Complications of type 1 diabetes mellitus during pregnancy include increased rates of congenital anomaly, stillbirth, neonatal death, preterm delivery, and macrosomia.1 Congenital anomalies are associated with poor glycemic control around the time of conception, whereas the other complications are associated with maternal hyperglycemia that persists during pregnancy.2-5
Pregnant women with type 1 diabetes face particular challenges in trying to maintain tight glycemic control. Insulin requirements typically increase by a factor of 2 to 3 during the second and third trimesters, with substantial day-to-day variability, making the need for dose adjustments and their required magnitude unpredictable.6,7 Even with regular glucose monitoring, intensive insulin therapy, and glycated hemoglobin levels below 7%, pregnant women with type 1 diabetes have glucose levels that are above the target range approximately half the time.8,9 They also have increased rates of hypoglycemia,6,10-12 with glucose levels that are below the target range for up to 3.5 hours per day,8 so the benefit of avoiding hyperglycemia for the infant must be weighed against the risk of hypoglycemia for the mother.
Technological advances in glucose monitoring and insulin delivery, including continuous glucose monitoring, insulin pumps, and sensor-augmented pump therapy, may allow for safer improvements in glycemic control.13 Closed-loop systems use a computer algorithm (a set of mathematical instructions) to adjust insulin-pump delivery in response to glucose measurements obtained from real-time continuous glucose monitors.14 These systems have been shown to improve glycemic control without increasing the risk of hypoglycemia under experimental conditions,15-17 in trials of supervised outpatient treatment,18-22 and in studies of unsupervised, self-administered treatment among patients who were not pregnant.23-25 Preliminary data suggest that closed-loop systems may maintain near-normal glucose levels and minimize the risk of nocturnal hypoglycemia among pregnant women with type 1 diabetes.26,27 We performed a 4-week, randomized, crossover study of an overnight closed-loop system, followed by a 14-week continuation phase of day-and-night closed-loop therapy, which encompassed pregnancy-related challenges, including antenatal hospital admission, labor, and delivery, as well as postnatal adaptation.
We recruited pregnant women who had a history of type 1 diabetes mellitus for at least 12 months. Participants were 18 to 45 years of age, with a pregnancy between 8 and 24 weeks of gestation and a glycated hemoglobin level between 6.5 and 10.0%. The women were receiving intensive insulin therapy administered by means of either multiple daily injections or an insulin pump. Women were excluded if they had conceived with the use of assisted reproductive technologies, were receiving concurrent treatment that might influence glucose control, had a multiple-gestation pregnancy, or had clinically significant nephropathy, neuropathy, or proliferative retinopathy, as judged by the investigator. Detailed inclusion and exclusion criteria are provided in theSupplementary Appendix, available with the full text of this article at NEJM.org. All participants provided written informed consent.
The study protocol was approved by the East of England Research Ethics Committee of the Health Research Authority, with notification of no objection provided by the U.K. Medicines and Healthcare Products Regulatory Agency. The study was overseen by an independent data and safety monitoring board.
Abbott Diabetes Care provided discounted continuous-glucose-monitoring devices and consumables. Company representatives had no role in the design of the study; in the collection, handling, analysis, or interpretation of data; or in the decision to submit the manuscript for publication. The National Institute for Health Research and Abbott Diabetes Care received a copy of the manuscript before submission as a formality but did not provide input on the content.
The study was an open-label, multicenter, randomized, crossover trial. Participants were recruited from three U.K. National Health Service (NHS) sites.
After enrollment, participants were trained to use the study devices: a DANA Diabecare R Insulin Pump (SOOIL) and the FreeStyle Navigator II (Abbott Diabetes Care). After a run-in period of 2 to 4 weeks for device training and optimization of insulin doses, participants were randomly assigned in permuted blocks of 4 to either the overnight closed-loop system (intervention) or sensor-augmented pump therapy (control). Participants underwent a 2-week washout period after completing the first assigned intervention and before starting the second intervention. During the washout phase, participants used finger-stick testing, with or without continuous glucose monitoring or pump therapy, but could not use the closed-loop system.
