Efficacy of an Overnight Predictive Low-Glucose Suspend System in Relation to Hypoglycemia Risk Factors in Youth and Adults With Type 1 Diabetes

Abstract

Background:

We developed a system to suspend insulin pump delivery overnight when the glucose trend predicts hypoglycemia. This predictive low-glucose suspend (PLGS) system substantially reduces nocturnal hypoglycemia without an increase in morning ketosis. Evaluation of hypoglycemia risk factors that could potentially influence the efficacy of the system remains critical for understanding possible problems with the system and identifying patients that may have the greatest benefit when using the system.

Methods:

The at-home randomized trial consisted of 127 study participants with hemoglobin A1c (A1C) of ≤8.5% (mmol/mol) for patients aged 4-14 years and ≤8.0% for patient aged 15-45 years. Factors assessed included age, gender, A1C, diabetes duration, daily percentage basal insulin, total daily dose of insulin (units/kg-day), bedtime BG, bedtime snack, insulin on board, continuous glucose monitor (CGM) rate of change (ROC), day of the week, time system activated, daytime exercise intensity, and daytime CGM-measured hypoglycemia.

Results:

The PLGS system was effective in preventing hypoglycemia for each factor subgroup. There was no evidence that the PLGS system was more or less effective in preventing hypoglycemia in any one subgroup compared with the other subgroups based on that factor. In addition, the effect of the system on overnight hyperglycemia did not differ in subgroups.

Conclusions:

The PLGS system tested in this study effectively reduced hypoglycemia without a meaningful increase in hyperglycemia across a variety of factors.

An important goal of a closed loop system for individuals with type 1 diabetes is prevention of hypoglycemia, particularly overnight. We developed a system to suspend insulin pump delivery overnight when the glucose trend predicts hypoglycemia.^1,2 The predictive low-glucose suspend (PLGS) system involves a continuous glucose monitor (CGM) device communicating with an insulin pump via a laptop computer that contains the hypoglycemia prediction algorithm. In an at-home randomized trial of 81 children aged 4-14 ¹ and 45 adults aged 15-45 ² with type 1 diabetes, we found that the PLGS system reduced the frequency of nocturnal hypoglycemia by 25% and 36%, respectively, compared with a control arm. Median time less than 70 mg/dL was reduced by more than 50%, prolonged nocturnal hypoglycemia (≤60 mg/dl for >2 hours) was reduced more than 3-fold, and there was no meaningful increase in hyperglycemia or ketosis. While the PLGS system greatly reduced hypoglycemia overall, it is possible subject characteristics or activities could influence the effectiveness of the PLGS system. During control nights, lower age, lower hemoglobin A1c (A1C) level, preceding hypoglycemia, and preceding exercise were associated with higher rates of nocturnal hypoglycemia. ³ We undertook the analyses reported herein to assess whether the efficacy of the PLGS system is related to the presence of these or other factors shown to be associated with nocturnal hypoglycemia.

Methods

The study was conducted at 3 clinical centers in the United States and Canada. The protocol was approved by each Institutional Review Board and informed consent and assent was obtained as appropriate. The study has been described elsewhere^1,2 and listed on the ClinicalTrials.gov website (clinical trial registration numbers NCT01823341 and NCT01591681); key aspects of the study protocols are described below.

The pump suspension system consisted of a MiniMed Paradigm® REAL-Time Veo™ System and Enlite™ glucose sensor (Medtronic Diabetes, Northridge, CA), in which the CGM and pump communicated with a bedside laptop computer which contained the hypoglycemia prediction algorithm developed by the investigators. The PLGS system used a Kalman filter to estimate the glucose level and ROC and suspended basal insulin delivery if glucose was predicted to fall below 80 mg/dl in the next 30 minutes. ⁴ Additional suspension/restart rules included a threshold suspend override once sensor glucose increased to 70 mg/dl, no suspension if CGM glucose concentration >230 mg/dl or if a pressure-induced sensor attenuation was suspected based on glucose ROC, ⁵ and restoration of basal insulin on the first CGM rise following a suspension. Audible alarms were set at 60 mg/dl. Additional details about the system have been published.^4,5

