What difference does sleep make? Continuous glucose monitoring metrics during fixed-overnight time versus sleep periods among older adults with type 1 diabetes
[Correction added on 8 December 2023, after first online publication: All instances of mg dl−1 were changed to mg/dL throughout the article.]
Summary
Hypoglycaemia during sleep is a common and clinically important issue for people living with insulin-treated diabetes. Continuous glucose monitoring devices can help to identify nocturnal hypoglycaemia and inform treatment strategies. However, sleep is generally inferred, with diabetes researchers and physicians using a fixed-overnight period as a proxy for sleep–wake status when analysing and interpretating continuous glucose monitoring data. No study to date has validated such an approach with established sleep measures. Continuous glucose monitoring and research-grade actigraphy devices were worn and sleep diaries completed for 2 weeks by 28 older adults (mean age 67 years [SD 5]; 17 (59%) women) with type 1 diabetes. Using continuous glucose monitoring data from a total of 356 nights, fixed-overnight (using the recommended period of 00:00 hours–06:00 hours) and objectively-measured sleep periods were compared. The fixed-overnight period approach missed a median 57 min per night (interquartile range: 49–64) of sleep for each participant, including five continuous glucose monitoring-detected hypoglycaemia episodes during objectively-measured sleep. Twenty-seven participants (96%) had at least 1 night with continuous glucose monitoring time-in-range and time-above-range discrepancies both ≥ 10 percentage points, a clinically significant discrepancy. The utility of fixed-overnight time continuous glucose monitoring as a proxy for sleep–awake continuous glucose monitoring is inadequate as it consistently excludes actual sleep time, obscures glycaemic patterns, and misses sensor hypoglycaemia episodes during sleep. The use of validated measures of sleep to aid interpretation of continuous glucose monitoring data is encouraged.
1 INTRODUCTION
Sleep is a constant challenge for a person with type 1 diabetes. It represents the predominate time they cannot actively monitor or manage their glucose with insulin, food or exercise. This matters because, during sleep, hypoglycaemia is common (Juvenile Diabetes Research Foundation Continuous Glucose Monitoring Study G, 2010), counter-regulatory responses are reduced (Jauch-Chara et al., 2007; Schultes et al., 2007), and the detrimental impact of hypoglycaemia on glucose metabolism, quality of life and diabetes management is clinically significant (Nefs et al., 2015). The use of continuous glucose monitoring (CGM) devices provides insight, for both researchers and clinicians alike, into glucose levels throughout the day and night. However, current consensus recommendations for clinical interpretation of CGM data advise that a fixed-overnight time block of 00:00 hours–06:00 hours should be used to approximate sleep (Danne et al., 2017). This is in contrast to sleep researchers who optimize sleep measurement accuracy with subjective and objective measures (e.g. diaries and actigraphy). These methodologies have highlighted the importance of sleep for glucose. For example, sleep restriction has been shown to decrease insulin sensitivity and is associated with poorer glycaemic outcomes (Donga et al., 2010). These studies are the exception, however, with clinical trials only considering treatment effects on CGM during sleep via a fixed-overnight period (e.g. 00:00 hours–06:00 hours). The use of standardized reports, known as ambulatory glucose profiles, is common in clinician–patient consultations, which typically involve interpreting at least 2 weeks of CGM data (Rodbard, 2021). Similar to clinical reports, interpretation of CGM data in these consultations is not aided by any metric of sleep behaviour.
This study compared CGM metrics during fixed-overnight time (00:00 hours–06:00 hours) versus systematically measured sleep periods (sleep diaries and wrist-worn actigraphy) data over a 2-week period among older adults with type 1 diabetes. Emphasis was placed on missed hypoglycaemic episodes, erroneously excluded sleep time and CGM metric discrepancies from using the fixed-overnight time approach.
