Physical Activity Capture Technology With Potential for Incorporation Into Closed-Loop Control for Type 1 Diabetes

Abstract

Physical activity is an important determinant of glucose variability in type 1 diabetes (T1D). It has been incorporated as a nonglucose input into closed-loop control (CLC) protocols for T1D during the last 4 years mainly by 3 research groups in single center based controlled clinical trials involving a maximum of 18 subjects in any 1 study. Although physical activity data capture may have clinical benefit in patients with T1D by impacting cardiovascular fitness and optimal body weight achievement and maintenance, limited number of such studies have been conducted to date. Clinical trial registries provide information about a single small sample size 2 center prospective study incorporating physical activity data input to modulate closed-loop control in T1D that are seeking to build on prior studies. We expect an increase in such studies especially since the NIH has expanded support of this type of research with additional grants starting in the second half of 2015. Studies (1) involving patients with other disorders that have lasted 12 weeks or longer and tracked physical activity and (2) including both aerobic and resistance activity may offer insights about the user experience and device optimization even as single input CLC heads into real-world clinical trials over the next few years and nonglucose input is introduced as the next advance.

Type 1 diabetes (T1D) is characterized by complete or near complete absence of endogenous insulin secretion resulting in the need to deliver exogenous insulin in an individualized, precise manner.¹ Current clinically approved technologies to replace insulin result in unacceptable risk for hypoglycemia, hyperglycemia, and glycemic variability (GV).² Changes in plasma glucose concentration are more pronounced in T1D, occur rapidly and are related to food intake, physical activity and imprecise insulin delivery.³

The maturation of technologies such as continuous glucose monitoring (CGM), continuous subcutaneous insulin infusion (CSII) or insulin pump, rapid acting insulin analogs and closed-loop control (CLC) algorithms that deliver insulin based on CGM data processing has resulted in CLC therapy becoming an area of active research with rapid therapeutic translation to clinical practice.⁴ Current CLC studies show significant improvement in hypoglycemia and a bigger burden of hyperglycemia relative to time spent in hypoglycemic range.^4,5 Occurrence of hyper and hypoglycemia with current CLC, the burden associated with current CLC and limitations of eligible populations to use current technologies indicate the need for further innovation. Important variables in this context with potentially significant effect sizes are physical activity and meals. Significant efforts have been made to model the aforementioned parameters and understand their effect on glucose–insulin dynamics by incorporating/adapting them into CLC simulators.^6,7

Even though the day to day effect of physical activity on glucose status poses major challenges for T1D, limited research in this area has been conducted.⁸ Since obesity and type 2 diabetes mellitus (T2D), 2 major disorders of whole body energy balance, continue to rise worldwide, most research to-date regarding physical activity capture device use has been pursued in these 2 disorders.^9,10 T2D is characterized by hyperglycemia for a significant proportion of the day with the majority of such subjects having limited hypoglycemia until the disorder is quite advanced in duration or associated with comorbidities.¹¹ Since weight control is one of the main clinical end points in both obesity and T2D, energy expenditure tracking in such patients requires devices and output quite different from those for T1D as discussed above and below.

Thus, a key challenge in this context for T1D is real-time quantification of physical activity; accomplishing this would enable development of more fully automated CLC. Here, we review the physical activity capture modalities currently utilized in CLC and the devices that have been/are being used for longer duration studies in T1D. Next, we review whether the physical activity capture devices used in CLC studies or other devices with potential for use in T1D have been used for longer periods of time in non T1D subjects. Finally, we discuss factors to be considered to enable physical activity capture device use for longer duration real-world CLC studies and clinical practice.

Motivations to use physical activity capture devices in T1D are to (1) limit hypoglycemia, (2) improve hyperglycemia, (3) achieve and maintain ideal body weight, and (4) promote Cardiovascular (CV) health.

