The Impact of Accelerometer Use in Exercise-Associated Hypoglycemia Prevention in Type 1 Diabetes

Transcription

1 551045DSTXXX / Journal of Diabetes Science and TechnologyStenerson et al research-article2014 Original Article The Impact of Accelerometer Use in Exercise-Associated Hypoglycemia Prevention in Type 1 Diabetes Journal of Diabetes Science and Technology 2015, Vol. 9(1) Diabetes Technology Society Reprints and permissions: sagepub.com/journalspermissions.nav DOI: / dst.sagepub.com Matthew Stenerson, MD 1, Fraser Cameron, PhD 2, Shelby R. Payne, BS 1, Sydney L. Payne, BS 1, Trang T. Ly, MBBS, FRACP, PhD 1, Darrell M. Wilson, MD 1, and Bruce A. Buckingham, MD 1 Abstract Background: Exercise-associated hypoglycemia is a common adverse event in people with type 1 diabetes. Previous in silico testing by our group demonstrated superior exercise-associated hypoglycemia mitigation when a predictive low glucose suspend (PLGS) algorithm was augmented to incorporate activity data. The current study investigates the effectiveness of an accelerometer-augmented PLGS algorithm in an outpatient exercise protocol. Methods: Subjects with type 1 diabetes on insulin pump therapy participated in two structured soccer sessions, one utilizing the algorithm and the other using the subject s regular basal insulin rate. Each subject wore their own insulin pump and a Dexcom G4 Platinum continuous glucose monitor (CGM); subjects on-algorithm also wore a Zephyr BioHarness 3 accelerometer. The algorithm utilized a Kalman filter with a 30-minute prediction horizon. Activity and CGM readings were manually entered into a spreadsheet and at five-minute intervals, the algorithm indicated whether the basal insulin infusion should be on or suspended; any changes were then implemented by study staff. The rate of hypoglycemia during and after exercise (until the following morning) was compared between groups. Results: Eighteen subjects (mean age 13.4 ± 3.7 years) participated in two separate sessions 7-22 days apart. The difference in meter blood glucose levels between groups at each rest period did not achieve statistical significance at any time point. Hypoglycemia during the session was recorded in three on-algorithm subjects, compared to six off-algorithm subjects. In the postexercise monitoring period, hypoglycemia occurred in two subjects who were on-algorithm during the session and four subjects who were off-algorithm. Conclusions: The accelerometer-augmented algorithm failed to prevent exercise-associated hypoglycemia compared to subjects on their usual basal rates. A larger sample size may have achieved statistical significance. Further research involving an automated system, a larger sample size, and an algorithm design that favors longer periods of pump suspension is necessary. Keywords accelerometer, exercise, hypoglycemia, predictive low glucose suspend, pump suspension, type 1 diabetes Hypoglycemia is a common and potentially dangerous side effect of exercise in people with type 1 diabetes (T1D). 1-3 Although suspension of insulin delivery at the beginning of moderate aerobic exercise reduces the risk of exercise-associated hypoglycemia, the risk of hyperglycemia is increased. 4 Glucose threshold-based insulin pump suspension has been shown to reduce the duration of exercise-induced hypoglycemia and can be performed in a user-independent manner. 5 Our group previously augmented our predictive low glucose suspend (PLGS) algorithm to incorporate physical activity using a combined accelerometer/heart rate monitor. 6 The activity-informed algorithm showed superior hypoglycemia mitigation compared to the PLGS algorithm alone using in silico testing. 7 The current study investigates the effectiveness of an accelerometer-augmented PLGS algorithm in the real-life setting of an outpatient exercise protocol. We hypothesized that 1 Division of Pediatric Endocrinology, Stanford University, Stanford, CA, USA 2 Department of Chemical and Biological Engineering, Rensselaer Polytechnic Institute, Troy, NY, USA Corresponding Author: Bruce A. Buckingham, MD, Division of Pediatric Endocrinology, Stanford University, 300 Pasteur Dr, Rm G-313, Stanford, CA , USA. buckingham@stanford.edu

