Sage Journals: Discover world-class research

Abstract

Objective:

This study was designed to investigate the switch between the open-source automated insulin delivery (OS-AID) system AndroidAPS (AAPS) and commercially available AID systems Control-IQ (CIQ) and MiniMed 780G (780G) conducted in a new extended follow-up study.

Research Design and Methods:

In this prospective open-label single-arm clinical trial, 41 adults with type 1 diabetes (age 35 ± 11 years, glycated hemoglobin [HbA1c] 6.4 ± 2.8% [46 ± 6.8 mmol/mol]) who have voluntarily used AAPS entered a total of three study phases. In the first phase, participants continued with AAPS for 3 months. In the second 3-month study phase, all participants initiated CIQ (n = 25) or 780G (n = 16). Finally, participants were switched back to the AAPS for the last 3 months phase. Results of the treatment with commercially available AID systems were compared with both AAPS phases.

Results:

Commercially available systems were comparable to AAPS in achieving time in range (TIR) (84.2 ± 7.6 vs. 85 ± 6.9%; P = 0.31) and in HbA1c (6.4 ± 3 vs. 6.3 ± 2.7% [46 ± 8.8 vs. 45.7 ± 6.2 mmol/mol]; P = 0.68). In contrast, time in tight range (TITR) was significantly higher in AAPS (66.38 ± 11.84 vs. 63.4 ± 11.77, P = 0.035). However, the time in hypoglycemia <70 mg/dL [<3.9 mmol/L] was significantly lower with commercially available AID systems (2.2 ± 1.2 vs. 3.8 ± 1.9%; P < 0.001). These results were consistent after switching back to AAPS.

Conclusion:

The extension of the Comparison of Different Hybrid Closed-Loop Systems—AndroidAPS and Control-IQ—in adults with Type 1 Diabetes study is the only prospective study to investigate switching between OS and commercially available AID systems. The switch from AAPS to commercially available systems was not associated with a change in TIR. However, the use of AAPS was associated with a higher TITR, but also with a higher risk of hypoglycemia.

Keywords

AndroidAPS automated delivery systems Control-IQ MiniMed 780G open-source type 1 diabetes

Introduction

Automated insulin delivery (AID) systems, also known as hybrid closed-loop or artificial pancreas systems, are regarded as the optimal treatment choice for people living with type 1 diabetes (PwT1D).¹

Many individuals with type 1 diabetes (T1D) utilize open-source AID (OS-AID) systems, otherwise known as do-it-yourself artificial pancreas systems.^2–8

Several studies have established that both OS-AID and commercially available AID systems demonstrate greater efficacy than sensor-augmented pumps (SAP).^5–7,9–13 However, no prospective studies have compared commercially available AID and OS-AID systems. This lack of evidence challenges health care providers and researchers. They must answer questions of PwT1D regarding whether OS-AID systems deliver equivalent, inferior, or superior results.

We recently reported results from the Comparison of Different Hybrid Closed-Loop Systems—AndroidAPS and Control-IQ—in adults with Type 1 Diabetes (CODIAC) study, which compared the certified Control-IQ (CIQ) technology with Tandem t:slim X2 insulin pump and the OS AndroidAPS (AAPS) with Dana R/RS or Accu-Chek Insight insulin pump in 25 people with T1D.¹⁴

In the current extended follow-up study, we evaluated switching between the OS-AID system AAPS and commercially available AID systems CIQ and MiniMed 780G (780G).

Research Design and Methods

Study design and participants

This nonrandomized, prospective, open-label, single-arm study compared the glycemic outcomes achieved by the commercially available CIQ or 780G technology and the OS AAPS in adults with T1D over 38 weeks. The study consisted of a 12-week AAPS phase, a 2-week run-in phase, a 12-week CIQ or 780G treatment phase, and a final 12-week AAPS phase (Fig. 1).

FIG. 1.

Study flow. V, visit; pV, phone visit.

PwT1D who had previously voluntarily used AAPS were recruited for this study.

The inclusion criteria were as follows: duration of T1D >2 years, age >18 years, and previous use of AAPS for at least 3 months before the start of the study.

Exclusion criteria included concomitant therapies affecting glucose metabolism, lactation, pregnancy, plans to become pregnant during the study, and nonadherence to the current treatment regimen.

The study was approved by an independent ethics review board and was conducted following the Declaration of Helsinki. The trial was registered with ClinicalTrials.gov (NCT 06709729). All participants provided their informed written consent before enrollment.

Procedures

Participants were scheduled for six study visits. At the first visit, participants who had already voluntarily used AAPS were screened for eligibility and informed about the study program. Eligible participants willing to participate in the study signed a written informed consent form. Parameters of glycemic outcomes and relevant medical history were collected. Participants were instructed to photograph and report skin problems at the sensor or cannula insertion sites to the investigators throughout the study.

Participants continued to use the AAPS between visits 1 and 2 (AAPS phase).

At visit 2, investigators obtained blood samples for laboratory analysis (glycated hemoglobin [HbA1c]), and other parameters of glycemic outcomes were collected using a unique uniform resource locator provided by Nightscout (an OS project that provides access to continuous glucose monitoring [CGM] data). Participants were switched to the CIQ or 780G system and trained by the study team on an outpatient basis. The first 25 participants entering the study were switched to CIQ, and the remaining 16 participants initiated 780G.

During the 2-week run-in phase, participants became familiar with the commercially available systems.

All participants were able to change settings between contacts throughout the study. However, staff were instructed to avoid monitoring outside of scheduled visits. This approach was designed to eliminate any effect on outcomes caused by additional planned contact with the study team.