Factors to Be Considered Prior to Using Activity Capture Technologies

A recent review has provided a framework to make a decision about which technology to use to capture physical activity in general.¹⁰ Extending this line of reasoning to T1D would require more information about representative cohorts of T1D.

Cohort Characteristics

Patients with T1D experience hypoglycemia more frequently than any other disorder in clinical practice with the risk being variable; some subjects experience considerably more hypoglycemia than others.¹¹ The heterogeneity may occur due to different factors with some patients predisposed by high daily physical activity. In recent years, with improvements in the management of T1D, an increasing number of patients have been undertaking extreme and/or competitive sports. It is thus important we have information about physical activity profiles from large population databases of T1D; such information is currently lacking. Brazeau et al administered a detailed questionnaire to 103 subjects with T1D and reported that most adult patients with T1D do not participate in regular physical activity due to multiple barriers: (1) fear of hypoglycemia, (2) work schedule, (3) poor fitness level, and (4) loss of control over diabetes.^12,13 These are important issues for prospective studies with patients afraid of hypoglycemia more challenging to recruit for clinical studies of tight glycemic control. It is well appreciated that moderate to high-grade aerobic physical activity can increase the risk of delayed hypoglycemia with a recent study reporting increase in overnight hypoglycemic events by 43% following moderate to vigorous exercise during the day and an additional 31% occurrence on the following day.¹⁴

Impact of Aerobic Versus Anaerobic Exercise on Plasma Glucose Concentrations

Aerobic and anaerobic/resistance activity may have contrasting effects on the duration and magnitude of acute and delayed postexercise glucose lowering. A recent meta-analysis has summarized results from 33 studies in T1D and reports large acute effects and smaller chronic effects on glycemic control.¹⁵ Larger glucose lowering effects result from aerobic rather than high intensity/resistance exercise. Thus, aerobic exercise is more favorable for glycemia improvement compared to anaerobic exercise in trained individuals. In addition, glucose lowering is rapid with aerobic exercise.^16,17 However, only 4 studies have studied the impact of both aerobic and anaerobic training in the same patients with T1D.^18-21 Reports from these studies to date have not provided estimates of rate of change in plasma or interstitial fluid glucose in the same individual with aerobic and resistance activity under otherwise comparable dietary and insulin managements conditions.

Physical Activity Input in Closed-Loop Studies to Date

So far, 3 groups have conducted such studies designed to prevent exercise induced hypoglycemia (Table 1).^22-24 Turksoy et al²⁴ utilized a multivariate adaptive artificial pancreas (MAAP) algorithm in an initial cohort of 6 patients and subsequently deployed an integrated multivariate adaptive artificial pancreas (IMAAP) in 3 patients. The IMAAP represents an augmentation of MAAP with an early alarm system for hypoglycemia. CGM readings were manually entered every 10 minutes with wireless transmission of SenseWear Pro3 into a laptop with the control algorithm. Insulin recommendations were manually entered by the study team into the insulin pump. A significant decrease (P < .01) in hypoglycemic events were observed with MAAP system, which further improved with IMAAP system with no hypoglycemia events.

Table 1.

Characteristics of Closed-Loop Studies in T1D With Physical Activity Incorporation to Date

