Exploring Blood Glucose Variation over Geographical Space

Abstract

Background:

Type 2 diabetes mellitus is known to be associated with environmental, behavioral, and lifestyle factors such as a sedentary lifestyle, overly rich nutrition, and obesity. However, the day-to-day human–environment interactions and real-life activities that cause an individual's blood glucose to fluctuate remain relatively unexplored, owing in part to data collection challenges. This article presents a novel data collection system that overcomes these challenges and allows exploration of the spatial correlates of blood glucose fluctuation.

Methods:

An automated monitoring system was developed combining a Global Positioning System (GPS) receiver with a continuous blood glucose monitor. The GPS was used to elicit a second-by-second accounting of an individual's daily activities alongside blood glucose measurement every 5 min. A pilot study of 40 diabetes patients was conducted over a 72-h period. Geographic Information System software was used to generate blood glucose maps, incorporating methods to deal with scale issues, overlapping data, and to protect subject identity.

Results:

Individual blood glucose variation maps revealed a variety of distinct patterns. Most subjects had at least two major anchor points in their life combined with a variety of other activity locations at varying distances from home, many associated with quite distinct low or high blood glucose values. Further statistical analysis revealed location and distance from home were significantly correlated with blood glucose variation—although the strength and direction of the effect was quite mixed.

Conclusions:

Results suggests that blood glucose and space/location are highly correlated and should be considered further as a lifestyle-related risk factor for diabetes patients. In the future, patients and caregivers may benefit from individualized visualization tools that help identify problematic locations that require special attention.

Introduction

T he number of people with type 2 diabetes mellitus (T2DM) is increasing dramatically around the world, with Canada being no exception.¹ T2DM is characterized by insulin resistance and/or abnormal insulin secretion, leading to hyperglycemia. Blood glucose (BG) is commonly managed by increasing exercise and dietary modification, followed by oral medications plus insulin or insulin alone if needed. Long-term complications from high BG can include increased risk of heart attacks, strokes, amputation, kidney failure, eye diseases, and nerve damage.

In conjunction with genetic susceptibility, T2DM is known to be associated with environmental, behavioral, and lifestyle factors such as a sedentary lifestyle, overly rich nutrition, and obesity,^2,3 changes closely linked to rapid rates of urbanization in the last century.⁴ Such associations are well known at the population level, but the day-to-day human–environment interactions, behavioral choices, activity spaces, and real-life events that cause an individual's BG to fluctuate remain relatively unexplored. Specific individual-level investigations of this kind are being called for to strengthen the search for the causality of disease.^5,6

The trouble is that such individual-level information, especially under free-living conditions, is costly, invasive, and burdensome to obtain using traditional survey methods and fingertip blood sampling.⁷ The recent development of wearable continuous glucose monitors overcomes part of this problem by automatically estimating BG at regular intervals such as every 5 min. In addition to successfully assisting patients self-manage their BG,^8,9 these devices are beginning to be used to examine lifestyle correlates of BG fluctuation, such as the effects of medications,¹⁰ food,¹¹ and exercise⁷ over time throughout the day.

Developments in wearable location sensing devices such as Global Positioning System (GPS) devices are opening new opportunities for tracking humans over space. When GPS coordinates are tracked continuously they provide a highly accurate accounting of the environmental conditions individuals are exposed and the types of activities and movements people engage in during everyday life. Both these factors have direct impacts on health. Recognizing this, growing numbers of studies have used GPS to automatically detect activity and travel patterns in urban areas,¹² including physical activity,¹³ traditionally tracked using self-reported diaries and accelerometers.¹⁴ This article explores how the pairing of GPS tracking with a biophysical sensing device such as a continuous glucose monitor opens entirely new opportunities for examining how a person's BG fluctuates over space and in identifying possible new spatial correlating factors and serves as a complement to traditional time-series analysis.

Geographic Information System (GIS) is ideally suited to spatial analysis of this type. A GIS is a computer system for the input, storage, maintenance, management, retrieval, analysis, synthesis, and output of geographic or location-based information.¹⁵ An entire field of geographical epidemiology has evolved focused on the use of GIS for creating disease maps and for ecological analysis,¹⁶ including sophisticated spatial analysis of disease occurrence and contributing environmental factors.⁶ Extensive texts on the advantages of GIS are available,¹⁷ foremost being the ability to convey a picture that “tells a thousand words.” Melnick and Fleming¹⁸ commented that GIS technology “has unleashed nearly unlimited potential for depicting relationships between environment, behaviour, health and disease.”