During phone visits (visits 3–4), participants were asked about adverse events and device problems and had the opportunity to consult with the study team.

At visit 5, all data were downloaded from the devices for analysis, and investigators collected blood samples for laboratory analysis (HbA1c). Participants were switched back to the AAPS system.

At visit 6, all data were downloaded from the devices for analysis, and investigators collected blood samples for laboratory analysis (HbA1c).

Hybrid closed-loop systems

AAPS employs the identical heuristic-based algorithm as OpenAPS, designed to adhere to the mathematical principles utilized by individuals with diabetes in their insulin adjustment decisions. AAPS establishes communication with devices powered by the Android operating system, such as smartphones and tablets.

The connection between the CGM and pump was established through a Bluetooth bridge. In this study, 27 participants utilized the Dana R/RS pump, whereas 14 opted for the Accu-Check Insight pump, all paired with Dexcom G6. The AAPS algorithm fine-tunes the basal rate and incorporates features such as unannounced meals (UAM) and super microbolus (SMB). The SMB feature enables the system to autonomously administer a correction bolus, advancing the peak of insulin activity and implementing a temporary zero basal rate to avert overdosing. The UAM feature equips AAPS to manage incorrect carbohydrate estimations or meals high in protein and fat. The UAM can identify rapid glucose spikes triggered by carbohydrates, adrenaline, and other factors and attempt to compensate with SMB. Conversely, if a rapid glucose decline is detected, SMB delivery can be halted sooner. These functionalities enhance glucose regulation, but do not create an entirely closed-loop system. Every participant in this study utilized both SMB and UAM functionalities on AAPS.

Real-time CGM data flow into AAPS every 5 min, activating the embedded algorithm to forecast a glucose trajectory by taking into account the current sensor glucose value, insulin on board, carbohydrates on board, duration of action, insulin sensitivity factor (ISF), insulin/carbohydrate ratio (CIR), and other pertinent factors. It subsequently provides a temporary basal rate or SMB to the pump based on the discrepancy between the projected glucose value and the established glycemic outcomes target, aiming to maintain glucose levels within that target range. Each new AAPS user is required to undergo training and achieve 10 goals before commencing hybrid closed-loop therapy.^9,15–17

The consensus statement for OS-AID systems¹⁸ and the AAPS website¹⁶ provide a detailed overview of AAPS setup options. Likewise, details about the CIQ algorithm can be found on the official Tandem website, whereas information about the 780G algorithm is provided on the official Medtronic website.¹⁹

Outcomes

The primary end point was the difference in percentage of time in range (%TIR, 70–180 mg/dL [3.9–10.0 mmol/L]) between commercially available systems and AAPS. Secondary outcomes were changes in HbA1c (% Diabetes Control and Complications Trial (DCCT) [mmol/mol]), percentage of time spent in hypoglycemia (%TBR, <70 mg/dL [<3.9 mmol/L] and <54 mg/dL [<3.0 mmol/L]), percentage of time spent in hyperglycemia (%TAR, >180 mg/dL [>10.0 mmol/L] and >250 mg/dL [>13.9 mmol/L]), percentage of time in a tight range (%TITR 70–140 mg/dL [3.9–7.8 mmol/L]), glycemic variability expressed as the coefficient of variation (%CV), and mean sensor glucose.

Safety outcomes included the incidence of severe hypoglycemia (requiring second-party assistance to treat), ketoacidosis requiring hospitalization, skin reaction, infection, or hematoma at the sensor insertion site.

Statistical analysis

In addition to the fundamental descriptive attributes, the examination used assessments to ascertain statistical significance for dependent and independent factors. Before utilizing these assessments, the normality of the data was evaluated using the Shapiro–Wilk examination. The null hypothesis of this examination is that the data exhibit a normal distribution. We used the Wilcoxon signed-rank assessment to identify the discrepancies within independent factors, and the paired t assessment was utilized to illustrate potential differences between dependent factors.

In both cases, the null hypothesis was rejected with less than a 5% probability of error, that is, in cases where the value of P was below 0.05.

Results

Baseline characteristics and adherence

Forty-nine adults with T1D who had been previous AAPS users by their own decision were screened for eligibility to participate in this study. Of them, 41 participants were eligible to enter this study. The other eight participants did not meet the inclusion criteria.

Table 1 lists participants’ baseline characteristics. Data at baseline (visit 1) were comparable with those at visit 2 when CIQ or 780G was initiated (Table 2). Data from 30 participants were available for final analysis. Two participants withdrew during the 780G phase (the first participant stated technical issues as the reason, and the second one was not satisfied with the algorithm). Nine participants did not enter the final phase of the study (extension) because they decided to switch from AAPS to commercially available AID systems before the final phase.

Table 1.

Baseline Characteristics

Male, n (%)	30 (73%)
Age (years)	35 ± 11.7
BMI (kg/m²)	25 ± 3.6
HbA1c (mmol/mol)	46 ± 8.1
HbA1c (% DCCT)	6.4 ± 2.9
Duration of diabetes (years)	19 ± 10.3
Time spent with AAPS (years)	1.9 ± 1.3

HbA1c, glycated hemoglobin; AAPS, AndroidAPS; BMI, Body Mass Index.

Table 2.