2 Stenerson et al 81 the accelerometer-augmented algorithm would significantly reduce the incidence of exercise-associated hypoglycemia compared to subjects usual basal rates, without an increase in hyperglycemia. Methods Study Design This was a pilot study employing a randomized crossover design. Subjects with type 1 diabetes participated in a structured soccer session on two separate dates at least one week apart. During one session a subject s basal insulin rate was continued ( off-algorithm ), while during the other session the algorithm (see algorithm monitoring below) was in effect ( on-algorithm ); the order of on- versus off-algorithm was chosen at random. The rate of hypoglycemia and change in blood glucose during exercise were compared between each group. Hypoglycemia was defined as (1) any meter blood glucose (BG) reading of 60 mg/dl, (2) two consecutive meter BG readings 70 mg/dl done within one hour, or (3) any instance in which carbohydrates were given at a subject s request for symptoms of hypoglycemia (regardless of corresponding meter BG reading). Eligibility Criteria and Assessment Subject recruitment took place at Stanford University. To be eligible for the study, each subject had to (1) be between 8 and 25 years of age, (2) have a clinical diagnosis of type 1 diabetes for 1-20 years, (3) use a downloadable insulin pump (with programmable carbohydrate to insulin ratios, correction doses, and insulin on board features) for insulin delivery, (4) own a continuous glucose monitor (CGM) or agree to wear a loaned CGM for purposes of study participation, and (5) have a hemoglobin A1c < 10%. Subjects were not eligible if they had an episode of severe hypoglycemia resulting in seizure or loss of consciousness in the prior four weeks. Clinic Visit Procedures At 1-3 days prior to the first soccer session, subjects had an initial visit with a study investigator for a baseline medical history and physical exam. A hemoglobin A1c was obtained using the Bayer DCA 2000 (Bayer HealthCare LLC, Mishawaka, IN). For those subjects not already using one, a Dexcom G4 Platinum CGM (Dexcom Inc, San Diego, CA) was placed on the abdomen, arm, or buttocks. Each subject also received a Bayer Contour Next EZ glucometer (Bayer HealthCare LLC, Mishawaka, IN) for calibration of the CGM. Pre-exercise Procedures Subjects were instructed to calibrate the CGM using their studyissued glucometer twice daily and on the morning of soccer. For breakfast on the morning of the soccer session, subjects bolused via their insulin pumps for all but g of carbohydrates, with the goal of avoiding hypoglycemia prior to the start of the session. This also mimics the common practice of subjects to not provide insulin coverage for g of carbohydrates in a meal prior to exercise ( exercise carbohydrates ). Subjects were instructed to consume the same meal before each session. Subjects arrived at the soccer field minutes prior to the session s start time. Upon arrival, initial meter BG and CGM glucose values were recorded. The CGM was recalibrated if there was a >20% difference between the glucometer reading and CGM value, unless the CGM demonstrated a rapid rate of change ( 2 mg/dl/min) in either direction. For meter BG < 120 mg/dl, subjects were given g carbohydrates and held back from starting exercise until meter BG measured > 120 mg/dl. For meter BG > 300 mg/dl, serum ketones were assessed using the Precision Xtra blood ketone meter (Abbott Laboratories, Abbott Park, IL). Infusion sets were changed if serum ketones were >0.3 mmol/l. For those subjects who were on-algorithm, the Zephyr BioHarness 3 (Zephyr Technology, Annapolis, MD) accelerometer was secured around the waist via a nylon strap. Activity data that were measured by the accelerometer were beamed via Bluetooth to an Android cell phone using the SenseView app (Mobili d.o.o., Ljubljana, Slovenia), resulting in a real-time Activity output graph that was followed using the phone s screen by an assigned study staff member. Intraexercise Procedures Subjects performed a final meter BG check just prior to the start of exercise. For those with BGs > 120 mg/dl, soccer practice began at 10:00 am. Each session involved skill-building activities across various ranges of exercise intensity (Table 1). Subjects had four segments of activity lasting approximately minutes each, with three rest periods of five minutes duration in between each Activity interval, during which subjects checked their meter BG. The CGM was recalibrated if there was a >20% difference between the glucometer reading and CGM value, unless the CGM demonstrated a rapid rate of change ( 2 mg/dl/min) in either direction. Unless hypoglycemic, subjects consumed only water throughout the duration of the session. However, in the event of hypoglycemia, subjects were given g of carbohydrates and their insulin pump was suspended for 30 minutes. Subjects were also removed from play and were not allowed to return until meter BG was >120 mg/dl. Additional carbohydrates beyond the g were given if there was not a substantial rise in the meter BG after 15 minutes. Algorithm Monitoring The pump suspension algorithm follows three rules based on a 30-minute prediction from a Kalman filter and the Activity readings from the accelerometer. First, insulin delivery is resumed if the CGM readings are rising. This rule trumps all other rules. Second, the pump is suspended if the 30-minute