2 METHODS
2.1 Study design
This is a post hoc analysis of data collected in the OldeR Adult Closed-Loop (ORACL) trial. Recruitment targeted adults aged 60 years and older, with type 1 diabetes duration of at least 10 years, and who were using an insulin pump (Chakrabarti et al., 2022; McAuley et al., 2022). Our analysis included 28 of the 30 participants who completed the ORACL trial run-in, 17 (59%) women and 11 (39%) men (mean age 67 years [SD 5], median diabetes duration 38 years [interquartile range, IQR: 21–46]). Sixteen (57%) were retired, with nine (32%) working part-time and three (11%) working full-time. Additional participant characteristics are presented in Table S1 Actigraphy watch failure during the assessment period led to one participant being excluded from the analysis; another participant was excluded as there were only 6 nights with sufficient CGM data. There was a total of 356 nights for analysis.
2.2 Sleep and glucose recording
The analysis evaluated a 2-week period of sleep assessment during real-time CGM use that all participants completed prior to randomization. Throughout the study period, participants used standardized sensor-augmented pump therapy (MiniMed 670G pump and Guardian 3 sensors; Medtronic, Northridge, CA, USA). The insulin pumps were used exclusively in manual delivery mode during the pre-randomization period. Consistent with international consensus guidelines on CGM use, the pre-defined fixed-overnight period was 00:00 hours–05:59 hours (Danne et al., 2017). Nightly sleep onset and offset were estimated via a consensus-based combination of wrist-worn research-grade actigraphy device (Actiwatch 2, Koninklijke Philips N.V, Amsterdam, The Netherlands) and sleep diaries. Actigraphy data were scored using the Actiware software (version 5.00, Philips Respironics, Pittsburgh, Pennsylvania, USA) with the default algorithm that uses 10 min of immobility as an indicator for sleep onset and offset. This output was compared with concurrent sleep diaries for any discrepancy larger than 30 min, prompting a manual inspection of the raw actigraphy data to reach a consensus estimation of sleep onset or offset (Chakrabarti et al., 2022). Consensus sleep periods were recorded and synchronized with CGM data. For inclusion in the analysis, at least 10 nights of sleep and CGM data needed to be recorded for a participant over the 2-week period, with at least 70% of usable CGM data from each night. This latter criterion was required for both approaches to facilitate direct comparison between time periods.
2.3 Data analysis
Our analysis was based on the premise that using different sleep estimation methods (i.e. objectively-measured sleep versus 00:00 hours–06:00 hours) with the same CGM dataset would yield varying metrics. The primary outcome of interest was the number of sensor hypoglycaemic episodes that occurred during sleep but outside of the fixed-overnight period (i.e. before 00:00 hours or after 06:00 hours). These episodes were defined as lasting ≥ 15 min (i.e. with a minimum of four consecutive sensor glucose readings) at < 70 mg/dL (Danne et al., 2017). Additional outcomes were missed objectively-measured sleep time (i.e. sleep before 00:00 hours and after 06:00 hours) and discrepancies in CGM metrics between fixed-overnight versus measured-sleep approaches. The latter concerns whether using fixed or objectively-measured sleep periods to examine the same glucose dataset can lead to clinically meaningful discrepancies. These discrepancies were calculated as percentage-point differences, with CGM metrics from the objectively-measured sleep period metric subtracted from the fixed-overnight period. For example, if the CGM metric time in range (TIR; 70–180 mg/dL) was estimated to be 70% (00:00 hours–06:00 hours) versus 50% (objectively-measured sleep), then the percentage-point difference would be +20, indicating an overestimation by the fixed-overnight time period. Absolute difference was employed here to indicate an overall discrepancy (either under- or overestimation) between the sleep estimation techniques. Additional CGM metrics were also similarly examined; time above range (TAR; > 180 mg/dL), time below range (TBR; < 70 mg/dL), mean glucose and glucose coefficient of variation (CV). Discrepancies above specified thresholds were chosen as being deemed clinically relevant (Danne et al., 2017), and comparable to reported overnight benefits from closed-loop insulin delivery systems (Jiao et al., 2022). Discrepancy thresholds of 5 and 10 absolute percentage points, respectively, were examined for TIR, TAR and TBR. Due to the clinical significance of hypoglycaemia during sleep, TBR discrepancy of 1 absolute percentage point was also examined. For mean glucose, an 18 mg/dL discrepancy between the two approaches was used, which was based on Jiao et al.'s meta-analysis of overnight mean glucose changes between closed-loop systems versus control (Jiao et al., 2022). For glucose CV, the discrepancy between approaches was examined at 10 absolute percentage points, which exceeds the overnight benefits observed from closed-loop insulin pump delivery studies (Jiao et al., 2022), while also spanning the gap between mean glucose CV (42%) in a sample of 200 people with type 1 diabetes (Marchand et al., 2019) and the suggested clinical target limit of 33% (Battelino et al., 2019). For each of the CGM discrepancy thresholds examined, the number and percentage of participants who had at least 1 night breaching the threshold was calculated.
