Altered sleep architecture in diabetes and prediabetes: findings from the Baependi Heart Study

Abstract

Study Objectives

People with diabetes and prediabetes are more likely to have sleep-disordered breathing (SDB), but few studies examined sleep architecture in people with diabetes or prediabetes in the absence of moderate-severe SDB, which was the aim of our cross-sectional study.

Methods

This cross-sectional sample is from the Baependi Heart Study, a family-based cohort of adults in Brazil. About 1074 participants underwent at-home polysomnography (PSG). Diabetes was defined as fasting glucose >125 mg/dL or HbA1c > 6.4 mmol/mol or taking diabetic medication, and prediabetes was defined as HbA1c ≥ 5.7 & <6.5 mmol/mol or fasting glucose ≥ 100 & ≤125 mg/dl. We excluded participants with an apnea-hypopnea index (AHI) ≥ 30 in primary analyses and ≥ 15 in secondary analysis. We compared sleep stages among the 3 diabetes groups (prediabetes, diabetes, neither).

Results

Compared to those without diabetes, we found shorter REM duration for participants with diabetes (−6.7 min, 95%CI −13.2, −0.1) and prediabetes (−5.9 min, 95%CI −10.5, −1.3), even after adjusting for age, gender, BMI, and AHI. Diabetes was also associated with lower total sleep time (−13.7 min, 95%CI −26.8, −0.6), longer slow-wave sleep (N3) duration (+7.6 min, 95%CI 0.6, 14.6) and higher N3 percentage (+2.4%, 95%CI 0.6, 4.2), compared to those without diabetes. Results were similar when restricting to AHI < 15.

Conclusions

People with diabetes and prediabetes had less REM sleep than people without either condition. People with diabetes also had more N3 sleep. These results suggest that diabetes and prediabetes are associated with differences in sleep architecture, even in the absence of moderate-severe sleep apnea.

Graphical Abstract

Graphical Abstract.

Statement of Significance.

Prior research has demonstrated an association between sleep quality and glucose control. In addition, people with diabetes have reported greater sleep disturbances and high prevalence of sleep disordered breathing. However, few studies have assessed sleep architecture via polysomnography in people with diabetes and pre-diabetes. In this study, we compared sleep architecture among those with diabetes, prediabetes and those with neither condition. We found that people with diabetes and prediabetes had less REM sleep after adjusting for age, sex, BMI and AHI and those with diabetes had more N3 after adjustment. Future research should examine whether these sleep stages are associated with glucose control in this patient populations.

Introduction

Type 2 diabetes mellitus is one of the most common chronic medical conditions in the world, affecting approximately 537 million people worldwide in 2021 [1]. It can lead to medical complications and early mortality, and reduced quality of life. Individuals with diabetes often report insufficient sleep duration or decreased sleep quality [2, 3]. Type 2 diabetes is also often associated with sleep disordered breathing, particularly obstructive sleep apnea (OSA) [4, 5]. Because of its association with glucose homeostasis, healthy sleep may be especially important in type 2 diabetes. For example, studies have observed associations between poor subjective sleep and higher HbA1c [6–9]. Fewer studies have examined objectively assessed sleep architecture via polysomnography in people with diabetes and prediabetes.

Polysomnography (PSG) provides detailed information on the distribution of time spent in each sleep stage across the night, known as sleep architecture. The two primary stages of sleep are rapid-eye movement (REM) and non-REM (NREM) sleep; N3 is the deepest stage of NREM. Alterations in sleep architecture have also been shown to cause impaired metabolic function. For example, three experimental studies that suppressed N3 in young healthy adults observed impairments in glucose metabolism and insulin sensitivity [10–12]. In addition, experimental sleep restriction which resulted in less REM sleep was associated with positive energy balance [13], which could increase risk of weight gain. Therefore, sleep architecture may be particularly important for people with diabetes since they rely so heavily on tight glycemic and weight control to avoid complications. A few studies have investigated whether sleep architecture is disturbed in type 2 diabetes, but results are conflicting [14–17]. These studies varied in sample sizes, with most having fewer than 200 participants [15–17] and only two adjusting for OSA in their analysis [15, 16]. OSA is commonly associated with disturbed sleep architecture, including decreased N3 and REM duration [18]. Therefore, it is unclear whether type 2 diabetes is associated with poorer sleep architecture, such as less N3 or REM sleep, particularly in the absence of moderate-severe sleep apnea.