3 82 Journal of Diabetes Science and Technology 9(1) Table 1. Typical Exercise Session, Which Included Skill-Building Activities of Various Intensity Levels and Three Rest Periods for Water and BG Checks. Activity Time (min) Intensity (1-10) Dynamic warm-up 10 5 Dribbling in a cone grid 10 6 Dribbling relays and figure 8s 5 4 rest 5 1 Speed ladder, cone slalom 15 7 Dribbling, shooting on goal 10 6 rest 5 1 Passing and moving 10 6 Shuttle run 5 10 Mini scrimmage 15 9 rest 5 1 Full scrimmage 25 9 Intensity levels are based on an independent scale in which 1 is equivalent to very light activity, 3 is light activity, 5 is moderate activity, 7 is hard activity, and 10 is very hard activity. prediction is below 80 mg/dl. Third, the pump is suspended if the CGM is below 180 mg/dl and the Activity averaged over 80 seconds is above 0.3, measured in Zephyr s proprietary units. (The values of 80 seconds, and 0.3 represent the retro-fitted parameters that best matched actual performance. A value of 0.3 represents the activity level of very light walking.) The assigned staff member judged Activity by eye using the screen of an Android phone. The algorithm was run using Excel (Microsoft Corporation, Redmond, WA) on a research laptop computer. Each subject who was on-algorithm had one assigned study staff member to follow his or her Activity and CGM trends. In five-minute intervals, the staff member manually entered into Excel the average observed Activity value over those five minutes, as well as the CGM glucose value at the end of that interval. Based on these data points, the algorithm then generated one of two messages for pump status: pump on or pump off. Algorithm determination of pump status was then implemented by the study staff: the assigned staff member manually suspended the subject s pump when commanded to turn the pump off, or manually resumed insulin delivery when commanded to turn the pump on (Table 2). At the conclusion of the soccer session, subjects basal rates were resumed. Postexercise Procedures Immediately following the completion of the soccer session, subjects on-algorithm had their accelerometers removed and all subjects were provided lunch. Subjects were instructed to resume their usual postexercise diabetes management, with no further intervention from study staff. However, subject monitoring was continued via continued use of the CGM and study-issued meter for all BG checks until breakfast the following morning to assess for postexercise hypoglycemia. Statistical Analysis Mean and standard deviation meter BG values were tabulated using Excel. The difference in rates of hypoglycemia between the on-algorithm and off-algorithm groups was analyzed by group analysis using a 2-sided Fischer s exact test. Results A total of 18 subjects (eight female) participated in two separate soccer sessions 7-22 days apart. Mean age was 13.4 years (±3.7 years, range 9.4 to 25 years) and mean hemoglobin A1c was 8.0% (±1.1%, range 5.5% to 9.9%). Eleven subjects were using a Medtronic pump, four were using an OmniPod pump, two an Animas pump, and one a Tandem pump. Of note, on arrival to the soccer field there were four subjects with meter BG values >300 (2 involving on-algorithm participants, the other two off-algorithm); a partial insulin bolus (generally half of the insulin pump s recommended bolus) was administered prior to the start of soccer for these subjects. For each soccer session, the dosage of each subject s breakfast-time insulin and intraexercise insulin as well as nadir glucose levels during exercise are shown in Table 3. The timing of insulin suspension and carbohydrate treatment for hypoglycemia in both groups is shown in Figure 1, including the aggregate number of patients whose pumps were suspended at any given five-minute interval throughout the soccer session. There was no difference in the meter BG values at the start of soccer between the on-algorithm group and the off-algorithm group (204.7 ± 63.7 mg/dl and ± 54.5 mg/dl, respectively; P =.40). Meter BG values at the scheduled checks throughout the session are shown in Figure 2. The difference in meter BGs between groups at each rest period did not achieve statistical significance at any time point. Hypoglycemia during the soccer session was recorded

