age (<65 vs ≥65 years) on the effect of transition to AID was also evaluated. In all statistical tests, a difference was considered statistically significant if the two-tailed P-value was <.001. Categorical data are reported as number of occurrences (frequency); continuous data are reported as mean ± standard deviation across individuals. Data analyses were performed using MATLAB R2022a; statistical analyses were performed using IBM SPSS Statistics 28.0.1.1.

Results

The initial data set contained matched baseline information and glycemic/insulin delivery metrics from 956 individuals with T2D. After filtering the data for sufficiency according to the conditions outlined above, 796 individual data sets met the criteria and were included in the analysis. There were 378 (47.5%) females and age was 60.7 ± 13.4 years. CGM metrics and TDI during PLGS use and each month of AID use are reported in Table 1. TIR increased from 63.2 ± 20.5% during PLGS use to 73.0 ± 15.7%, 72.5 ± 15.8%, and 72.1 ± 15.8% during first, second, and third month of AID use (P <.001 for each AID month vs PLGS month). There were 318 (39.9%) individuals meeting the recommended TIR goal (TIR >70%) ²² during PLGS use; this number increased to 493 (61.9%) during the third month of AID use. Mean of CGM readings, time >180 mg/dL, and time >250 mg/dL significantly decreased with AID use. TDI significantly increased with AID use, without resulting in increased time <70 or <54 mg/dL. The coefficient of variation of CGM readings was not significantly different between PLGS and AID. A statistically significant interaction was present between baseline GMI and effect of transition to AID on all metrics, except for time <54 mg/dL. As depicted in the left panel of Figure 1, larger increase in TIR with AID use was associated with higher baseline GMI (ie, poorer quality of glycemic control). In the group with the lowest baseline GMI, no significant change was detected in TIR with AID use; in the other three baseline GMI groups, TIR increased significantly, with average TIR increase from PLGS to first, second, and third month of AID being: 7.2%, 6.7%, and 6.2% for the group with 6.9%< GMI_PLGS ≤7.4%; 14.0%, 13.6%, 13.3% for the group with 7.4%< GMI_PLGS ≤8.0%; and 22.9%, 23.0%, 23.4% for the group with GMI_PLGS >8.0% (all P < .001). Similarly, larger increase in TDI was associated with higher baseline GMI (Figure 1, right panel), with average TDI increase for the group with the largest baseline GMI of 14.1, 13.4, and 13.0 U during first, second, and third month of AID use (all P < .001). No significant interaction was observed between sex or age and effect of transition to AID on any considered metric.

Table 1.

Glycemic Metrics (Computed From Continuous Glucose Monitoring Data) and Total Daily Insulin Comparing Automated Insulin Delivery (AID) Use to Predictive Low-Glucose Suspend (PLGS) Use.

	PLGS	AID Month 1	AID Month 2	AID Month 3	Mean difference [CI]
					AID month 1—PLGS	AID month 2—PLGS	AID month 3—PLGS
Time in 70 to 180 mg/dL (%)	63.2 ± 20.5	73.0 ± 15.7*	72.5 ± 15.8*	72.1 ± 15.8*	9.9 [9.1, 10.7]	9.4 [8.5, 10.2]	9.0 [8.0, 9.9]
Mean of CGM readings (mg/dL)	169.5 ± 32.0	156.5 ± 22.1*	157.2 ± 23.0*	157.8 ± 23.1*	−13.0 [−14.3, −11.7]	−12.3 [−13.6, −10.9]	−11.7 [−13.2, −10.2]
Time >180 mg/dL (%)	36.2 ± 20.7	26.3 ± 15.8*	26.8 ± 15.9*	27.2 ± 16.0*	−9.8 [−10.7, −9.0]	−9.3 [−10.2, −8.5]	−8.9 [−9.9, −8.0]
Time >250 mg/dL (%)	10.3 ± 12.6	5.5 ± 7.3*	5.8 ± 7.8*	5.9 ± 7.8*	−4.8 [−5.4, −4.2]	−4.6 [−5.2, −4.0]	−4.4 [−5.1, −3.8]
Time <70 mg/dL (%)	0.6 ± 1.1	0.6 ± 1.0	0.6 ± 1.0	0.6 ± 1.0	0.0 [−0.1, 0.0]	0.0 [−0.1, 0.0]	0.0 [−0.1, 0.0]
Time <54 mg/dL (%)	0.1 ± 0.2	0.1 ± 0.3	0.1 ± 0.3	0.1 ± 0.2	0.0 [0.0, 0.0]	0.0 [0.0, 0.0]	0.0 [0.0, 0.0]
Coefficient of variation of CGM readings (%)	24.8 ± 5.0	24.8 ± 5.2	24.8 ± 5.2	24.9 ± 5.0	0.0 [−0.2, 0.2]	0.0 [−0.2, 0.2]	0.1 [−0.2, 0.3]
Total daily insulin (U)	70.2 ± 38.9	76.2 ± 40.5*	76.1 ± 40.3*	76.7 ± 40.7*	6.0 [4.8, 7.2]	5.9 [4.6, 7.1]	6.5 [5.0, 7.9]