3 RESULTS
3.1 Discrepancy in sleep estimation
Sleep-onset and -offset times varied substantially between participants (Figure 1). During the assessment period, the mean sleep-onset time varied between participants by more than 3 hr (time range 21:59 hours–01:15 hours), and mean sleep-offset times varied by more than 4 hr (time range 04:50 hours–08:59 hours). For individual participants, a median of 57 min per night (IQR: 49–64) of sleep was missed by the fixed-overnight 00:00 hours–06:00 hours approach. Across all 356 nights, a median of 57 min per night (IQR: 35–78) of sleep was missed. Conversely, participants were awake during the fixed-overnight period for a median of 0 min per night (IQR: 0–14); the median across all nights was 0 min (IQR: 0–21).
Even though the 00:00 hours–06:00 hours period consistently includes some sleep, it comes at a cost of missing some glucose data during actual sleep. Figure 2 provides a visualization of this cost by plotting glucose levels for all sleep periods. What is immediately apparent is the amount of glucose data during sleep that is excluded with the 00:00 hours–06:00 hours approach.
3.2 Discrepancy in glucose metrics
During the monitoring period there were a total of 39 sensor hypoglycaemia episodes that occurred when participants were asleep. Importantly, five (13%) occurred outside of the 00:00 hours–06:00 hours period. All five missed hypoglycaemic episodes were within a clinically significant glucose range (54–70 mg/dL), with median duration 40 min (IQR: 30–50). Four out of the five missed hypoglycaemic episodes occurred before midnight. Absolute discrepancies, indicating either under- or overestimation by the 00:00 hours–06:00 hours period, were also assessed. During the 2-week period, 27 of 28 participants (96%) had at least 1 night where both their TIR and TAR measurements differed by at least 10 absolute percentage points (Figure 3a,b). Three people had at least 1 night with TBR under- or overestimated discrepancy ≥ 10 percentage points, and one-third of the sample (9/28) had at least 1 night with TBR discrepancy ≥ 5 percentage points (Figure 3c). Seven participants had discrepancies of at least 5 percentage points in both their TIR and TAR for at least half of the nights recorded. Twenty-one participants (75%) had at least 1 night with a mean glucose discrepancy of ≥ 18 mg/dL, and 18 (64%) exceeded the glucose variation threshold of 10 percentage points (Table S2).
4 CONCLUSIONS
Our analysis has highlighted how using 00:00 hours–06:00 hours to estimate sleep consistently excludes both the start and end of the sleep period, precluding any complete analysis of glycaemic patterns during sleep. This has implications for clinical trials of diabetes treatments that follow the consensus recommendation. For example, a recent trial by our group reported the therapeutic timeline of closed-loop insulin delivery during measured sleep (Chakrabarti et al., 2022). We found that improvements in CGM metrics with this treatment were not consistently seen until after the fourth hour of sleep. Such insights may allow future closed-loop insulin delivery algorithms to consider the timing of the sleep window. Furthermore, research that has shown differences in counterregulatory responses to hypoglycaemia between early-in-sleep (between 23:00 hours and 00:45 hours) and late-in-sleep (between 02:45 hours and 03:30 hours) also highlights the need to consider sleep periods when examining CGM data (Jauch-Chara et al., 2007).
