Abstract
Background: Data are limited on the need for and benefits of pump setting optimization with automated insulin delivery. We examined clinical management of a closed-loop control (CLC) system and its relationship to glycemic outcomes.
Materials and Methods: We analyzed personal parameter adjustments in 168 participants in a 6-month multicenter trial of CLC with Control-IQ versus sensor-augmented pump (SAP) therapy. Preset parameters (BR = basal rates, CF = correction factors, CR = carbohydrate ratios) were optimized at randomization, 2 and 13 weeks, for safety issues, participant concerns, or initiation by participants' usual diabetes care team. Time in range (TIR 70–180 mg/dL) was compared in the week before and after parameter changes.
Results: In 607 encounters for parameter changes, there were fewer adjustments for CLC than SAP (3.4 vs. 4.1/participant). Adjustments involved BR (CLC 69%, SAP 80%), CR (CLC 68%, SAP 50%), CF (CLC 44%, SAP 41%), and overnight parameters (CLC 62%, SAP 75%). TIR before and after adjustments was 71.2% and 71.3% for CLC and 61.0% and 62.9% for SAP. The highest baseline HbA1c CLC subgroup had the largest TIR improvement (51.2% vs. 57.7%). When a CR was made more aggressive in the CLC group, postprandial time >180 mg/dL was 43.1% before the change and 36.0% after the change. The median postprandial time <70 mg/dL before making CR less aggressive was 1.8%, and after the change was 0.7%.
Conclusions: No difference in TIR was detected with parameter changes overall, but they may have an effect in higher HbA1c subgroups or following user-directed boluses, suggesting that changes may matter more in suboptimal control or during discrete periods of the day.
Clinical Trials Registration number: NCT03563313.
Keywords: Type 1 diabetes, Automated insulin delivery, Closed-loop control, Continuous glucose monitor, Pump parameters
Introduction
Automated insulin delivery is a promising approach to improve glycemic outcomes for people with type 1 diabetes. Currently, two closed-loop systems, the Medtronic MiniMed 670G and the Tandem t:slim X2 with Control-IQ Technology, are approved by the Food and Drug Administration (FDA) for commercial use in the United States. Meta-analyses have found that automated insulin delivery systems can be effective; however, there are limited data on pump setting optimization and clinician input for these systems.1–3
The two commercially available systems use different algorithms, and their insulin delivery is affected by different preset personal parameters. While in automated mode, insulin delivery by the Medtronic MiniMed 670G system is affected by the previous 2–6 days' total daily insulin (TDI), continuous glucose monitor (CGM) trends, and current target (120 or 150 mg/dL). User-initiated correction boluses are affected by active insulin time. Automated insulin delivery by the Control-IQ system is affected to some extent by tracked TDI, preset insulin delivery parameters (i.e., basal rates [BR], bolus settings), and current and predicted CGM trends. The target glucose range for the algorithm is 112.5–160 mg/dL under usual conditions with tighter ranges of 112.5–120 mg/dL during sleep and 140–160 mg/dL during exercise.
The system can also deliver an automatic correction dose up to once per hour if indicated. Automatic correction doses are not affected by the preset active insulin time but are affected by the preset correction factor (CF). User-initiated correction boluses are calculated using the preset CF and a fixed target of 110 mg/dL. Both the MiniMed 670G and Control-IQ systems use preset insulin to carbohydrate ratios (CR) for meal boluses.
We have previously reported the results of the International Diabetes Closed Loop (iDCL) trial, which assessed the efficacy and safety of the Control-IQ system. The mean time in range (TIR, defined at 70–180 mg/dL) for the closed-loop control (CLC) group was 11 percentage points higher than the sensor-augmented pump (SAP) control group (CLC: 61% ± 17% at baseline to 71% ± 12% during the 6-month trial vs. SAP group: 59% ± 14% at baseline and unchanged during the 6-month trial).4 Participants were in closed-loop with a median of 90% (interquartile range [IQR] 86%–94%) during the trial.4
In the pivotal trial of the Medtronic MiniMed 670G system, TIR increased from 60.4% ± 10.9% to 67.2% ± 8.2% in adolescents and from 68.8% ± 11.9% to 73.8% ± 8.4% in adults.5 The median time in automated mode was 75.8% (IQR 68.0%–88.4%) for adolescents and 88% (IQR 77.6%–92.7%) for adults.5 Real-world observations of the 670G system found some users had difficulty remaining in the automated mode, with time of CLC use declining to 72%–80%,6–9 or discontinuing automated mode altogether.10
To optimize use and efficacy of CLC systems, providers must understand how to set expectations for users, how to train them on appropriate use of the system, and which parameter adjustments may be used to optimize TIR. Therefore, the purpose of this article was to report on the observed clinical management and participant use of a CLC system (Control-IQ) compared with SAP during the iDCL trial and its relationship to glycemic outcomes.