Major eligibility criteria included having type 1 diabetes with use of daily insulin therapy for ≥1 year and use of insulin infusion pump for ≥6 months. Patients aged 4-14 and 15-45 years old required an A1C level measured with a point-of-care device of ≤8.5% and ≤8.0%, respectively, at enrollment. Included patients also had a minimum amount of nocturnal hypoglycemia during a run-in phase. Additional criteria are listed in previously published articles.^1,2 The 127 study participants included in this analysis ranged in age from 4 to 45 years; half were female and 94% were Caucasian. Supplemental Table S1 gives the participant characteristics at enrollment. Children aged 4-12, adolescents aged 13-21, and adults aged 22-45 (FDA-defined age groups) had a median duration of type 1 diabetes of 4, 8, and 22 years, respectively, and median A1C of 7.8%, 7.6%, and 6.6%, respectively.

Following a run-in phase, the randomized trial consisted of 42 nights with at least 4 hours of sensor glucose data, during which approximately half the nights were intervention nights (PLGS algorithm active) and half control (PLGS algorithm inactive) nights. The order of intervention versus control nights was randomized nightly and blinded to the participant. At the time of system activation each night, the participant entered the level of exercise intensity for that day (none, low, medium, or high) and whether or not a bedtime snack was consumed. CGM device, pump downloads, and blood glucose meter downloads provided data used in analysis for insulin delivery and occurrence of hypoglycemia during both day and night as well as capillary glucose levels when the system was activated each night and turned off in the morning.

Statistical Methods

The time period for outcome assessment each night was from activation until deactivation the following morning. Randomized nights with at least 6 hours of CGM data were included in the analysis. A hypoglycemia night was defined as having at least 30 minutes of CGM values below 60 mg/dl. Both subject-level and night-level factors were assessed for possible influence on the efficacy of the system in preventing hypoglycemia. Subject-level factors included age, gender, A1C at enrollment, diabetes duration, daily percentage basal insulin, and total daily dose of insulin (units/kg-day). Night-level factors included bedtime BG, bedtime snack, insulin on board at activation, CGM rate of change (ROC) before activation, day of the week, time system activated, daytime exercise intensity preceding nighttime system use, and daytime CGM-measured hypoglycemia preceding nighttime system use. A previous article found the system was less effective in the first 3 hours; ² a secondary analysis was performed to assess factors influencing the effectiveness of the system to prevent hypoglycemia in the first 3 hours.

To assess preceding daytime CGM-measured hypoglycemia, time below 60 mg/dl was measured from noon until 8pm. Nights activated between 8pm and 8am following at least 6 hours of daytime CGM data were included in this analysis (2031 out of 2524 [80%] control nights and 1997 out of 2504 [80%] intervention nights analyzed).

The percentage of control nights and intervention nights with hypoglycemia was tabulated for each factor, giving equal weight to each night. An interaction effect was tested to assess if the PLGS system was more or less effective preventing hypoglycemia within the subgroups for a given factor. A repeated-measures regression model with random subject effects and spatial power autocorrelation structure was fitted to test for the interaction effect. Factors with a P value < .01 were considered significant to adjust for multiple hypothesis testing. A linear relationship was fitted for continuous factors. A 99% confidence interval of the odds ratio was calculated for each subgroup. Odds ratios greater than 1 indicated hypoglycemia was more likely to occur on control nights than intervention nights. All P values are 2-tailed and analyses were performed using SAS 9.4.

Results

Nocturnal hypoglycemia with at least 30 minutes of CGM values below 60 mg/dl occurred on 515 of 2524 (20%) control nights versus 299 of 2504 (12%) intervention nights (odds ratio 1.91 [99% CI 1.57-2.32]; P value < .001).