Hypoglycaemia during sleep was missed with the 00:00 hours–06:00 hours approach. This is of particular clinical importance, as the risks from hypoglycaemia and its sequelae are increased during sleep when the individual cannot adjust their insulin delivery or carbohydrate intake to compensate. As identification of nocturnal hypoglycaemia via inspection of ambulatory glucose profile daily views is encouraged in clinical practice (Aleppo & Webb, 2019), fixed time proxies for sleep (e.g. 00:00 hours–06:00 hours) make retrospective interpretation of glycaemia during sleep challenging. To our knowledge, there is no option for people with diabetes or their healthcare teams to incorporate their typical sleep habits into commonly used CGM reports, such as Medtronic's CareLink™ Weekly Review Report (Medtronic CareLink, 2023). While arguments for enhancement of ambulatory glucose profiles have included the use of actigraphy to measure physical activity (Rodbard, 2021), and that CGM data can be presented on consumer-grade wearables that track sleep (Olsen, 2022), the present study now highlights the additional clinical significance of identifying sleep periods via actigraphy for people with type 1 diabetes.
In conclusion, the growing body of research regarding the interplay between sleep and glucose metabolism supports analysis of glucose data in the context of the individual's sleep–wake status. By measuring sleep–wake status, our findings show that sleep does make a difference for contextualizing the clinical implications of glucose excursions in type 1 diabetes. Clinical recommendations for diabetes research and practice should encourage the interpretation of glucose levels during systematically measured sleep.
AUTHOR CONTRIBUTIONS
Steven Trawley: Conceptualization; funding acquisition; writing – original draft; formal analysis; data curation; methodology; visualization; writing – review and editing; investigation; validation. Erin Kubilay: Writing – review and editing; data curation; conceptualization; formal analysis. Peter G. Colman: Writing – review and editing; conceptualization; funding acquisition. Melissa H. Lee: Writing – review and editing; investigation; funding acquisition. David N. O'Neal: Writing – review and editing; investigation; conceptualization. Vijaya Sundararajan: Writing – review and editing; investigation; conceptualization; formal analysis; funding acquisition. Sara Vogrin: Formal analysis; writing – review and editing; conceptualization; methodology; data curation; validation. Sybil A. McAuley: Writing – review and editing; writing – original draft; funding acquisition; investigation; conceptualization; methodology; data curation.
ACKNOWLEDGEMENTS
The authors thank the study volunteers for their participation. The authors acknowledge the ORACL Co-Investigators for their co-design of the ORACL trial and for acquiring financial support for the trial. The authors acknowledge support by the research nurses, diabetes educators and dietitians at St Vincent's Hospital Melbourne and The Royal Melbourne Hospital, Melbourne, Australia. Open access publishing facilitated by The University of Melbourne, as part of the Wiley - The University of Melbourne agreement via the Council of Australian University Librarians.
FUNDING INFORMATION
The ORACL trial was funded by JDRF (3-SRA-2018-667-M-R), the Diabetes Australia Research Program, and St Vincent's Hospital (Melbourne) Research Endowment Fund. Medtronic supplied discounted insulin pumps and glucose monitoring devices for the study. SAM is supported by a JDRF Research Award and University of Melbourne Paul Desmond Clinical Research Fellowship. MHL is supported by a National Health and Medical Research Council (NHMRC) postgraduate scholarship, co-funded by Diabetes Australia. The funders of this study had no role in trial design, data collection, data analysis, data interpretation or writing of the report.
CONFLICT OF INTEREST STATEMENT
S.T. reports a research grant from Insulet Corporation. M.H.L. reports speaker honoraria from Medtronic. D.N.O. reports serving on advisory boards for Abbott, Medtronic, MSD, Novo, Roche, and Sanofi; receiving research support from Medtronic, Novo, Roche, Lilly, and Sanofi; and receiving travel support from Novo. S.A.M. reports support for research from Medtronic to her institution; receiving speaker honoraria from Eli Lilly Australia, Roche Diabetes Care Australia, and Sanofi-Aventis Australia; serving on advisory boards for Medtronic and Ypsomed; and facilitating workshops for the Australian Diabetes Society. No other potential conflicts of interest relevant to this article were reported.
Open Research
DATA AVAILABILITY STATEMENT
The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.
