Multidimensional sleep health and diabetic retinopathy: Systematic review and meta-analysis.

INTRODUCTION

Diabetic retinopathy (DR) is one of the most prevalent microvascular diabetic complications. Worldwide, DR affected 103 million people in 2020, and this number is projected to rise to 160 million by 2045, with a prevalence of ~33% in North America1. It is the leading cause of blindness in working-age adults in the US and costs ~$500 million annually2. Diabetes duration, chronic hyperglycemia, hypertension, dyslipidemia and nephropathy are well-known risk factors for DR3–5. The Early Treatment Diabetic Retinopathy Study (ETDRS) categorized DR into nonproliferative (NPDR, namely mild, moderate, severe) and proliferative (PDR) stages6. Tight glycemic control, optimization of blood pressure and lipid control can reduce the risk or slow the progression of DR3. Early detection and treatment of more severe DR, utilizing panretinal photocoagulation and intravitreal injection of anti-vascular endothelial growth factor agents, can reduce the risk of vision loss 7,8.

Sleep is increasingly recognized as an important factor for health outcomes. The concept of multidimensional sleep health incorporates aspects of sleep that have been clearly associated with health outcomes (physical, mental and neurobehavioral well-being), including sleep duration, efficiency, timing, alertness and satisfaction/quality9. Sleep disturbances, including short or long sleep duration, poor sleep quality or mistimed sleep, are now well-known risk factors for incident diabetes, particularly type 2 diabetes (T2D)10. In patients with T2D, short sleep duration and poor self-reported sleep quality were associated with poor glycemic control11. Further, short sleep duration was shown to be associated with hypertension12. Therefore, it is conceivable that these sleep disturbances could be risk factors for DR. Obstructive sleep apnea (OSA) is a common sleep disorder observed in those with T2D, with the prevalence between 58–86%13. Besides contributing to insulin resistance and elevated blood pressure, intermittent hypoxia in OSA leads to endothelial dysfunction, increased systemic inflammation and oxidative stress13 which could contribute to progression of DR14. Studies have explored the link between sleep disturbances and DR including a few meta-analyses. Zheng et al performed a meta-analysis of 7 cross sectional studies and found that short and long self-reported sleep duration were both associated with DR, with a respective odds ratio (OR) of 1.49 (95% confidence interval (CI) 1.15, 1.94) and 1.83 (95% CI 1.36, 2.47)15. In the same review, the author reported that self-reported sleep quality (Pittsburg Sleep Quality Index, PSQI, and Epworth Sleepiness Scale, ESS) was worse in patients with DR as compared to those without DR15. In addition, a meta-analysis of 10 studies also found that OSA was associated with DR, OR=1.57, 95%CI 1.09, 2.2714. While this evidence supports the association between sleep disturbances and DR, causality could not be established due to a cross-sectional design of the studies. A bidirectional relationship between sleep and DR is also possible as patients with diabetes complications may have poorer general health status than those without complications, which could result in sleep disturbances. A few recent studies have explored the effects of continuous positive airway pressure (CPAP) treatment on DR progression; however, the results were mixed16,17.

Emerging evidence suggests that melatonin regulation was altered in DR, possibly as a consequence of a dysfunction of intrinsically photosensitive retinal ganglion cells (ipRCGs) located in the retina18–20. The ipRGCs contain the photopigment melanopsin and detect light. These cells project directly to the central circadian clock through the retinohypothalamic tract to synchronize (entrain) the sleep-wake cycle and melatonin secretion. Using salivary melatonin, overnight urinary secretion of 6-sulfatoxymelatonin (aMT6s, a major melatonin metabolite) or plasma melatonin, researchers have shown that melatonin production was lower or mis-timed in DR compared to no-DR patients 21–23, although some found this to be significant in patients with PDR only24. As melatonin is known to play a role in sleep and circadian regulation, melatonin dysregulation in DR could be one of the mediators linking DR and sleep disturbances.

While previous studies have explored various aspects of sleep in DR, none have comprehensively examined multidimensional sleep health. Therefore, the aim of this systematic review and meta-analysis was to first comprehensively explore the link between multidimensional sleep health, both objectively and subjectively assessed, and DR and its severity. We included OSA in this review as it is a very common sleep disorder in diabetes13. Melatonin regulation was also evaluated to determine whether it was disturbed in DR. Secondly, we aimed to review the effects of sleep intervention, particularly CPAP treatment, on DR incidence or progression which could help establish causality. The result of this review will broaden our understanding of sleep health in DR.

