A composite measure of sleep health is associated with glycaemic target achievement in young adults with type 1 diabetes

We investigated whether sleep health (each individual dimension and a composite measure) was associated with better glycaemia among a cohort of young adults with type 1 diabetes (mean age 21.5 years, mean body mass index 24.55 kg m−2). Multiple validated self‐report questionnaires were completed, and raw continuous glucose monitor data were shared. One self‐reported sleep characteristic for each of the five sleep health dimensions was selected. A composite score was calculated by summing the number of “good” sleep health dimensions. We evaluated the associations between sleep health and glycaemia, and whether covariates, including age, type 1 diabetes duration and sleep apnea risk, influenced the relationships among the study variables using multivariable linear regression. Individual dimensions of sleep satisfaction (β = 0.380, p = 0.019; β = −0.414, p = 0.010), timing (β = 0.392, p = 0.015; β = −0.393, p = 0.015) and sleep efficiency (β = 0.428, p = 0.007) were associated with higher achievement of glycaemic targets (J‐index and time in range); however, these associations did not persist after considering covariates. A better Sleep Health Composite score was associated with higher achievement of glycaemic targets even after considering covariates. Using a multidimensional framework can guide future research on causal pathways between sleep and diabetes health, interventions to target sleep health profiles, and may improve sleep screening in routine diabetes care.

type 1 diabetes (T1D; Donga et al., 2010). Specifically, alterations to these two prominent sleep health dimensions, duration and timing, lead to an imbalance between glucose production by the liver and glucose utilization by the insulin-dependent tissues (e.g. muscle and adipose) and non-insulin-dependent tissues (e.g. brain ;Spiegel et al., 1999;Spiegel et al., 2004;Van Cauter et al., 2007). Impairments in both sleep health and glucoregulation are independent predictors of all-cause mortality (Buysse, 2014;Sondrup et al., 2022).
Type 1 diabetes is an autoimmune T-cell-mediated condition characterized by destruction of beta cells, absolute insulin deficiency and a lifelong dependency on exogenous sources (e.g. insulin and glucose) that requires regular self-monitoring to achieve glycaemic targets (Burrack et al., 2017). Glycated haemoglobin (HbA1c) is the gold-standard long-term (chronic) glycaemic target measurement for individuals with T1D. Suboptimal achievement (HbA1c < 7%) is a major predictor of long-term vascular complications (Nathan & DCCT EDIC Research Group, 2014). Short-term glycaemic target achievement is often monitored day-to-day during wake and sleep through continuous glucose monitors (i.e. target time in range 70 mg-180 mg dl À1 > 80% and Glucose Management Indicator [GMI], < 7%). Self-monitoring of glucose is a unique challenge for young adults with T1D as they transition away from their childhood homes and providers to adult care and responsibilities (Morrissey et al., 2021).
Promoting modifiable dimensions of sleep health including regularity, timing, efficiency and duration improves health outcomes among adolescents and young adults without chronic conditions. However, less is known about the impact of modifying sleep health dimensions on diabetes health outcomes in individuals with T1D (Beebe et al., 2008(Beebe et al., , 2014(Beebe et al., , 2015Van Dyk et al., 2017). In prior research of adolescents or young adults, only one sleep health dimension at a time, predominantly sleep duration or sleep efficiency, has been examined with little attention given to other domains (Farabi et al., 2017;Martyn-Nemeth et al., 2018;Patel et al., 2018;Perfect et al., 2012). A combination of sleep health dimensions is likely associated with better diabetes health outcomes. Therefore, there is a need to examine the unique and cumulative contributions of sleep health dimensions on glycaemic target achievement in this at-risk population (e.g. multidimensional sleep health studies). The aim of this study was to examine the association between sleep health and glycaemia. Each sleep health dimension and a composite measure were the independent variables of interest. We hypothesized that better sleep health (individual dimensions and composite) would be associated with better achievement of glycaemic targets (less glucose variability and more time in range).