Comparison of Visit 1 (Baseline) and Visit 2 in the AndroidAPS Phase

AAPS	V1 (baseline)	V2	P value	Δ mean (95% CI)
HbA1c, mmol/mol (%DCCT)	46 ± 8.06 (6.4 ± 2.9)	45.98 ± 8.82 (6.4 ± 3)	0.862264	0.02 (−3.74, 3.78)
TIR (3.9–10.0 mmol/L [70–180 mg/dL]), %	84.61 ± 8.13	84.15 ± 7.7	0.124234	0.47 (−3.01, 3.95)
TBR (<3.9 mmol/L [<70 mg/dL]),%	3.83 ± 2.23	3.82 ± 1.94	0.463814	0.01 (−0.91, 0.93)
TBR (<3.0 mmol/L [<54 mg/dL]),%	0.66 ± 0.59	0.69 ± 0.65	0.618641	−0.02 (−0.29, 0.25)
TAR (>10.0 mmol/L [>180 mg/dL]),%	11.37 ± 8.45	11.96 ± 8.48	0.071418	−0.59 (−4.31, 3.14)
TAR (>13.9 mmol/L [>250 mg/dL]),%	1.84 ± 2.69	1.83 ± 2.62	0.284463	0.007 (−1.16, 1.18)
Mean sensor glucose, mmol/L	7.11 ± 0.81	7.16 ± 0.85	0.122209	−0.043 (−0.41, 0.32)
SD	2.22 ± 0.52	2.29 ± 0.53	0.085760	−0.06 (−0.29, 0.16)
%CV	30.98 ± 4.81	31.62 ± 4.39	0.295273	−0.64 (−2.66, 1.39)
BMI, kg/m²	24.89 ± 3.65	24.97 ± 3.84	0.461085	−0.07 (−1.74, 1.59)
Total daily dose of insulin, units	51.48 ± 15.82	51.06 ± 15.66	0.968998	−0.19 (−7.62, 7.25)

TIR, time in range; TBR, time spent in hypoglycemia; TAR, time spent in hyperglycemia; %CV, coefficient of variation; CI, Confidence Interval; SD, Standard Deviation.

Primary and secondary end points

The results at the end of the AAPS phase (V2) were compared with those after the 3-month CIQ or 780G phase (V5). In the study extension, the results at the end of the CIQ or 780G phase (V5) were compared with those after the final AAPS phase (V6).

Commercially available systems (CIQ and 780G) were comparable with AAPS in achieving TIR (84.98 ± 6.86 vs. 84.15 ± 7.61%; P = 0.31). No significant difference in TIR was observed in a separate analysis for each AID system—CIQ versus AAPS (85.72 ± 7.64 vs. 84.24 ± 8.46%; P = 0.12) and 780G versus AAPS (83.81 ± 5.54 vs. 83.99 ± 6.38%; P = 0.83).

%TBR (<70 mg/dL [<3.9 mmol/L]) was lower when using commercially available systems than AAPS (2.17 ± 1.17 vs. 3.82 ± 1.92%; P < 0.001), in separate analyses for CIQ versus AAPS (2.4 ± 1.22 vs. 3.88 ± 2.04%; P = 0.001) and 780G versus AAPS (1.81 ± 1.01 vs. 3.74 ± 1.75%; P = 0.003).

In addition, following the transition from AAPS to CIQ, %TBR (<70 mg/dL [<3.9 mmol/L]) decreased in 16 participants, increased in 2 cases, and was unchanged in 7 participants. In the transition from AAPS to 780G, %TBR decreased in 12 participants, increased in 2 participants, and remained unchanged in 2 cases.

%TBR2 (<54 mg/dL [<3.0 mmol/L]) was also lower with commercially available AID systems compared with AAPS (0.27 ± 0.44 vs. 0.69 ± 0.65%; P < 0.001) in separate analyses for CIQ versus AAPS (0.4 ± 0.5 vs. 0.76 ± 0.77%; P = 0.02) and 780G versus AAPS (0.06 ± 0.24 vs. 0.59 ± 0.38%; P = 0.002).

% TITR (70–140 mg/dL [3.9–7.8 mmol/L]) was lower using commercially available systems than AAPS (63.4 ± 11.77 vs. 66.38 ± 11.84; P = 0.13). However, there was no difference in separate analyses for CIQ versus AAPS (65.1 ± 12.97 vs. 67.65 ± 12.44; P = 0.28) and 780G versus AAPS (61.13 ± 9.48 vs. 64.69 ± 10.75; P = 0.13) (Tables 3, 4).

Table 3.

Comparison of AndroidAPS (Visit 2) and Commercially Available Automated Insulin Delivery Systems Control-IQ and MiniMed 780G in a Separate Analysis (Visit 5)

V2–V5	AAPS	CIQ	P value	Δ mean (95% CI)
HbA1c, mmol/mol (% DCCT)	46.88 ± 10.32 (6.4 ± 3.1)	45.48 ± 7.06 (6.3 ± 2.8)	0.594548	1.4 (−3.65, 6.45)
TIR (3.9–10.0 mmol/L [70–180 mg/dL]),%	84.24 ± 8.46	85.72 ± 7.64	0.117272	−1.48 (−6.07, 3.11)
TBR (<3.9 mmol/L [<70 mg/dL]),%	3.88 ± 2.04	2.4 ± 1.22	0.001246	1.48 (0.52, 2.44)
TBR (<3.0 mmol/L [<54 mg/dL]),%	0.76 ± 0.77	0.4 ± 0.5	0.021831	0.36 (−0.01, 0.73)
TAR (>10.0 mmol/L [>180 mg/dL]),%	11.68 ± 9.31	11.56 ± 8.03	0.832868	0.12 (−4.83, 5.07)
TAR (>13.9 mmol/L [>250 mg/dL]),%	2.04 ± 3.24	1.56 ± 2.58	0.272103	0.48 (−1.19, 2.15)
TITR (3.9–7.8 mmol/L [70–140 mg/dL]),%	67.65 ± 12.44	65.1 ± 12.97	0.278965	2.55 (−5.79, 10.89)
Mean sensor glucose, mmol/L	7.13 ± 0.94 [128.47 ± 16.94]	7.24 ± 0.77 [130.45 ± 13.87]	0.212402	−0.11 (0.6, 0.38)
SD	2.27 ± 0.55	2.18 ± 0.48	0.284579	0.09 (−0.21, 0.39)
%CV	31.48 ± 4.35	29.99 ± 3.92	0.031612	1.49 (−0.89, 3.87)
Total daily dose of insulin, units	51.48 ± 15.91	48.77 ± 14.15	0.169940	5.35 (−6.44, 11.79)

