Improved Open-Loop Glucose Control With Basal Insulin Reduction 90 Minutes Before Aerobic Exercise in Patients With Type 1 Diabetes on Continuous Subcutaneous Insulin Infusion

In order to maintain blood glucose homeostasis during exercise, circulating insulin concentration normally decreases and glucagon level increases to promote hepatic glucose production to a level that matches glucose utilization (1). However, in individuals with type 1 diabetes, due to the absolute destruction of the insulin-producing β-cells in the pancreas, patients become dependent on exogenous insulin via multiple daily injections or continuous subcutaneous insulin infusion (CSII) (2). CSII therapy uses only short-acting insulin for both continuous basal insulin infusion (regulates glucose levels in between meals) and bolus insulin infusion (often taken before a meal or to correct hyperglycemia) (3). In this patient population, specifically during aerobic exercise, insulin levels tend not to drop fast enough, even if insulin delivery is suspended on their pump, and there may be an attenuated glucagon response (4). In fact, individuals living with type 1 diabetes can have increased circulating exogenous insulin levels during exercise, either when subcutaneous basal insulin infusion rates are held constant (5) or even if a temporary basal rate reduction (BRR) is set 1 h before exercise (6).

For active individuals with type 1 diabetes, common strategies to help reduce the risk of hypoglycemia include increasing carbohydrate ingestion before or during exercise (7,8), reducing the prandial dose before exercise (9,10), and/or performing a BRR at some point before exercise onset (6,11,12). Exercise guidelines for open-loop insulin delivery (4,13–15) suggest BRRs at least 60–90 min before the onset of exercise, with little evidence to substantiate these recommendations. In reality, patients often disconnect or suspend insulin infusion at exercise start and resume basal insulin immediately postexercise (16). This may be in part due to forgetting to set a BRR pre-exercise, participating in unplanned exercise, or trying to prevent postexercise rebound hyperglycemia. Pump suspension at exercise onset appears to be effective in limiting hypoglycemia during circuit-based exercise (17), but this strategy may not be that effective in eliminating the drop in glycemia during aerobic exercise (17,18). The DirecNet study group showed that in a pediatric population (n = 49), insulin pump suspension (PS) at the start of a 75-min aerobic exercise bout attenuated the drop in glycemia compared with no PS, but it did not completely eliminate hypoglycemia risk and promoted increased hyperglycemia risk postexercise (19). Franc et al. (11) proposed that without additional carbohydrates, the most effective strategy to reduce hypoglycemia for 30 min of spontaneous exercise is an 80% BRR (i.e., 20% of the usual basal rate) implemented at exercise start time throughout the activity and for 2 h postexercise, although a significant drop in glycemia should still be expected. Thus, PS at exercise onset may attenuate the drop in glycemia, but does not sufficiently protect against hypoglycemia during aerobic exercise. Also, leaving the pump suspended or disconnected in recovery in many cases appears to contribute to postexercise hyperglycemia (11).

The study objective was to determine which BRR strategy best attenuates the drop in glycemia during prolonged aerobic exercise in patients with type 1 diabetes using CSII. The change in glucose was examined for the exercise period and for 2 h following a standardized meal postexercise. This study is referred to as the Omnipod Type 1 diabetes Insulin Management for Exercise (OmniTIME) Study.

The experimental protocol conformed to the standards set by the Declaration of Helsinki and was approved by the Research Ethics Board at York University (Toronto, Ontario, Canada). The study was registered at ClinicalTrials.gov in 2017 (identifier NCT03130101). Enrollment criteria included participants 17–65 years of age, duration of diabetes >1 year, on CSII >1 month (total daily dose of at least 0.25 units/kg), and last HbA_1c ≤9.9% or 85 mmol/mol. All participants were using the Omnipod Insulin Management System (Insulet Corporation, Billerica, MA) and either insulin lispro (Humalog; Eli Lilly and Company, Indianapolis, IN) (n = 5) or insulin aspart (NovoRapid; Novo Nordisk, Bagsværd, Denmark) (n = 12). Exclusion criteria included frequent and unpredictable hypoglycemia, inability to exercise regularly due to an injury, having conditions that may make exercise unsafe (i.e., high blood pressure, late pregnancy, etc.), and/or physician diagnosis of active diabetic retinopathy or neuropathy. Written informed consent was obtained from all participants before study initiation.