Study	Subjects/centers (n)	Device	Duration of closed loop	Crossover design	Type of exercise	Type of algorithm	Results
Turksoy et al24	3/1	SenseWear Pro3 Armband (BodyMedia Inc, Pittsburgh, PA)	32 hours	No	One 20-min treadmill running; moderate to intense intensity. Speed and incline gradually increased based on target heart rate.	Adaptive generalized predictive control (GPC)	Hypoglycemic events: 0 CGM (mg/dl) <70 = 0% 70-180 = 50.2% 180-250 = 31.1% >250 = 18.7%
Turksoy et al24	9/1	SenseWear Pro3 Armband (BodyMedia Inc, Pittsburgh, PA)	32 hours	No	One 20-min treadmill running; moderate to intense intensity. Speed and incline gradually increased based on target heart rate.	GPC N = 6 Multivariable adaptive artificial pancreas (MAAP) N = 3 Integrated multivariable adaptive artificial pancreas (IMAAP)	Severe hypoglycemic events (<55): MAAP = 0 IMAAP = 0 Hypoglycemic events (55-70): MAAP = 3 IMAAP = 0
Breton et al23	12/1	Polar HR monitor (Polar, Lake Success, NY)	26 hours	Yes	One 30-min cycle ergometer; work exertion 9-10 on Borg scale. Work load adjusted every 5 min to maintain exertion.	Control to range (CTR) vs CTR + HR	Hypoglycemic events in exercise: CTR = 2 CTR + HR = 0 Hypoglycemic events in recovery: CTR = 2 CTR + HR = 1 Hypoglycemic events overnight: CTR = 1 CTR + HR = 0
Stenerson et al25	18/1	Zephyr BioHarness 3 Chest strap (Zephyr Technology, Annapolis, MD)	2 hours	Yes	Four 25-30-min soccer-related activities; intensity varied; 5-min rest between periods.	Accelerometer augmented predictive low glucose suspend (PLGS) algorithm with 30-minute prediction horizon	Hypoglycemic events in exercise: Off PLGS = 6 On PLGS = 3 Hypoglycemic events in recovery: Off PLGS = 4 On PLGS = 2

Breton et al augmented their control to range (CTR) CLC algorithm with heart rate (HR) information (CTR-HR) and compared this to CTR in 12 T1D subjects in a crossover randomized clinical trial (RCT).²³ HR was incorporated manually by pressing the exercise button on the Diabetes Assistant when HR was 125% above resting thus triggering the CTR + HR algorithm. When HR was below 125% resting, the exercise button was pressed to return to the CTR algorithm. HR input into the CTR significantly reduced the decrease in blood sugar during exercise (P = .022). Whereas with CTR alone there were 2 hypoglycemic events during exercise, with HR input there were no hypoglycemic events during exercise (P = .16).

Extending their previous work on predictive low glucose suspend (p-LGS),²² Buckingham used Zephr BioHarness 3 accelerometer data with a predetermined threshold to turn the p-LGS algorithm on based on preset increased physical activity data in 18 T1D patients during 2 separate soccer sessions.²⁵ CGM and average observed activity was manually entered every 5 minutes into the control system and insulin suspended if CGM glucose was projected to decrease <80 mg/dl during a 30 min prediction horizon. Insulin was also suspended if (1) CGM glucose was <180 mg/dl during the activity period and (2) activity was ≥0.3. Study team members watched screens displaying data continuously. Hypoglycemia occurred in 3 subjects on-algorithm compared to 6 subjects off-algorithm during the exercise bout (P = .45). Hypoglycemia occurred in 2 subjects on-algorithm compared to 4 off-algorithm postexercise (P = .66).

The reduction in hypoglycemic events is an important strength in all 3 CLC studies. In addition, one of the CLC studies also investigated a variety of exercise intensities. These studies also have several potential limitations that need to be considered to aid in the advancement of future closed-loop studies especially for real-world studies and clinical translation. Each of these studies (1) involved only 1 center, (2) involved small sample sizes, (3) was short in duration, and (4) focused primarily on aerobic exercises.^23-25 These studies were also limited by manual entry of physical activity output into the controller^23,25 and manually entered CGM data into the controller.²⁵ Two of the 3 studies used laptops to run the control algorithm.^24,25 Different types of control algorithms incorporating physical activity have not been tested head to head in any study.