After completion of the randomly assigned interventions, participants could choose to continue sensor-augmented pump therapy or the day-and-night closed-loop system with manually administered boluses before meals until delivery. This continuation phase provided a longer-term feasibility assessment of 24-hour closed-loop therapy while addressing the ethical questions that would be raised by withdrawal of an effective treatment during pregnancy.
Participants were advised to perform finger-stick glucose testing at least seven times daily, with standard glucose targets in both groups (63 to 99 mg per deciliter [3.5 to 5.5 mmol per liter] before a meal and <140 mg per deciliter [7.8 mmol per liter] 1 hour after a meal). Routine antenatal clinic visits were scheduled every 2 weeks, with fetal ultrasonographic assessments performed at 12, 20, 28, 32, and 36 weeks of gestation. There were no restrictions on physical activity, meals, or overseas travel, and no remote monitoring was performed. Participants had access to a 24-hour telephone line for assistance with technical difficulties.
C-peptide levels were measured when the serum glucose level was within the target range (63 to 140 mg per deciliter [3.5 to 7.8 mmol per liter]) at baseline, and glycated hemoglobin levels were measured at baseline, after each intervention phase, and at 28, 32, and 36 weeks of gestation. The study design is shown in Fig. S1 in the Supplementary Appendix.
During closed-loop therapy, a computer program, housed on a tablet computer, used continuous glucose measurements to determine an appropriate insulin dose. The insulin was delivered by means of an insulin pump every 12 minutes (Fig. S3 in the Supplementary Appendix). Premeal boluses were administered manually (15 to 30 minutes before the meal) as clinically indicated.28 To initialize closed-loop therapy, the participant’s weight and total daily insulin dose were entered in the computer program. During the 4-week randomized phase, participants started closed-loop therapy after their evening meal and stopped before breakfast. During the day-and-night continuation phase, closed-loop therapy was used continuously, with manually administered boluses before meals. The device had to be within approximately 30 m of the participant in order to maintain connectivity. There were no programming changes in anticipation of antenatal glucocorticoid use, labor, or delivery.
Study End Points
The primary efficacy end point was the percentage of time that glucose was in the target range of 63 to 140 mg per deciliter overnight, as recorded by means of continuous glucose monitoring during each 4-week study phase. Secondary efficacy end points were the mean glucose level, the percentages of time overnight and during the day and evening that glucose levels were above and below relevant thresholds, glucose variability, insulin dose, and glycated hemoglobin level.
Safety end points included the number and duration of hypoglycemic episodes (moderate or severe). Moderate hypoglycemia was defined as a glucose level of less than 63 mg per deciliter for 20 minutes or longer, as measured by continuous glucose monitoring. A severe hypoglycemic episode was defined as an episode requiring third-party assistance.
The feasibility of day-and-night closed-loop therapy in the continuation phase (from the end of the crossover phase until delivery) was assessed on the basis of glucose measurements during sequential 4-week intervals and over the period as a whole. The same glucose targets and study end points were used during the crossover and continuation study phases.
In our previous study of a closed-loop system with the use of sensor-augmented pump therapy in pregnant women with type 1 diabetes (median glycated hemoglobin level, 6.4%), the mean (±SD) percentage of time that glucose levels were in the target range was 61.7±24.9%.27 We calculated that we would need to enroll 16 women for the current study to have a power of 80% to detect a 30% relative increase in the percentage of time that glucose levels were in the target range (from 62% with sensor-augmented pump therapy to 80% with the closed-loop system), at an alpha level of 0.05 (two-tailed). The standard deviation for the primary outcome was assumed to be 25%.26,27
Statistical analyses were performed on an intention-to-treat basis, with data analyzed according to the study phase to which the participant had been assigned, regardless of adherence to the assigned intervention. We used linear mixed-effects models to estimate the percentage of overnight time that glucose levels were in the target range (response variable). The fixed effect of interest was whether there was a difference between sensor-augmented pump therapy and closed-loop therapy. Since the response variable was a repeated measure, we included nested random effects for the average time-in-target value for each study participant and for each 4-week time period for each participant. The fit of the model was not improved by including a term for either study phase-by-intervention interaction or autocorrelation of the response variable over time, and the estimated difference between study phases was not materially altered. Functional analysis of the continuous glucose data29 was performed and adjusted for weeks of gestation and period effect. Sequential glucose measurements were modeled as trajectories by calculating continuous mathematical functions of glucose measurements. These trajectories were modeled by fitting B-splines to the repeated measures.30 A two-sided significance level of 0.05 was used for both primary and secondary outcomes, without adjustment for multiple comparisons.