^*Indicates P < .001 compared with PLGS.

Data are reported as mean ± standard deviation and mean difference [95% confidence interval for the difference (CI)].

Figure 1.

Evolution of the percent time spent in 70 to 180 mg/dL (left) and total daily insulin (right), as technology users transition from predictive low-glucose suspend (PLGS) use to automated insulin delivery (AID) use. Day 0 is the day of system transition. Each panel displays four user groups each referring to a different baseline (ie, during PLGS use) glucose management indicator (GMI_PLGS) range, per figure legend, with the following number of users in each group: 219 (blue), 250 (purple), 193 (red), and 134 (yellow). Continuous lines are metric averages across subjects for each day of analysis; shaded areas are 95% confidence intervals.

Discussion

In this technology report, new results from real-life, off-label use of AID in individuals with T2D are presented. The retrospective analysis shows that AID use is safe and effective in T2D as it rapidly improves glycemic control, increasing time spent in normoglycemia and reducing exposure to hyperglycemia, without significant changes in exposure to hypoglycemia. Glycemic outcomes achieved after transition to AID are maintained across the three months considered in this analysis. Improvement of quality of glycemic control with AID use was most prominent in individuals who demonstrated poor glycemic control at baseline, whereas no significant differences in the benefit of transitioning to AID were noted between females and males, and between individuals <65 and ≥65 years of age. When analyzing the total daily dose of insulin delivered by the system, it appears that AID up-titrated insulin rapidly and again most prominently in individuals with poor baseline glycemic control, suggesting the possibility of using AID as a fast way to titrate insulin in individuals who need to initiate insulin therapy. An analysis of bolusing behaviors indicates that manual correction boluses became less frequent with the use of AID and were replaced by automated corrections delivered by the AID system, especially in those with higher baseline GMI (results not shown).

The main limitation of this work is that the analysis is based on retrospective real-life data and not on data collected as part of a randomized controlled trial, with potential bias in the group of individuals providing data (eg, related to technology access and use of the t:connect web application) and possible confounding factors not controlled for in the analysis (eg, more frequent interactions with the health care provider upon technology transition). Availability of baseline characteristics beyond sex and age is also lacking and does not allow to infer about result generalizability across sociodemographic categories. Similarly, conclusions from this work cannot be generalized to individuals not using an insulin pump before initiation of AID therapy, as all individuals providing data in this analysis were previous users of a PLGS system. The time evolution of metrics beyond those quantifying quality of glycemic control and insulin delivery, such as body weight, is also not available in this data set, whereas body weight change represents an important factor to consider in the overall evaluation of AID use benefit in T2D, because of the critical role of body weight in this patient population. Finally, as this is a real-life data set and possible adverse events associated with the technology use are not known, the presented analysis does not allow a complete characterization of technology safety in individuals with T2D, and safety assessment in this work is solely based on the evaluation of exposure to hypoglycemia.

Conclusions

According to this analysis of real-life data from individuals with T2D who transitioned from using a PLGS system to using an AID system, therapy based on AID is associated with a rapid and sustained improvement in the quality of glycemic control, without increased exposure to hypoglycemia, as compared with previous PLGS therapy. AID up-titrated insulin quickly in this patient population, suggesting the possibility of using this technology to find the total daily insulin dose required by individuals with T2D initiating insulin therapy.