Abstract
Background:
The within-person and between-sensor variability of metrics from different interstitial continuous glucose monitoring (CGM) sensors in adults with type 2 diabetes not taking insulin is unclear.
Methods:
Secondary analysis of data from 172 participants from the HYPNOS randomized clinical trial. Participants simultaneously wore Dexcom G4 and Abbott Libre Pro CGM sensors for up to 2-weeks at baseline and, again, at the 3-month follow-up visit.
Results:
At baseline (up to 2-weeks of CGM), mean glucose for both the Abbot and Dexcom sensors was approximately 150 mg/dl and time-in-range (70-180 mg/dL) was just below 80%. When comparing the same sensor at two different time points (two 2-week wear periods, 3 months apart), the within-person variability (CVw) in mean glucose was 17.4% (Abbott) and 14.2% (Dexcom). CVw for percent time-in-range: 20.1% (Abbott) and 18.6% (Dexcom). At baseline, the Pearson’s correlation of mean glucose from the two sensors worn simultaneously was r=0.86, root mean squared error (RMSE), 13 mg/dL; for time-in-range, r=0.88, RMSE 8%-points.
Conclusions:
Substantial variation was observed within sensors over time and across two different sensors worn simultaneously on the same individuals. Clinicians should be aware of this variability when using CGM technology to make clinical decisions.
Keywords: type 2 diabetes, continuous glucose monitoring, with-person variability, method comparison
There is growing interest in the use of interstitial continuous glucose monitoring (CGM) technology for the management of diabetes. CGM devices have improved in accuracy over the past decade and provide detailed summaries of day-to-day glycemic patterns and information on biochemical hypoglycemia. Few studies have rigorously documented the within-person variability of CGM metrics. Total within-person error and between-sensor variability in CGM data are important for the interpretation of CGM data by patients and providers and the use of CGM metrics to guide management decisions in diabetes.
The Hyperglycemic Profiles in Obstructive Sleep Apnea (HYPNOS) randomized clinical trial enrolled 186 participants with type 2 diabetes who were not receiving insulin therapy and who had sleep apnea (1). HYPNOS participants underwent baseline and 3-month examinations that each included two weeks of CGM data collection using two different CGM devices (worn simultaneously). These data provide a unique opportunity to assess within- and between-sensor variability in CGM metrics in persons with diabetes who are not on insulin.
The objective of this study was to characterize how CGM metrics (mean glucose, time-in-range, and time-above-range) compare within and across sensors in adults with type 2 diabetes who were not receiving insulin.
METHODS
Study Population
The HYPNOS trial was undertaken to evaluate the effectiveness of positive airway pressure on glycemic variability assessed by CGM in a population of patients with type 2 diabetes and sleep apnea who were not taking insulin. The main trial results were null (https://clinicaltrials.gov/ct2/show/NCT02454153) and information on the design of the HYPNOS trial has been previously published (1). Study participants were enrolled from 2014 to 2019. Participants were eligible if they were 21 to 75 years of age, had type 2 diabetes and untreated obstructive sleep apnea, were not taking insulin therapy, and had an HbA1c ≥6.5%. Participants underwent a screening visit which included overnight home self-monitoring to confirm the diagnosis of sleep apnea.
Participants provided written informed consent and all study procedures were approved by the Institutional Review Board at the Johns Hopkins School of Medicine (IRB NA_00093188).
Measurements
Age, sex, race/ethnicity, and smoking status were self-reported. Body mass index (BMI) was calculated (kg/m2) from measured height and weight. Diabetes medications were assessed at the screening visit. History of cardiovascular disease (congestive heart failure, stroke, coronary artery bypass surgery, myocardial infarction, or angina) was self-reported. Hypertension was defined as measured blood pressure ≥130/80 mm Hg, or currently taking hypertension medication. HbA1c was measured using the DCA Vantage Analyzer (Siemens), a point-of-care device, at baseline and the 3-month study visit.