A total of 20 participants were recruited to participate in the study. Of these participants, 3 withdrew during the run-in training phase and 17 underwent randomization. One participant withdrew during her first study phase (sensor-augmented pump therapy) because of termination of pregnancy for trisomy 13 (a chromosomal anomaly unrelated to diabetes). Sixteen participants completed both study phases and were included in the analyses (Table 1TABLE 1Baseline Characteristics of the 16 Study Participants.). Six participants were receiving multiple daily insulin injections, and 14 participants had no experience with continuous glucose monitoring before the study.
The percentage of overnight time that glucose values were within the target range was significantly higher with closed-loop therapy than with sensor-augmented pump therapy (74.7% vs. 59.5%; absolute difference, 15.2 percentage points; 95% confidence interval, 6.1 to 24.2; P=0.002) (Table 2TABLE 2Comparison of Sensor-Augmented Pump Therapy and Closed-Loop Insulin Delivery during the Overnight Period in the Crossover Phase of the Study.and Figure 1FIGURE 1Median Sensor-Recorded Glucose Values over a 24-Hour Period with Sensor-Augmented Pump Therapy and Closed-Loop Insulin Delivery.). The mean glucose level was significantly lower during closed-loop therapy than during sensor-augmented pump therapy, both overnight (119 vs. 133 mg per deciliter [6.6 vs. 7.4 mmol per liter], P=0.009) (Table 2) and over a 24-hour period (128 vs. 137 mg per deciliter [7.1 vs. 7.6 mmol per liter], P<0.001) (Table 2 and Table 3TABLE 3Comparison of Sensor-Augmented Pump Therapy and Closed-Loop Insulin Delivery during the Day and Evening in the Crossover Phase of the Study.).
The incidence of maternal hyperglycemia was lower during closed-loop therapy than during sensor-augmented pump therapy, both overnight and over a 24-hour period. The incidence of substantial nocturnal hyperglycemia (glucose level, >180 mg per deciliter [10.0 mmol per liter]) was significantly lower during overnight closed-loop therapy than during sensor-augmented pump therapy (Table 2). Functional data analysis showed that overnight closed-loop therapy was associated with a significantly lower glucose level for a total time of 7 hours and 20 minutes (between 1:50 a.m. and 9:10 a.m.), with no significant effect for gestational age or study phase (Fig. S4 in the Supplementary Appendix).
The percentage of time that glucose levels were in the hypoglycemic range (<63 mg per deciliter) was low (<2%), with no significant differences between the two phases of the crossover study. There were no episodes of severe hypoglycemia during either phase.
Glycated hemoglobin levels declined from baseline to the end of both phases, with no significant difference in changes between the two phases of the crossover study (P=0.67). Total daily insulin doses were similar in the two phases, although insulin delivery was significantly more variable during closed-loop therapy than during sensor-augmented pump therapy (P<0.001) (Table S1 in the Supplementary Appendix). The results were similar among experienced pump users and those with no previous experience in pump use (Table S2 in the Supplementary Appendix).
There were 26 adverse events (14 skin reactions and 12 minor illnesses), with no significant differences between closed-loop and sensor-augmented pump therapy. There were 95 device deficiencies (18 during sensor-augmented pump therapy, 21 during closed-loop therapy, and 56 during the run-in and continuation phases), none of which resulted in severe hypoglycemia or hyperglycemia (Table S3 in the Supplementary Appendix). There were 8 serious adverse events. One of these events, which occurred during the randomized phase of closed-loop therapy, was a hospital admission after an episode of self-treated hypoglycemia resulting from recurrent vomiting; however, this event occurred during the daytime, when the closed-loop system was not operational. Another serious adverse event (vomiting due to gastroenteritis) occurred during the run-in training phase, and 6 events occurred during the continuation phase (Table S4 in the Supplementary Appendix). No serious adverse events were considered by the study investigators to be device-related.
