Advertisement

Monatshefte für Chemie - Chemical Monthly

, Volume 149, Issue 9, pp 1659–1669 | Cite as

The accuracy of continuous glucose monitoring system by the athlete with diabetes mellitus type 1

  • Šárka Vokounová-Honsová
  • Tomáš Navrátil
  • Eva Kohlíková
Original Paper
  • 53 Downloads

Abstract

This article deals with assessment of accuracy of the widely used glucose sensors MMT-7002 (Medtronic, USA), for continuous monitoring of glucose levels. The samples were gained from a 35-year-old sportswoman with diagnosed DM type 1. The sampling was carried out by her common daily as well as extreme physical and sport activities. The evaluated data were collected during 45 months, 56 sensors were used and 603 calibrations were carried out. The tested glucose sensor operates on the principle of a needle sensor, which electrochemically evaluates the reaction products catalyzed by glucose oxidase. Accuracy was evaluated by Clarke error grid analysis and the results were compared with the accuracy given by the sensor manufacturer. The measurements carried out by the tested sensor in the range of the normal glucose levels, realized under usual life conditions and by sport activities were even more accurate than those given by the manufacturer. However, in the regions of hypoglycemia or hyperglycemia, the number of errors was higher than specified by the manufacturer. The continuous glucose monitoring proved to be sufficiently accurate for prevention of acute complications in patients with DM type 1 during their common as well as extreme sport and physical activities.

Graphical abstract

Keywords

Continuous glucose monitoring Continuous glucose-error grid analysis Diabetes mellitus Glucose oxidase response Practical accuracy Sensor 

Introduction

Diabetes mellitus (DM) is a chronic metabolic disease. At present, approximately 8% of the total population of the Czech Republic (i.e., from 10 millions) are diabetics. According to the Institute of Health Information and Statistics of the Czech Republic, DM type 1 (DM1) was confirmed in case of 55,542 patients [1]. People without clinical symptoms exhibit blood glucose (BG) level (blood sugar concentration) under physiological conditions in the range from 3.9 to 5.6 mmol dm−3 fasting and less than 10 mmol dm−3 post-meal (for more details see Ref. [2]).

Continuous glucose monitoring (CGM) is one of fundamental options how to improve the quality of their life [3, 4, 5, 6]. To receive the sufficient amount of data for good compensation, these people should measure levels of their BG 4 times a day at least. However, the use of CGM yields required information on BG levels 24 h a day. It contributes significantly to its on-line or off-line compensation. CGM plays very important role in the prevention of hypoglycemia, which represents a principal complication during sport and physical activities of these persons. CGM is often based on electrochemical principle, in some types of sensors the measurements are realized by application of microdialysis, reverse ionophoresis, fluorescence, viscosimetry, and recently Raman spectroscopy has been used as well [7, 8]. The most commonly used CGMs operate on the principle of a needle sensor which electrochemically evaluates the reaction products catalyzed by glucose oxidase [9]. The sensors produced by companies Medtronic (USA) and DexCom (USA) are the most frequently used in the Czech Republic. The sensor is mainly composed of a platinum electrode coated with glucose oxidase (EC 1.1.3.4), which is then covered by a semipermeable membrane. d-glucose and oxygen from the interstitial fluid diffuse through the semipermeable membrane and react with the glucose oxidase on the surface of the sensor [10]. Sensor sends every 10 s the registered values of voltage and of generated current to the monitor (pump) via MiniLink. When 30 measurements are collected (i.e., every 5 min), the pump calculates the average value. This is stored in memory of the pump and is displayed to the user. Therefore, the monitor stores 288 values of the glucose concentrations in 24 h [11].

However, due to properties of the interstitial fluid, this system of BG determination is more complicated and the occurring problems can significantly affect the measurement accuracy. First, it is necessary to take into account that a delay can be observed between the glucose levels measured in interstitial fluid and BG levels. Some authors have indicated the range of this delay from 7 to 20 min [12, 13, 14, 15]. Besides, extreme short-term fluctuations in BG levels may not be registered by evaluation of glucose concentrations in interstitial fluid at all. In practice this means that if the patient calibrated his sensor during rapid changes in blood glucose levels (e.g., during sports activities) or in high levels of blood glucose, his sensor exhibits inadequate glucose values. Moreover, the patient can further influence the accuracy of measurement in particular by frequency of calibration, by point of obtaining blood sample for calibration, and by time interval between the determination of glucose level and by entering this value into the pump [16].