At baseline and the 3-month follow-up visit trained research technicians placed two different CGM devices on each participant; both were worn simultaneously for up to 14 days. The Abbott FreeStyle Libre Pro (Abbott Diabetes Care, Inc, Alameda, California) was placed on the back of the upper arm and the Dexcom G4 Platinum (Dexcom, San Diego, California) was placed on the right lateral abdominal wall. The Abbott Libre Pro is a masked device (participants cannot see the glucose readings) that does not require fingerstick calibration and automatically measures and stores glucose readings every 15 minutes for up to 14 days. The Dexcom G4 sensor measures and records glucose readings every 5 minutes for up to 7 days. The Dexcom G4 software was modified by manufacturer to mask participants to the glucose readings. Dexcom sensors were replaced after 7 days to obtain up to 14 days of continuous readings. Participants performed fingerstick calibrations using a FreeStyle InsuLinx glucometer (Abbott Diabetes Care, Inc) for the Dexcom G4 sensor at least twice per day per the manufacturer’s instructions. Sensors that detached or malfunctioned were replaced (n=5 at baseline and n=2 at the 3-month follow-up visit).
Statistical Analyses
We summarized clinical and demographic characteristics for the study population at baseline. We analyzed the ~2 weeks of CGM data from baseline and the 3-month follow-up visits for the Abbott and Dexcom CGM systems and calculated the following summary parameters: mean glucose, time-in-range (glucose 70 to 180 mg/dL) and percent time above and below clinically relevant glucose cut-points (≥250 and ≥180 mg/dL; <54 and <70 mg/dL). Hypoglycemic events were defined as a glucose <70 mg/dl (level 1) or <54 mg/dl (level 2) for at least 15 minutes (at least two consecutive readings for the Abbott CGM; and at least three consecutive readings for the Dexcom CGM).
We examined variability within and across the two different CGM sensors using scatterplots and Bland-Altman plots (difference in the measurements plotted against the mean of the measurements). We calculated the Pearson’s correlations and used errors-in-variables regression (2) to quantify the three-month (within-person) variability (baseline vs follow-up CGM wear periods) and also the baseline between-sensor variability. Errors-in-variable regression is used to account for biases that arise when the independent variable is measured with error (3). We reported the root mean square error (RMSE) from the regression analyses. The RMSE is a measure of accuracy and tells us the typical deviation (difference) from the predicted value. To compare total (3-month) within-person variability (baseline vs follow-up) in CGM mean glucose, time-in-range, and time-above range for the Abbott and Dexcom sensors, we also calculated with the within-person coefficient variation (CVw) (4, 5). We conducted secondary analyses of the individual glucose readings from the two sensors obtained at the same moment and also calculated the % of glucose readings from one sensor that were within +/− 20% of the other sensor (6).
RESULTS
The mean age of HYPNOS participants included in our study was 59.6 years, 48.8% were female, 35.5% self-identified as Black, 55.8% self-identified as White, and 8.7% self-identified as another race (Supplemental Table 1). The majority of participants were obese (77.3%) or overweight (16.9%). The mean HbA1c at baseline was 7.5% (SD, 1.0) and most participants were currently taking one or more glucose-lowering medications (90.1%), most commonly metformin (83.1%).
The mean glucose during the 2-week baseline CGM data collection period was approximately 150 mg/dL according to both the Abbott and Dexcom sensors (Table 1). The average CV was just below 30% and the percent time-in-range was just below 80%. Patterns were similar when examining the full 4-weeks of data collection (baseline and 3-month follow-up) (Supplemental Table 2). The rates of hypoglycemic events were low (e.g., 1 to 2 per week for level 1 events; and 0.3 to 0.6 per week for level 2 events) but with large standard deviations (Table 1 and Supplemental Table 2).
Table 1.