| Design
We followed the World Medical Association Declaration of Helsinki for research involving human subjects, and received approval from the

| Outcome
Glycaemic target achievement (GMI, J-index, coefficient of variation [CV] and time in range) was determined from the objective raw continuous glucose monitor (CGM) data that were shared from each participant's existing CGM to capture glucose patterns. J-index was calculated as 0.001 Â (mean + SD) 2 , and time in range was calculated as % time spent in target range 70-180 mg dl À1 . CGM systems provide real-time, dynamic glucose information every 5 min-up to 288 readings in a 24-hr period (Danne et al., 2017). CGMs are accurate across a wide range of test-retest reliability levels ranging from 0.77 to 0.95 (Danne et al., 2017). Glucose variability indices were calculated across the days of monitoring (Danne et al., 2017).

| Sleep health characteristics and dimensions
Obstructive sleep apnea risk was assessed with the validated Berlin questionnaire, and categorized as low versus high risk (Netzer et al., 1999).
Each item is ranked using a five-point Likert scale (not at all to very much; Yu et al., 2012). Scores range from 8 to 40, with higher scores indicating greater severity of sleep disturbance (Yu et al., 2012). The raw scores on the eight items are summed to obtain a total raw score.
The raw scores are then transformed into a T-score. The Cronbach's alpha for the PROMIS Sleep Disturbance in the current study was 0.875.

Alertness was measured with the eight-item Epworth Sleepiness
Scale (ESS; Cronbach's α = 0.88; Johns, 1992). Scores range from 0 to 24, with higher scores indicating higher sleepiness (Johns, 1992). The scores on the eight items are summed to obtain a total score. The Cronbach's alpha for the ESS in the current study was 0.771.
Sleep duration was measured with one item on the PSQI ("during the past month, how many hours of actual sleep did you get at night? This may be different than the number of hours you spend in bed"; Buysse et al., 1989).

| Derivation of the Sleep Health Composite
Derivation of the Sleep Health Composite was based on several considerations. We decided to dichotomize each dimension as "good" or "poor". This approach is more feasible and deployable in clinical practice, and has been documented in prior studies (Dong et al., 2019).
The Sleep Health Composite was coded as 1 = good and 0 = poor, with scores ranging from 0 to 5, and higher scores indicating better sleep health (Dong et al., 2019). We chose the cut-off point for each dimension based on the extant literature and data from previously published studies in normative samples (Dong et al., 2019;Ohayon et al., 2017). The cut-off points for the five dimensions were: satisfaction (PROMIS < 56.36, i.e. < 1 SD above the mean); alertness (ESS ≤ 7.5); timing (MEQ 42-58); efficiency ≥ 85%; duration 7-9 hr.

| Non-sleep risk factors
Participants completed a sociodemographic and clinical form and other study instruments about their diabetes management via an online survey. They shared either their raw continuous glucose monitor data or a share code from Dexcom Clarity so that the research team could access and export the raw glucose data.

| Statistical analysis
Prior to the analysis, we screened data for missing or out-of-range values, and examined distributions of continuous variables. Data were managed using the REDCap site and exported into the Statistical Package for the Social Sciences (SPSS) version 28 for analysis. CGM data were calculated with Glyculator v. 3.0 software (Czerwoniuk et al., 2011).
A quantitative descriptive approach was used to characterize sleep health dimensions and glycaemia. Glucose variability indices were calculated based on data across the days of monitoring. Descriptive statistics were used to summarize each of the variables, including the scores for multi-item scales. Self-report A1c was used to determine chronic glycaemia, and CGM data were used to calculate shortterm day-to-day glycaemia (Suh & Kim, 2015). A series of t-tests and correlations were conducted to determine differences in sleep health dimensions and glucose variables by covariates, including age, sex, T1D duration, race and body mass index (BMI).
Bivariate and multivariable linear regression models were used to examine the relationships between sleep health dimensions and glycaemia. A series of linear regression models was conducted for each individual sleep health dimension and for the Sleep Health Composite score to evaluate the explanatory contributions of sleep health dimensions to glycaemia. Covariates were included in multivariable models if they were significantly associated with sleep health (composite measure or individual dimension) or glycaemia (GMI, J-index, CV or time in range; p < 0.05). We also controlled for sleep apnea risk due to a priori knowledge of an independent effect of sleep apnea on glycaemia (Priou et al., 2012). Statistical significance was set at p < 0.05.