Anthropometric measurements, including height, body mass, blood pressure, and body fat percentage, were completed during the initial assessment. Participants were asked to complete a Physical Activity Readiness Questionnaire for Everyone to screen for any cardiometabolic complications (20). Once cleared to exercise, participants completed the International Physical Activity Questionnaire - Short Form based on self-reported activity. Participants were also asked to complete questionnaires by Gold et al. (21) and Clarke et al. (22) to categorize hypoglycemia awareness. Using an incremental-to-maximum effort treadmill protocol, peak oxygen consumption (VO_2peak) and peak heart rate were measured with a portable metabolic unit (K5; COSMED, Rome, Italy) and heart rate monitor (Polar Electro, Kempele, Finland).

Participants wore a continuous glucose monitoring (CGM) device (G4 or G5; Dexcom, San Diego, CA) on the abdomen or arm according to the manufacturer’s instructions and were instructed to use their Personal Diabetes Manager (PDM) (Omnipod; Insulet Corporation) for self-monitoring of blood glucose (SMBG) with the associated blood glucose test strips (FreeStyle Lite; Abbott Laboratories, Chicago, IL) and for CGM calibrations whenever necessary throughout the study.

Participants were asked to avoid alcohol and caffeine consumption and refrain from all forms of vigorous exercise (i.e., activities of more than six metabolic equivalents) for 24 h prior to each visit. Female participants were not tested during the luteal phase because research suggests that estrogen levels are higher particularly during this phase of the menstrual cycle, and this may affect fuel selection by increasing the rate of fat utilization (23). Participants were asked to refrain from consuming additional food or drink following their lunchtime meal, unless hypoglycemia (<70 mg/dL) developed. Participants completed the following BRR strategies in a randomized crossover fashion: 1) an 80% BRR (i.e., 20% of basal insulin), set 90 min in advance of exercise for the duration of activity; 2) a 50% BRR (i.e., 50% of basal insulin), set 90 min in advance of exercise for the duration of activity; and 3) PS (i.e., 0% of basal insulin) for the duration of activity, starting at the onset of exercise. Based on the recent exercise consensus statement (4), blood glucose targets for exercise were between 70 and 360 mg/dL. Using SMBG, participants were asked to monitor their glucose levels 90 min before exercise. If blood glucose level was between 70 and 90 mg/dL, 8 g of dextrose was taken to achieve glycemic target by the onset of exercise. If blood glucose was >270 mg/dL, blood ketones were tested (FreeStyle Precision Neo; Abbott Laboratories), and exercise could proceed if β-hydroxybutyrate levels were <0.6 mmol/L. Each participant was contacted at ∼1:30 p.m. as a reminder to set the appropriate BRR, depending on the visit. Exercise began at 3:00 p.m. for all conditions. Participants had little to no prandial insulin at exercise onset, and this was confirmed by examining their pump history to ensure that no additional insulin was given following the lunchtime bolus. The experimental visits consisted of 60 min of treadmill walking/light jogging at 45–55% of the participant’s predetermined VO_2peak, and all visits were separated by at least 24 h. The exercise was divided into four 15-min bouts with 5-min rest periods in between, similar to previous research (24). Substrate oxidation rates were calculated using the expired gas collection during the last 5 min of the first and last bouts of exercise (25).

Capillary glucose was measured 30 min and 10 min before exercise, just before exercise start (time = 0), and every 15 min throughout exercise using a standardized laboratory glucose meter (Contour Next Link; Ascensia Diabetes Care, Parsippany, NJ). This blood glucose monitoring device has a high level of accuracy as compared with a laboratory standard (26), and we regularly performed quality control tests using the control solution provided by the manufacturer. All measurements were completed in duplicate, and in some cases in triplicate, if the duplicate values differed by >10 mg/dL. If a participant developed hypoglycemia (<70 mg/dL) based on SMBG, they were instructed to stop exercising (if it was during exercise), and 16 g of oral dextrose (Dex4; A.M.G. Medical Inc., Montreal, Quebec, Canada) was provided. If the initial 16 g failed to raise glucose level sufficiently, additional dextrose was provided as necessary to restore blood glucose level. Participants returned to the treadmill to complete the exercise portion only once blood glucose concentration reached >81 mg/dL. For each participant, the change in glucose was calculated based on the participants’ glucose at the start of exercise and the last value measured at the end of exercise, according to the following:

Note: if exercise was stopped early because of hypoglycemia, then the last exercise glucose value was used for analysis (27).