Materials and Methods
Trial design
We analyzed personal parameter adjustments in 168 participants of the 6-month, multicenter iDCL trial. A total of 168 participants were randomly assigned to CLC (112 participants) or SAP (56 participants) in a 2:1 randomization scheme. The groups appeared balanced with regard to baseline characteristics with mean age 33 years; mean HbA1c at randomization at 7.4% and TDI 0.6 U/kg in both groups with the proportion in CLC versus SAP, respectively, for ages 14–24 years (36% vs. 41%); multiple daily injection users (20% vs. 23%); HbA1c > 8.5% (13% vs. 7%) and female (48% vs. 54%).4 The study was approved by a central institutional review board. Further details of the study protocol can be found in Brown et al.4
The study began with a run-in phase to assess adherence with study procedures, introduce the study CGM to sensor-naive participants, and introduce insulin pump use to participants who were previously using multiple daily injections or different pump. At the run-in review visit, optimization of personal pump parameters was performed for all participants based on investigator recommendations. Investigators were allowed to change any parameters based on clinical judgment. Additional pump setting optimization was performed at 2 and 13 weeks during the study. Changes were only made at other times for safety issues, participant concerns, or if initiated by participants' usual diabetes care team.
Pump downloads were required for the CLC group but were optional for the SAP group. Therefore, data on pump parameter adjustments were captured on electronic case report forms at defined study visits and in an ad hoc manner if identified outside study visits.
Standard pump parameters that are preset and could be adjusted by participants and study teams include BR, CR, and CF, and an analysis of these adjustments are the focus of this article. For the CLC system, additional inputs that could be adjusted during the trial included: weight, TDI, and sleep times. Weight between 20 and 140 kg was entered at initialization, and the upper bound was input if participants were outside those ranges as there were no enrollment criteria based on weight.
TDI between 10 and 100 U/day was entered at initialization. Participants were not enrolled if TDI was <10 U, whereas TDI was entered as 100 U at initialization when the actual TDI exceeded 100 U/day. The manually entered TDI is only relevant upon starting a new system or after a system reboot. After the first day, the TDI is then tracked and updated automatically without user input with a rolling 6-day average. For this reason, sites were allowed to adjust TDI if they wished but were not required to do so during the trial. Study guidelines encouraged CLC participants to set sleep times as well as activate exercise mode when appropriate but did not require it.
Bolus targets and duration of insulin action could be adjusted for both the SAP and CLC groups. For the CLC system, however, those settings were only relevant when automated insulin delivery is turned off and the pump was operating as a standard pump (Control-IQ toggled off). The CLC system does not allow changes to any system targets (described in the Introduction section). In the case of CLC, the system uses a different insulin action curve when the pump is being used as a standard pump than when it is being used as a CLC system. The automated insulin delivery uses an insulin action curve of approximately 5 h, but this curve is not linear and much of the insulin action decay occurs in the first few hours.
Algorithm description
The CLC algorithm has several distinct features: (1) automated insulin delivery through continual BR modulation as well as periodic automated correction boluses, (2) attenuation of insulin delivery to avoid hypoglycemia, and (3) gradual lowering of the target to a more narrow range overnight. These goals are attained by optimizing insulin on board to achieve desired targets based on predictions of CGM excursions.
Although the insulin parameters are preset in a similar manner as a standard pump, the CLC system can vary the insulin infusion rate. Insulin delivery is increased when CGM value is predicted to be above glucose targets in 30 min with a maximum rate that varies but can never exceed four times the preprogrammed BR. Conversely, insulin delivery is decreased when CGM value is predicted to be below glucose targets in 30 min, and insulin delivery will cease if CGM value is predicted to be ≤70 mg/dL in 30 min (or ≤80 mg/dL during exercise). The target glucose range for the algorithm varies by activity (sleep or exercise) as described in the Introduction section and cannot be adjusted by the user. TDI is tracked and serves as boundaries for the calculation of a CF that is used by the algorithm for BR modulation.
In addition to these continual adjustments in basal insulin delivery, there are periodic automated correction boluses that can occur once an hour. The automated corrections can occur when CGM value is predicted to be >180 mg/dL in 30 min despite increased BR delivery as long as there was no prior user-initiated bolus within the hour and sleep setting is not activated. The automated corrections are calculated by giving 60% of a usual correction dose based on the user's preset CF using a fixed target of 110 mg/dL and adjusted for insulin-on-board. The automated corrections as well as the automated BR adjustments are fed into insulin-on-board to limit future bolus actions by either the user or the algorithm.
The system also has several alerts and alarms that are specific to automated insulin delivery. Two relevant alerts include a hypoglycemic risk alert that is triggered when predicted CGM value is expected to be <70 mg/dL in 15 min (or <80 mg/dL if exercise is activated). In addition, there is a hyperglycemic risk alert that is triggered when CGM value is predicted to be >200 mg/dL and not decreasing in 30 min despite increased insulin delivery.