The pitfalls of using GIS technology for public health are of equal concern, especially in light of new types of data available for analysis in this article. To avoid hiding critical information in a fog of detail, maps inevitably offer a selective, incomplete view of reality.¹⁹ Visual distortions may result, and interpretative mistakes made, including inferring causation from apparent spatial correlation, incorrectly make inferences about individuals from population data, or incorrectly identifying hot-spots of high disease rates in small areas.^15,20 Scale effects may also arise, wherein large-scale maps may exhibit or inhibit different spatial patterns compared with small-scale maps.⁶ Rytkonen⁶ cautioned, however, that use of high-resolution spatial data may endanger the privacy of individuals, with this being especially so for GPS data that are precise enough to reveal a subject's home or work location. Armstrong et al.²¹ proposed several key methods for geographically masking subject identity in such cases, including record-transforming and attribute-transforming masks that involve strategic aggregation, suppression, sampling, or displacement of records and/or attributes.

Objectives

The objectives of this article are both methodological and empirical. First, it presents a novel data collection methodology combining a physiological sensor (BG) with a geographic location sensor (GPS) for continuous observation of environmental exposure and human activity under free-living conditions, overcoming the costs and burden issues associated with past techniques. Empirically, the objective is to provide the first spatial map of an individual's BG as it varies throughout the day, allowing speculation of potentially correlating factors. In the process, this article also contributes several specific methods for dealing with well-known GIS mapping challenges, including dealing with overlapping GPS points, masking data to protect subject identity, and handling missing/distorted GPS points. Overall, this article contributes to a recent growing body of research attempting to explore more closely the relationship between human geographies and health in everyday life.

Subjects and Methods

Patient monitoring system

The automated patient monitoring system developed for this study includes a GPS receiver and the CGMS^® System Gold™ from Medtronic (Minneapolis, MN). The CGMS is an approved medical device for outpatient use in the United States and Canada and has been shown useful in exploring daily stressors on BG.²² The CGMS consists of a small clip-on storage device (90×70×22 mm) and a wired under-the-skin sensor that provides automatic measurement of interstitial fluid glucose every 5 min, for up to a 72-h period. The sensor is inserted under the skin in the lower torso area by a nurse and secured with IV3000^® tape (Smith & Nephew, London, UK). Calibration by finger stick BG is required four times per day, done manually, and entered into the device by the patient. Data were downloaded from the device at the end of the study period and manually uploaded to a computer server.

A Bluetooth^® (Kirkland, WA)-enabled GlobalSat^® BT-338 SiRF Star III GPS receiver (72.5×40.4×23 mm) (Philips, Eindhoven, The Netherlands) with clip-on holster was used to track each subject's location second by second. This wearable device broadcasts a National Marine Electronics Association 0183 standard sentence that includes longitude, latitude, altitude, speed, heading, and signal quality variables (number of satellites, displacement). This sentence was transmitted via Bluetooth to a BlackBerry^® (Waterloo, ON, Canada) 7520 smartphone (114×75×28 mm) with 1,600-mAh extended battery. A custom Java program was developed to log these data until they reached a set size (the default is 50 kilobytes), compress them, and then transmit them to a remote project server at regular intervals (typically, every 5–10 min), where they are stored in a MySQL database.

An automated activity detection algorithm on the server was used to initially detect a subject's stops/activities (with location, start/end time) and trips/movements (by travel mode, start/end time). An interactive web-based diary was used to display these results, as described in Clark and Doherty.¹² A research assistant helped subjects to review the results on screen, confirm the accuracy of diary events, modify them if needed, and add more specific information such as location names (e.g., Work, Home). Most importantly, for this article, this allowed subjects to account for their location when the GPS signal was lost because of signal outage in dense cover—mostly within buildings.

Pilot study

A pilot study using the above multisensor data collection system was carried out with 40 diabetes patients from the Toronto Rehabilitation Institute (Toronto, ON, Canada). The study was approved by the Institute's Research Ethics Board. Patients were recruited via advertisements and word-of-mouth. Patients provided informed consent and were given incentives valued at $30 CAD.