Each sensor is characterized by its lifetime, which depends on the so-called “inflammatory response” to the presence of the sensor in the body, on the battery lifetime, on the rate of enzyme and of electrodes degradation, on biofilm formation, and on many other factors. The main lifetime and accuracy-determining factor of a sensor is the inflammation at the injection site and the biofilm formation. Therefore, there have been made attempts to develop so-called “tissue response modifiers”, by which the sensor surfaces could be coated. Dexamethasone, vasodilator nitric oxide, vascular endothelial growth factor (VEGF), and other anticoagulants have been the most frequently used for these purposes.

Fabrication of glucometers is regulated by ISO 15197:2003 [17], according to which 95% of the measured results must be within ± 20% interval around the true value of BG (glycemia) above 4.17 mmol dm−3, and the maximum allowed deviation is 0.83 mmol dm−3 in BG below 4.17 mmol dm−3. Similar standards for the CGM have been developed. However, there has not been accepted any final consensus on how to optimally assess the accuracy of continuous glucose sensors [15]. The mathematical methods of linear regression, of error analysis, and of mean absolute deviation of the value evaluated by the sensor from the reference value determined by a glucometer belong to the most frequently used, similarly as Clarke error grid analysis (EGA). It was developed in 1987 to quantify clinical accuracy of patient estimates of their current blood glucose as compared to the blood glucose value obtained in their analyzer [16]). Accuracy of the method is dependent on location of recorded points in different regions of the graph. The highest accuracy reaches the method in case of the highest percentage of points located within regions A and B. If the measured value is situated within the regions D or E, the method is clinically inapplicable [18].

The aim of this study was, in correspondence with previously published results [19, 20], to compare the data collected using two different methods, i.e., using CGM and using glucometer. They were registered mostly during extreme physical and sport activities of the person with diagnosed DM1. We wanted to evaluate the achieved accuracy with that given by the manufacturer and to confirm the applicability of this method for cases of extreme physical and sport activities.

Results and discussion

In the case of all data collected during experiments realized by us, total deviation shows that the results registered by MMT-7002 sensors are lower than those registered by the reference glucometer. The value of the mean absolute percent difference was affected by extreme values (the maximum difference reached 250%). Nevertheless, these cases belonged to erroneous measurements (calibration error, etc.). In comparison with the manufacturer’s data [21], we found that the mean absolute difference achieved by us was higher by 3.1% (Table 1).
Table 1

Overview of achieved results

 

a

b

c

d

e

f

g

h

i

Number of paired BG measurements

616

213

10

203

403

205

410

205

206

Mean absolute percent difference

22.8

21.7

29.3

21.2

23.3

20.1

23.7

26.8

20.7

± Standard deviation (SD)/%

1.9

2.0

4.6

1.8

1.9

1.9

1.9

2.0

1.9

Total sum of mean absolute differences/mmol dm−3

− 151.5

− 22.6

+0.4

− 23.0

− 128.9

− 49.3

− 116.2

− 66.9

− 35.3

Bias/mmol dm−3

− 0.25

− 0.11

0.04

− 0.11

− 0.32

− 0.24

− 0.28

− 0.33

− 0.17

(a) All evaluated data from the described experiments; (b) data evaluated during more extreme conditions (windsurfing course, snowboarding, skiing, climbing, etc.); (c) data evaluated during Matterhorn climbing; (d) data evaluated during conditions given in point b excluding data in point c; (e) data evaluated under standard conditions, i.e., after elimination of data in point b; (f) first third of the evaluated data (sorted according to time); (g) two-thirds of the evaluated data (sorted according to time); (h) second third of the evaluated data (sorted according to time); (i) third third of the evaluated data (sorted according to time)

Five values registered by glucose meter (from the total number of 616 measurements) were below the level of 2.2 mmol dm−3. Their values had to be omitted as too low (out of range), i.e., five pairs had to be excluded from the total number of evaluated measurements.

Due to the fact that the manufacturer performed the data analysis according to the original Clarke EGA [16] without taking into account the rate of BG changes, we also include all pairs of measurements regardless of the rates of BG changes. Moreover, we also registered the rate of BG changes. 18% of calibrations were carried out under faster BG changes than 1.1 mmol dm−3 min−1. This was caused by the fact that in the real life, especially during a physical activity, it is not always possible to wait for BG steady-state.