METHODS

This systematic review was performed and reported according to the Preferred Reporting Items for Systematic Review and Meta-analyses (PRISMA) guideline25. The protocol of this review is registered in the PROSPERO website (CRD42022370142).

Investigators identified relevant studies from searches of PubMed, Embase and Scopus databases since their inception through October 2022. Reference lists of included studies were explored for identifying additional relevant studies. Search terms and search strategies are described in Appendix 1.

Two reviewers (Y.L., M.S.) independently selected studies, following the PRISMA flow diagram (Figure 1). Disagreements between the two reviewers were resolved by discussion and consultation with senior advisors (S.R., T.A..).

For the first aim exploring the association between multidimensional sleep health, OSA, melatonin and DR, observational studies (i.e., cross-sectional, case-control, prospective and retrospective cohort studies) were eligible for this review if they met the inclusion criteria as follows: 1) Participants were adults with type 1 or type 2 diabetes, 2) The outcome of interest was diabetic retinopathy, which was assessed objectively on clinical examination or verified from the medical records, as defined below, 3) The factors of interest were sleep parameters (i.e., sleep duration, sleep satisfaction, timing, alertness or OSA) and melatonin level. Sleep parameters were assessed by objective or subjective measurements. Melatonin level was measured from salivary, serum or urine melatonin metabolites.

For the second aim exploring the effects of OSA treatment and DR incidence/progression, randomized controlled trials or intervention studies (non-randomized) were eligible for this review if 1) Participants were type 1 or type 2 diabetes patients with OSA or sleep-disordered breathing, 2) Compared the effect between OSA treatments (e.g. CPAP) and non-OSA treatment, and 3) Measured the outcomes as incidence of diabetic retinopathy or progression of diabetic retinopathy.

Three reviewers independently extracted the data using a standardized data record form, with at least two reviewers extracting data independently for each study. Study characteristics (author, year, location) and participants characteristics (age, sex distribution, body mass index (BMI), type of diabetes, hemoglobin A1C), sleep characteristics, melatonin assessments and DR staging were extracted. Discrepancies were resolved by discussion and consensus with senior advisors (S.R. and T.A.). The corresponding authors were contacted if there were missing data. Overall, we contacted 11 authors and received responses from 4 authors 26–29.

Multidimensional sleep health included the following:

Sleep duration, as assessed by questionnaires or objectively measured [actigraphy or polysomnography (PSG)]. Short sleep or long sleep was defined per the original studies. Sleep duration as a continuous variable was included if available.

Satisfaction or/ quality, as assessed by questionnaires (mostly PSQI30) or a diagnosis of insomnia. For PSQI, poor sleep quality was defined as a score of >5, or >7 as per population specific cutoff 31. PSQI score as a continuous variable was included if available.

Efficiency, as assessed by sleep efficiency (percentages of time in bed spent sleeping) from actigraphy or PSG

Alertness, as assessed by Epworth sleepiness scale (ESS)32. Poor alertness was defined as a score of >10. ESS as a continuous variable was included if available.

Sleep timing (e.g. bed and waketime, sleep midpoint)/regularity (variation of sleep duration or timing), as assessed by questionnaire or objective sleep recording.

For OSA, this was either assessed by questionnaires (Berlin Questionnaire or STOP-Bang33,34) or objectively by PSG, oximetry or other portable monitors such as WatchPAT, Apnea Link, or Embletta. The participants were categorized as having high risk for OSA (per questionnaire cutoff) or OSA (per cutoffs of the diagnostic tools, for example, an apnea hypopnea index (AHI) of ≥5 from PSG, or a cutoff using an oxygen desaturation index (ODI)) per the original studies. In addition, AHI as a continuous variable was analyzed.

For melatonin, serum or salivary measurements as well as overnight urine 6-sulfatoxymelatonin, a major melatonin metabolite, were included. In addition, dim light melatonin onset (DLMO) as assessed by salivary or plasma melatonin was also included.

We included studies using CPAP, a gold standard treatment for OSA. CPAP compliant vs. non-compliant was defined according to the original studies.

Retinopathy was assessed according to the original studies using standard objective measures including ophthalmologic (fundus) examination or fundus photography, and a diagnosis as documented in medical records. The grading of retinopathy was categorized according to the original studies (no DR, mild/moderate/severe non-proliferative DR (NPDR), proliferative DR (PDR)).