In all conditions, basal insulin was resumed to the usual rate immediately postexercise. Participants rested for 30 min and then consumed a standardized meal. Meals were consistent within each participant; however, they differed between participants depending on food restrictions. Meals consisted of ∼30–50 g of carbohydrates, ∼10–20 g of protein, and ∼5–15 g of fat (Lean Cuisine; Nestlé, Glendale, CA). The amount of prandial insulin administered at the postexercise meal was based on the carbohydrate content of the meal and the patient’s own individualized insulin-to-carbohydrate ratio, insulin sensitivity index, and glycemic targets preprogrammed into their PDM. Prandial insulin was reduced by 25% of the total dose to account for the increase in insulin sensitivity postexercise (28). If an insulin correction was suggested based on the patient’s PDM settings (i.e., usual care), the full correction dose was given. Bolus insulin was administered ∼10 min before meal consumption. Following the meal, SMBG was determined every 30 min for a total of 90 min before participants were sent home.

All participants consumed a standardized snack (20 g carbohydrates, 3 g fiber, and 10 g protein) (Glucerna bar; Abbott Laboratories) before bedtime, with or without bolus insulin as per usual practice, and set a 20% BRR for 6 h to reduce the likelihood of postexercise nocturnal hypoglycemia.

Plasma samples were collected just before exercise, 30 min into exercise, immediately before meal consumption, and 90 min postmeal for circulating free insulin concentrations (Insulin ELISA; Crystal Chem, Elk Grove Village, IL) as per the manufacturer’s recommendations. To remove antibody-bound insulin, polyethylene glycol 6000 (BioShop, Burlington, Ontario, Canada) was used to create a precipitate (modified from Nakagawa et al. [29]). A total of 50 μL of 25% aqueous polyethylene glycol 6000 and phosphate buffer solution (pH 7.4) was added to 50 μL of each sample and lightly spun with a mixer (Vortex-Genie 2; Scientific Industries, Inc., Bohemia, NY) for 10 s. Samples were then centrifuged at 10,000g for 5 min, and the supernatant was extracted and analyzed for circulating free insulin concentration.

Statistical significance was set at P < 0.05 unless otherwise indicated, and a Tukey post hoc test was used if significance was found. All statistical analyses were conducted using GraphPad Prism Version 7.0 (GraphPad Software, La Jolla, CA). The changes in glucose concentration, carbohydrate and fat oxidation rates, and respiratory exchange ratio from the first to the last bout of exercise were compared using two-way repeated-measures ANOVA (condition by time), after confirming normality using a residuals analysis. A one-way repeated-measures ANOVA was used to compare each of the following: baseline glucose, change in blood glucose from start to end of exercise, nadir glucose, decremental area under the curve (AUC), and mealtime bolus insulin across all three conditions. Data are reported as mean (SD) and median (interquartile range [IQR]). Differences in the percent time spent in euglycemia (70–180 mg/dL), hyperglycemia (>180 mg/dL), and hypoglycemia (<70 mg/dL) were compared using a Friedman test.

A total of 17 participants (13 females) were recruited for this study. Participants were all adults (mean ± SD age 31 ± 10 years [median 28 years; IQR 25–35]), with a BMI of 25.3 ± 2.5 kg/m² and HbA_1c of 6.5 ± 0.5% (47 ± 5 mmol/mol). The total daily insulin dose was 31 ± 8 units (0.43 ± 0.1 units/kg), and diabetes duration was 14 ± 10 (10; 5–21) years. Based on two questionnaires used to assess hypoglycemia awareness (Gold score = 3 ± 1 and Clarke score), 5 participants had reduced hypoglycemia awareness, and 12 were hypoglycemia aware. Based on the International Physical Activity Questionnaire - Short Form, participants were moderate to highly active (metabolic equivalent of task-minutes per week = 2,805 ± 1,365 [2,782; 1,937–3,746]); as such, the participant group could be categorized as active, but not highly fit based on BMI or VO_2peak (41.6 ± 5.9 mL/kg/min).