Accuracy of the measurement system MiniMed Paradigm REAL-Time was also characterized by evaluation of relative number of pairs within 20 and 30%, respectively, of the reference BG sensor value in particular regions of the Clarke EGA (Table 2). The relative numbers of paired measurements within 20% are equal to those given by producer of the reference BG sensor. In case of data within 30%, the difference amounts to 2%. The measurements in the range of the normal BG are more accurate. On the other hand, the accuracy of determined values in ranges of hypoglycemia and hyperglycemia, respectively, are significantly worse. The difference smaller than 20% was achieved in 38% of evaluated pairs in BG level over 13.3 mmol dm−3 for all evaluated data in our experiments (according to the manufacturer 61% [21]) and the difference smaller than 30% was achieved in 54% of evaluated pairs in BG level over 13.3 mmol dm−3 for all evaluated data in our experiments (according to the manufacturer 82% [21]) (Table 2a). For comparison in the case of data evaluated during more extreme conditions, the difference smaller than 20% increased to 58% of evaluated pairs in BG level over 13.3 mmol dm−3 (Table 2b). The highest percentages of data pairs with difference smaller than 20% were found in group of data collected during more extreme conditions excluding data collected during Matterhorn climbing and in group of data collected during 3rd third of the evaluated data (sorted according to time) (64% of data pairs in both cases) (Table 2d, i). Similar results were achieved in the case of differences smaller than 30%, i.e., the highest percentages were achieved in the case of data collected during more extreme conditions excluding data collected during Matterhorn climbing (82%) (Table 2d). It is possible to conclude that the lowest percentage of differences smaller than 20% was achieved in BG level over 13.3 mmol dm−3 under “standard” conditions (17%) and smaller than 30% in BG level over 13.3 mmol dm−3 (33%). Therefore, it could be supposed that the highest number of differences should be registered in case of data collected during Matterhorn climbing. Nevertheless, the total number of these data pairs is too small to yield statistically significant information.
Table 2

Comparison of the relative number of pairs of measurements with deviation within 20 and 30% respectively, evaluated under different conditions and with the manufacturer’s data

Glucose range/mmol dm−3

All evaluated data from the described experiments (a)

Data given by the manufacturer

No. of paired measurements

Percentage within 20%/%

Percentage within 30%/%

No. of paired measurements

Percentage within 20%/%

Percentage within 30%/%

Total

616

62

77

3941

62

79

2.2–4.4

155

43

60

356

68

68

4.5–6.7

174

65

84

769

60

77

6.8–13.3

258

74

86

2362

62

81

> 13.3

24

38

54

454

61

82

Glucose range/mmol dm−3

Data evaluated during more extreme conditions (windsurfing course, snowboarding, skiing, climbing, etc.) (b)

Data evaluated during Matterhorn climbing (c)

No. of paired measurements

Percentage within 20%/%

Percentage within 30%/%

No. of paired measurements

Percentage within 20%/%

Percentage within 30%/%

Total

213

64

77

10

60

80

2.2–4.4

61

44

61

3

100

100

4.5–6.7

59

66

86

2

0

100

6.8–13.3

80

79

85

4

75

75

> 13.3

12

58

75

1

0

0

Glucose range/mmol dm−3

Data evaluated during conditions given in point (b) excluding data in point (c)

(d)

Data evaluated under standard conditions, i.e., after elimination of data in point (b)

(e)

No. of paired measurements

Percentage within 20%/%

Percentage within 30%/%

No. of paired measurements

Percentage within 20%/%

Percentage within 30%/%

Total

203

64

77

403

60

77

2.2–4.4

58

41

59

94

43

60

4.5–6.7

57

68

86

115

64

83

6.8–13.3

76

79

86

178

71

86

> 13.3

11

64

82

12

17

33

Glucose range/mmol dm−3

First third of the evaluated data (sorted according to time) (f);

First two-thirds of the evaluated data (sorted according to time); (g)

All evaluated data from the described experiments (a)

No. of paired measurements

Percentage within 20%/%

Percentage within 30%/%

No. of paired measurements

Percentage within 20%/%

Percentage within 30%/%

Total

205

61

78

410

59

76

 

2.2–4.4

44

36

55

103

39

57

 

4.5–6.7

65

63

88

119

68

87

See Table 2

6.8–13.3

87

76

86

172

69

84

 

> 13.3

7

29

43

13

15

38

 

Glucose range/mmol dm−3

First third of the evaluated data (sorted according to time) (f);

Second third of the evaluated data (sorted according to time) (h)

Third third of the evaluated data (sorted according to time) (i)

 

No. of paired measurements

Percentage within 20%/%

Percentage within 30%/%

No. of paired measurements

Percentage within 20%/%

Percentage within 30%/%

Total

 

205

57

75

206

67

79

2.2–4.4

 

59

41

59

52

52

65

4.5–6.7

See Table 2f

54

74

87

55

58

78

6.8–13.3

 