Other important points from these studies are (1) 2 CGM per subject,^23,24 (2) used different physical activity capture devices, (3) heterogeneity of type and duration of exercise thus limiting comparison between studies, (4) different control algorithms used across studies, (5) heterogeneity of methodology by which activity data modulated insulin delivery, and (6) heterogeneity in effect size.^23-25

Table 2 shows characteristics of the devices used to capture physical activity data in patients with T1D in closed-loop studies so far. In summary, (1) site of placement varied for each device utilized,^23-25 (2) accelerometers were used in 2 of 3 devices,^24,25 (3) range of capture differed between 2 accelerometers, with the 16 g sensor being capable of activity capture at a high intensity of exercise,²⁵ (4) activity capture for water-based activities is limited,^23-25 (5) data storage capabilities varied considerably,^23-25 and (6) although all devices have real-time display, only 2 devices allow for wireless transfer.^24,25

Table 2.

Devices Used in Patients With T1D in Closed-Loop and Non-Closed-Loop Studies.

Device	Site of placement	Measurement variables	Accelerometer characteristics	Waterproof	Storagea	Transmission type	Real-time display	Incorporated in closed loop
SenseWear Pro3 Armband (BodyMedia Inc, Pittsburgh, PA)68,69	Back of upper right arm	Activity	Two-axis MEMS ±2 g	Splash- resistant	10 days	USB or wireless station—ABS	Yes	Yes
Zephyr BioHarness 3 (Zephyr Technology, Annapolis, MD)70,71	Chest strap, compression shirt, or Biomodule holder	Heart rate 25-240 BPM activity	Three-axis MEMS ±16 g 100 Hz	Water-resistant up to 1 m	20 days	802.15.4 (Zigbee) Bluetooth 2.1 Bluetooth LE	Yes	Yes
Polar RS800CX HR monitor (Polar, Lake Success, NY)72,73	Wrist	Heart rate 15-240 BPM	NA	Water-resistant, but HR capture is not functional in water	40 hr, 30 min	Infrared transfer with Polar IrDA USB Adapter	Yes	Yes
ActiGraph wGT3X-BT wireless activity monitor (ActiGraph, Pensacola, FL)74,75	Ankle, thigh, waist, wrist	Activity	Three-axis MEMS ±8 g 30 Hz to 100 Hz	Water-resistant to 1 m for 30 min	120 days	USB, Bluetooth LE	No	No
Actiheart (Cambridge Neurotechnology, Cambridge, UK)76,77	Chest (V1 or V2 and the second electrode is placed approximately 10cm away on the left side at V4 or V5)	Heart rate 30-250 BPM activity	Single-axis accelerometer Range: >±2.5g Frequency range: 1Hz to 7Hz (3dB) Sampling rate: 32Hz	Waterproof	21 days	USB reader/charger	No	No
GENEActiv device model 1.1 (Activinsights Ltd)78,79	Wrist	Activity	Three-axis MEMS ±16 g	Water-resistant to 10 m	45 days	USB, Bluetooth compatible	No	No
Actical (Phillips Respironics, Bend, OR)80	Waist, wrist, ankle	Activity	Three-axis 0.05 to 2 G	Water-resistant to 1 m for 30 min	12 days raw; 194 days for epoch mode 1 s + steps	9-pin RS-232 serial port (standard with ActiReader)	No	No

^aStorage capabilities can decrease when measurement variables and sampling rate increases.

Current Studies Including Physical Activity Signal in CLC

We searched multiple global clinical trial registries with combinations of terms ensuring maximal retrieval of active studies. In a study being conducted by the University of Virginia, 20 adolescent T1D patients have been undergoing CLC using HR as the exercise indicator. This is a crossover RCT consisting of an experimental arm using the CTR system enhanced with HR data stream during exercise compared with standard CTR. The CTR HR button will be activated when HR ≥ 140 beats per minute (bpm) during exercise and deactivated when ≤ 140 bpm.

Subsequent sections of this manuscript provide insight into longer term studies of (1) physical activity in T1D and (2) physical activity capturing device use in any cohort studied for metabolic reasons. We believe that such studies will enable better design of real-world CLC studies of T1D incorporating real time physical activity data.