Feasibility of Day-and-Night Closed-Loop Therapy throughout Pregnancy
Fourteen participants chose to continue using the closed-loop system after they finished the crossover studies, which provided up to an additional 14.6 weeks (median, 11.6 weeks [interquartile range, 7.1 to 12.7]) of day-and-night closed-loop use for feasibility assessment. Throughout the continuation phase with the closed-loop system, glucose levels were in the target range a mean of 68.7% of the time, and the mean glucose level was 126 mg per deciliter (7.0 mmol per liter).
The median glucose levels were 124, 128, 124, and 115 mg per deciliter (6.9, 7.1, 6.9, and 6.4 mmol per liter) at gestational ages of 24 weeks to 27 weeks 6 days, 28 weeks to 31 weeks 6 days, 32 weeks to 35 weeks 6 days, and 36 weeks or more, respectively (Tables S5 and S6 in theSupplementary Appendix). The median percentages of time that glucose levels were in the target range were 70.9%, 67.6%, 67.8%, and 77.3% at the respective gestational ages. The median percentages of time that glucose levels were below the target ranged from 1.2 to 2.1% throughout pregnancy.
Closed-Loop Therapy during Labor and Delivery
Fourteen women continued closed-loop therapy during labor and delivery (Figure 2FIGURE 2Glycemic Control during Labor and Delivery in the 14 Participants Who Continued to Use Closed-Loop Insulin Delivery.). In the 24 hours before delivery, these women had a median glucose level of 110 mg per deciliter (interquartile range, 104 to 128 [6.1 mmol per liter; interquartile range, 5.8 to 7.1]), with glucose levels in the target range of 63 to 140 mg per deciliter 86.8% of the time (interquartile range, 59.6 to 94.1) and glucose levels below the target range 0.5% of the time (interquartile range, 0 to 1.8). In the first 48 hours after delivery, these women had a median glucose level of 117 mg per deciliter (interquartile range, 104 to 137 [6.5 mmol per liter; interquartile range, 5.8 to 5.6]), with glucose levels in the target range 73.7% of the time (interquartile range, 61.4 to 86.0) and values below the target range 0% of the time (interquartile range, 0 to 0.5). The median total daily insulin dose was 53.6% of the predelivery dose (interquartile range, 48.6 to 73.6), with substantial variation among the participants (Table S7 in theSupplementary Appendix). There were no episodes of maternal hypoglycemia during the 24 hours before or 48 hours after delivery.
Obstetrical and Neonatal Outcomes
The median gestational age at delivery was 36.9 weeks (interquartile range, 34.5 to 37.7). Preeclampsia developed in 5 participants, including 1 with the HELLP syndrome (hemolysis, elevated liver enzyme levels, and a low platelet count). Cesarean section was performed in 15 participants (2 of whom were under general anesthesia); 10 of the procedures were performed before the onset of labor. One participant underwent an ovarian cystectomy during cesarean section. Seven participants delivered before 37 weeks of gestation (with 4 of the 7 participants delivering before 34 weeks of gestation), and in 6 participants, antenatal glucocorticoids were administered for fetal lung maturation. On days 1, 2, and 3 after glucocorticoid administration, glucose levels were in the target range a median of 58.1%, 59.1%, and 70.9% of the time, respectively. The closed-loop system delivered a median of 169 to 178% of the pre-glucocorticoid insulin dose, although there was substantial variation among the participants (Tables S8 and S10 in the Supplementary Appendix).
The median birth weight was 3588 g (interquartile range, 2670 to 3998). Thirteen of the 16 infants had a birth weight (corrected for sex and gestational age) that was above the 90th percentile, according to population-based U.K. data.31 Twelve infants received neonatal intensive care, 11 of whom were treated with intravenous dextrose for neonatal hypoglycemia. Details regarding the obstetrical outcomes are provided in Table S11 in the Supplementary Appendix.