85

62

81

86

83

90

> 13.3

 

6

0

33

11

64

73

Characterization of clinical relevance of measurement deviations between the sensor and the glucometer meter using Clarke EGA are depicted in correlograms (Fig. 1) and summarized in Table 3. In all variants of the evaluated data (Fig. 1a–i), there can be found the highest number of points in regions A and B, which lead the patient to the correct and innocuous decisions. This is in fair correspondence with the manufacturer’s data [21] and with results published in some other studies [22, 23, 24]. In our case, 100% (according to the manufacturer 99.9% [21]) of measured pairs from the interval 4.5–6.7 mmol dm−3 are situated in regions A and B, i.e., if the patient was in this ideal region of glycemia, the sensor provided him always correct information for his decision.
Fig. 1

Clarke error grid analysis. a All evaluated data from the described experiments; b data evaluated during more extreme conditions (windsurfing course, snowboarding, skiing, climbing, etc.); c Data evaluated during Matterhorn climbing; d data evaluated during conditions given in point b excluding data in point c; e data evaluated under standard conditions, i.e., after elimination of data in point b; f first third of the evaluated data (sorted according to time); g two-thirds of the evaluated data (sorted according to time); h second third of the evaluated data (sorted according to time); i Third third of the evaluated data (sorted according to time)

Table 3

Paired measurements in regions A..E in dependence on glucose range (there is not any point in the E-region, therefore, this region is not given in the tables)

Glucose range/mmol dm−3

No. of data points evaluated

Percentage of data points evaluated/%

A + B

A + B/%

A

A/%

B

B/%

C

C/%

D

D/%

(a) All evaluated data from the described experiments

 < 2.2

5

0.8

4

80.0

3

60.0

1

20.0

0

0.0

1

20.0

 2.2–4.4

155

25.2

140

90.3

99

63.9

41

26.5

1

0.6

14

9.0

 4.5–6.7

174

28.2

157

90.2

111

63.8

46

26.4

1

0.6

16

9.2

 6.8–13.3

258

41.9

233

90.3

165

64.0

68

26.4

1

0.4

24

9.2

 > 13.3

24

3.9

21

87.5

15

62.5

6

25.0

0

0.0

3

12.5

 Total

616

100.0

555

 

393

 

162

 

3

 

58

 

(b) Data evaluated during more extreme conditions (windsurfing course, snowboarding, skiing, climbing, etc.)

 < 2.2

1

0.5

1

100.0

1

100.0

0

0.0

0

0.0

0

0.0

 2.2–4.4

61

28.6

53

86.9

39

63.9

14

23.0

1

1.6

7

11.5

 4.5–6.7

59

27.7

52

88.1

38

64.4

14

23.7

1

1.7

6

10.2

 6.8–13.3

80

37.6

70

87.5

52

65.0

18

22.5

1

1.3

9

11.2

 > 13.3

12

5.6

11

91.7

8

66.7

3

25.0

0

0.0

1

8.3

 Total

213

100.0

187

 

138

 

49

 

3

 

23

 

(c) Data evaluated during Matterhorn climbing

 < 2.2

0

0.0

0

0.0

0

0.0

0

0.0

0

0.0

0

0.0

 2.2–4.4

3

30.0

3

100.0

2

66.7

1

33.3

0

0.0

0

0.0

 4.5–6.7

2

20.0

1

50.0

1

50.0

0

0.0

0

0.0

1

50.0

 6.8–13.3

4

40.0

3

75.0

2

50.0

1

25.0

1

25.0

0

0.0

 > 13.3

1

10.0

1

100.0

1

100.0

0

0.0

0

0.0

0

0.0

 Total

10

100.0

8

 

6

 

2

 

1

 

1

 

(d) Data evaluated during conditions given in point b excluding data in point c

 < 2.2

1

0.5

1

100.0

1

100.0

0

0.0

0

0.0

0

0.0

 2.2–4.4

58

28.6

51

87.9

38

65.5

13

22.4

1

1.7

6

10.4

 4.5–6.7

57

28.1

50

87.7

37

64.9

13

22.8

0

0.0

7

12.3

 6.8–13.3

76

37.4

67

88.2

49

64.5

18

23.7

1

1.3

8

10.5

 > 13.3

11

5.4

10

90.9

7

63.6

3

27.3

0

0.0

1

9.1

 Total

203

100.0

179

 

132

 

47

 

2

 

22

 