Physical Activity Capture Device Use for Longer Periods in T1D

Newton et al enrolled 78 T1D adolescent patients in a 12 week RCT to assess whether pedometer use could increase physical activity and reduce weight.²⁶ Results did not show increase in PA or significant reduction in BMI and A1C. Suboptimal adherence to intervention (pedometer use was terminated by 37% before the end of the study) may explain these results to an extent. Faulkner et al recruited 12 T1D sedentary adolescents for a 16 week personalized exercise prescription study with the prescription provided based on current fitness level for each subject.²⁷ During the study period, subjects were advised to perform 60min of moderate to vigorous physical activity each day wearing an ActiGraph™ Accelerometer (model GT1M, Pensacola, FL, USA). This intervention improved VO₂ max in the cohort.

Low-Grade Physical Activity May Impact CGM Data

We and others have previously shown that low-grade physical activity including standing and walking at 1.2 miles per hour has a significant impact on postprandial laboratory glucose measurements and CGM data.^28,29 Since most of the energy expended during daily life is related to nonexercise activity rather than formal exercise, it is important to capture low- and high-grade physical activity and incorporate both into closed-loop control systems.

Physical Activity Capture in T1D in Special Populations

Pregnancy in T1D is characterized by increased fetal and maternal morbidity with the morbidity linked to suboptimal glucose control.^30-33 Whereas early pregnancy is characterized by insulin sensitivity and hypoglycemia, advanced pregnancy results in increased insulin resistance and increasing hyperglycemia. Tracking physical activity offers an opportunity to increase insulin sensitivity by behavioral change and incorporation into CLC. Kumareswaran et al studied 10 pregnant T1D women for two 24-hour periods at 20 weeks pregnancy in a prospective crossover open trial.³⁰ All subjects were on open-loop therapy and self-managed T1D. During the 24 hour in clinic period, subjects performed 3 and 2 bouts of light and moderate physical activity respectively. Quantitative details of food intake during the 2 periods were not provided in the manuscript. Energy expenditure during the 2 periods was not different. Subjects performed longer duration light physical activity during free living compared to more moderate activity during in clinic periods. Glucose control overnight was better following the in clinic period with less hyperglycemia and lower mean glucose reflecting the delayed glucose lowering effect of more intense physical activity.

Devices Used in Long Duration Clinical Studies Irrespective of Disease Type

Physical activity was assessed in 47 obese and sedentary T2D subjects in a prospective 24 week RCT.³⁴ All subjects were randomized to intervention or control groups. In the intervention group subjects wore a pedometer (Yamax SW-200, Yamax Corporation, Tokyo, Japan) for the first 16 weeks and were followed for 24 more weeks without the device. Physical activity was observed to be increased during the 16 weeks of intervention but declined at the end of the next 24 weeks. Decrease in FPG and HBA1C was observed in subjects who were on antihyperglycemic medications but not on diet alone.

General Considerations Important in Translating Physical Activity Data Capture to Real-World Studies

Choice of Accelerometer

Currently although nearly all accelerometers used in sensing technologies are based on microelectromechanical systems chips,³⁵ they present fundamental differences in performance and accuracy.^35-45 Most devices measure all 3 axes, X, Y, and Z. Low amplitude movement needs to be detected with an accelerometer with a low G setting (nonexercise activity is generally around 1 to 2 G).⁴⁶ High intensity movements such as exercise are detected with a high G setting (10 to 15 G). For subjects undertaking intensive exercise, an accelerometer with a high G setting is mandatory. A solution would be an accelerometer with self-regulating G values⁴⁷ where the accelerometer electronically self-senses the degree of movement and adjusts its own range