Summary Statistics* of Abbott Libre Pro and Dexcom G4 Platinum Continuous Glucose Monitoring (CGM) Metrics Assessed Simultaneously During up to Two Weeks, Baseline, N=172
| Abbott Libre Pro | Dexcom G4 Platinum | |
|---|---|---|
| Days of CGM wear time, mean (SD) | 13.0 (2.9) | 12.4 (3.0) |
| Mean glucose, mg/dL, mean (SD) | 150.8 (38.8) | 153.2 (35.1) |
| Median glucose, mg/dL, mean (SD) | 145.0 (39.0) | 147.9 (35.4) |
| SD, mg/dL, median (p25, p75) | 38.5 (31.0, 47.6) | 37.3 (29.3, 44.8) |
| CV, %, mean (SD) | 26.9 (6.5) | 25.2 (5.7) |
| Percent of time in range (70-180 mg/dL), median (p25, p75) | 79.2 (65.6, 90.3) | 81.5 (67.5, 90.5) |
| Percent of time of time glucose >180 mg/dL, median (p25, p75) | 17.4 (7.2, 35.1) | 17.7 (8.3, 32.7) |
| Percent of time of time glucose >250 mg/dL, median (p25, p75) | 1.5 (0.1, 5.0) | 0.9 (0.0, 4.7) |
| Percent of time of time glucose <70 mg/dL, median (p25, p75) | 0.4 (0.0, 2.0) | 0.2 (0.0, 0.9) |
| Percent of time of time glucose <54 mg/dL, median (p25, p75) | 0.0 (0.0, 0.2) | 0.0 (0.0, 0.1) |
| Abbott Libre Pro, Hypoglycemic Events** |
Dexcom G4 Platinum Hypoglycemic Events† |
|
| Rates of hypoglycemic events (per week), mean (SD) | ||
| Level 1: glucose < 70 mg/dL | 2.1 (3.4) | 1.2 (2.2) |
| Level 2: glucose < 54 mg/dL | 0.6 (1.4) | 0.3 (1.1) |
Mean number of total glucose readings: 1251 for Abbott, 3574 for Dexcom
Abbott Libre Pro: Hypoglycemia Level 1: glucose <70 mg/dL for at least 15 minutes (at least two consecutive readings). Hypoglycemia Level 2: glucose <54 mg/dL for at least 15 minutes (at least two consecutive readings).
Dexcom G4 Platinum: Hypoglycemia Level 1: glucose <70 mg/dL for at least 15 minutes (at least three consecutive readings). Hypoglycemia Level 2: glucose <54 mg/dL for at least 15 minutes (at least three consecutive readings).
We compared mean glucose, time-in-range, and time-above-range at baseline to the 3-month follow-up visit (two, 2-week wear periods) within the two sensors (total within-person variability) (Table 2). Mean glucose at baseline was positively correlated with mean glucose at follow-up (3-months later) for both sensors (r = 0.69 Abbott; r = 0.73 Dexcom), but with substantial variability: RMSEs of 28.0 and 23.7 mg/dL, for Abbott and Dexcom, respectively (Figure 1). The corresponding within-person coefficient of variations (CVw) for Abbott and Dexcom, respectively, were 17.4% and 14.2% for mean glucose and 20.1% and 18.6% for time-in-range (Table 2). For time above range, correlations remained high (~0.70) but with poor reliability: CVws >50% and RMSEs of +/− 14.3% to 14.8% (Table 2 and Supplemental Figure 1). Correlations of estimate of time below range were low with extremely high variability over time (Table 2).
Table 2.
Pearson’s correlation and within-person coefficient of variation (CVw) for CGM parameters from the Abbott Libre Pro and Dexcom G4 Platinum sensors comparing two-week wear periods, assessed at baseline and the 3-month follow-up visit, N=153
| Abbott Libre Pro | Dexcom G4 Platinum | |||
|---|---|---|---|---|
| Pearson’s Correlation (r) |
CVw (95% CI)* | Pearson’s Correlation (r) |
CVw (95% CI)* | |
| Mean glucose, mg/dL | 0.69 | 17.4% (14.4, 20.4) | 0.73 | 14.2% (12.0, 16.5) |
| Percent time-in-range glucose 70-180 mg/dL | 0.71 | 20.1% (16.9, 23.2) | 0.73 | 18.6% (15.6, 21.6) |
| Percent of time above glucose >180 mg/dL | 0.72 | 53.0% (44.4, 61.7) | 0.71 | 52.5% (41.5, 63.5) |
| Percent of time above glucose >250 mg/dL | 0.70 | 122.5% (85.7, 159.3) | 0.72 | 118.1% (90.3, 146.0) |
| Percent of time below glucose <70 mg/dL | 0.38 | 157.8% (121.2, 194.3) | 0.43 | 160.6% (93.9, 227.3) |
| Percent of time below glucose <54 mg/dL | 0.26 | 263.5% (187.7, 339.2) | 0.24 | 298.8% (166.3, 431.3) |
The 95% CIs were bootstrapped with 200 replications
Figure 1. Scatterplots and Bland Altman Plots of the within-person variability in CGM mean glucose (comparison of two, 2-week wear periods, 3-months apart), Abbott Libre Pro and Dexcom G4 Platinum sensors, N=153.