| Participant characteristics
Seventy-five participants (74.7% female sex assigned at birth, 9.3% gender minority) between the ages of 18 and 25 years (mean 21.47 years, SD = 2.06) participated in the study (64.9% on an insulin pump). Participants reported a mean T1D duration of 9.4 years (SD = 5.8 years) and mean HbA1c of 6.8% (SD = 1.04%). Mean GMI was 7.01 (SD = 0.63%) and mean time in range was 70% (SD = 17.2%) from raw CGM data. We present demographic, clinical profile and sleep characteristics in Table 1.

| Individual sleep health dimensions
In the first set of models, we examined the associations between each individual sleep health dimension and glycaemia as measured by the GMI. Higher satisfaction and lower sleep efficiency were significantly associated with a lower GMI (better achievement of target), accounting for 18.6% and 14.9% of the variance, respectively.
Satisfaction and efficiency had a medium effect on GMI (f 2 = 0.23 and f 2 = 0.18), and timing had a small effect on GMI (f 2 = 0.04).
Alertness and duration did not have an effect on GMI. However, the associations were no longer significant after considering covariates (age, T1D duration, race, insulin mode of delivery, and sleep apnea risk; Table 2).
In the second set of models, we examined the associations between each individual sleep health dimension and glycaemia as measured by J-index. In these models, lower satisfaction, later timing and lower sleep efficiency were significantly associated with lower achievement of glycaemic targets as measured by J-index accounting for 14.5%, 15.3% and 10.3%, respectively; however, these relationships were no longer significant when adjusting for covariates (age, T1D duration, insulin mode of delivery, race and sleep apnea risk; Table 3). Satisfaction and timing had a medium effect on J-index (f 2 = 0.17 and f 2 = 0.18), and efficiency had a small effect on J-index (f 2 = 0.11). In the next set of models (Table 4), we examined the associations between individual sleep health dimensions and glycaemia as measured by CV, and none of the associations were statistically significant.
In the next set of models, we examined the associations between duration, insulin mode of delivery, race and sleep apnea risk; Table 5).

| Sleep Health Composite score
In the last set of models, we examined the associations between the Sleep Health Composite and all glycaemia indices (GMI, J-index, CV and time in range; Table 6). Better sleep health was significantly associated with better achievement of glycaemic targets even after  significantly associated with achieving glycaemic targets in the unadjusted models; however, these were not statistically significant after Although in previous studies, poorer glycaemia was associated with individual dimensions of sleep health such as short duration and variability in sleep duration (Borel et al., 2009;Chontong et al., 2016;Griggs et al., 2020;Larcher et al., 2016;Patel et al., 2018), we suggest based on our findings that these sleep  (Borel et al., 2009;Chontong et al., 2016;Griggs et al., 2020;Griggs et al., 2022;Perfect et al., 2012;Rechenberg et al., 2020;Reutrakul et al., 2016). In these studies, poorer individual dimensions of sleep health (e.g. lower satisfaction, lower daytime alertness, variability in timing, lower efficiency and/or shorter duration) were associated with lower achievement of glycaemic targets (Borel et al., 2009;Chontong et al., 2016;Griggs et al., 2020;Griggs et al., 2022;Perfect et al., 2012;Rechenberg et al., 2020;Reutrakul et al., 2016).