The cardiometabolic and glycemic outcome variables are shown in Table 1. The relative exercise intensity across all three experimental conditions was 50.3 ± 3.3% of VO_2peak, and heart rate was 123 ± 12 beats per min. The respiratory exchange ratio values did not differ significantly across the three conditions (P > 0.05). In the 80% and 50% BRR arms, the carbohydrate oxidation rate tended to decrease from the start to end of exercise (both P = 0.06), but remained constant during PS (P = 0.45). Fat oxidation rate did not change from the start to end of exercise across any of the conditions (P > 0.05).

Figure 1A represents the absolute glucose concentrations from pre-exercise until the end of the meal challenge across all three conditions. Prior to exercise start (minus 10 min), blood glucose was higher in 50% BRR compared with 80% BRR (P < 0.001). Blood glucose at exercise onset was also significantly higher with 50% BRR compared with the other two conditions (Fig. 1A and Table 1) (both P < 0.001). From 10 min pre-exercise until the end of the meal challenge, glucose concentration was higher with 50% BRR compared with PS (all P < 0.001). From 30 min into exercise until 90 min into the meal challenge, blood glucose was also higher with 80% BRR versus PS (P < 0.01). During the meal challenge, blood glucose concentration was lower with PS as compared with the two other arms except for the last time point (all P < 0.01) (Fig. 1A). In addition, a greater rise in glycemia occurred with PS as compared with the other two other arms (P < 0.001). Blood glucose level also rose at mealtime with PS and 50% BRR (both P < 0.001), but failed to rise significantly with 80% BRR (P = 0.16). The mean ± SD mealtime bolus insulin was 2.9 ± 1.1 (median 2.6; IQR 2.2–3.5) units with 80% BRR, 3.1 ± 1.2 (3.8; 1.9–3.9) units with 50% BRR, and 2.6 ± 1.1 (2.3; 1.7–3.5) units with PS (not significantly different) (Table 1).

Figure 1B shows the change in blood glucose concentrations from 30 min pre-exercise to 90 min postmeal challenge, normalized to starting exercise glucose level. There was a significant trial by time interaction for the change in blood glucose during exercise (P < 0.001). More specifically, by the end of exercise, the smallest drop in glycemia was −31 ± 58 mg/dL (median −32; IQR −74 to 11) with 80% BRR, as compared with −47 ± 50 mg/dL (−45; −88 to −14) with 50% BRR (P = 0.04) and −67 ± 41 mg/dL (−65; −90 to −38) with PS (P < 0.001). The decremental AUC was greater with PS as compared with 80% BRR (P = 0.007). Also, nadir blood glucose during exercise was lower with PS as compared with 50% BRR (P = 0.02). The highest number of hypoglycemic events occurred with PS (n = 7 out of 17[41%]), as compared with 80% and 50% BRR (both n = 1 out of 17[6%]; P < 0.02, χ²). The average carbohydrate intake required to return blood glucose level to target range following a hypoglycemic event was 19 ± 6 g (16; 16–16), and the average time before resuming exercise was 19 ± 4 min (17; 16–21). The time to the first hypoglycemic event ranged between 29 and 51 min during exercise in the nine hypoglycemic events that occurred during the 51 total exercise sessions.

Figure 2 represents circulating free insulin concentrations across all conditions. At exercise start, circulating free insulin concentration was higher with PS compared with 80% BRR (P = 0.02) and 50% BRR (P = 0.04). Insulin concentration fell during exercise in all conditions (P = 0.01, main effect of time). However, there was no significant difference in insulin concentration among the three treatment arms by the end of the meal challenge. In examining the pre- and postmeal insulin levels specifically, we observed a rise in insulin concentrations in 82% of all cases (in all treatment conditions combined). However, we noted a drop in insulin levels in the other 18% of cases. The mean plasma insulin concentration premeal was 67 pmol/L, increasing to 94 pmol/L (i.e., 52% increase; P = 0.02).