4 Stenerson et al 83 Table 2. Sample Algorithm Flow Sheet (Subject 6). Time (HH:MM) CGM glucose (mg/dl) Activity reading 30-min glucose prediction (mg/dl) Desired pump state 9: : Off 10: Off 10: Off 10: Off 10: Off 10: On 10: Off 10: Off 10: Off 10: Off 10: Off 10: Off 11: On 11: On 11: On 11: On 11: On 11: On 11: Off 11: Off 11: Off 11: Off 11: Off 11: On 12: Off CGM glucose and average activity were entered in five-minute intervals, prompting the algorithm to generate a 30-minute glucose prediction and desired pump state (on vs off). Based on the algorithm s suggestion, staff pulled a subject aside to make recommended changes in pump status. in three on-algorithm subjects, compared to six off-algorithm subjects (P =.45). In future studies we plan to have the algorithm communicating directly with the insulin pump to initiate pump suspensions. We therefore performed a post hoc analysis of the data to determine the difference in pump suspensions if they had occurred automatically compared to manually. If pumps had been suspended automatically, the amount of time the pumps would have been suspended would have been decreased by 3%. In the postexercise monitoring period (during which subjects made their own insulin adjustments and were not restricted on food consumption), hypoglycemia occurred in two subjects who were on-algorithm during the soccer session and four subjects who were off-algorithm (P =.66). The distribution of CGM glucose values throughout the postexercise monitoring period was not significantly different between the two groups. Discussion Exercise-associated hypoglycemia is a common adverse event in people with T1D, 1-3 in part because of inappropriate insulin adjustments to account for the increased insulin sensitivity associated with exercise. 8,9 Children and adolescents are at particular risk due to notoriously suboptimal adherence to their insulin regimens. 10,11 Thus, as a user-independent means of insulin management, the artificial pancreas a system of closed-loop communication between a CGM and an insulin pump under algorithmic control 12 offers a unique opportunity for hypoglycemia prevention/mitigation. However, although a control-to-range algorithm augmented with heart rate data and a low glucose suspend algorithm have shown encouraging results, the use of algorithmic controls to prevent hypoglycemia in the exercise setting is still a fledgling venture in artificial pancreas research. 5,13 To our knowledge, the current study is the first in which an accelerometer-augmented insulin pump suspension algorithm has been tested in the setting of an outpatient exercise protocol. In previous in silico testing, the activity-informed algorithm showed superior hypoglycemia mitigation compared to a PLGS algorithm alone. 7 However, in the real-life setting of the current study, there was no statistically significant difference in the on-algorithm versus off-algorithm groups with regard to average final BG, nor in the number of hypoglycemia events either during or postexercise. We suspect that the

5 84 Journal of Diabetes Science and Technology 9(1) Table 3. Dosage of Each Subject s Breakfast-Time Insulin (Cumulative Insulin Boluses Within Four Hours of Session Start Time), Intraexercise Insulin, and Nadir Glucose Levels at Each Soccer Session. Breakfast insulin Exercise insulin Nadir glucose (meter/sensor) Subject ID On-algorithm Off-algorithm On-algorithm Off-algorithm On-algorithm Off-algorithm /56 49/ /NA 83/ / / /56 57/ / / /76 77/ /139 66/ / / / / /105 67/ /NA 161/ /148 66/ /83 59/ /131 80/ a a /97 a 53/ /129 87/ /95 189/ / /164 a On-algorithm data counted as off-algorithm because pump erroneously stayed on despite attempts to suspend. NA, not able to be determined due to extended periods of sensor failure during exercise. Figure 1. Overview of insulin delivery status for all subjects whose pumps were suspended at least once during the soccer session. The bar graph in the top pane details the percentage of pumps suspended at any given time in five-minute intervals. Lines in the bottom pane demonstrate each subject s pump status throughout the two-hour session (black = off-algorithm, blue = on-algorithm). For subjects in the off-algorithm group, pumps were suspended only if meter BG (not shown) was 60 mg/dl or the subject request treatment for hypoglycemia. Carbohydrate interventions for hypoglycemia are depicted as squares. lack of superior hypoglycemia prevention in this real-life setting may in part be due to a small sample size and an insufficient number of hypoglycemic events in the off-algorithm group. The latter was partly the result of our use of free or exercise carbohydrates that subjects were instructed to consume prior to the soccer session (10-30 g); this resulted in an average starting BG >180 mg/dl, so that the incidence of hypoglycemia in the control arm was 33%, compared to an expected incidence of about 43%. 4 We also suspect that the lack of superior hypoglycemia prevention may in part be due to the algorithm s over-aggressiveness in favoring the pump on status even with BGs bordering hypoglycemia, as seen with Subject 6 in Table 2. Future adjustments to the algorithm may include a threshold above zero on the glucose rate of change for resumption of insulin delivery when activity has occurred in the recent past and glucose levels are less than 80 or 90 mg/dl. These changes would favor more pump suspension, decreasing the risk of hypoglycemia both during and after exercise. Although adjustments are needed, the resumption of insulin delivery using the current algorithm did not result in any difference in hyperglycemia between the on-algorithm and off-algorithm groups. This highlights the benefits of algorithmic control over the arbitrary suspension of insulin delivery at the beginning of exercise; in a study carried out by the DirecNet group, such arbitrary suspension demonstrated hypoglycemia prevention in 57% of subjects, but hyperglycemia in 27% of subjects (compared to 4% of controls) shortly after the completion of exercise. 4 An additional limitation of the study is related to the potential for operator error in regard to both triggering the