Two weeks of data at baseline (mean wear time, 12.4 days, SD, 3.0 days) is compared to two weeks of data at follow-up (mean wear time, 12.3 days, SD, 3.1 days). Mean glucose from the Dexcom sensor was calculated, on average, from 3572 readings (every 5 minutes) at baseline and 3542 readings (every 5 minutes) at follow-up. Mean glucose from the Abbott sensor at baseline was calculated from 1190 readings (every 15 minutes) at baseline and from 1181 readings (every 15 minutes) at follow-up.
In the scatterplots, the errors-in-variables regression line (dotted blue) and line of identity (solid blue) are shown.
Panel A (Abbott), errors-in-variables regression line: Mean glucose, mg/dl (at 3-months) = 1.304 x Mean glucose, mg/dl (at baseline) – 36.012. Bland-Altman: Mean (SD) of difference in mean glucose: 10.1 (36.8) mg/dl.
Panel B (Dexcom), errors-in-variables regression line: Mean glucose, mg/dl (at 3-months) = 1.263 x Mean glucose, mg/dl (at baseline) – 31.712. Bland-Altman: Mean (SD) of difference in mean glucose: 8.7 (31.4) mg/dl.
When comparing mean glucose (2-weeks) from the two different CGM sensors worn simultaneously for up to 2 weeks at baseline, there was a strong correlation (r=0.86) but high variability (RMSE, 13.3 mg/dL) (Figure 2, Panel A). The standard deviation (SD) of the differences was 20.0 mg/dL (Figure 2, Panel B). High variability was also observed when we examined interstitial glucose readings from the two sensors matched on time of occurrence (Supplemental Figure 2). Overall, 83% of the CGM mean glucose values from the Abbott sensor were within 20% of the Dexcom sensor and 71% of the individual readings matched on time from the Abbott sensor were within 20% of the Dexcom sensor. Overall, 87% of the CGM mean glucose values from the Dexcom sensor were within 20% of the Abbott sensor and 71% of the individual readings matched on time from the Dexcom sensor were within 20% of the Abbott sensor. The correlation of percent time-in-range was high for the Abbott and Dexcom sensors worn simultaneously (r=0.88) but high variability (RMSE, 8%-points) (Figure 2, Panel C). The standard deviation (SD) of the differences was 11 %-points (Figure 2, Panel D). Among individuals who were classified as having time in range ≥70% according to the Abbott sensor, 7% were classified as <70% of the time by the Dexcom sensor; among those participants classified has having time in range ≥70% according to the Dexcom sensor, 10% were below 70% according to the Abbott sensor.
Figure 2. Scatterplots and Bland Altman plots of mean glucose and percent time-in-range (glucose 70-180 mg/dL) from two CGM devices worn simultaneously (~2 weeks), N=172.
Two weeks of data (mean wear time, 12.4 days, SD, 3.0 days). Mean glucose and % time in range (70-180 mg/dL) from the Dexcom sensor was calculated, on average, from ~3572 readings (every 5 minutes). Mean glucose and % time-in-range (70-180 mg/dL) from the Abbott sensor was calculated, on average, from ~1190 readings (every 15 minutes). Errors-in-variables regression line (dotted blue) and line of identity (solid blue) are shown.
(A) Errors-in-variables regression: Mean glucose (Dexcom), mg/dl = 0.904 x [Mean glucose (Abbott), mg/dl] +16.906
(B) Bland-Altman: Mean (SD) of difference in mean glucose (Abbott minus Dexcom) = −2.4 (20.0) mg/dl
(C) Errors-in-variables regression: % Time-in-range (Dexcom), % = 0.993 x [% Time-in-range (Abbott), %] + 1.981
(D) Bland-Altman: Mean (SD) of difference in % Time-in-range (Abbott minus Dexcom)= −1.4 (11.2) %
DISCUSSION
In this population of adults with type 2 diabetes and obstructive sleep apnea who were not receiving insulin, there was meaningful within-person variability (same sensor, two assessments, ~3 months apart) across the systems for mean glucose, time-in-range, and time-above-range. The mean glucose obtained from two different CGM devices were strongly correlated but with a range of mean glucose values on one sensor as compared to the other sensor. For the two systems, substantial variability was observed for individual glucose readings aligned by time of measurement. The Abbott and Dexcom sensors generally gave similar overall results, with no major differences in baseline summaries or within-person variability CGM clinical parameters across the two systems.