V2–V5	AAPS	780G	P value	Δ mean (95% CI)
HbA1c, mmol/mol (%DCCT)	44.56 ± 6.01 (6.2 ± 2.7)	46.13 ± 4.87 (6.3 ± 2.8)	0.140146	−1.56 (−5.6, 2.48)
TIR (3.9–10.0 mmol/L [70–180 mg/dL]),%	83.99 ± 6.38	83.81 ± 5.54	0.834336	0.18 (−4.28, 4.64)
TBR (<3.9 mmol/L [<70 mg/dL]),%	3.74 ± 1.75	1.81 ± 1.01	0.003205	1.95 (0.85, 3)
TBR (<3.0 mmol/L [<54 mg/dL]),%	0.59 ± 0.38	0.06 ± 0.24	0.002093	0.53 (0.29, 0.77)
TAR (>10.0 mmol/L [>180 mg/dL]),%	12.38 ± 7.03	14.38 ± 5.83	0.125573	−2 (−6.82, 2.82)
TAR (>13.9 mmol/L [>250 mg/dL]),%	1.53 ± 1.04	1.53 ± 1.11	0.983167	−0.006 (−0.81, 0.79)
TITR (3.9–7.8 mmol/L [70–140 mg/dL]),%	64.69 ± 10.75	61.13 ± 9.48	0.132299	3.56 (−4.29, 11.39)
Mean sensor glucose, mmol/L	7.19 ± 0.69 [129.55 ± 12.43]	7.52 ± 0.5 [135.5 ± 9]	0.059709	−0.33 (−0.78, 0.13)
SD	2.31 ± 0.47	2.24 ± 0.29	0.410198	0.07 (−0.23, 0.36)
%CV	31.83 ± 4.46	29.73 ± 2.91	0.026058	2.09 (−0.73, 4.93)
Total daily dose of insulin, units	49.86 ± 17.19	53.54 ± 20.02	0.310897	−3.68 (−17.61, 10.26)

CIQ, Control-IQ; TITR, time in tight range; 780G, MiniMed 780G.

Table 4.

Comparison of AndroidAPS (Visit 2) and Commercially Available Automated Insulin Delivery Systems Control-IQ and MiniMed 780G in Combined Analysis (Visit 5)

V2–V5	AAPS	CIQ + 780G	P value	Δ mean (95% CI)
HbA1c, mmol/mol (%DCCT)	45.98 ± 8.81 (6.4 ± 3)	45.73 ± 6.16 (6.3 ± 2.7)	0.67823	0.24 (−3.15, 0.24)
TIR (3.9–10.0 mmol/L [70–180 mg/dL]),%	84.15 ± 7.61	84.98 ± 6.86	0.3100298	−0.83 (−4.05, 2.39)
TBR (<3.9 mmol/L [<70 mg/dL]),%	3.82 ± 1.92	2.17 ± 1.17	0.000038	1.65 (0.94, 2.35)
TBR (<3.0 mmol/L [<54 mg/dL]),%	0.69 ± 0.65	0.27 ± 0.44	0.000248	0.42 (0.17, 0.66)
TAR (>10.0 mmol/L [>180 mg/dL]),%	11.96 ± 8.38	12.66 ± 7.27	0.240121	−0.7 (−4.19, 2.78)
TAR (>13.9 mmol/L [>250 mg/dL]),%	1.83 ± 2.58	1.55 ± 2.09	0.50114	0.28 (−0.77, 1.33)
TITR (3.9–7.8 mmol/L [70–140 mg/dL]),%	66.38 ± 11.84	63.4 ± 11.77	0.035035	2.98 (−2.73, 8.69)
Mean sensor glucose, mmol/L	7.16 ± 0.84 (129 ± 15.14)	7.35 ± 0.68 (132.43 ± 12.25)	0.04245	−0.19 (−0.53, 0.14)
SD	2.29 ± 0.52	2.2 ± 0.41	0.11008	0.08 (−0.13, 0.29)
%CV	31.62 ± 4.35	29.81 ± 3.49	0.001659	1.81 (0.05, 3.56)
Total daily dose of insulin, units	50.85 ± 16.44	50.78 ± 16.86	0.844515	0.7 (−7.34, 7.48)

After returning to AAPS in the last study phase, there was no significant difference in TIR, TAR, mean sensor glucose, or HbA1c compared with commercially available systems.