6 Stenerson et al 85 Declaration of Conflicting Interests The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article. Funding The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was supported in part by the Lilly Endocrine Scholars Award; the Ernest and Amelia Gallo Endowed Postdoctoral Fellowship; the Child Health Research Institute; and the Stanford Clinical & Translational Science Award, grant UL1 RR Figure 2. Mean meter BG values from the start of the exercise session to the end. For subjects who met hypoglycemia criteria prior to a rest period, the meter BG for those subjects is reflected in the immediately subsequent rest period; any meter BGs thereafter were removed, resulting in a progressively lower N. algorithm and carrying out the algorithms commands. Using the SenseView app, staff members observed Activity readings in 5-minute segments and entered into Excel an observed average Activity number over that time. As an estimate, this number was less reliable than an automated system. In addition, with staff members being responsible for manually suspending and resuming delivery on a subject s pump, delays in pump suspension and resumptions of delivery also occurred, which could have been avoided using an automated system. Conclusions In this pilot study investigating the effectiveness of an accelerometer-augmented pump suspension at hypoglycemia prevention in an outpatient exercise protocol, the algorithm did not prevent exercise-associated hypoglycemia. This lack of statistical significance may in part have been related to an overaggressive bias toward insulin delivery. However, small sample size and the requirement for manual interpretation of data and carrying out of algorithm commands may also have played a role. We therefore remain optimistic about the potential for accelerometer-augmented algorithms to provide a meaningful contribution in an artificial pancreas. Further research involving an automated system, a larger sample size, and an algorithm design that favors longer periods of pump suspension is necessary. Abbreviations BG, blood glucose; CGM, continuous glucose monitor; PLGS, predictive low glucose suspend; T1D, type 1 diabetes. References 1. Tansey MJ, Tsalikian E, Beck RW, et al. The effects of aerobic exercise on glucose and counterregulatory hormone concen-trations in children with type 1 diabetes. Diabetes Care. 2006;29: MacDonald MJ. Postexercise late-onset hypoglycemia in insulindependent diabetic patients. Diabetes Care. 1987;10: Shehadeh N, Kassem J, Tchaban I, et al. High incidence of hypoglycemic episodes with neurologic manifestations in children with insulin dependent diabetes mellitus. J Pediatr Endocrinol Metab. 1998;11(suppl 1): Tsalikian E, Kollman C, Tamborlane WB, et al. Prevention of hypoglycemia during exercise in children with type 1 diabetes by suspending basal insulin. Diabetes Care. 2006;29: Garg S, Brazg RL, Bailey TS, et al. Reduction in duration of hypoglycemia by automatic suspension of insulin delivery: the in-clinic ASPIRE study. Diabetes Technol Ther. 2012;14: Cameron F, Wilson DM, Buckingham BA, et al. Inpatient studies of a Kalman-filter-based predictive pump shutoff algorithm. J Diabetes Sci Technol. 2012;6: Stenerson M, Cameron F, Wilson DM, et al. The impact of accelerometer and heart rate data on hypoglycemia mitigation in type 1 diabetes. J Diabetes Sci Technol. 2014;8: Devadoss M, Kennedy L, Herbold N. Endurance athletes and type 1 diabetes. Diabetes Educ. 2011;37: Bernardini AL, Vanelli M, Chiari G, et al. Adherence to physical activity in young people with type 1 diabetes. Acta Biomed. 2004;75: Burdick J, Chase HP, Slover RH, et al. Missed insulin meal boluses and elevated hemoglobin A1c levels in children receiving insulin pump therapy. Pediatrics. 2004;113:e221-e O Connell MA, Donath S, Cameron FJ. Poor adherence to integral daily tasks limits the efficacy of CSII in youth. Pediatr Diabetes. 2011;12: Peyser T, Dassau E, Breton M, Skyler JS. The artificial pancreas: current status and future prospects in the management of diabetes. Ann N Y Acad Sci. 2014;1311: Breton MD, Brown SA, Karvetski CH, et al. Adding heart rate signal to a control-to-range artificial pancreas system improves the protection against hypoglycemia during exercise in type 1 diabetes. Diabetes Technol Ther. 2014;16:1-6.