Given that 1) experimental sleep restriction or N3 suppression can impair glucose metabolism or alter energy intake, and 2) prior work has demonstrated greater sleep disturbances, particularly sleep-disordered breathing, among people with diabetes: our primary objective was to compare sleep architecture, accounting for OSA, among those with diabetes, prediabetes and neither condition in a large, observational study from Brazil. Our exploratory hypothesis was that people with prediabetes and diabetes will have less N3 and less REM sleep than those with neither condition.

Methods

Sample

This cross-sectional sample is from an ancillary study to the Baependi Heart Study (BHS), a family-based cohort of adults in a rural town in the state of Minas Gerais in southeastern Brazil. The recruitment methodology for the BHS has been described previously [19]. Briefly, adult probands were selected from the community at random across 11 out of the 12 census districts in Baependi. Once probands were enrolled, all of their relatives, who were at least 18 years old, were invited to participate. All BHS participants were eligible to participate in this ancillary study, which aimed to collect full ambulatory PSG in the cohort. Collection and processing of PSG data from the BHS is ongoing and the PSG recordings included in these present analyses were collected between August 2019 and November 2022. PSG recordings were scored within 2.5 months of recording, on average. The BHS subsample used herein includes participants whose PSG data was collected and processed as of November 10, 2022.

As of November 11, 2022, the BHS sleep ancillary sample included 1608 participants. Participants were excluded from these analyses if the PSG recording was invalid due to poor quality (n = 280), if the PSG recorded less than 4 hours of total sleep time (n = 112), if systolic blood pressure < 50 mmHg or diastolic blood pressure < 30 mmHg (n = 4), and if AHI ≥ 30 events/hour or AHI was missing (n = 138). Our final analytic sample included 1074 participants (see Supplemental Figure S1). We compared the analytic sample to those excluded (see Supplemental Table S1), and the excluded participants tended to be older, were more likely to be female, and had higher mean glucose, HBa1c, BMI and AHI on average.

Measures

Diabetes status

Our primary exposure was diabetes status, grouped into diabetes, prediabetes or neither condition. We used the fasting blood sample collected in the morning after the PSG to assay both fasting blood glucose (FBG) and HbA1c levels. Blood collection generally occurred within 20 min to 2 h of waking. We also asked participants to report their medications. Using this information, we classified participants into three diabetes groups: diabetes, prediabetes, or neither condition. Diabetes was defined as meeting at least one of the following criteria: (1) an FBG of > 125 mg/dL, (2) an HbA1c > 6.4, or (3) taking diabetic medication. We defined prediabetes as: 1) not in the diabetes category, 2) FBG ≥ 100 and ≤ 125 mg/dl or HbA1c ≥ 5.7 and ≤ 6.4, and 3) not taking diabetic medication. All other individuals who were not missing FBG or HbA1c values were classified as having neither condition.

Sleep architecture

Our primary outcome was sleep architecture. Participants underwent one night of full ambulatory at-home PSG and provided blood samples before and after sleep. The PSG recording included the following channels: 8-channel EEG (F3, F4, C3, C4, O1, O2, A1, A2), EOG, EMG, nasal airflow, respiratory effort from two RIP belts, and pulse oximetry. The PSG recordings were staged and scored by qualified polysomnographic technologists using the American Academy of Sleep Medicine (AASM) scoring criteria [20]. The scored PSG data was used to calculate the following characteristics of sleep: total sleep time (TST; hours), total time in NREM stage 2 (N2; minutes), total time in NREM stage 3 (N3; minutes), total time in REM sleep (minutes), and total wake after sleep onset (WASO; minutes). We also calculated the percentages of total sleep time spent in N3 and REM. Finally, we calculated the apnea-hypopnea index (AHI) as an indicator of sleep disordered breathing. The AHI was calculated as the average number of apneas and hypopneas that occur per hour of sleep. Hypopneas were scored if the peak signal excursions dropped by ≥ 30%, the duration of the drop was ≥ 10 s and there was a ≥ 3% oxygen desaturation or the event was associated with an arousal (AASM definition 1A) [20].

Covariates

Covariates in these analyses include age, gender, and body mass index (BMI), which were selected because of their association with sleep and diabetes. Age was calculated based on date of birth and gender (man/woman) was self-reported. Height and weight were measured by the technicians and used to calculate BMI (kg/m²).