The data gathering rate of accelerometer is a second important consideration.⁴⁰ To illustrate, if daily movement was only catalogued once every 10 seconds, theoretically a person could have a short sprint (50 or 100 yards) which would not be detected by the device and would have no effect on the insulin delivery algorithm. Based on this, it could be argued that gathering the maximum amount of data at the maximum rate is the key. However, if data are gathered at very high frequencies, the volume of data is high and speed of acquisition is fast, thus limiting filtering to avoid noise and interference signals. Also with very high data acquisition rates, the internal buffers on the electronic systems need to be large (adding unnecessary cost).^48-50 Overall the rate of data gathering needs to be appropriate for the signal being traced. When walking briskly, a full stride (right and left heel strike) occurs about once per second which is 2 Hz. The duration of a heel strike and the subsequent z axis (vertical) elevation is about one-tenth of that. Hence 20 Hz is necessary to fully characterize walking which is the predominant bipedal activity of free living people.

Attachment to the Patient

Although the current proliferation of wearable activity tracking devices might suggest that the activity monitor can be worn in a variety of places such as the around the neck or on the waistband, studies do not confirm the reliability of such devices with the trunk representing the best site and adherence to skin essential for a high quality signal.^39,40 An implantable accelerometer⁵¹ could be inserted under the skin attached to the glucose sensing electrode or alternatively integrated into the insulin delivery pump.⁵²

Data Integrity

Corruption of the activity data stream when activity data is transmitted via a low-power Bluetooth (BT) signal to the integrated system may affect the ability to use the information safely to improve insulin delivery.^53,54 The signal may also be interfered with by other BTLE devices or malicious interference. Currently as CLC systems are being developed, there is a need to share data within and between sub components of the CLC system and the clinical team. With the growing movement toward gathering multiplex data, the need for patient data protection is critical.⁵⁵ A person being very active could be geo-traced in real time, raising privacy concerns.⁵⁵

Future Technologies, Patient Factors, and Prospects for Future Studies

Long-term incorporation of physical activity data as a nonglucose input is necessary for CLC. Since adherence to single function device is suboptimal, it is vital that physical activity monitoring devices be incorporated into 1 of the devices required for CLC or daily subject functioning: CSII, a CLC algorithm device, a CGM, or a smartphone. Cellnovo has developed an insulin pump approved for use in the United Kingdom and France that incorporates a 3-axis accelerometer capable of monitoring physical activity.^52,56

With the computational power enabling both integration of data and high-level analysis now available in smartphones, there has been great interest in using these as an intelligent hub^57,58 for the integration of physiological signals in CLC.^59-64 Furthermore, cloud-based data systems enable the health care team to access and analyze CGM, CSII, and CLC data. However, the use of smartphone as the data center and the core of a CLC system has some additional challenges with respect to cybersecurity, power consumption and the human nature of misplacing the device or losing it.⁵⁵ Hekler et al compared 3 Android smartphones (HTC MyTouch, Google Nexus One, and Motorola Cliq) to the ActiGraph GT3X+ accelerometer in laboratory and free-living studies.⁶⁵ The principal findings include (1) raw counts of the Android smartphone and ActiGraph correlate in laboratory and free-living settings, (2) raw counts of the Android smartphone correlate with each other, and (3) the phones can be worn on the hip or in a pocket and still provide reliable estimates.

Considerations for Future Studies

Patient factors to be considered in longer term studies of CLC include (1) physical activity profile of patient: swimming, vigorous walking/running, cycling, anaerobic exercises, (2) specific context such as pregnancy, (3) comorbidities such as cardiovascular disease and autonomic dysfunction, and (4) willingness to use specific devices.

The NIH has increased its support of closed-loop studies studying physical activity capture with additional grants initiated in the second half of 2015 to accelerate progress.⁶⁶. The NIH will also fund real word clinical trials CLC through consortia to conduct real-world RCT starting in 2016.⁶⁷ Thus the stage is set for further acceleration of nonglucose physical activity data input into CLC over the next decade resulting in more advanced, safer and more effective CLC for T1D.