(e) Data evaluated during conditions given in point b excluding data in point c

 < 2.2

4

1.0

4

100.0

3

75.0

1

25.0

0

0.0

0

0.0

 2.2–4.4

94

23.3

86

91.5

60

63.8

26

27.7

0

0.0

8

8.5

 4.5–6.7

115

28.5

105

91.3

72

62.6

33

28.7

0

0.0

10

8.7

 6.8–13.3

178

44.2

162

91.0

112

62.9

50

28.1

0

0.0

16

9.0

 > 13.3

12

3.0

11

91.7

8

66.7

3

25.0

0

0.0

1

8.3

 Total

403

100.0

368

 

255

 

113

 

0

 

35

 

(f) First third of the evaluated data (sorted according to time)

 < 2.2

2

1.0

2

100.0

1

50.0

1

50.0

0

0.0

0

0.0

 2.2–4.4

44

21.5

39

88.6

27

61.4

12

27.3

0

0.0

5

11.3

 4.5–6.7

65

31.7

58

89.2

41

63.1

17

26.2

0

0.0

7

10.7

 6.8–13.3

87

42.4

78

89.7

55

63.2

23

26.4

0

0.0

9

10.4

 > 13.3

7

3.4

6

85.7

4

57.1

2

28.6

0

0.0

1

14.3

 Total

205

100.0

183

 

128

 

55

 

0

 

22

 

(g) Two-thirds of the evaluated data (sorted according to time)

 < 2.2

3

0.7

3

100.0

2

66.7

1

33.3

0

0.0

0

0.0

 2.2–4.4

103

25.1

93

90.3

64

62.1

29

28.2

0

0.0

10

9.7

 4.5–6.7

119

29.0

106

89.1

73

61.3

33

27.7

1

0.8

12

10.2

 6.8–13.3

172

42.0

154

89.5

106

61.6

48

27.9

1

0.6

17

9.9

 > 13.3

13

3.2

12

92.3

8

61.5

4

30.8

0

0.0

1

7.7

 Total

410

100.0

368

 

253

 

115

 

2

 

40

 

(h) Second third of the evaluated data (sorted according to time)

 < 2.2

1

0.5

1

100.0

1

100.0

0

0.0

0

0.0

0

0.0

 2.2–4.4

59

28.8

53

89.8

36

61.0

17

28.8

1

1.7

5

8.5

 4.5–6.7

54

26.3

49

90.7

33

61.1

16

29.6

0

0.0

5

9.3

 6.8–13.3

85

41.5

77

90.6

52

61.2

25

29.4

1

1.2

7

8.2

 > 13.3

6

2.9

5

83.3

3

50.0

2

33.3

0

0.0

1

16.7

 Total

205

100.0

185

 

125

 

60

 

2

 

18

 

(i) Third third of the evaluated data (sorted according to time)

 < 2.2

2

1.0

2

100.0

1

50.0

1

50.0

0

0.0

0

0.0

 2.2–4.4

52

25.2

47

90.4

35

67.3

12

23.1

0

0.0

5

9.6

 4.5–6.7

55

26.7

50

90.9

38

69.1

12

21.8

0

0.0

5

9.1

 6.8–13.3

86

41.7

78

90.7

59

68.6

19

22.1

0

0.0

8

9.3

 > 13.3

11

5.3

10

90.9

7

63.6

3

27.3

1

9.1

0

0.0

 Total

206

100.0

187

 

140

 

47

 

1

 

18

 

The most important information for the patient yielded by the sensor is situated in the region of hypoglycemia. In this range in regions A and B, respectively, can be found 90.3% of pairs of all evaluated data. The manufacturer indicates 76.1% in these regions. It is the lowest registered percentage of the evaluated data points (Table 3a). Given that hypoglycemia is directly life-threatening, the accuracy of the glucose sensor is still insufficient. The smallest relative number of data points in the regions A and B (87.5%) exhibit the pairs of measurements which were recorded in BG above 13.3 mmol dm−3 for the case of all evaluated data (the data recorded in BG below 2.2 mmol dm−3 are not reliable). These values are in very good correspondence with manufacturer’s data (he published the relative frequency of 86.8% [21]).

Generally, our results are in fair correspondence with manufacturer’s data [21] and with results published in other studies [22, 23, 24]. According to Mráz et al. [25], the measurement accuracy increases with increasing glycemia. However, according to the results summarized in Table 3a–i the results are relatively stable, mostly over 85%, almost independent of glycemia level. The only exception was observed in data collected during Matterhorn climbing. However, as it was mentioned above, the number of data pairs is too low to yield statistically significant information.