A 72-h monitoring period was chosen for the pilot study, reflecting the limits of the CGMS. This period covered a 4-day period: on the first day, patients were set up at the rehab clinic at a time convenient to them, resulting in a partial day of monitoring; the next 2 days (day 2 and 3) were full days of monitoring; and on day 4, they returned to the clinic for debriefing, again resulting in a partial day of monitoring. Two or three patients were set up and monitored per week over a 5-month period.

A 2-h upfront interview on the first day was used to hook up the various sensors and train subjects on their use. A full hour of this time was spent with the CGMS, owing to its invasive nature. Patients were given the option of wearing devices on their belt, in a small fanny pack, or in their own purse/bag. A one-sheet pamphlet provided instructions on device usage and what to do on each day. Outside of recharging the smartphone and GPS receiver on a nightly basis, the only time patients were instructed to manually interact with the devices was if the smartphone issued a long “buzz.” In this case, an automated message would appear instructing patients to call a research assistant (by pressing “OK”) or to help deal with a hardware problem that had been remotely detected (e.g., low battery, disabled device). A full-time research assistant monitored all incoming data on the server, addressed any equipment problems, assisted patients, and took detailed field notes on any issue that arose.

Results

Subject and data characteristics

Of the 40 original subjects, 34 ended up providing complete data, including 68 full days of tracking. Two subjects ceased CGMS monitoring after day 2, reporting irritation with the tape-on sensor, and for four other subjects GPS data were deemed unusable, owing to an early technical problem related to data storage. The 34 subjects included in this analysis ranged in age from 32 to 75 years (mean of 56 years), weighed between 45 and 147 kg (mean of 88.4 kg), and included equal numbers of men and women. The duration since diagnosis with T2DM ranged from 3 months to 46 years (mean of 9.5 years). Glycated hemoglobin levels were available from 24 subjects and ranged from 5.1% to 10.1% (mean of 7.1%).

The CGMS provided 288 measurements per subject per full day of tracking (one every 5 min), with a mean of 7.3 mmol/L and SD of 2.9. The mean absolute difference between the CGMS and manual glucometer readings taken by subjects at least four times per day was 0.76 mmol/L (SD=0.97), and the associated correlation coefficient (Pearson's) was 0.91. An average of 20.2 h of GPS data points per subject per day was recorded for the 34 subjects across 68 full days of tracking. The subsequent use of the GPS-supported time diary resulted in a full 24-h accounting of the subject's location, activity, and travel patterns.

Glucose data analysis

Significant creative efforts were needed to build a BG variation map that minimized distortion of the data and protected subject anonymity. BG values measured every 5 min for each subject were first merged with GPS coordinates (longitude and latitude) by time of day. These data were then imported into the GIS (both Esri^® [Toronto] ArcGIS and Quantum GIS [qgis.org] softwares were used), and a shape file was created for each subject depicting GPS points (longitude and latitude) with BG value as an attribute. Graduated colors were then assigned to each point on the following 5-point BG scale (in mmol/L):

• Low=2.20–3.99 (blue)

• Normal=4.00–6.99 (green)

• Normal–high (if user has eaten)=7.00–11.99 (yellow)

• Medium–high=12.00–16.99 (orange)

• High=17.00–22.20 (red)

This scale is based in part on information provided in the Canadian Diabetes Association Clinical Practice Guidelines for the Prevention and Management of Diabetes.²³

From this point, several key challenges needed to be addressed:

1. How to display multiple glucose values stacked up at a single location

2. How to represent BG values when the subject is on the move

3. How to maintain subject anonymity (so their data can be shown here)

The first challenge relates to the desire to avoid hiding critical information in a fog of detail,¹⁹ with the fog being created by huge clusters of points at single locations, such as home and work. The example in Figure 1a clearly shows this. Not only do the points cluster, but they end up overlapping, such that earlier values are hidden by latter values. The visual distribution of BG values at these locations is difficult to determine, if not severely biased. One solution to this involves creating pie graphs depicting the distribution of glucose values in a single clusters of GPS points, with the overall diameter of the pie representing the number of points in the cluster (and hence relative duration of time spent at the location), as shown in Figure 1b. These were created in an ad hoc fashion, wherein any cluster of 12 points or more (1 h) within a very small area (less than about 50 m) was selected and replaced by a pie graph with centroid equal to the spatial median of the original cluster of points. Conveying the relative size of the pie chart is also an effective defense against incorrect identification of hot-spots.

FIG. 1.