Statistical methods
For this secondary analysis, the aim was to describe the parameter changes that occurred throughout the randomized trial. Given the study was not powered for this analysis, descriptive statistics are provided with no statistical tests to compare groups. Overnight changes were defined as parameters from midnight to 6 a.m. For the purposes of this analysis, participants were stratified into two age categories: adolescents and young adults (AYA), defined as those aged 14–24 years, and adults, defined as those aged >24 years. Analysis by enrollment HbA1c divided participants into three categories: lowest HbA1c (<7%), mid-range HbA1c (7%–8.4%), and highest HbA1c (≥8.5%). Analysis of glycemic changes in response to parameter changes in CF and CR that directly affect participant-directed boluses was confined to the CLC group as SAP bolus data were not available.
Results
Parameter adjustments
There were 607 encounters for parameter adjustments throughout the 6-month study. The mean number of parameter changes per participant was 3.4 ± 2.1 in the closed-loop group and 4.1 ± 2.1 in the control group (Table 1). Number of changes per participant by site ranged from 1.9 ± 1.5 to 5.7 ± 2.2 in the CLC group and from 2.8 ± 2.1 to 7.5 ± 1.6 in the SAP group. Changes by gender were female CLC 3.6 ± 2.3, female SAP 4.4 ± 2.2, male CLC 3.2 ± 2.0, and male SAP 3.7 ± 2.1 changes.
Table 1.
Parameter Changes, Insulin Doses, and Time in Range Overall and Stratified by Age and HbA1c
| Comparison |
Overall study |
Comparison of AYA and adults |
Comparison by HbA1c categories |
|||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Study group |
CLC |
SAP |
CLC |
SAP |
CLC |
SAP |
||||||
| Parameter | Overall (n = 112) | Overall (n = 56) | AYA (n = 40) | Adults (n = 72) | AYA (n = 23) | Adults (n = 33) | <7% (n = 38) | 7%–8.4% (n = 60) | ≥8.5% (n = 14) | <7% (n = 19) | 7%–8.4% (n = 33) | ≥8.5% (n = 4) |
| Mean number of parameter changes/participant | 3.4 ± 2.1 | 4.1 ± 2.1 | 3.7 ± 1.8 | 3.2 ± 2.3 | 4.0 ± 1.5 | 4.2 ± 2.5 | 3.0 ± 2.1 | 3.6 ± 2.1 | 3.5 ± 2.1 | 3.7 ± 2.3 | 4.2 ± 2.2 | 4.3 ± 1.0 |
| % of changes involving BR | 69 | 80 | 74 | 65 | 70 | 86 | 73 | 65 | 73 | 83 | 81 | 59 |
| Median % change in BR | +0.28 | +0.38 | +0.37 | +0.28 | +1.14 | −0.26 | +0.82 | +0.28 | −0.25 | −0.90 | +0.73 | +8.47 |
| % of changes involving CR | 68 | 50 | 71 | 67 | 52 | 49 | 59 | 74 | 63 | 48 | 50 | 59 |
| Median % change in CR | −1.66 | −1.64 | −1.90 | −1.56 | −1.76 | −1.30 | −1.64 | −1.54 | −2.27 | −0.38 | −2.11 | −4.53 |
| % of changes involving CF | 44 | 41 | 45 | 43 | 48 | 36 | 43 | 39 | 67 | 41 | 41 | 41 |
| Median % change in CF | −1.62 | −0.68 | −1.79 | −1.33 | −0.44 | −1.55 | −1.92 | −0.43 | −1.96 | +0.09 | −1.55 | +15.91 |
| % of changes involving an overnight parameter | 62 | 75 | 65 | 60 | 71 | 78 | 65 | 58 | 73 | 70 | 78 | 76 |
| % of changes involving a daytime parameter | 97 | 96 | 98 | 97 | 95 | 98 | 97 | 98 | 96 | 97 | 96 | 100 |
| Median total daily insulin dose (IQR), Ua | ||||||||||||
| Week before change | 50 (35–67) | — | 62 (52–75) | 40 (32–56) | — | — | 43 (34–56) | 52 (35–70) | 55 (45–65) | — | — | — |
| Week after change | 49 (37–67) | — | 63 (53–75) | 40 (32–54) | — | — | 42 (35–56) | 54 (37–71) | 50 (41–65) | — | — | — |
| Median total daily basal dose (IQR), U | ||||||||||||
| Week before change | 24 (17–31) | — | 30 (22–35) | 21 (14–27) | — | — | 21 (15–27) | 25 (16–32) | 30 (22–35) | — | — | — |
| Week after change | 24 (17–31) | — | 30 (23–36) | 21 (14–26) | — | — | 21 (16–26) | 25 (17–32) | 29 (20–34) | — | — | — |
| Median CGM % TIR 70–180 mg/dL (IQR) | ||||||||||||
| Week before change | 71.2 (61.8–79.4) | 61.0 (49.6–69.8) | 64.8 (56.3–71.7) | 76.0 (66.0–82.1) | 56.0 (46.2–66.3) | 63.5 (52.1–70.7 | 78.9 (74.3–84.5) | 69.9 (62.0–76.6) | 51.2 (43.7–63.8) | 69.8 (64.5–76.1) | 55.0 (46.2–65.2) | 48.2 (35.5–56.9) |
| Week after change | 71.3 (62.7–79.7 | 62.9 (50.9–70.9) | 66.5 (58.2–72.2) | 75.7 (67.8–83.9) | 58.7 (46.2–67.7) | 66.0 (52.9–72.8) | 79.8 (72.5–85.4) | 69.9 (62.6–76.8) | 57.7 (51.0–66.0) | 72.0 (65.6–78.3) | 56.8 (46.5–67.1) | 50.1 (44.3–59.4) |
Insulin doses reported in this table are delivered doses based on pump downloads and are only available for the CLC group.