However, time in hypoglycemia was again significantly higher with AAPS (3.9 ± 2.2 vs. 2.2 ± 1.2%; P < 0.001) and in the separate analysis for AAPS versus CIQ (3.8 ± 2.3 vs. 2 ± 0.94, P = 0.004) and for AAPS versus 780G (3.91 ± 2.24 vs. 1.71 ± 1.03, P = 0.005). %CV was significantly lower with commercially available systems (29.51 ± 3.37 vs. 31.5 ± 4.31%; P = 0.005) and in the separate analysis for 780G versus AAPS (29.37 ± 2.88 vs. 32.35 ± 4.34%; P = 0.004), but not for CIQ versus AAPS (29.63 ± 3.75 vs. 30.76 ± 4.14%; P = 0.21).

%TITR (70–140 mg/dL [3.9–7.8 mmol/L]) was higher using AAPS than commercially available systems (65.78 ± 11.83 vs. 62.73 ± 11.57; P = 0.048). However, there was no difference in separate analyses for CIQ versus AAPS (63.75 ± 12.93 vs. 67.12 ± 11.79; P = 0.25) and 780G versus AAPS (61.57 ± 9.66 vs. 64.24 ± 11.69; P = 0.22) (Tables 5, 6).

Table 5.

Comparison of Commercially Available Automated Insulin Delivery Systems—Control-IQ and MiniMed 780G (Visit 5) and Extension with AndroidAPS in Separate Analysis (Visit 6)

V5–V6	CIQ	AAPS	P value	Δ mean (95% CI)
HbA1c, mmol/mol (%DCCT)	46.31 ± 7.16 (6.4 ± 2.8)	46.06 ± 9.81 (6.4 ± 3)	0.43277	0.25 (−6.18, 6.68)
TIR (3.9–10.0 mmol/L [70–180 mg/dL]),%	85.44 ± 8.26	85.89 ± 4.79	0.55093	−0.46 (−5.54, 4.63)
TBR (<3.9 mmol/L [<70 mg/dL]),%	2 ± 0.94	3.8 ± 2.33	0.00417	−1.8 (−3.15, 0.45)
TBR (<3.0 mmol/L [<54 mg/dL]),%	0.31 ± 0.46	0.74 ± 0.54	0.03285	−0.43 (−0.81, −0.06)
TAR (>10.0 mmol/L [>180 mg/dL]),%	12.13 ± 8.89	10.4 ± 5.53	0.61552	1.73 (−3.84, 7.29)
TAR (>13.9 mmol/L [>250 mg/dL]),%	1.81 ± 3	1.05 ± 0.81	0.47553	0.76 (−0.99, 2.39)
TITR (3.9–7.8 mmol/L [70–140 mg/dL]),%	63.75 ± 12.93	67.12 ± 11.79	0.24549	−3.37 (−12.59, 5.86)
Mean sensor glucose, mmol/L	7.32 ± 0.79	7.01 ± 0.58	0.13230	0.31 (−0.21, 0.82)
SD	2.19 ± 0.49	2.16 ± 0.37	0.64956	0.025 (−0.29, 0.35)
%CV	29.63 ± 3.75	30.76 ± 4.14	0.21460	−1.13 (−4.07, 1.82)

V5–V6	780G	AAPS	P value	Δ mean (95% CI)
HbA1c, mmol/mol (%DCCT)	45.71 ± 4.77 (6.3 ± 2.6)	45.71 ± 5.15 (6.3 ± 2.6)	0.78365	0 (4, 4)
TIR (3.9–10.0 mmol/L [70–180 mg/dL]),%	84.14 ± 5.77	84.93 ± 6.49	0.87848	0.79 (5.75, 4.16)
TBR (<3.9 mmol/L [<70 mg/dL]),%	1.71 ± 1.03	3.91 ± 2.24	0.00518	−2.2 (−3.64, −0.76)
TBR (<3.0 mmol/L [<54 mg/dL]),%	0.07 ± 0.26	0.76 ± 0.91	0.01172	−0.69 (−1.25, −0.13)
TAR (>10.0 mmol/L [>180 mg/dL]),%	14.14 ± 6.15	11.08 ± 7.41	0.080739	3.06 (−2.44, 8.57)
TAR (>13.9 mmol/L [>250 mg/dL]),%	1.46 ± 1.11	1.8 ± 1.69	0.28600	−0.34 (−1.49, 0.83)
TITR (3.9–7.8 mmol/L [70–140 mg/dL]),%	61.57 ± 9.66	64.24 ± 11.69	0.220899	−2.67 (−11.34, 5.99)
Mean sensor glucose, mmol/L	7.51 ± 0.53	7.19 ± 0.68	0.19554	0.32 (−0.17, 0.81)
SD	2.21 ± 0.3	2.33 ± 0.44	0.22402	−1.3 (−0.43, 0.18)
%CV	29.37 ± 2.88	32.35 ± 4.34	0.00429	−2.97 (−5.96, 0.01)

Table 6.

Comparison of Commercially Available Automated Insulin Delivery Systems—Control-IQ and MiniMed 780G (Visit 5) and Extension with AndroidAPS in Combined Analysis (Visit 6)