Clustering of 5-min blood glucose readings at a single location and their conversion to a pie graph. (a) All Global Positioning System (GPS) points plotted by blood glucose values (in mmol/L), highlighting the magnified area representing a cluster/fog of many overlapping points in any are the size of a typical suburban home footprint. (b) Final map showing the pie graph that replaced the cluster of points, directional arrows, and street network.

The second challenge relates to the depiction of GPS points when subjects are on the move, which become difficult to interpret when subjects cross paths or use the same route twice (see Fig. 1a). This is most easily remedied by connecting subsequent points by lines and adding a marker that indicates direction of travel, as shown in Figure 1b and beyond.

A solution to the third problem presented itself once the pie graphs were implemented. Plotting of raw GPS data, especially if overlaid with the road network or aerial photos, makes identification of a subject's home very obvious because it will most likely be the location of the biggest cluster of points. Two types of data transformations were implemented to deal with this. First, replacing the GPS clusters with a single large pie obscures the actual location, given the much larger pie symbol—a form of record-transformation mask, as suggested by Armstrong et al.²¹ To further obscure the location, without jeopardizing the pattern in any significant way, the pie centroid can be manually adjusted to obscure the actual location—a type of attribute-transformation mask. This was done by choosing an undisclosed random degree and random distance (>100 m) to move the pie centroid. Map base layers were also kept simple and to a minimum, so that buildings and exact locations were not displayed.

The first map generated displayed data from all 34 subjects, but this proved unusable owing to significant data overlap and scale issues that largely obscured patterns. A more effective and revealing approach is to generate separate maps for each subject. An illustrative example is shown in Figure 2. The significant variation in BG readings at these locations, and during the course of travel, is quite effectively conveyed, without overly obscuring the data or breaking patient anonymity. Note, however, the scale effects that emerge when examining Figure 2. When a section of the map is expanded in scale as shown in the low portion of Figure 2, an important BG glucose walk trip emerges for analysis (at top of map), obscured in the full-scale map.

FIG. 2.

Example of an individual blood glucose map at two scales of analysis. Note that the smallest points represent single color-coded blood glucose readings (in mmol/L) taken every 5 min, most prevalent during periods of travel or walking, and that increasingly larger circles are indicative of multiple blood glucose readings at stationary activity locations of increasing periods of time. GPS, Global Positioning System.

Visual inspection of the maps for each of the 34 subjects revealed four typical patterns of activity and BG variation:

1. Dispersed activity pattern: with more than three main locations often chained together. Figure 2 is a typical example highlighting several high BG locations compared with Figure 3, which has several low BG locations.

2. Dual anchored activity patterns: typical of full-time commuters, consisting of two main locations (work and home) with several radiating short-duration activities stemming from home, with mixed low/high BG values at outlying locations (home often being most stable). Figure 4 is a typical example.

3. Home-centered activity patterns, with more than four return-home trips to outlying activities of modest duration and mixed BG levels. Figure 5 is a typical example.

4. Local neighborhood focused: with only occasional activities in close proximity and typically stable BG values with a few lows.

FIG. 3.

Example of a “highly dispersed” blood glucose map. Note the dispersed locations, some having low blood glucose values (in mmol/L), others more stable. GPS, Global Positioning System.

FIG. 4.

Example of a “dual-anchored” blood glucose map. Note that the center (home) location has fairly stable blood glucose values (in mmol/L), whereas the outlying main location has a mix of lows and highs. GPS, Global Positioning System.

FIG. 5.

Example of a “home-centered” blood glucose map. Note the various round trip activities radiating from home, with locations having mixed high and low blood glucose values (in mmol/L). GPS, Global Positioning System.

The point here is not to draw conclusions about activity patterns so much as to visually illustrate the wide variety of BG values that occur over space and at distinct locations.

As a complement to this visual analysis, two types of statistical analysis were conducted. First, an analysis of variance was conducted on the mean BG values at locations where subjects spent a meaningful amount of time. Locations were identified as spatial clusters of 48 or more BG values (i.e., more than 4 h in duration) within a small radius typical of an activity stop (usually well within 40 m). This approach yielded several distinct locations for each subject of meaningful duration. Results revealed that all but one subject had at least one location that exhibited significantly different mean BG values, and most had several, with often dramatically varying results. For example, the patient depicted in Figure 2 had three distinct activity locations of more than 4 h. The most northerly location had a mean BG of 16.3 mmol/L (SD=1.8, n=217), the most southerly had 8.1 mmol/L (SD=3.5, n=104), and a midway location had 13.2 mmol/L (SD=6.4, n=356), values that are very significantly different (F=98.9, P<0.000), in addition to being visually distinct on the map.