Conclusion

In correspondence with our presumptions, we could conclude that CGM is accurate enough to help significantly in prevention of acute complications in patient with DM1 during her intensive physical activities. The measurements carried out by the tested sensor in the range of the normal glucose, realized under standard life conditions and under more extreme conditions, were even more accurate than those given by the manufacturer. Similarly, in the regions of hypoglycemia or hyperglycemia, the number of errors was lower than specified by the manufacturer. On basis of the evaluated data, it is possible to conclude that no statistically significant difference was registered among data yielded by both methods in criteria: mean absolute percent differences, biases, and slopes and intercepts of the dependences sensor vs. glucometer data in groups of data collected under more extreme physical conditions, standard conditions, during first third, second third and third third of data collection, during first two-thirds, and during the whole time interval of data collection. The only difference was registered in data collected during Matterhorn climbing, however, the number of these data is too small (10) and they did not exhibit normal distribution.

The method of CGM, despite of its observed weaknesses, represents a significant progress in the treatment of diabetes. It is possible to conclude that the instructed patients are able to utilize fully this method, which can yield them reliable and accurate information on BG levels.

Experimental

Ethics statement

Authors declare that research reported in this manuscript has been conducted in an ethical and responsible manner, in full compliance with all relevant codes of experimentation and legislation. The investigated person collected evaluated data individually and after finishing of their colleting, they were yielded to the authors of this manuscript for scientific evaluation and for publishing on the base agreement between this person and Charles University, Faculty of Physical Education and Sport. This agreement was accepted and confirmed by Ethics committee of the Faculty of Physical Education and Sport, Charles University. Authors confirm that any participant in any research and experiment, who is described in the manuscript, has given written consent to the inclusion of material pertaining to themselves, and that he/she acknowledges he/she they cannot be identified via the manuscript. Authors have anonymized the investigated person.

Authors confirm that all mandatory laboratory health and safety procedures have been complied with in the course of conducting any experimental work reported in the manuscript; and that the manuscript contains all appropriate warnings concerning any specific and particular hazards that may be involved in carrying out experiments or procedures described in the manuscript or involved in instructions, materials, or formulae in the manuscript. All relevant safety precautions has been explicitly included and cited.

Participant and data collection

All experiments were realized on one person (woman) with diagnosed DM1, who was 32 years old at the start of data registration (body mass 66.5 ± 1.0 kg (height 173 cm, body mass index (BMI) 22.2 ± 0.3 kg m−2). The evaluated data were collected during about 44 months.

However, the data were not collected continuously during this time, but in about 13 time intervals, in which could be the results substantially affected realized activities (windsurfing course, snowboarding, skiing, climbing, etc.). Her average time of physical activity amounted from 10 to 15 h per week and the average intensity 70–75% of maximum heart rate (105 beats per minute). Her organism was standardly hydrated by supplied fluids, i.e., the results could not be influenced by dehydration [26]. She realized windsurfing course, snowboarding, skiing, and climbing (Matterhorn) during the evaluated time interval. Altogether, 56 continuous glucose sensors MMT-7002 (Medtronic, USA) were applied for these measurements and 603 sensor calibrations were carried out. 68 calibrations (9%) were excluded from the evaluation process, because the times between the last BG estimation and calibration measurement by the glucometer were longer than 10 min. In sum, 616 pairs of BGs were analyzed (sensor vs. glucometer).

Evaluation of bioanalytical performance of continuous glucose monitoring system

The analysis was based on data from the “Medtronic Guardian® REAL-Time Continuous Glucose Monitoring System” (Medtronic, MiniMed, USA).

The CGM system MiniMed Paradigm REAL-Time is the part of the insulin pump. Therefore, the patient does not wear it on his body. Moreover, this device is waterproof. That is why this system is suitable for all sport activities.

Two MiniLink Transmitters, four insulin pumps MiniMed Paradigm [three MiniMed Paradigms 722, and one MiniMed Paradigm Veo 754 (all Medtronic, USA)] were used for the measurements. The declared life time of the used insulin pump MiniMed Paradigm is 4 years. Nevertheless, two warranty replacements of the insulin pumps were performed during the evaluation period. Based on the technical information from the manufacturer [21], it is possible to assume that the accuracy of CGM is not affected by the insulin pump. The patient performed measurements at non-standard situations, e.g., under increased physical exertion or when the patient failed to achieve the target levels of BG. Patient’s glycosylated hemoglobin (HbA1c) ranged from 4.8 to 5.7% during the whole period.