Pairwise meta-analysis was performed if there were more than two studies that measured similar sleep parameters and outcomes. For continuous data, mean difference of OSA severity (AHI), sleep duration (hours), sleep satisfaction scores (PSQI), alertness (ESS) and sleep efficiency between DR and no-DR were pooled using unstandardized mean difference (USMD). For melatonin, mean serum and saliva melatonin levels, mean overnight urine 6-sulfatoxymelatonin between two groups were pooled using SMD, while the subgroup analysis with urine 6-sulfatoxymelatonin levels was pooled using USMD.

For dichotomous data, odds of having DR and no-DR were estimated and compared between OSA vs non-OSA, sleep ≤ 6 hours (short sleep) vs 7–8 hours vs > 8 hours (long sleep), poor sleep satisfaction vs good sleep satisfaction, and poor alertness vs. normal alertness. ORs of each study were estimated and pooled using inverse variance method. Heterogeneity between studies was estimated using the Q-test and I2 statistic. Heterogeneity between studies was considered, if the P-value from the Q-test was less than 0.10 or the I2 statistic was greater than 25%. Random effect model using DerSimonian & Laird method was applied for both continuous and dichotomous data35, if heterogeneity between studies was presented; otherwise, a fixed effect model was applied.

If there was heterogeneity between studies, the possible sources of heterogeneity were explored by fitting covariates (i.e., sex, mean age, and BMI) one by one in a meta-regression. Subgroup analysis according to severity of DR was performed for OSA, sleep duration, and sleep satisfaction, efficiency and alertness analyses. Severity of DR was classified as moderate DR or greater and no-DR. Additionally, Subgroup analysis according to methods for sleep assessment (objective vs. subjective) was performed wherever applicable. Potential publication bias was examined using the Egger test, and a funnel plot. A contour-enhanced funnel was performed if asymmetrical funnel plot was present to explore whether the cause of asymmetry was resulted from heterogeneity between studies or small study effect. . All statistical analyses were performed using STATA version 18 (College Station, Texas). A P-value<0.05 was considered as statistical significance for all tests, except for the heterogeneity test in which a P-value<0.10 was used.

Risk of bias was assessed for case-control and cohort studies using the Newcastle-Ottawa Scale (NOS)36. For each study design, nine domains were assessed categorically by selection, comparability, and outcome. A maximum of four, two and three points can be awarded for each category of NOS and correspond to numbered domains (D1-D9). For case-control studies, these domains included: adequacy of the case definition (D1), representativeness of the cases (D2), selection of controls (D3), definition of controls (D4), comparability of cohorts on the basis of the design or analysis (D5-D6), ascertainment of exposure (D7), same method of ascertainment for cases and controls (D8), and the nonresponse rate (D9). Analysis domains included for cohort studies were as follows: representativeness of cohorts (D1), selection of the nonexposed cohorts (D2), ascertainment of exposure (D3), demonstration that outcome of interest was not present at start of study (D4), comparability of cohorts on the basis of the design or analysis (D5-D6), assessment of outcome (D7), whether follow-up was long enough for outcomes to occur (D8), and adequacy of follow up of cohorts (D9). Assessment of risk of bias for cross-sectional studies was accomplished using a modified Newcastle-Ottawa Scale. The modified scale was derived from the NOS for cohort studies and included analysis of domains D1-D7 as described earlier while domains D8-D9 were excluded, which resulted in a maximum point breakdown of four, two, and one for the NOS categories described previously. The randomized controlled trial (RCT) was assessed for risk of bias by utilizing the Cochrane Risk of Bias Tool 2 37. Seven domains were assessed with each domain having the capability to be awarded one point, i.e., random sequence generation (D1), allocation concealment (D2), selective reporting (D3), other bias (D4), blinding of participants and personnel (D5), blinding of outcome assessment (D6), and incomplete outcome data presence (D7). Risk of bias was evaluated by two independent reviewers (M.S. and Y.L.). A third reviewer (B.Z) was brought in to settle disagreements when there was uncertainty over disputed domains. Overall risk of bias was determined by point totals with a higher score suggesting a higher quality. The NOS allows a score from 0–9 points, while the modified NOS and Cochrane Risk of Bias Tool ranges from 0–7.