The percentage of time below range was very low, with most time outside of range reflecting hyperglycemia (glucose readings above 180 mg/dL). However, most participants experienced at least one level 1 hypoglycemic event during 2 weeks of CGM monitoring. Time below range and hypoglycemic events did not track well over time, meaning that hypoglycemic events during this initial period of monitoring was not strongly related to having hypoglycemic events again at the 3-month follow-up visit. These results are consistent with a prior study in adults with type 1 diabetes in an inpatient setting which demonstrated that the time below range differed substantially between two CGM systems worn by the same individuals (7). The clinical significance of episodes of biochemical hypoglycemia in our population of adults with type 2 diabetes is unclear.
For the same sensor is worn at two different time points, 3-months apart, the typical differences in mean glucose were +/− 29 and 23 mg/dL, with CVws of 17.4% and 14.2% for the Abbott and Dexcom sensors, respectively. Similarly, the 3-month within person variability (CVw) for time-in-range for both sensors was approximately 20%. These estimates of within-person variability are much higher than those observed for standard laboratory assays of hyperglycemia. In a study of community-dwelling adults with type 2 diabetes, the short-term (<2 month) within-person variability (CVw) for glycated albumin and fructosamine were both less than 5%, HbA1c was 2%, and fasting glucose was 9.6% (5). In the HYPNOS study population, the 3-month CVw for point-of-care HbA1c was 8.1%. The short-term within-person variability in CGM summary measures over time has implications for the use of this technology in clinical practice.
We observed that 71% of the Abbott (or Dexcom) CGM individual sensor readings were within 20% of the Dexcom (or Abbott) values when matched on time of occurrence. FDA standards of accuracy for CGM sensors are roughly based on readings being within +/− 15 to 20% of reference glucose values (8, 9). Against reference glucose measurements, 89.9% of readings from the Abbott FreeStyle Libre sensor were within 20% (10) and 82% of the Dexcom G4 readings were within 20% of reference glucose values (11). Because we did not evaluate a reference method here, we can only comment on reliability and cannot provide information on the accuracy of the individual systems.
The HYPNOS data inform our understanding how much CGM summary metrics are likely to vary and how well they track over time in patients with type 2 diabetes. We showed that if an individual wore a different sensor during the same period, typical differences in mean glucose were +/− 13 mg/dL and typical differences in percent time-in-range were +/− 8%-points. Among people who were classified as having good glycemic control (percent time in range ≥70%) by one CGM sensor, approximately 7 to 10% would be classified as above range by the other sensor worn by the same person, at the same time.
Our results are consistent with emerging studies on the reliability of CGM technology. Prior studies have reported moderate to high correlations of CGM measures within and across sensors (12-14). In a study of adults with type 2 diabetes and a high prevalence of reduced kidney function, 2-week within person variability in CGM mean glucose was much higher as compared to laboratory biomarkers (HbA1c, fructosamine, and glycated albumin) (12).
We extend the prior literature by providing a rigorous examination and presentation of measures of within-person variability for common CGM summary metrics used in clinical practice. Our study also highlights the weakness of using the Pearson’s correlation coefficient as the main measure of association when evaluating the variability within and across CGM systems. We found high correlations between mean glucose between the Abbott and Dexcom sensors worn on the same persons at the same time (r = 0.86), but this belied a high degree of variability. Using an errors-in-variables regression approach, we observed typical deviations from the line of best fit of +/− 13 mg/dL. The correlation coefficient is highly sensitive to the range of the data, is strongly influenced by extreme values, and, for these reasons, caution should be exercised when comparing correlations across different study populations or population subgroups. Going forward, comparison studies evaluating CGM sensors (15) should report the RMSE along with the correlation coefficient (and scatterplot) to help document typical deviations (spread of residuals) from the regression line. Bland-Altman plots are also complementary and extremely useful in method comparison studies (16, 17).