Statistical analysis

All statistical analyses were conducted using R Statistical Software version 4.1.2 [21]. Sample characteristics were summarized as either mean (standard deviation) or n (%), as appropriate. We used linear regression to model the association between diabetes groups (diabetes, prediabetes, neither condition) and each PSG sleep characteristic. For each sleep outcome, we fit unadjusted models (Model 1) and fully adjusted models (Model 2) that adjusted for age, gender, body mass index (BMI), and AHI. For each outcome variable, linear modelling assumptions were assessed with residual diagnostics. No violations of the modeling assumptions were found. We conducted sensitivity analyses using the same modeling procedure described above to assess whether results differed in a sample restricted to participants with moderate-severe sleep apnea (AHI < 15; N = 890). Finally, we examined potential effect modification by age since those with prediabetes and diabetes tended to be older. To do this, we included age*diabetes interaction terms in the fully adjusted models. All significance tests were two sided, and statistical estimates are reported with accompanying 95% confidence intervals and p-values.

Ethical approval

The Baependi Heart Study protocol conformed to the tenets of the Declaration of Helsinki and was approved by the Ethics Committee of the Hospital das Clínicas, University of São Paulo, Brazil (approval number 0494/10).

Results

The characteristics of the 1074 participants are summarized in Table 1. The majority of the participants were female (64.2%), and the percentage of females was similar across all three diabetes groups (prediabetes, diabetes, neither condition). Figure 1 plots the distributions of the PSG measures by diabetes group: TST, WASO, N3, and REM. In unadjusted models (Table 2), both prediabetes and diabetes were associated with shorter TST, more WASO, less REM and lower REM percentage. People with prediabetes also had less N3 and lower N3 percentage in unadjusted models.

Table 1.

Descriptive characteristics of the sample with AHI < 30 events/hour.

Characteristic	Overall	Neither condition	Prediabetes	Diabetes
N	1,074	700	261	113
Age (y)	48.8 ± 14.0	44.6 ± 12.8	55.7 ± 12.6	59.1 ± 12.1
Gender (% male)	35.8	37.1	34.5	31.0
FBG (mg/dL)	86 ± 22	80 ± 9	87 ± 12	121 ± 47
HbA1c (%)	5.60 ± 0.88	5.24 ± 0.30	5.86 ± 0.22	7.31 ± 1.82
BMI	26.7 ± 5.1	25.8 ± 4.6	28.0 ± 5.2	29.7 ± 6.1
AHI	6.9 ± 7.6	5.7 ± 7.1	8.7 ± 7.7	10.4 ± 8.8
TST (h)	6.6 ± 1.0	6.7 ± 1.0	6.5 ± 1.1	6.2 ± 1.0
WASO (min)	59 ± 41	54 ± 38	65 ± 43	73 ± 45
N2 (min)	225 ± 52	225 ± 50	226 ± 57	217 ± 57
N3 (min)	46.1 ± 34.8	48.5 ± 33.6	41.4 ± 35.8	42.5 ± 38.3
N3 %	11.7 ± 8.9	12.1 ± 8.4	10.8 ± 9.3	11.5 ± 10.4
REM (min)	79.8 ± 31.1	84.5 ± 30.2	71.9 ± 32.3	68.9 ± 29.0
REM %	20.0 ± 6.8	20.9 ± 6.3	18.3 ± 7.4	18.4 ± 7.3

All measurements reported as mean ± SD or percentage (%). Abbreviations are defined as follows: FBG = fasting blood glucose, HbA1c = hemoglobin A1c, BMI = body mass index, AHI = apnea-hypopnea index, TST = total sleep time, WASO = wake after sleep onset.

Figure 1.

Comparison of the distribution of PSG measures among individuals with neither condition, prediabetes and diabetes with AHI < 30. Asterisk denotes significant unadjusted differences (p < .05) between neither condition (referent) and prediabetes or diabetes group.

Table 2.

Regression models comparing PSG measures among individuals with neither condition, prediabetes and diabetes in the sample with AHI < 30 events/hour.