AYA, adolescents and young adult; BR, basal rates; CGM; continuous glucose monitor; CLC, closed-loop control; CF, correction factors; CR, carbohydrate ratios; IQR, interquartile range; SAP, sensor-augmented pump; TIR, time in range.
Changes were made in 52% of CLC and 59% of SAP participants at 2 weeks and in 71% of CLC and 71% of SAP participants at 13 weeks. There were 1.6 ± 1.4 changes per CLC participant and 2.0 ± 1.3 changes per SAP participant in the first 3 months of the study. In the last 3 months of the study, there were 1.8 ± 1.1 changes per CLC participant and 2.0 ± 1.2 changes per SAP participant.
Regarding types of parameter changes, 43% of changes affected only one type of parameter, 36% affected two parameters, and 20% involved changes to all the three parameters. BR were changed in 69% of changes made in the CLC group and 80% of changes made in the SAP group. In the CLC group, changes involved CR in 68% and CF in 50% of changes; SAP changes involved CR in 44% and CF in 41%. Overnight changes (basal or bolus) were made in 62% of CLC changes and 75% of SAP changes. Insulin doses were neither systematically increased nor decreased.
When assessing impact of parameter changes on glycemic control, the median overall TIR for the week before any parameter changes was similar to the week after any parameter changes in the CLC group (71.2% vs. 71.3%) and in the SAP group (61.0% vs. 62.9%) (Tables 1 and 2). This indicates that in the overall cohort, the parameters did not overall improve glycemic control in the immediate weeks following.
Table 2.
Glycemic Measures Before and After Parameter Changes
| Study group |
CLC (n = 112) |
SAP (n = 56) |
||
|---|---|---|---|---|
| Period | Week before change | Week after change | Week before change | Week after change |
| LBGI | 0.4 (0.2–0.7) | 0.4 (0.2–0.7) | 0.5 (0.3–0.9) | 0.6 (0.3–0.9) |
| HBGI | 6.0 (4.1–8.2) | 5.8 (4.0–7.9) | 8.1 (5.8–11.0) | 7.4 (5.3–10.9) |
| CGM % time | ||||
| <54 mg/dL | 0.1 (0.0–0.4) | 0.1 (0.0–0.4) | 0.2 (0.0–0.5) | 0.1 (0.0–0.5) |
| <70 mg/dL | 1.0 (0.4–2.3) | 1.0 (0.4–2.2) | 1.7 (0.7–3.2) | 1.9 (0.7–3.4) |
| 70–180 mg/dL | 71.2 (61.8–79.4) | 71.3 (62.7–79.7) | 61.0 (49.6–69.8) | 62.9 (50.9–70.9) |
| >180 mg/dL | 26.7 (19.2–36.4) | 27.1 (19.2–35.8) | 36.9 (27.1–49.2) | 33.9 (26.0–46.5) |
| >300 mg/dL | 1.0 (0.0–3.4) | 0.8 (0.0–2.7) | 2.1 (0.7–5.4) | 1.8 (0.3–4.8) |
Data are expressed in median (IQR).
HBGI, high blood glucose index; LBGI, low blood glucose index.
During CLC use, we assessed the impact of changing CR on the TIR 4 h after a meal bolus that did not include any associated correction insulin. The median TIR for the 4 h after a meal bolus the week before making a CR more aggressive was 54.2%, and the TIR for the week after such change was 62.1%. The median time >180 mg/dL was 43.1% before and 36.0% after the change. The TIR before making a CR less aggressive was 65.5%, and the TIR after the change was 64.6%. The median time <70 mg/dL before making a CR less aggressive was 1.8%, and after the change was 0.7%.
In addition, we analyzed the impact of a change in CF in the 4 h after a correction bolus without any associated carbohydrates. The median TIR for the 4 h after a correction bolus was 49.6% the week before making a CF more aggressive and 55.5% during the week after the change. The median time >180 mg/dL was 49.8% before and 44.5% after the change. The median TIR before making a CF less aggressive was 52.1% and the TIR after the change was 55.7%. The median time <70 mg/dL before making a CF less aggressive was 1.7%, and after the change was 0.8%.