There are several limitations of this study. First, the Dexcom G4 Platinum represents an older generation device. The Abbott Libre Pro used here is the ‘professional’ device currently used in practice but is not equivalent to the Abbott FreeStyle Libre 2, more frequently used in clinical practice. Nonetheless, upgrades to these systems have primarily been to the physical systems and the calibration algorithms. The fundamental underlying biochemical technology of the latest generation of Abbott and Dexcom systems are unchanged from these earlier generation devices. Further, the devices used in the HYPNOS trial are identical to the systems used in major clinical trials of type 1 diabetes that have formed the basis for consensus recommendations for clinical use of CGM technology. Second, direct comparisons of the Dexcom and the Abbott sensors, especially for individual values, are challenging since the Abbott device was factory calibrated whereas the Dexcom was calibrated by fingerstick glucose by participants in the trial. Third, the study participants all had obstructive sleep apnea so caution is warranted in generalizing these results to the general population of patients of type 2 diabetes not receiving insulin therapy. Fourth, a point-of-care device was used to assess HbA1c, likely explaining the higher within-person variability in HbA1c observed here as compared to prior studies (5).
Strengths of the study included the relatively large sample of adults with type 2 diabetes, diverse study population (34% Black adults), measurement of glucose simultaneously from two different CGM sensors for up to 4 weeks, calculation of standard measures to quantify within-person variability, and the controlled setting with rigorous assessment of clinical measures by trained personnel using standardized study protocols.
Our data provide essential information on expected random variability in CGM data in adults with type 2 diabetes not receiving insulin therapy. It is important to understand total within-person variability in any measure used in clinical practice so as place any therapy-related changes in the context of expected random fluctuations. We observed high overall within-person variability over time, reflecting both inherent variability in CGM systems and glycemic changes over this period. Our results support the need to reassess mean glucose and time-in-range in persons with type 2 diabetes more frequently, as substantial variation is expected over a 3-month period. We also found that CGM summaries from two different sensors were similar, but with substantial discordance in values from the same person at the same time, even for mean glucose and percent time in range (calculated from ~3,500 individual interstitial glucose readings for the Dexcom sensor and ~1,200 readings for the Abbott sensor).
The real-world limitations and variability within and across CGM sensors are not well-recognized. CGM systems are known to be particularly error-prone in the low range (8, 10, 11), although we observed variability across the range of glucose values. The underlying reasons for this high variability are not clear but reflect the characteristics and limitations of CGM technology including of a lack of standardization of interstitial glucose methods, different proprietary algorithms used to generate glucose readings, and non-glycemic factors that may influence interstitial glucose (e.g., sensor placement, blood flow, and other local conditions that influence glucose diffusion from plasma into the interstitial space) (18) (19, 20). Ultimately, our study demonstrates clinically significant random error in CGM summary metrics.
CGM technology has revolutionized diabetes management for many patients on insulin therapy, especially patients with type 1 diabetes. CGM summaries from two different sensors are highly correlated but substantial variation was observed within and across sensors. Clinicians should be aware of this variability when using CGM technology to make clinical decisions.
Supplementary Material
Acknowledgements:
The authors would like to thank Tim Dunn, Abbott Diabetes Care, for his helpful comments. Abbott Diabetes Care and its employees had no role in the study design, data collection, or the decision to submit the manuscript for publication. Mr. Dunn provided helpful, non-binding comments during the revision of this manuscript.
Funding:
Dr. Selvin was also supported by NIH/NHLBI grant K24 HL152440. This work was also supported by NIH/NIDDK grants R01 DK128837 and R01 DK128900 and NIH/NIA grant RF1 AG074044. Dr. Echouffo-Tcheugui was supported by NIH/NHLBI grant K23 HL153774. The HYPNOS Trial was supported by NIH/NHLBI grant R01HL117167. Dr. R. Nisha Aurora was supported by K23 HL118414. Dr. Tang was supported by NIH/NHLBI grant T32 HL007024.
Footnotes
DISCLOSURES: Dr. Selvin is a Deputy Editor of Diabetes Care. Abbott Diabetes Care provided continuous glucose monitoring systems and self-monitoring blood glucose supplies for this investigator-initiated research. Dexcom provided continuous glucose monitoring systems at a discount.