		Unadjusted Model (N = 1,074)	Adjusted Modela (N = 1,068)
PSG Measure	Diabetes Group	Beta (95% CI)	Beta (95% CI)	p-value
TST (h)	Neither condition	Referent	Referent	Referent
	Prediabetes	−0.23 (−0.37, −0.09)	−0.06 (−0.21, 0.10)	0.50
	Diabetes	−0.47 (−0.67, −0.26)	−0.24 (−0.45, −0.01)	0.04
WASO (min)	Neither condition	Referent	Referent	Referent
	Prediabetes	10.8 (5.1, 16.6)	−1.2 (−7.1, 4.7)	0.70
	Diabetes	18.8 (10.7, 26.8)	3.4 (−4.8, 11.7)	0.40
N3 duration (min)	Neither condition	Referent	Referent	Referent
	Prediabetes	−7.0 (−12.0, −2.1)	2.6 (−2.3, 7.6)	0.30
	Diabetes	−6.0 (−12.9, 0.9)	7.6 (0.6, 14.6)	0.03
REM duration (min)	Neither condition	Referent	Referent	Referent
	Prediabetes	−12.6 (−17.0, −8.3)	−5.9 (−10.5, −1.3)	0.01
	Diabetes	−15.6 (−21.7, −9.5)	−6.7 (−13.2, −0.1)	0.045
N3 %	Neither condition	Referent	Referent	Referent
	Prediabetes	−1.3 (−2.6, −0.1)	0.8 (−0.5, 2.1)	0.20
	Diabetes	−0.6 (−2.4, 1.2)	2.4 (0.6, 4.2)	0.01
REM %	Neither condition	Referent	Referent	Referent
	Prediabetes	−2.5 (−3.5, −1.6)	−1.4 (−2.4, −0.4)	0.01
	Diabetes	−2.4 (−3.8, −1.1)	−0.9 (−2.4, 0.5)	0.20

^aModel adjusted for: age, gender, BMI, AHI; six participants were excluded due to missingness.

In regression models adjusted for age, gender, BMI, and AHI (Table 2), participants with diabetes had shorter TST (−14 min on average). Also, both participants with diabetes and participants with prediabetes averaged approximately 6–7 min less REM compared to those with neither condition. People with prediabetes also had lower REM percentage (−1.4%) than those with neither condition, although people with diabetes did not differ. Finally, in the fully adjusted models, people with diabetes had longer average N3 duration (+7.6 min) and higher N3 percentage (+2.4%) than those with neither condition. There were no differences in WASO among the three groups.

We also conducted secondary analyses to exclude participants with moderate sleep apnea (AHI ≥ 15). The subsample with AHI < 15 was significantly younger, had a lower proportion of men, had lower glucose levels, lower HbA1c, lower AHI, and fewer participants with diabetes and prediabetes (see Supplemental Table S2). The unadjusted comparisons of the PSG characteristics are depicted in Supplemental Figure S1 and the regression results are presented in Supplemental Table S2. The results were similar to the primary analyses. Diabetes was associated with lower TST and higher N3 duration and percentage, and prediabetes was associated with lower REM duration and percentage. Finally, we conducted sensitivity analyses to investigate effect modification by age, but there was no significant effect modification.

Discussion

Our goal was to compare sleep architecture in people with diabetes or prediabetes to those with neither condition after mitigating potential confounding bias due to sleep-disordered breathing. Our hypothesis was that people with prediabetes and diabetes would have less N3 and less REM sleep than those with neither condition, and it was only partially supported by the analyses. Consistent with our hypothesis, both participants with prediabetes and diabetes had less REM sleep compared to those with neither condition. This association persisted even after accounting for potential confounders of this association, including age, gender and BMI. However, contrary to our hypothesis, we observed a longer duration of N3 sleep in participants with diabetes after adjusting for covariates.