Insulin delivery
In the CLC group, insulin delivery was assessed for the week before and week after each pump parameter change using downloaded pump data (Table 1). These pump data were not systematically collected from participants in the SAP group. The median TDI and total basal insulin were generally similar during both periods for the CLC group overall as well as for the age and HbA1c subgroups.
Use of exercise or sleep setting
The exercise setting was activated a median of 0.4 days/week for a median of 2.1 h (IQR 1–4.3). During the first 3 months of the study, it was used 0.5 days/week, and during the last 3 months of the study, it was used 0.2 days/week. TIR during exercise was high with a median TIR of 93.1% (IQR: 67.6%–100.0%). The mean percent time <70 mg/dL was 3.4% ± 10.7%.
The majority of participants preset the sleep setting 6–7 days/week (78%, 6–7 days/week; 21%, 1–5 days/week; median length 8 h; one participant with no use of sleep setting). The median TIR during sleep mode was 85.8% (IQR 61.8%–100.0%). The mean percent time <70 mg/dL was 1.39% ± 4.82%.
System alerts and alarms
System alarms occurred for events of higher priority related to insulin delivery (e.g., occlusion alarm, manual suspension of insulin) compared with system alerts that were issued (e.g., hyperglycemia alert). There was a median of 0.8 (IQR 0.5–1.6) alarms per participant per day. There was a median of 6.0 (IQR 4.7–7.3) alerts per participant per day with the majority of alerts being Control-IQ high glucose alert (50%) and Control-IQ low glucose (26%).
Adolescents and young adults
The AYA group using CLC had an average of 3.7 ± 1.8 changes, and the adult group using CLC had 3.2 ± 2.3. In SAP, AYA had 4.0 ± 1.5 changes and adults had 4.2 ± 2.5 (Table 1). There were 36 automated corrections for AYA in the week before the change and 40 in the week after the change; in adults, there were 27 in the week before the change and 31 in the week after the change. AYA used the exercise setting for a median of 0.3 days/week for 3.4 h, and adults used it for a median of 0.5 days/week for 1.8 h. The AYA group had 1.0 alarms/day and 7.5 alerts/day; the adult group had 0.7 alarms/day and 5.3 alerts/day.
Baseline HbA1c
When comparing HbA1c groups, the lowest HbA1c group had the fewest parameter changes (Table 1). The highest HbA1c group had the most pronounced change in BR and CF, especially in the SAP group (Table 1). Notably, the highest HbA1c group had TIR of 51.2% versus 57.7% in the CLC group and 48.2% versus 50.1% in the SAP group during the weeks before and after parameter changes, respectively. Automated corrections based on the HbA1c group were: lowest (21 and 24), mid-range (30 and 34), highest (51 and 57) in the week before and after the change, respectively. There were 0.7 alarms/day in the lowest HbA1c group, 0.9 in the mid-range, and 1.4 in the highest.
Recommendations
Table 3 summarizes recommendations for Control-IQ initiation and use based on this subanalysis and our expert opinion using the CARES paradigm: how the system Calculates insulin delivery, what parameters should be Adjusted, when to Revert to standard therapy, how to Educate users, and Sensor characteristics.11
Table 3.
Recommendations for Control-IQ Based on This Subanalysis and Our Expert Opinion Using the CARES Paradigm11
| Calculate | • Control-IQ modulates programmed basal insulin delivery to target glucose levels 112.5–160 mg/dL. |
| • Delivers hourly automatic correction doses for predicted glucose >180 mg/dL in certain circumstances. | |
| Adjust | • Users/health care providers can adjust BR, CR, and CF. |
| • Users can activate exercise, which targets 140–160 mg/dL, and sleep, which targets 112.5–120 mg/dL. | |
| Revert | • During the pivotal trial, users were asked to revert to open loop for 4 h if they gave themselves a manual injection outside the pump for suspected infusion occlusion. When users revert to open loop, they are able to use temporary BR. |
| Educate | • Set the user expectation that Control-IQ should be used like a traditional SAP in terms of user-initiated boluses; however, the system will be working to increase TIR by adjusting automated doses. Optimizing CGM use will optimize time that Control-IQ is active. |
| • Bolus setting changes (CR and CF) may be more useful in Control-IQ than basal setting changes and were changed in this trial more than in the SAP group. | |
| • Delivering meal boluses before eating will likely improve postprandial hyperglycemia. | |
| • Setting a sleep mode every night will maintain increased TIR overnight. | |
| • Although the system provides automated corrections, manual corrections before meals and when hyperglycemia is prolonged are still advised. | |
| • Some participants reported they ingested a smaller amount of carbohydrates to prevent or treat hypoglycemia during use of system. This is likely due to lower insulin-on-board following decreased automated insulin delivery in the setting of hypoglycemia. | |
| Sensor | • Control-IQ uses the DexCom G6 sensor, which means the CGM values can be used for insulin bolus calculations. |
| • The participants were able to stay in CLC 92% of the time. |
Discussion
This is the first report of the clinical use of the Control-IQ system, and this analysis highlights important considerations when comparing use of CLC compared with SAP. Participants using the CLC system received fewer basal dose adjustments than SAP users, but more adjustments to both CR and CF. TIR was similar before and after parameter changes in the overall cohort; however, the highest HbA1c subgroup among CLC participants showed a trend toward increased TIR, indicating parameter changes may matter more in those with the most suboptimal glucose control. Total daily and total daily basal insulin delivery did not increase after the parameter changes in this cohort, so there may be another explanation for the change in glycemic control.