ClinicalTrials.gov Identifier: NCT02454153
References
- 1.Rooney MR, Aurora RN, Wang D, Selvin E, Punjabi NM. Rationale and design of the hyperglycemic profiles in obstructive sleep apnea (hypnos) trial. Contemp Clin Trials 2021;101:106248. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Linnet K Evaluation of regression procedures for methods comparison studies. Clin Chem 1993;39:424–32. [PubMed] [Google Scholar]
- 3.Cornbleet PJ, Gochman N. Incorrect least-squares regression coefficients in method-comparison analysis. Clinical Chemistry 1979;25:432–8. [PubMed] [Google Scholar]
- 4.Juraschek SP, Coresh J, Inker LA, Rynders GP, Eckfeldt JH, Selvin E. The effects of freeze-thaw on beta-trace protein and beta2-microglobulin assays after long-term sample storage. Clin Biochem 2012;45:694–6. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Parrinello CM, Lutsey PL, Couper D, Eckfeldt JH, Steffes MW, Coresh J, Selvin E. Total short-term variability in biomarkers of hyperglycemia in older adults. Clin Chem 2015;61:1540–1. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Freckmann G, Pleus S, Grady M, Setford S, Levy B. Measures of accuracy for continuous glucose monitoring and blood glucose monitoring devices. J Diabetes Sci Technol 2019;13:575–83. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Freckmann G, Pleus S, Schauer S, Link M, Jendrike N, Waldenmaier D, et al. Choice of continuous glucose monitoring systems may affect metrics: Clinically relevant differences in times in ranges. Exp Clin Endocrinol Diabetes 2021. [DOI] [PubMed] [Google Scholar]
- 8.US Food and Drug Administration (FDA). Code of federal regulations - 21 cfr rules, 2018. Integrated continuous glucose monitoring system, https://wwwaccessdatafdagov/cdrh_docs/pdf17/DEN170088pdf; Last Accessed: February 17, 2022.
- 9.Bailey TS, Alva S. Landscape of continuous glucose monitoring (cgm) and integrated cgm: Accuracy considerations. Diabetes Technol Ther 2021;23:S5–S11. [DOI] [PubMed] [Google Scholar]
- 10.Abbott Diabetes Care Inc. Freestyle libre 14 day flash glucose monitoring system, summary of safety and effectiveness data (ssed). FDA Center for Devices and Radiological Health, 2018, https://wwwaccessdatafdagov/cdrh_docs/pdf16/P160030S017Bpdf; Last accessed February 23, 2022. [Google Scholar]
- 11.Dexcom Inc. Dexcom g4 platinum continuous glucose monitoring system, summary of safety and effectiveness data (ssed). FDA Center for Devices and Radiological Health, 2012, https://wwwaccessdatafdagov/cdrh_docs/pdf12/P120005bpdf; Last accessed February 23, 2022. [Google Scholar]
- 12.Zelnick LR, Batacchi ZO, Ahmad I, Dighe A, Little RR, Trence DL, et al. Continuous glucose monitoring and use of alternative markers to assess glycemia in chronic kidney disease. Diabetes Care 2020;43:2379–87. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Merino J, Linenberg I, Bermingham KM, Ganesh S, Bakker E, Delahanty LM, et al. Validity of continuous glucose monitoring for categorizing glycemic responses to diet: Implications for use in personalized nutrition. Am J Clin Nutr 2022;115:1569–76. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Howard R, Guo J, Hall KD. Imprecision nutrition? Different simultaneous continuous glucose monitors provide discordant meal rankings for incremental postprandial glucose in subjects without diabetes. Am J Clin Nutr 2020;112:1114–9. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Damiano ER, El-Khatib FH, Zheng H, Nathan DM, Russell SJ. A comparative effectiveness analysis of three continuous glucose monitors. Diabetes Care 2013;36:251–9. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Bland JM, Altman DG. Measuring agreement in method comparison studies. Stat Methods Med Res 1999;8:135–60. [DOI] [PubMed] [Google Scholar]
- 17.Bland JM, Altman DG. Statistical methods for assessing agreement between two methods of clinical measurement. Lancet 1986;1:307–10. [PubMed] [Google Scholar]
- 18.Siegmund T, Heinemann L, Kolassa R, Thomas A. Discrepancies between blood glucose and interstitial glucose-technological artifacts or physiology: Implications for selection of the appropriate therapeutic target. J Diabetes Sci Technol 2017;11:766–72. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Rossetti P, Bondia J, Vehi J, Fanelli CG. Estimating plasma glucose from interstitial glucose: The issue of calibration algorithms in commercial continuous glucose monitoring devices. Sensors (Basel) 2010;10:10936–52. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Acciaroli G, Vettoretti M, Facchinetti A, Sparacino G. Calibration of minimally invasive continuous glucose monitoring sensors: State-of-the-art and current perspectives. Biosensors (Basel) 2018;8. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.