A second more exploratory analysis involved calculating the distance (straight line) between the location of every out-of-home BG reading and the subject's home location and correlating this distance with the respective BG value, using Pearson product-moment correlation coefficients (r value). The results were quite interesting: 10 patients had no significant correlation, whereas an equal number of patients had significant positive correlations (n=12) as negative (n=12), ranging from weak (0.11) to very strong (0.67). Thus, for most subjects, it appears that the farther one is from home, the more varied is the BG.

Discussion

Methodologically, the combination of GPS tracking and continuous BG monitoring appears to be an effective solution to obtaining sought-after individual-level environmental exposure, activity, and biophysical data under free-living conditions. The passive nature of the sensors is inherently less burdensome compared with manual reporting methods, and the frequency of observation leads to a much finer level of detail. With respect to analytical methods, several data transformation masks were effectively implemented to deal with the fog of data and scale issues and to protect subject identity, while maintaining the ability to discern patterns in the data using a GIS. These approaches afforded a unique new ability to explore individual-level variations in BG at a fine spatial/temporal scale, revealing potentially new correlating factors.

The resulting individual BG variation maps revealed a variety of distinct patterns. Most subjects had at least two major anchor points in their daily life, such as home and work. Others had several radiating activities from the home, had highly dispersed activities with considerable travel, or stayed close to home. Many of these locations were associated with quite distinct low or high BG values, whereas other locations were more stable. All but one subject had activity locations that were associated with significantly different average BG levels. Trips were similar in that BG ranged in values and changed during trips. Further statistical analysis revealed that the location of activities in one's life and the distance they are from the comfort of home are significantly correlated with BG variation—although the strength and direction of the effect are quite mixed and unique to each subject. Overall, this analysis strongly suggests that BG and space/location and are highly correlated and should be considered further as a potentially significant lifestyle-related risk factor for diabetes patients.

The causes of this association are open to speculation. Conducting activities at different locations and/or traveling further distances to get there exposes people to different environments, experiences, breaks in routine, and stressors at the destination or along the way that likely affect BG. These factors most likely interact with other well-known causes of BG fluctuation, thereby compounding the impacts, such as food intake, physical activity, medications, and other factors that vary by location or have accumulated up to that point in time. In particular, activities and routines inside the home are obviously much different than outside the home, and the further one goes from home, the more so it appears; people must drive to further locations (which can be stressful, limits movement, affects food intake, etc.) and are likely to conduct distinct activities (e.g., shopping, socializing, exercise) and consume different foods than in the comfort of their home, all likely contributing to this effect. Exploring causation further will require multivariate statistical modeling to sort out these effects, as well as more distinct identification of indoor activities such as exercise and small movements. A good start would be development of time-series models of BG as a function of the spatial variables afforded by this study (activity type, location, distance from home, etc.), controlling for food, medicine, and other lagged effects. More detailed accounting of physical activities and movements within buildings will require more precise GPS tracking (currently, the signal is often too inaccurate indoors for this purpose) or additional sensor data such as that provided by accelerometer or heart-rate monitors.

In terms of practical applications, spatial analysis studies such as this may eventually lead to refinements in diabetes self-management guidelines. For instance, in addition to time-dependent suggestions to manually measure BG before/after meals, ad hoc measurement after long trips, during long activities, or at problematic locations could be encouraged. However, the varied results of this article suggest that such recommendations need be highly individualized. To that extent, patients and caregivers may benefit from data visualization tools that add a spatial dimension to traditional time-series analysis.

Footnotes

Acknowledgments

The author acknowledges funding support from the Canadian Institutes of Health Research, The Health Technology Exchange, and Innovations at the University of Toronto. The following industry partners in this grant also provided valuable in-kind support in the form of equipment donation: Research in Motion, Telus Mobility, and Medtronic. Special thanks also go to Dr. Paul Oh and Susan Marzolini of Toronto Rehab and to research assistants Sheena Archie, Luke Cwik, and Sarah Resendes.

Author Disclosure Statement

No competing financial interests exist.

References

Canadian Diabetes Association: The Prevalence and Costs of Diabetes. Toronto: Canadian Diabetes Association, 2011.