V5–V6	CIQ + 780G	AAPS	P value	Δ mean (95% CI)
HbA1c, mmol/mol (%DCCT)	46.03 ± 6.17 (6.4 ± 2.7)	45.9 ± 7.98 (6.3 ± 2.9)	0.72098	0.13 (−3.62, 3.89)
TIR (3.9–10.0 mmol/L [70–180 mg/dL]),%	84.83 ± 7.23	85.45 ± 5.67	0.73662	−0.61 (−4.03, 2.81)
TBR (<3.9 mmol/L [<70 mg/dL]),%	1.87 ± 0.99	3.85 ± 2.29	0.000057	−1.99 (−2.92, −1.05)
TBR (<3.0 mmol/L [<54 mg/dL]),%	0.2 ± 0.4	0.79 ± 0.78	0.00129	−0.59 (−0.92, −0.26)
TAR (>10.0 mmol/L [>180 mg/dL]),%	13.07 ± 7.79	10.72 ± 6.48	0.09197	2.95 (−0.86, 6.76)
TAR (>13.9 mmol/L [>250 mg/dL]),%	1.65 ± 2.33	1.47 ± 1.33	0.59218	0.18 (−0.82, 1.18)
TITR (3.9–7.8 mmol/L [70–140 mg/dL]),%	62.73 ± 11.57	65.78 ± 11.83	0.04754	−3.04 (−9.19, −3.11)
Mean sensor glucose, mmol/L	7.41 ± 0.69	7.09 ± 0.63	0.05461	0.31 (−0.03, 0.66)
SD	2.19 ± 0.41	2.24 ± 0.41	0.27433	−0.05 (−0.26, 0.17)
%CV	29.51 ± 3.37	31.5 ± 4.31	0.00532	−1.99 (−4.03, 0.05)

Safety outcomes

No episodes of severe hypoglycemia or diabetic ketoacidosis requiring hospitalization were reported. Despite 4-step bolus administration in CIQ, one participant mistakenly administered 25 units of bolus insulin. However, this mistake did not result in severe hypoglycemia. No skin reactions, hematomas, or infections at the sensor or cannula insertion sites were reported during this study.

Discussion

The extension of the CODIAC study is the only prospective study investigating the switch between OS and commercially available AID systems.

Commercially available AID systems and AAPS were comparable in TIR (70–180 mg/dL [3.9–10.0 mmol/L]). In addition to TIR, we did not observe any significant difference in secondary outcomes such as HbA1c and TAR (both >180 mg/dL [>10.0 mmol/L] and >250 mg/dL [>13.9 mmol/L]). However, the use of AAPS was associated with higher TITR, but also with risk of both levels of hypoglycemia (<70 mg/dL [<3.9 mmol/L] and <54 mg/dL [<3.0 mmol/L]). Although TBR was higher in AAPS, all participants using AAPS or commercial systems met consensus guidelines for TBR (<4%).²⁰

Consistent with hypoglycemia, glycemic variability (%CV) was also lower when using commercially available AID systems. The observed differences can probably be linked to the diverse algorithms utilized by these systems. AAPS, specifically, allows a high level of customization, which includes the option to set a target glycemia as low as 4.5 mmol/L. These features may, in turn, be associated with a higher incidence of hypoglycemia compared with commercially available systems. However, AAPS has more settings that can be adjusted and this makes it challenging to directly compare with commercially available AID systems, which have fewer settings other than basal rate, ISF, and CIR that a user can adjust.

In this study, we also assessed the TITR as a new, emerging metric for the assessment of normoglycemia.^20,21

When using AAPS, the average glycemia and TITR appeared to be significantly better compared with a combined set of commercial systems. This difference might be explained by the numerically higher, although not statistically significant, mean sensor glucose and lower TITR observed with the 780G system. In contrast, the 780G system was associated with the lowest risk of hypoglycemia. Similar trends were noted in the extension study, suggesting a consistent pattern in the outcomes.

Although the occurrence of hypoglycemia in all systems aligns with expert consensus,²² it is essential to recognize that the evaluation of hypoglycemia can differ in specific clinical situations, particularly in individuals who experience hypoglycemia unawareness or in elderly PwT1D with diabetes-related complications. Moreover, meeting the existing recommendations does not imply that we should not strive for even more optimal outcomes.

Clinicians frequently encounter the use of OS-AID in their daily practice. However, there is yet to be any established structured training for clinicians regarding OS-AID, and significant gaps remain in the evidence from clinical studies.

The majority of studies rely on self-reported data, which present a considerable risk of bias,^23–26 with a few exceptions.^5–8,27 A retrospective observational trial conducted by Jeyaventhan et al.²⁸ compared commercially available AID Medtronic MiniMed 670G with OS-AID. However, the specific type of OS-AID utilized in that study was not detailed.

In contrast to our study, we included only participants utilizing the single type of OS-AID system, which minimizes the potential for bias.

Moreover, 670G was the first generation of commercially available AID with many limitations.^29,30 Burnside et al. conducted a clinical trial with AAPS. However, in contrast with our study, their trial evaluated the effectiveness of AAPS compared with SAP.^6,7 In addition, a similar design was also used in other studies.^5–8 However, we believe that comparing the AID system with SAP will not provide many new insights, and there is a need to concentrate on research involving diverse AID systems, as these are recognized as the standard treatment for type 1 diabetes.^31,32 While significant variations among AID systems are typically not anticipated,³² our study indicates that there might be differences that can be considered clinically relevant. It is also important to note that our trial involved adults with already optimal glycemic outcomes experienced in using AID, in contrast to other studies that included participants lacking AID experience. These aspects could have affected the inferior average outcomes observed in AAPS users compared with our findings.^5–7 Our study is not without its limitations.

Participants were trained on commercial systems as part of the study design. No training by the study team or evaluation of settings or skills was available for the AAPS, this may have influenced the resulting rates of hypoglycemia in AAPS. However, AAPS training requires several weeks of self-training, and users must complete a series of 10 objectives before initiating AAPS. In contrast, the study team provided participants with standard training for CIQ or 780G, which lasted around 60 min.