The reference values were determined using glucometer “FreeStyle Mini” (Abbott Diabetes Care Česká Republika, Czech Republic). The measurements are based on the principle of the electrochemical coulometer. It evaluates the anodic charge, which is proportional to the amount of glucose in the original sample. According to the study published by “Institut für Diabetes-Technologie GmbH“, these blood glucose meters exhibit the accuracy of 98–100% [11]. As clinically accurate can be considered the results, in which the deviation is smaller than ± 20% from the reference value. However, in the case of this glucometer, 95% of result deviations were smaller than ± 10%. Due to the fact that the calibrations of sensors were carried out mostly using diabetes test strips from different batches, the systematic determination error can be omitted. The reference values for calibration purposes were determined from capillary blood from fingertips.

Evaluation method using Clarke error grid analysis

The accuracy of sensor measurements was evaluated using the method of Clarke EGA, as this method was used in accuracy determination by the CGM producer [21], and as it has been the most frequently used method in many of recently realized studies [16, 23, 27].

Clarke EGA for continuous measurement of BG, so-called CG-EGA (continuous glucose-error grid analysis) in its modified form analyzes blood glucose levels registered by the sensor and the reference values (in our case gained from the glucometer) at chosen times. It takes into account trends of increasing and decreasing BG. The grid breaks down a scatterplot of a reference glucose meter and of an evaluated glucose sensor into five regions A-E [15] in dependence on the clinical impact of the patient decision [28]. Results in zones A and B are considered as clinically acceptable, while results in zones C, D, and E are potentially dangerous and, therefore, they correspond to clinically significant errors [28]. In region A, there are glucose values within 20% of the reference BG sensor values. These values represent the clinically accurate results and they would lead to the correct treatment decisions of the patient. The values differ by more than 20% from the reference method in region B. However, the clinical decisions based on these measurements could not cause serious problems. In region C are those points which do not lead to necessity of BG level treatment. The points in region D would lead to failure to detect and treat hyperglycemia or hypoglycemia. Points in region E are completely erroneous and of dangerous values. These could lead the patient to wrong decisions. If the rate of changes in BG is higher or lower by more than 1.1 mmol dm−3 min−1, it is necessary to shift the grids of the above mentioned regions [14].

The correspondences among paired values were evaluated as the differences between the values registered by CGM sensor and those measured by the glucometer. The difference was calculated as percentage of the value given by the reference glucometer (absolute percent difference). Furthermore, the absolute difference was calculated. This is defined as the difference between the values registered by CGM sensor and glucometer.

Data groups

To characterize the ability of the user to utilize the BG sensor under different conditions and in different time intervals, the collected data were divided into following groups:
  1. (a)

    All evaluated data from the described experiments;

     
  2. (b)

    Data evaluated during more extreme conditions (windsurfing course, snowboarding, skiing, climbing, etc.);

     
  3. (c)

    Data evaluated during Matterhorn climbing;

     
  4. (d)

    Data evaluated during conditions given in point (b) excluding data in point (c);

     
  5. (e)

    Data evaluated under standard conditions, i.e., after elimination of data in point (b);

     
  6. (f)

    First third of the evaluated data (sorted according to time);

     
  7. (g)

    Two-thirds of the evaluated data (sorted according to time);

     
  8. (h)

    Second third of the evaluated data (sorted according to time);

     
  9. (i)

    Third third of the evaluated data (sorted according to time).

     

Notes

Acknowledgements

The article was written in the framework of the scientific branch development program UK FTVS PRVOUK No. P38 Biological-Sciences Aspects of Human Movement Studies (E. Kohliková), PRVOUK No. P39 Social-Sciences Aspects of Human Movement Studies (S. Vokounová-Honsová) and under financial support of the Czech Science Foundation (GA ČR No. 17-03868S) (T. Navrátil).