The PLGS system was effective in preventing hypoglycemia for each patient-level factor subgroup (Table 1) and each night-level factor subgroup (Table 2). There was not strong evidence in any subgroup that the PLGS system was more or less effective in preventing hypoglycemia than in other subgroups based on that factor (all interaction P values > .01).

Table 1.

Patient-Level Factors Influence on the PLGS System to Prevent Hypoglycemia.

a

Nights with hypo had at least 6 CGM values below 60 mg/dl over entire night.

b

Linear relationship fitted for continuous factor.

Table 2.

Night-Level Factors Influence on the PLGS System to Prevent Hypoglycemia.

a

Nights with hypoglycemia had at least 60 CGM values below 60 mg/dl over entire night.

b

Linear relationship fitted for continuous factor.

c

Exercise intensity and snacks entered by the patients.

d

CGM rate-of-change (ROC) calculated 15min before activation.

Supplemental Table S2 provides the nocturnal hypoglycemia rates stratified by age. Nocturnal hypoglycemia rates during the first 3 hours on intervention and control nights were similar (Supplemental Table S3). There was a substantial benefit of the PLGS system after 3 hours of system use. Nocturnal hyperglycemia with at least 30 minutes of CGM values above 250 mg/dl was similar on control and intervention nights (19% vs 21%, respectively) with no evidence any factor subgroup had higher rates of hyperglycemia on intervention nights compared with control nights.

Conclusion

Several factors have been associated with the frequency of nocturnal hypoglycemia including lower age, ³ lower hemoglobin A1c level, ⁶ occurrence of prior hypoglycemia^6,7 or an exercise session,^8,9 no bedtime snack, ¹⁰ and lower bedtime BG. ¹¹ A reasonable hypothesis is that a PLGS system would be more effective on nights with higher risk of hypoglycemia, but there was no evidence this was true for the PLGS system we tested; the system was equally beneficial irrespective of the patient’s activity or characteristics.

There is little research devoted to factors influencing the efficacy of closed-loop systems in preventing hypoglycemia. High insulin on board has been shown to affect the ability of glucagon to prevent hypoglycemia. ¹² In a preliminary study, ⁴ we found changing the PLGS algorithm to a longer projection horizon with slower restoration of basal insulin following suspension led to worse performance in preventing hypoglycemia.

The PLGS system was ineffective in reducing hypoglycemia in the first 3 hours after activation. We speculated that high insulin on board contributed to the inability of the PLGS system to reduce early hypoglycemia, but the ineffectiveness of the system did not differ for any factor subgroup. This could be attributed to the low rates of early hypoglycemia on control nights (only 8% of control nights had hypoglycemia in the first 3 hours), making it difficult to reduce the nocturnal hypoglycemia rate further. Many studies utilize a warm-up period^13,14 or have patients wear the system continuously. ¹⁵ Further research is warranted to assess whether continuously worn systems are more effective in the early part of the night.

A limitation of the study was that exercise intensity and bedtime snack were self-reported. A guideline for exercise intensity was given to subjects, but exercise duration, snack size, and snack composition were not reported. Three-way interactions were not assessed, so it is possible that a combination of 2 factors could influence the efficacy of the PLGS system. The study consisted of patients aged 4 to 45 years old with an A1C ≤8.5%, and the results may not be generalizable outside this domain. In addition, the findings may not generalize to closed-loop or other PLGS systems. Sensor inaccuracies potentially diluted the treatment effect, but would not affect the probability of a type 1 error. A major strength of the study is the large sample size. The data consisted of 5028 randomized nights from 127 patients thereby making it possible to assess the efficacy and safety of the system under a variety of conditions. The novel randomization scheme where the treatment assignment (control or intervention) was randomized each night minimized bias that could be attributed to a patient being aware of the treatment assignment or changing their diabetes management.

The PLGS system safely and effectively reduced hypoglycemia for all patient-level and night-level factor subgroups. The ability of the PLGS system to reduce hypoglycemia without increasing overnight hyperglycemia under a variety of conditions illustrates the system is robust and comprehensive.