Figure 3A represents interstitial glucose from 10:00 p.m. to 7:00 a.m. postexercise across all conditions. Interstitial glucose remained relatively stable overnight with no difference across any of the three conditions (P > 0.05). Interstitial glucose level overnight was 131 ± 12 (122; 97–152) mg/dL, 136 ± 10 (131; 103–160) mg/dL, and 140 ± 13 (135; 105–169) mg/dL with 80% BRR, 50% BRR, and PS conditions, respectively (P > 0.05). Figure 3B represents the mean percent time in euglycemia (70–180 mg/dL), hyperglycemia (>180 mg/dL), and hypoglycemia (<70 mg/dL). Percent time in euglycemia was 83% (100%; 66–100%), 83% (97%; 67–100%), and 78% (91%; 69–100%) with 80% BRR, 50% BRR, and PS arms, respectively (not significantly different). Percent time in hyperglycemia was similar across all treatment arms at 15% (0; 0–25%), 16% (0; 0–31%), and 17% (0; 0–18%) with 80% BRR, 50% BRR, and PS arms, respectively (not significantly different). Percent time in hypoglycemia was 2% (0; 0 to 0%), 1% (0; 0 to 1%), and 5% (0; 0 to 0%) with 80% BRR, 50% BRR, and PS (not significantly different).

Regular physical activity is at the cornerstone of care for people living with type 1 diabetes for a number of health and fitness reasons (4). However, maintaining reasonable glucose control for prolonged moderate-intensity aerobic activities remains a major challenge, even for patients on CSII therapy (16). If prolonged exercise is to be performed soon after a meal, then prandial insulin dose reductions by 25–75%, depending on the intensity and duration of the activity, helps to limit hypoglycemia (9). However, if the activity is ≥3 h after a meal with bolus insulin, then a BRR may be a more effective approach for those using CSII (4). Previous studies have shown that even with pump disconnect (17,18), or with a temporary BRR (by 50–80%) set up to 40 min before exercise (6,12), hypoglycemia remains a concern during aerobic exercise. Although recent guidelines recommend a BRR set 60–90 min pre-exercise for patients with type 1 diabetes on CSII (4,13–15), we know of only one study published to date (6) that has examined the efficacy of a 50% BRR set 60 min pre-exercise until 60 min postexercise. Although efficacy was demonstrated, the intensity of exercise (i.e., 65–70% of age-predicted heart rate maximum) may have explained why blood glucose levels did not drop markedly. Therefore, the aim of the current study was to determine the efficacy of three different BRR strategies (one for spontaneous exercise and two for preplanned exercise) on attenuating the drop in glycemia during and after prolonged aerobic exercise in adults with type 1 diabetes on CSII.

Roy-Fleming et al. (12) recently published that an 80% BRR up to 40 min pre-exercise is insufficient to reduce hypoglycemia during a 45-min bout of submaximal aerobic exercise. One of the main differences in our study was that the BRR was set earlier (i.e., 90 min pre-exercise) in an attempt to lower circulating insulin before exercise onset. We show that 50–80% BRR set 90 min before prolonged aerobic exercise better protects against hypoglycemia than PS at exercise onset, likely because this forecast strategy results in lower circulating insulin levels at the start of exercise as compared with PS at exercise onset. In the PS condition at exercise onset, 7 out of 17 (41%) participants developed hypoglycemia during exercise, whereas only 1 out of 17 (6%) developed hypoglycemia with both the 80% BRR and 50% BRR arms (P < 0.05). Pump suspension at exercise onset helps with attenuating the drop in blood glucose concentration as compared with leaving the pump at the usual basal rate (i.e., 100%) (19), but this approach is clearly not sufficient to eliminate hypoglycemia risk if patients initiate exercise with a normal or only slightly elevated blood glucose level, as seen in our study (i.e., 41% of subjects developed hypoglycemia with PS).

A more aggressive BRR set well in advance of exercise may eliminate the risk of hypoglycemia, but may also cause pre-exercise hyperglycemia. However, we found that 50–80% BRR did not lead to overt hyperglycemia before the onset of exercise, nor did it result in hyperglycemia in early recovery. For unknown reasons, 50% BRR resulted in slightly higher starting blood glucose level as compared with 80% BRR and PS. Moreover, from the start to end of the meal challenge, the greatest rise in glycemia was observed in PS and 50% BRR. PS causing the largest rise in glycemia is likely attributable to the 7 out of 17 participants who were treated for hypoglycemia during exercise. Thus, treating hypoglycemia during exercise with a carbohydrate snack (∼15–20 g), although essential for safety reasons, may compromise glucose control at the upcoming meal. Interestingly, the 80% and 50% BRR conditions did not cause overt hyperglycemia during the meal challenge. Moreover, overall glucose levels in late recovery were not different among all conditions, resulting in reasonable control overnight (Fig. 3A and B).