| DISCUSSION
In the current study, poor sleep satisfaction (specifically sleep disturbance) was significantly associated with lower achievement of glycaemic targets. This finding is consistent with several cross-sectional studies of small clinical samples and epidemiological studies (Matejko et al., 2015;Perfect et al., 2012;Reutrakul et al., 2016;von Schnurbein et al., 2018). In other studies of adolescents with T1D, sleep satisfaction was not associated with glycaemia (Patel et al., 2018;Perfect et al., 2012). However, measures of sleep satisfaction differ among these studies. Satisfaction in the current study was measured with the PROMIS sleep disturbance scale, which measures perceptions of sleep quality, sleep depth and restoration associated with sleep over the past 7 days. The PSQI global score or an individual sleep quality item was used in the other studies (Patel et al., 2018;Reutrakul et al., 2016).
Of the individual sleep health dimensions, sleep timing had the strongest association with glycaemia. This finding is consistent with some previous studies Reutrakul et al., 2016), but not others (Perfect et al., 2012;Siwasaranond et al., 2016). Consistent with our findings in the current study, the F I G U R E 1 Scatterplots of sleep satisfaction, timing, sleep efficiency (%) and time in range 70-180 mg dl À1 % association between better or more stable timing and lower glycaemia in individuals with T1D has been highlighted in numerous studies Griggs et al., 2020;Griggs et al., 2022;Patel et al., 2018;Perfect et al., 2012;Rechenberg et al., 2020;Reutrakul et al., 2016;Reutrakul & Van Cauter, 2014).
Beyond cross-sectional studies, in a recent clinical trial, the chronic sleep restriction with recurrent circadian disruption condition had significantly elevated postprandial plasma glucose levels, whereas the condition with chronic sleep restriction and minimized circadian disruption had no adverse glycaemic effects after 3 weeks of exposure among adults without chronic conditions (Yuan et al., 2021).
Neither sleep duration nor daytime alertness as individual dimensions were significantly associated with any of the glycaemic target measures in the current study. This finding was not expected given the evidence of impaired glucose metabolism in adults without chronic conditions undergoing experimental sleep deprivation (Donga et al., 2010;Knutson & Van Cauter, 2008;Spiegel et al., 1999;Van Cauter et al., 2007). In previous comparison studies of adults with T1D, this association has been mixed, with some researchers reporting an association between a shorter sleep duration and poorer glycaemia (Borel et al., 2013;Denic-Roberts et al., 2016;Reutrakul et al., 2016;von Schnurbein et al., 2018), or between a lower daytime alertness and poorer glycaemia Perfect et al., 2012;Zhu et al., 2021). In other studies, the associations between either sleep duration or daytime alertness and glycaemia have not been significant Patel et al., 2018;Rechenberg et al., 2020).
The current cross-sectional study and analyses have several limitations to consider when interpreting the results. First, the sample was not representative of community-dwelling young adults with T1D, as a majority were Non-Hispanic White (87%) and female (75%). Therefore, results cannot be generalized to the general T1D population, to other age groups (adolescents, middle or older adults), or to clinical populations with sleep disorders.
However, the current sample did achieve glycaemic targets at comparable rates than the T1D Exchange, a national comparison sample (mean GMI 7.0% versus 7.3%; Foster et al., 2019). Also, neither sex-nor race-based differences could be determined. There is a paucity of studies focused on determining whether modifying sleep has an effect on glycaemia in young adults with T1D.
Promoting sleep duration had a positive impact on glucose targets in two pilot studies of adolescents with T1D in the short term (3 months; Jaser et al., 2020;Perfect et al., 2016;Perfect et al., 2018). Specifically, in a randomized-controlled trial, Perfect and colleagues established that a behavioural intervention aimed at increasing sleep duration by 30 min per day led to an improvement in time in range after 3 months. In a small experimental hyper-insulinaemic euglycaemic clamp study of seven middle-aged adults with T1D (mean age 44 years, SD 7 years), sleep restriction reduced the glucose disposal rate reflecting decreased peripheral sensitivity (Donga et al., 2010). It is unknown if extending sleep duration over time is sustainable, or what the long-term impact is on clinical outcomes specifically in young adults with T1D who have unique developmental and social needs.
The current study provides support of the utility of a Sleep Health Composite score. The results can help to establish validity in the context of chronic illness, and may translate to other related populations such as those with type 2 diabetes, hypertension or obesity. Future researchers may consider a similar approach in other populations and contexts. Linear relationships were examined in the current study, and in the future it may be warranted to examine both linear and non-linear associations between sleep health dimensions and glycaemia. The current study was cross-sectional, therefore the mechanistic pathways between sleep health and glucose variability could not be determined. Future longitudinal and experimental studies can provide further insight into the findings presented here and whether poor sleep precedes greater glucose variability, or vice versa, or acts bidirectionally. In addition, age effects on multidimensional sleep health and whether relationships are consistent in broader age ranges, such as adolescents and middle-to older-aged adults with T1D should be examined. Promoting healthy sleep through a multidimensional framework addresses the entire diabetes population rather than only those with sleep disorders.

AUTHOR CONTRIBUTIONS
Stephanie Griggs, PI on the grant (R00NR018886), secured the funding, designed the study, collected, analysed and interpreted the data, and wrote the manuscript. Grant Pignatiello interpreted the data and cowrote the manuscript. Senior author Ronald L. Hickman Jr contributed to the study design, interpreted the findings and co-wrote the manuscript. All authors have seen and approved the final version of this manuscript.