References

  1. 1.
    UZIS (2011) Diabetes care management (In Czech). ÚZIS ČR, PragueGoogle Scholar
  2. 2.
    Friedecky B, Zima T, Kratochvil A, Springer D (2012) Diabetes mellitus—laboratory diagnosis and patient status monitoring (in Czech). Czech Society of Clinical Biochemistry. http://www.cskb.cz/res/file/doporuceni/DM_verze%202012.pdf
  3. 3.
    Mueller AJ, Hasslacher C, Krivanek R, Knuth M, Nikolaus KS, Kuester F, Theinert S (2013) Diabetes Technol Ther 15:A68Google Scholar
  4. 4.
    Vashist SK (2012) Anal Chim Acta 750:16CrossRefPubMedGoogle Scholar
  5. 5.
    Vashist SK (2013) Diagnostics (Basel) 3:385CrossRefGoogle Scholar
  6. 6.
    Vashist SK, Luong JHT (2016) Point-of-care glucose detection for diabetic monitoring and management. CRC Press, FloridaCrossRefGoogle Scholar
  7. 7.
    English BE (2012) Eur Endocrinol 8:18CrossRefGoogle Scholar
  8. 8.
    Broz J (2009) DIAstyl 5:24Google Scholar
  9. 9.
    Vashist SK, Zheng D, Al-Rubeaan K, Luong JHT, Sheu FS (2011) Anal Chim Acta 703:124CrossRefPubMedGoogle Scholar
  10. 10.
    Jirkovska A (2009) Remedia 2:1Google Scholar
  11. 11.
    Pickova K (2013) Accuracy of glucometers (in Czech). Aidia. http://www.mojecukrovka.cz/clanek/presnost-mereni-glukometru/
  12. 12.
    Boyne MS, Silver DM, Kaplan J, Saudek CD (2003) Diabetes 52:2790CrossRefPubMedGoogle Scholar
  13. 13.
    Anonymous (2011) Diabetes mellitus (in Czech). http://cukrovka-ocima-biochemie.blog.cz/
  14. 14.
    Kovatchev B, Anderson S, Heinemann L, Clarke W (2008) Diabetes Care 31:1160CrossRefPubMedPubMedCentralGoogle Scholar
  15. 15.
    Wentholt IM, Hoekstra JB, DeVries JH (2006) Diabetes Care 29:1805CrossRefPubMedGoogle Scholar
  16. 16.
    Clarke WL, Cox D, Gonderfrederick LA, Carter W, Pohl SL (1987) Diabetes Care 10:622CrossRefPubMedGoogle Scholar
  17. 17.
    Commission European (2003) ISO 15197:2003 In vitro diagnostic test systems—requirements for blood-glucose monitoring systems for self-testing in managing diabetes mellitus. European Committee for Standardization, BrusselsGoogle Scholar
  18. 18.
    Zourek M (2012) Zdravotnicke Noviny 3:2Google Scholar
  19. 19.
    Honsova S, Navratil T (2011) Evaluating the Accuracy of Continuous Glucose Monitoring by the Patient through Continuous Glucose Error Grid Analysis. In: Navratil T, Barek J (eds) XXXI Moderni Elektrochemicke Metody (Modern Electrochemical Methods XXXI), Jetrichovice. Best Servis, p 59Google Scholar
  20. 20.
    Honsova S, Navratil T (2013) Continuous Glucose Monitoring by Real Time Sensor in Interstitial Fluid. In: Navratil T, Fojta M, Peckova K (eds) XXXIII Moderni Elektrochemicke Metody (Modern Electrochemical Methods XXXIII), Jetrichovice. Best Servis, p 63Google Scholar
  21. 21.
    Anonymous (2011) Guardian Real Time User Guide. http://www.minimed.com/pdf/guardian_real_time_user_guide.pdf
  22. 22.
    Hasanbegovic S (2010) HealthMED 4:615Google Scholar
  23. 23.
    Clarke WL, Anderson S, Farhy L, Breton M, Gonder-Frederick L, Cox D, Kovatchev B (2005) Diabetes Care 28:2412CrossRefPubMedGoogle Scholar
  24. 24.
    Clarke W, Stacey A, Kovatchev B (2008) Cur Diabetes Rev 4:7Google Scholar
  25. 25.
    Mraz M, Svacina S, Haluzik M (2009) Diabetol Metabol Endokr Výživa 12:71Google Scholar
  26. 26.
    Cauchy E (2009) Vade-mecum of mountain and travel medicine (in Czech) (trans: Hoschlova K). Cohen Property & Devolopment, PragueGoogle Scholar
  27. 27.
    Clarke W, Kovatchev B (2009) Diabetes Technol Ther 11:S45CrossRefPubMedGoogle Scholar
  28. 28.
    Medtronic (2011) Guardian® REAL-time continuous glucose monitoring system. (4. 3. 2011)Google Scholar

Copyright information

© Springer-Verlag GmbH Austria, part of Springer Nature 2018

Authors and Affiliations

  1. 1.Faculty of Physical Education and SportCharles UniversityPrague 6Czech Republic
  2. 2.J. Heyrovský Institute of Physical Chemistry of the Czech Academy of SciencesPrague 8Czech Republic
  3. 3.Institute of Medical Biochemistry and Laboratory Diagnostics, First Faculty of MedicineCharles University and General University Hospital in PraguePrague 2Czech Republic

Personalised recommendations