Because this was a nonrandomized study, there is a possibility that certain unmeasured variables could influence our results. However, numerous countries mandate national regulatory approval for each device and software used in Randomized Control Trials (RCTs) and additional regulatory clearance for every version of those devices and software. These constraints make RCTs with OS-AID systems challenging to conduct in many countries, with a few exceptions.^5–8 In the Czech Republic, where this trial was conducted, national regulatory authorities approved RCTs solely for studies conducted within hospital environments.²⁷ However, they did not allow RCTs to be conducted in home settings. The design of our trial represents the most that we could achieve within the existing regulatory landscape while still keeping the system OS.

Conclusions

The results showed that switching from AAPS to commercially available systems was not associated with a change in TIR and HbA1c. However, this study suggests that TITR may be higher when using AAPS, but also with a higher risk of hypoglycemia compared with both commercially available AID systems (Fig. 2).

FIG. 2.

(A) Column charts displaying percentage time in ranges for AAPS (visit 2) and 780G (visit 5). (B) Column charts displaying percentage time in ranges for AAPS (visit 2) and CIQ (visit 5). (C) Column charts displaying percentage time in ranges for AAPS (visit 2) and CIQ + 780G (visit 5). (D) Column charts displaying percentage time in ranges for CIQ (visit 5) and AAPS (visit 6). (E) Column charts displaying percentage time in ranges for 780G (visit 5) and AAPS (visit 6). (F) Column charts displaying percentage time in ranges for CIQ + 780G (visit 5) and AAPS (visit 6). AAPS, AndroidAPS; 780G, MiniMed 780G; CIQ, Control-IQ.

Authors’ Contributions

J.Š. was responsible for the protocol design. J.Š., L.R., A.H., J.K., Q.D.D., E.H., G.G., C.G.P., M.F., and M.P. wrote and revised the article. Q.D.D., L.R., J.Š., A.H., J.K., M.F., E.H., and M.P. were responsible for study implementation and administration. Q.D.D., L.R., J.Š., A.H., J.K., C.G.P., and E.H. reviewed the data. J.Š. is the guarantor of this work and, as such, had full access to all the data in the study and had final responsibility for the decision to submit for publication.

References

Holt

RIG

, DeVries

, Hess-Fischl

, et al. The management of Type 1 diabetes in adults. A Consensus Report by the American Diabetes Association (ADA) and the European Association for the Study of Diabetes (EASD). Diabetes Care, 2021; 44(11):2589–2625; doi: 10.2337/dci21-0043

Asarani

NAM

, Reynolds

, Elbalshy

, et al. Efficacy, safety, and user experience of DIY or open-source artificial pancreas systems: A systematic review. Acta Diabetol, 2021; 58(5):539–547; doi: 10.1007/s00592-020-01623-4

OpenAPS Outcomes—OpenAPS.org. Available from: https://OpenAPS.org/outcomes/ [Last accessed: August 26, 2024].

Lum

, Bailey

, Barnes-Lomen

, et al. A real-world prospective study of the safety and effectiveness of the loop open source automated insulin delivery system. Diabetes Technol Ther, 2021; 23(5):367–375; doi: 10.1089/dia.2020.0535

Burnside

, Lewis

, Crocket

, et al. CREATE (Community deRivEd AutomaTEd insulin delivery) trial. Randomized parallel arm open-label clinical trial comparing automated insulin delivery using a mobile controller (AnyDANA-loop) with an open-source algorithm with sensor-augmented pump therapy in type 1 diabetes. J Diabetes Metab Disord, 2020; 19(2):1615–1629; doi: 10.1007/s40200-020-00547-8

Burnside

, Lewis

, Crocket

, et al. Open-source automated insulin delivery in Type 1 Diabetes. N Engl J Med, 2022; 387(10):869–881; doi: 10.1056/NEJMoa2203913

Burnside

, Lewis

, Crocket

, et al. Extended use of an open-source automated insulin delivery system in children and adults with Type 1 Diabetes: The 24-week continuation phase following the CREATE randomized controlled trial. Diabetes Technol Ther, 2023; 25(4):250–259; doi: 10.1089/dia.2022.0484

Nanayakkara

, Sharifi

, Burren

, et al. Hybrid closed loop using a do-it-yourself artificial pancreas system in adults with Type 1 diabetes. J Diabetes Sci Technol, 2024; 18(4):889–896; doi: 10.1177/19322968231153882

, Luo

, Zheng

, et al. The use of a do-it-yourself artificial pancreas system is associated with better glucose management and a higher quality of life among adults with type 1 diabetes. Ther Adv Endocrinol Metab, 2020; 11:2042018820950146; doi: 10.1177/2042018820950146

10.

Brown

, Kovatchev

, Raghinaru

, et al.; iDCL Trial Research Group. Six-month randomized, multicenter trial of closed-loop control in type 1 diabetes. N Engl J Med, 2019; 381(18):1707–1717; doi: 10.1056/NEJMoa1907863

11.

Breton

, Kanapka

, Beck

, et al.; iDCL Trial Research Group. A randomized trial of closed-loop control in children with Type 1 Diabetes. N Engl J Med, 2020; 383(9):836–845; doi: 10.1056/NEJMoa2004736

12.

Tauschmann

, Thabit

, Bally

, et al.; APCam11 Consortium. Closed-loop insulin delivery in suboptimally controlled type 1 diabetes: A multicentre, 12-week randomized trial. Lancet, 2018; 392(10155):1321–1329; doi: 10.1016/S0140-6736(18)31947-0

13.

Tauschmann

, Allen

, Nagl

, et al.; KidsAP Consortium. Home use of day-and-night hybrid closed-loop insulin delivery in very young children: A multicenter, 3-week, randomized trial. Diabetes Care, 2019; 42(4):594–600; doi: 10.2337/dc18-1881

14.