Despite TIR not changing overall, there appear to be some differences when analyzing the 4-h time period after the CR or CF parameter changes. In this analysis, there is a suggestion that the glycemic profiles are improved in that window of time, such that making a CR more aggressive results in greater postprandial TIR, whereas making a CR less aggressive may result in lower postprandial hypoglycemia. The CF changes are not as clear, but it is possible a less aggressive CF may lower post-correction hypoglycemia.
One of the reasons that the parameter changes in the overall CLC cohort may not have improved glycemic control is the robustness of the CLC algorithm. Whereas SAP therapy uses the same BR regardless of glucose levels, with the CLC system BR delivery can vary by 0–4 times the programmed BR. Changes, especially to BR, were likely much smaller than the four times changes allowed by the algorithm. In the highest HbA1c group, however, changes may have more of an impact. This could be due to also having the most pronounced change in parameters BR and CF. Users who have better optimized settings before initiating CLC may need fewer parameter changes compared with those with higher HbA1c who may need more changes.
These findings are in line with a recent analysis of two different methods of initializing the Control-IQ system in adolescents who showed similar TIR overall with a suggestion that suboptimal baseline parameter characteristics have some influence on outcomes.12 In that study, adolescents with higher bolus insulin to TDI ratios at baseline achieved greater glycemic target range improvements than those with lower ratios when given a systematic approach to initialization.
The two FDA-approved systems in use are hybrid CLC systems that do not automate meal boluses and therefore changes in CR may make a meaningful difference in insulin delivery. Messer et al. previously reported that AYA using the MiniMed 670G system frequently strengthened CR during the MiniMed 670G pivotal trial,7 and this was also observed in this pivotal trial of Control-IQ. The Control-IQ system may additionally mitigate hyperglycemia after meals with automated correction boluses, which are not available in MiniMed 670G.
With both CLC systems, however, CR that are too aggressive may still cause post-bolus hypoglycemia even with automated suspension of insulin delivery. For this reason, automated insulin delivery system can more easily compensate for meal boluses that are under-bolused (either due to inaccurate carbohydrate counting estimation or suboptimal CR) rather than those that are over-bolused. Even in the extremes of missed boluses, the system can partially compensate for missed meal boluses.13
CF (also known as insulin sensitivity) were also frequently adjusted in this trial. This parameter change directly influences insulin delivery with the Control-IQ system, as it will influence insulin delivery for both user-initiated boluses and automatic correction doses. CF changes would not influence insulin delivery in the MiniMed 670G system, so this is an important distinction between the two systems.
Control-IQ users can enable the exercise and sleep settings to change target blood glucose. Study participants were instructed to use the exercise setting in a similar manner to how they would have used temporary BR with SAP before and/or after exercise. There was a decrease in use of the exercise setting across the study, but it cannot be determined by this analysis if this was due to increased trust in the automated insulin delivery system to prevent hypoglycemia or user dissatisfaction with the performance of the exercise setting. TIR while using the exercise setting was high.
Overall, clinicians should understand the key principles of automated insulin delivery for CLC systems. This can be conceptualized through the CARES paradigm.11 This subanalysis of one particular CLC system and our expert opinion is summarized in Table 3.
The strengths of this subanalysis are the large participant sample and longitudinal data related to clinical use of the Control-IQ system. This analysis was limited in that it is a secondary analysis, which was not powered to find significant differences in number and type of parameter changes. This applies in particular to the highest HbA1c group of SAP participants; more study of this high-risk group is needed. Instead, we aimed to describe the changes made by clinicians and participants during the trial to contextualize outcomes for real-world use. We previously reported that overall satisfaction with the system was high during the study.4
Conclusions
Parameter changes likely have a minimal effect on overall TIR when using Control-IQ, but they may have an effect in higher HbA1c subgroups or following user-directed boluses, suggesting that changes may matter more in suboptimal control or during discrete periods of the day. Appropriate expectation setting, training, and optimization of the system are necessary to transition of these findings to real-world use. Understanding how the Control-IQ system works and how parameter adjustments are different from SAP can give clinicians meaningful guidance for how to optimize glycemic outcomes for people with type 1 diabetes.