Our finding that prevalent diabetes and prediabetes were associated with less REM sleep is consistent with prior work. With over 1000 participants in our sample, ours is one of the largest studies to investigate this relationship. One other large study [14], also reported a decrease in REM percentage among individuals with type 2 diabetes (they did not examine prediabetes). Other smaller studies vary in their findings when comparing those with diabetes to those without; two found no differences in REM sleep [15, 17] while another (n = 22 in each group) found a higher REM percentage in people with type 2 diabetes [16]. Our finding that N3 duration was greater among people with diabetes was contrary to other studies that either found less N3 among those with diabetes [17] or no difference [14, 15]. The discrepancies in findings among these studies, including ours, may be due to smaller sample sizes in most other studies, demographic differences between studies, including age, differences in covariate adjustment and our exclusion of those with moderate and severe sleep disordered breathing. There are many potential physiological mechanisms linking diabetes to less REM sleep, beyond the presence of sleep disordered breathing. One potential mechanism is inflammation. It is well-known that diabetes and obesity, which is common in diabetes and prediabetes, cause chronic inflammation [22], and inflammatory markers have been associated with greater REM latency (time between sleep onset and first REM epoch) [23], although the latter study did not observe a significant association between inflammation and REM percentage. Another study examined several inflammatory markers and PSG characteristics and found that greater REM percentage was associated with higher levels of complements but also lower levels of IL-6 and CFI [24], all of which are proinflammatory. Therefore, the association between sleep stages and inflammation, particularly in the setting of diabetes, requires further investigation. Diabetes also causes a multitude of vascular complications, and microvascular complications in diabetes have been associated with decreased REM duration and lower sleep efficiency [25]. However, since participants with prediabetes also have less REM sleep, it is less likely that the difference in sleep architecture is due to cerebrovascular or atherosclerotic differences. Another common clinical sequela of diabetes is peripheral neuropathy, and neuropathic pain has a strong association with sleep disturbance [26]. We did not have measures of neuropathy or pain in this study, so we could not explore these associations.

The finding that diabetes is associated with greater N3 duration was unexpected. One possible explanation for this finding is also related to inflammation, as some of the studies discussed above also found that greater levels of proinflammatory makers are associated with higher N3 [23, 24]. In addition, if people with diabetes are chronically sleep deprived (as suggested by their lower total sleep time), they may have more N3 and less REM due to the homeostatic regulation of sleep. It is well known that N3 sleep is homeostatically controlled, with increased sleep deprivation and associated increased sleep pressure leading to more N3 sleep [27]. Furthermore, it may be possible that sleep need is altered in diabetes resulting in greater sleep pressure as well. Another possible explanation is that perhaps N3 is increase as a response to impaired glucose regulation to control glucose levels since prior experimental studies suggest N3 plays an important role in glucose metabolism [10–12]. Finally, the temporal distribution of sleep stages across the night may be somewhat altered in diabetes. Normally, N3 occurs primarily in the first half of the night, while REM occurs primarily in the second half of the night. Nonetheless, there are periods of overlap and it is possible that N3 is prioritized over REM at these times, resulting in both higher N3 and lower REM. Understanding whether these alterations in sleep architecture have implications for disease management is an important next step.

This study had several strengths, which included a large sample size, full PSG recordings, as well as our examination of individuals with prediabetes. Our study is the first to our knowledge to observe poorer sleep architecture (i.e. less REM) among people with prediabetes. There are some important limitations of our study as well. These include the cross-sectional study design, which does not allow us to determine whether sleep architecture changes after the development of prediabetes or diabetes. In addition, we do not know the duration of disease for each person. Also, we did not collect subjective measures of sleep, such as sleep duration and daytime sleepiness. Having these other measures would allow for a more comprehensive evaluation of sleep health. In addition, our one-night of PSG may not have adequately captured inter-night variability in sleep architecture. Finally, the use of diabetic medication was self-reported, which raises the possibility that medications were under- or over-reported.

Since sleep architecture has been linked to impaired glucose homeostasis, it is important to understand how these sleep stages differ among people with type 2 diabetes or prediabetes. We found that diabetes and prediabetes are both associated with less REM sleep, even in the absence of moderate-severe sleep apnea. However, while the differences in REM sleep were statistically significant, the differences in means were not that large, that is 6–7 minutes less sleep, which is approximately 18%–23% of a standard deviation. It is unclear whether this difference is clinically significant. The scope of our study does not allow us to determine the clinical sequelae of less REM sleep in people with pre-diabetes and diabetes. Therefore, future work should determine whether these differences, that is less REM sleep or more N3, are associated with glucose control, or other health outcomes, in these patients. Similarly, the association with N3 should be examined further to see if it is associated with disease management or habitual sleep patterns. Sleep health, including sleep architecture, may play an important role in the health and well-being of patients with type 2 diabetes or prediabetes and warrants further consideration.

Disclosure statements

Financial disclosure. This work was supported by National Institues of Health grant 1R01HL141881. The authors have no other financial disclosures.

Non-financial disclosure. A preprint of this article was submitted to medrxiv: 2023.03.23.23287631

Data Availability. The data underlying this article will be shared on reasonable request to the corresponding author.