Zimmet

, Alberti

, Shaw

. Global and societal implications of the diabetes epidemic. Nature, 2001; 414:782–787.

Mozaffarian

, Kamineni

, Carnethon

, Djousse

, Mukamal

, Siscovick

. Lifestyle risk factors and new-onset diabetes mellitus in older adults: the Cardiovascular Health Study. Arch Intern Med, 2009; 169:798–807.

Perdue

, Stone

, Gostin

. The built environment and its relationship to the public's health: the legal framework. Am J Public Health, 2003; 93:1390–1394.

Curtis

, Lee

. Spatial patterns of diabetes related health problems for vulnerable populations in Los Angeles. Int J Health Geogr, 2010; 9:43.

Rytkonen

. Not all maps are equal: GIS and spatial analysis in epidemiology. Int J Circumpolar Health, 2004; 63:9–24.

MacDonald

, Philp

, Harrison

, Bone

, Watt

. Monitoring exercise-induced changes in glycemic control in type 2 diabetes. Med Sci Sports Exerc, 2006; 38:201–207.

Yoo

, An

, Park

, Ryu

, Kim

, Seo

, Hong

, Shin

, Kim

, Choi

, Park

, Yu

, Baik

. Use of a real time continuous glucose monitoring system as a motivational device for poorly controlled type 2 diabetes. Diabetes Res Clin Pract, 2008; 82:73–79.

Hoeks

, Greven

, de Valk

. Real-time continuous glucose monitoring system for treatment of diabetes: a systematic review. Diabet Med, 2011; 28:386–394.

10.

Alemzadeh

, Loppnow

, Parton

, Kirby

. Glucose sensor evaluation of glycemic instability in pediatric type 1 diabetes mellitus. Diabetes Technol Ther, 2003; 5:167–173.

11.

Colette

, Ginet

, Boegner

, Benichou

, Pham

, Cristol

, Monnier

. Dichotomous responses of inter and postprandial hyperglycaemia to short-term calorie restriction in patients with type 2 diabetes. Eur J Clin Invest, 2005; 35:259–264.

12.

Clark

, Doherty

. A multi-instrumented approach to observing the activity rescheduling decision process. Transportation, 2010; 37:165–181.

13.

Elgethun

, Yost

, Fitzpatrick

, Nyerges

, Fenske

. Comparison of global positioning system (GPS) tracking and parent-report diaries to characterize children's time-location patterns. J Expo Sci Environ Epidemiol, 2007; 17:196–206.

14.

Doherty

. Emerging methods and technologies for tracking physical activity in the built environment. Bonnel

, Lee-Gosselin

, Zmud

, Madre

J-L

. Transport Survey Methods: Keeping up with a Changing World. Bingley, UK: Emerald, 2009; 153–190.

15.

Richards

, Croner

, Rushton

, Brown

, Fowler

. Geographic information systems and public health: mapping the future. Public Health Rep, 1999; 114:359–373.

16.

Kistemann

, Dangendorf

, Schweikart

. New perspectives on the use of Geographical Information Systems (GIS) in environmental health sciences. Int J Hyg Environ Health, 2002; 205:169–181.

17.

Cromley

, McLafferty

. GIS and Public Health. New York: Guilford, 2002.

18.

Melnick

, Fleming

. Modern geographic information systems—promise and pitfalls. J Public Health Manag Pract, 1999; 5:viii–x.

19.

Monmonier

. How to Lie with Maps, 2 ^nd. Chicago: University of Chicago Press, 1996.

20.

Green

, Hoppa

, Young

, Blanchard

. Geographic analysis of diabetes prevalence in an urban area. Soc Sci Med, 2003; 57:551–560.

21.

Armstrong

, Rushton

, Zimmerman

. Geographically masking health data to preserve confidentiality. Stat Med, 1999; 18:497–525.

22.

Derosa

, Salvadeo

, Mereu

, Ciccarelli

, Piccinni

, Ferrari

, Gravina

, Maffioli

, Tinelli

. Continuous glucose monitoring system in free-living healthy subjects: results from a pilot study. Diabetes Technol Ther, 2009; 11:159–169.

23.

Canadian Diabetes Association Clinical Practice Guidelines Expert Committee: Canadian Diabetes Association 2008 clinical practice guidelines for the prevention and management of diabetes in Canada. Can J Diabetes, 2008; 32,Suppl 1:S1–S201.