Acknowledgments
iDCL Trial Research Group: University of Virginia, Center for Diabetes Technology, Charlottesville, VA: S.A.B. (PI), Boris Kovatchev (Grant PI), Stacey Anderson (I), Emma Emory, Mary Voelmle, Katie Conshafter, Kim Morris, Mary Oliveri, Linda Gondor-Fredrick, Harry Mitchell, Kayla Calvo, Christian Wakeman, Marc Breton. Joslin Diabetes Center, Harvard Medical School, Boston, MA: Lori Laffel (PI), E.I. (I), Louise Ambler-Osborn (I), Emily Flint, Kenny Kim, Lindsay Roethke. Sansum Diabetes Research Institute, Santa Barbara, CA: J.E.P. (PI), Mei Mei Church (I), Camille Andre, Molly Piper. Division of Endocrinology, Diabetes, Icahn School of Medicine at Mount Sinai, New York City, NY: C.J.L. (PI), David Lam (I), G.O. (I), Camilla Levister (I), Selassie Ogyaadu, Jessica Lovett. Division of Endocrinology, Diabetes, Metabolism, and Nutrition, Department of Internal Medicine, Mayo Clinic, Rochester MN: Y.C.K. (PI), Vinaya Simha (I), Vikash Dadlani, Shelly McCrady-Spitzer, Corey Reid, Kanchan Kumari. Barbara Davis Center for Diabetes, University of Colorado, Anschutz Medical Campus, Aurora, CO: R. Paul Wadwa (PI), Gregory P. Forlenza (I), G. Todd Alonso (I), Robert Slover (I), Emily Jost, L.H.M., Cari Berget, Lindsey Towers, Alex Rossick-Solis. Department of Pediatrics, Division of Pediatric Endocrinology and Diabetes, Stanford University School of Medicine: Bruce Buckingham (PI), L.E. (I), Tali Jacobson, Marissa Town, Ideen Tabatabai, Jordan Keller, Evalina Salas. John A. Paulson School of Engineering and Applied Sciences, Harvard University, Cambridge, MA: Francis Doyle III, Eyal Dassau. Jaeb Center for Health Research: J.L., Roy Beck, Samantha Passman, Tiffany Campos, D.R., Craig Kollman, Carlos Murphy, Nandan Patibandla, Sarah Borgman. National Institute of Diabetes, Digestive, and Kidney Diseases (NIDDK): Guillermo Arreza-Rubin (Project Scientist), Thomas Eggerman (Program Officer), Neal Green. iDCL Steering Committee Members: Boris Kovatchev, S.A.B., Stacey Anderson, Marc Breton, Lori Laffel, J.E.P., C.J.L., Y.C.K., R. Paul Wadwa, Bruce Buckingham, Francis Doyle III, Eric Renard, Claudio Cobelli, Yves Reznik, Guillermo Arreza-Rubin, J.L., Roy Beck.
Contributor Information
Collaborators: for the iDCL Trial Research Group, Sue A. Brown, Boris Kovatchev, Stacey Anderson, Emma Emory, Mary Voelmle, Katie Conshafter, Kim Morris, Mary Oliveri, Linda Gondor-Fredrick, Harry Mitchell, Kayla Calvo, Christian Wakeman, Marc Breton, Lori Laffel, Elvira Isganaitis, Louise Ambler-Osborn, Emily Flint, Kenny Kim, Lindsay Roethke, Jordan E. Pinsker, Mei Mei Church, Camille Andre, Molly Piper, Carol J. Levy, David Lam, Grenye O'Malley, Camilla Levister, Selassie Ogyaadu, Jessica Lovett, Yogish C. Kudva, Vinaya Simha, Vikash Dadlani, Shelly McCrady-Spitzer, Corey Reid, Kanchan Kumari, R. Paul Wadwa, Gregory P. Forlenza, G. Todd Alonso, Robert Slover, Emily Jost, Laurel H. Messer, Cari Berget, Lindsey Towers, Alex Rossick-Solis, Bruce Buckingham, Laya Ekhlaspour, Tali Jacobson, Marissa Town, Ideen Tabatabai, Jordan Keller, Evalina Salas, John A. Paulson, J.L. Roy Beck, Samantha Passman, Tiffany Campos, D.R. Craig Kollman, Carlos Murphy, Nandan Patibandla, Sarah Borgman, Guillermo Arreza-Rubin, Thomas Eggerman, Neal Green, Boris Kovatchev, Sue A. Brown, Stacey Anderson, Marc Breton, Lori Laffel, Jordan E. Pinsker, Carol J. Levy, Yogish C. Kudva, R. Paul Wadwa, Bruce Buckingham, Eric Renard, Claudio Cobelli, Yves Reznik, Guillermo Arreza-Rubin, and J.L. Roy Beck.
Authors' Contributions
G.O., D.R., J.L., L.H.M., and S.A.B. developed the concept for the article. D.R. analyzed the data. G.O. and S.A.B. wrote the article. All the authors are responsible for reviewing and revising this article and assume responsibility and accountability for the results.