, Hásková

, Radovnická

, et al. Comparison of Control-IQ and open-source AndroidAPS automated insulin delivery systems in adults with type 1 diabetes: The CODIAC study. Diabetes Obes Metab, 2024; 26(1):78–84; doi: 10.1111/dom.15289

15.

Frequently Asked Questions—OpenAPS.org. Available from: https://OpenAPS.org/frequently-asked-questions/ [Last accessed: December 26, 2024].

16.

OpenAPS features—AndroidAPS 3.2 Documentation. Available from: https://wiki.aaps.app/en/latest/Usage/Open-APS-features.html [Last accessed: December 26, 2024].

17.

Completing the Objectives — AndroidAPS 3.2 Documentation. Available from: https://wiki.aaps.app/en/latest/Usage/completing-the-objectives.html [Last accessed: December 26, 2024].

18.

Braune

, Lal

, Petruželková

, et al.; OPEN International Healthcare Professional Network and OPEN Legal Advisory Group. Open-source automated insulin delivery: International consensus statement and practical guidance for health-care professionals. Lancet Diabetes Endocrinol, 2022; 10(1):58–74.

19.

What are Automated Insulin Delivery Systems? Medtronic. 2024. Available from: https://www.medtronicdiabetes.com/treatments/automated-insulin-delivery [Last accessed: December 11, 2024].

20.

Battelino

, Alexander

, Amiel

, et al. Continuous glucose monitoring and metrics for clinical trials: An international consensus statement. Lancet Diabetes Endocrinol, 2023; 11(1):42–57; doi: 10.1016/S2213-8587(22)00319-9

21.

Akturk

, Battelino

, Castañeda

, et al. Future of time-in-range goals in the era of advanced hybrid closed-loop automated insulin delivery systems. Diabetes Technol Ther, 2024; 26(S3):102–106; doi: 10.1089/dia.2023.0432

22.

Battelino

, Danne

, Bergenstal

, et al. Clinical targets for continuous glucose monitoring data interpretation: Recommendations from the international consensus on time in range. Diabetes Care, 2019; 42(8):1593–1603; doi: 10.2337/dci19-0028

23.

Lewis

, Leibrand

, #OpenAPS Community. Real-world use of open source artificial pancreas systems. J Diabetes Sci Technol, 2016; 10(6):1411; doi: 10.1177/1932296816665635

24.

Lee

, Newman

, Gebremariam

, et al. Real-world use and self-reported health outcomes of a patient-designed do-it-yourself mobile technology system for diabetes: Lessons for mobile health. Diabetes Technol Ther, 2017; 19(4):209–219; doi: 10.1089/dia.2016.0312

25.

Street

. Review of self-reported data from UK Do-It-Yourself Artificial Pancreas System (DIYAPS) users to determine whether demographic of population affects use or outcomes. Diabetes Ther, 2021; 12(7):1839–1848; doi: 10.1007/s13300-021-01071-5

26.

Litchman

, Lewis

, Kelly

, et al. Twitter Analysis of #OpenAPS DIY artificial pancreas technology use suggests improved HbA1c and quality of life. J Diabetes Sci Technol, 2019; 13(2):164–170; doi: 10.1177/1932296818795705

27.

Petruzelkova

, Neuman

, Plachy

, et al. First use of open-source automated insulin delivery androidaps in full closed-loop scenario: Pancreas4all Randomized Pilot Study. Diabetes Technol Ther, 2023; 25(5):315–323; doi: 10.1089/dia.2022.0562

28.

Jeyaventhan

, Gallen

, Choudhary

, et al. A real-world study of user characteristics, safety and efficacy of open-source closed-loop systems and Medtronic 670G. Diabetes Obes Metab, 2021; 23(8):1989–1994; doi: 10.1111/dom.14439

29.

Saunders

, Messer

, Forlenza

. MiniMed 670G hybrid closed loop artificial pancreas system for treating type 1 diabetes mellitus: Overview of its safety and efficacy. Expert Rev Med Devices, 2019; 16(10):845–853; doi: 10.1080/17434440.2019.1670639

30.

Messer

, Forlenza

, Sherr

, et al. Optimizing hybrid closed-loop therapy in adolescents and emerging adults using the MiniMed 670G System. Diabetes Care, 2018; 41(4):789–796; doi: 10.2337/dc17-1682

31.

Weisman

, Bai

, Cardinez

, et al. Effect of artificial pancreas systems on glycaemic control in patients with type 1 diabetes: A systematic review and meta-analysis of outpatient randomized controlled trials. Lancet Diabetes Endocrinol, 2017; 5(7):501–512; doi: 10.1016/S2213-8587(17)30167-5

32.

Phillip

, Nimri

, Bergenstal

, et al. Consensus recommendations for the use of automated insulin delivery technologies in clinical practice. Endocr Rev, 2023; 44(2):254–280; doi: 10.1210/endrev/bnac022

Switch of Open-Source Automated Insulin Delivery (AID) System—AndroidAPS to Commercially Available AID Systems in Type 1 Diabetes: The Extension of the CODIAC Study

Abstract

Objective:

Research Design and Methods:

Results:

Conclusion:

Keywords

Introduction

Research Design and Methods

Study design and participants

Procedures

Hybrid closed-loop systems

Outcomes

Statistical analysis

Results

Baseline characteristics and adherence

Primary and secondary end points

Safety outcomes

Discussion

Conclusions

Authors’ Contributions

References