At exercise onset, lower circulating free insulin concentrations were apparent in the 80% and 50% BRR conditions as compared with PS. Insulin levels continued to decline across all conditions from the start to end of exercise, thereby demonstrating that all strategies lower circulating insulin levels during exercise. Previous literature has shown a modest exercise-related increase in circulating insulin, followed by an accelerated decline, as the exercise continues (6,11). However, we did not find this initial rise in insulin levels during the first 30 min of exercise, possibly because not enough time points were examined. Interestingly, insulin levels did not rise markedly at mealtime in our study (Fig. 2), although the majority of participants did have a small rise in insulin levels. The failure to observe a meal-related rise in circulating insulin levels may have been because of the prolonged reduction in basal insulin infusion before the meal (lasting up to 165 min) and the relatively small dose of insulin that was administered at mealtime.

Importantly, and for the first time, we show that a 50–80% BRR set 90 min before prolonged aerobic exercise does not cause overt hyperglycemia pre-exercise and also attenuates the drop in glycemia during aerobic exercise. However, we also acknowledge that no singular approach is likely to work for all individuals. Specifically, Supplementary Fig. 1 shows the individual and mean participant change in glycemia during exercise in each of the three BRR strategies. As can be observed, one participant developed hypoglycemia in all three conditions, and some developed a significant rise in glucose, whereas others had minimal excursions. Thus, BRR strategies attenuate the drop in glycemia during prolonged aerobic exercise, but may still need to be individualized for any given patient.

This study had a number of strengths, including randomized crossover design, standardization of the meal and snack following exercise, and the measurement of circulating free insulin. However, this study also had a number of limitations that need to be mentioned. There are multiple factors that may limit our ability to translate the present findings across a wider population with type 1 diabetes. For example, the participants enrolled in this study were primarily female, with good HbA_1c, and all participants were using Omnipod Insulin Management System (Insulet Corporation) CSII therapy. In addition, we did not have information on potential lipodystrophy and lipohypertrophy, and we did not do a run-in period to assess participants’ usual basal insulin rates prior to study enrollment. We also only examined exercise at one time of day (late afternoon) and with one type and intensity of exercise (prolonged moderate intensity walking/light jogging). Finally, we failed to examine if other factors, such as menstrual phase, fitness, or disease duration, influenced our findings. Future studies should examine if these results translate to a larger patient population.

In conclusion, we found that a 50–80% BRR performed 90 min before the onset of prolonged aerobic exercise improves open-loop glucose control and decreases hypoglycemia risk during exercise while not compromising the postexercise meal glucose control in patients living with type 1 diabetes. These findings support recent guidelines that preplanned BRR well in advance of prolonged exercise is an effective strategy to reduce hypoglycemia risk and minimize the need for carbohydrate feeding during the activity (4). Whether the community with type 1 diabetes can adopt this preplanning strategy effectively still needs to be determined.

Acknowledgments. The authors thank Dr. Jennifer Kuk, York University (Toronto, Ontario, Canada), for statistical consultation and contribution and Dr. Julian Aiken and Trevor Teich, York University, for technical expertise and assistance with the insulin assay. The authors also thank all of the participants for the time and dedication to the study.

Duality of Interest. This study was funded by Insulet Canada Corporation and Insulet Corporation. D.P.Z. has received speaker’s honoraria from Medtronic Diabetes and Ascensia Diabetes Care. T.V. and T.L. are both employees and shareholders of Insulet Corporation. M.C.R. has received speaker’s honoraria from Medtronic Diabetes, Insulet Corporation, Ascensia Diabetes Care, Novo Nordisk (via JDRF T1D Performance in Exercise and Knowledge Program), Xeris Pharmaceuticals, Lilly Diabetes, and Lilly Innovation. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. D.P.Z. was responsible for data collection, interpretation of data, and analysis. D.P.Z., S.M., R.P., T.V., T.L., and M.C.R. contributed feedback and revisions for the final manuscript. D.P.Z. and M.C.R. designed the study and wrote the manuscript. S.M. and R.P. assisted in the data collection. M.C.R. is the guarantor of this work and, as such, had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Prior Presentation. Parts of this study were presented in oral form at the 78th Scientific Sessions of the American Diabetes Association, Orlando, FL, 22–26 June 2018.