Author Disclosure Statement
G.O. receives research support from Tandem Diabetes, DexCom, and Abbot. L.H.M. has received speaking/consulting honoraria from Tandem Diabetes and DexCom, Inc., and Capillary Biomedical; her institution receives research grants from Medtronic, Tandem Diabetes, DexCom, Beta Bionics, Abbott, and Insulet Corp. C.J.L. has received research support from Insulet, Abbott Diabetes, Tandem Diabetes, and Dexcom and has received consulting fees from Dexcom. J.E.P. reports receiving grant support, provided to his institution, and consulting fees and speaker fees from Tandem Diabetes Care; grant support, provided to his institution, and advisory board fees from Medtronic; grant support, provided to his institution, and consulting fees from Eli Lilly; grant support and supplies, provided to his institution, from Insulet; and supplies, provided to his institution, from Dexcom. G.P.F. conducts research sponsored by Medtronic, Dexcom, Abbott, Insulet, Tandem, Lilly, and Beta Bionics; he has been a consultant/speaker for Medtronic, Dexcom, Abbott, Insulet, Tandem, Lilly, and Beta Bionics. E.I. reports no disclosures. Y.C.K. receives research support from Tandem Diabetes, Dexcom, and Roche Diabetes. L.E. reports receiving consultancy fees from Tandem Diabetes Care and Ypsomed. D.R. and J.L. report no disclosures. S.A.B. reports nonfinancial support from Tandem Diabetes Care, nonfinancial support from Dexcom, nonfinancial support from Roche Diagnostics, grants from the National Institute of Health, during the conduct of the study; grants and nonfinancial support from Tandem Diabetes Care, nonfinancial support from Dexcom, nonfinancial support from Roche Diagnostics, grants from Insulet, grants from Tolerion, outside the submitted work.
Funding Information
This study was funded by a grant from the NIDDK to UVA (UC4 108483). Tandem Diabetes Care provided the experimental CLC systems used in the trial, system-related supplies, and technical expertise. Tandem Diabetes Care was not involved in data analysis and was provided a copy of the article for review before publication.
References
- 1. Bekiari E, Kitsios K, Thabit H, et al. : Artificial pancreas treatment for outpatients with type 1 diabetes: systematic review and meta-analysis. BMJ 2018;361:k1310. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2. Karageorgiou V, Papaioannou TG, Bellos I, et al. : Effectiveness of artificial pancreas in the non-adult population: a systematic review and network meta-analysis. Metabolism 2019;90:20–30 [DOI] [PubMed] [Google Scholar]
- 3. Weisman A, Bai J-W, Cardinez M, et al. : Effect of artificial pancreas systems on glycaemic control in patients with type 1 diabetes: a systematic review and meta-analysis of outpatient randomised controlled trials. Lancet Diabetes Endocrinol 2017;5:501–512 [DOI] [PubMed] [Google Scholar]
- 4. Brown SA, Kovatchev BP, Raghinaru D, et al. : Six-month randomized, multicenter trial of closed-loop control in type 1 diabetes. N Engl J Med 2019;381:1707–1717 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5. Garg SK, Weinzimer SA, Tamborlane WV, et al. : Glucose outcomes with the in-home use of a hybrid closed-loop insulin delivery system in adolescents and adults with type 1 diabetes. Diabetes Technol Ther 2017;19:155–163 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6. Stone MP, Agrawal P, Chen X, et al. : Retrospective analysis of 3-month real-world glucose data after the MiniMed 670G system commercial launch. Diabetes Technol Ther 2018;20:689–692 [DOI] [PubMed] [Google Scholar]
- 7. Messer LH, Forlenza GP, Sherr JL, et al. : Optimizing hybrid closed-loop therapy in adolescents and emerging adults using the MiniMed 670G system. Diabetes Care 2018;41:789–796 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8. Lal RA, Basina M, Maahs DM, et al. : One year clinical experience of the first commercial hybrid closed-loop system. Diabetes Care 2019;42:2190–2196 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9. Berget C, Messer LH, Vigers T, et al. : Six months of hybrid closed loop in the real-world: an evaluation of children and young adults using the 670G system. Pediatr Diabetes 2020;21:310–318 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10. Messer LH, Berget C, Vigers T, et al. : Real world hybrid closed-loop discontinuation: predictors and perceptions of youth discontinuing the 670G system in the first 6 months. Pediatr Diabetes 2020;21:319–327 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11. Messer LH, Berget C, Forlenza GP: A clinical guide to advanced diabetes devices and closed-loop systems using the CARES paradigm. Diabetes Technol Ther 2019;21:462–469 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12. Schoelwer MJ, Robic JL, Gautier T, et al. : Safety and efficacy of initializing the control-IQ artificial pancreas system based on total daily insulin in adolescents with type 1 diabetes. Diabetes Technol Ther 2020;22:594–601 [DOI] [PubMed] [Google Scholar]
- 13. Cherñavvsky DR, DeBoer MD, Keith-Hynes P, et al. : Use of an artificial pancreas among adolescents for a missed snack bolus and an underestimated meal bolus. Pediatr Diabetes 2016;17:28–35 [DOI] [PubMed] [Google Scholar]