RESULTS

Figure 1 shows the flow of study selection. We identified 205 studies from PubMed, 415 from Scopus and 446 from Embase, yielding 588 studies. After excluding duplicates, we screened a total of 543 studies. A total of 42 studies (44 cohort) with a total of 2,419,768 participants qualified for aim 1 exploring the association between sleep dimensions, OSA and DR; characteristics of these studies are shown in Tables 1 and 2. For aim 2, 5 studies with a total of 562 participants qualified for the analysis exploring the effects of OSA treatment on DR incidence or progression. Characteristics of these studies are shown in Table 3.

Of the 42 studies (44 cohorts), 33 included participants with T2D, 2 included those with T1D, 3 with both types of diabetes and 4 did not specify types of diabetes. All studies utilized objective assessments of DR (ophthalmologic (fundus) examination, or fundus photography) or documented diagnosis codes in medical records. For sleep variables, a total of 28 studies explored association between OSA and DR, 12 studies explored sleep duration, 8 explored sleep satisfaction (self-reported questionnaires or insomnia diagnosis), 4 explored sleep efficiency (by actigraphy), 5 explored alertness, 7 studies explored melatonin or its metabolite levels, and 2 studies explored sleep timing. The majority of the studies (n=22) were conducted in the Asia-Pacific region, 18 were conducted in Europe or North America, and 2 studies were conducted in the Middle East. Characteristics of participants, e.g., age, sex, BMI and hemoglobin A1C, are shown in Table 1.

A total of 28 studies (30 cohorts, n=201,567) were included in the meta-analyses (Figures 2 and 3)20,21,26,27,29,38–60. Twenty-eight cohorts were cross-sectional in design, 2 were retrospective and 1 was prospective. Three studies utilized questionnaires to assess OSA risk42,45,58 while others utilized objective sleep assessment (PSG, oximetry, other portable monitors) or documented diagnosis codes for OSA. Data used for pooling the effect of OSA on risk of DR are presented in Table S1. The results revealed that OSA was significantly associated with having DR, with an OR of 1.34, 95% confidence interval (CI) 1.07, 1.69 (Figure 2). The degree of heterogeneity among studies was high (I2= 86.48%). When focusing on those with moderate DR or greater, the results showed no significant association with OSA, (OR 1.89, 95% CI 0.82, 4.35) (Figure S1). Similarly, heterogeneity was high (I2= 81.93%).

Meta-regression was conducted to explore the source of heterogeneity including age (<60 vs. ≥60 years), BMI (<30 vs. ≥30 kg/m2), and methods of OSA assessment (questionnaire, objective sleep assessments or diagnosis codes). Subgroup analysis by age reduced heterogeneity and revealed a significant association between OSA and DR in those <60 years, OR 1.60 (95%CI 1.25, 2.06), I2=45.10%, but not in those ≥60 years, OR 1.17 (95%CI 0.79, 1.73), I2=65.77%, Figure S2. BMI also contributed to heterogeneity. Associations between OSA and DR were significant in both those with BMI <30 kg/m2 and ≥30 kg/m2, OR 1.45 (95%CI 1.07, 1.96, I2=44.02%) and 1.39 (95%CI 1.04, 1.84, I2=61.75%), Figure S3, respectively. Methods of OSA assessments also contributed to heterogeneity. When using questionnaires, OSA was not significantly associated with DR, OR 1.95 (0.94, 4.06), I2=30.58%, while the association was significant when objective sleep assessments were utilized, OR 1.40 (95%CI 1.07, 1.85), I253.24%, Figure S4. The association between OSA and DR was not significant in studies using diagnosis codes and heterogeneity remained high (I2=97.92%).

Association between severity of OSA and DR was explored by comparing mean differences of AHI between those with DR vs. no-DR. This revealed non-significant differences in AHI between the two groups (MD 1.66 event/h, 95%CI −1.04, 4.35), with moderate heterogeneity (I2= 58.55%), Figure 3.However, those with moderate DR or greater had significantly higher AHI when compared to those with no-DR, MD 1.86 events/h, 95%CI (0.49, 3.24), with moderate heterogeneity (I2= 58.55%), Figure S5. Meta-regression was conducted to explore the source of heterogeneity. Age contributed to heterogeneity as subgroup analysis showed low heterogeneity in those ≥60 years (I2=0.0%) but remained high in those <60 years (I2=65.13%), Figure S6.

A total of 12 studies (n=10,908) were included in meta-analyses20–22,26,29,61–67. Data used for pooling the effect of short sleep and long sleep duration on risk of DR are shown in Tables S2 and S3. The results revealed that short sleep (≤6 hours) was not significantly associated with DR (OR 1.09, 95% CI 0.86, 1.38), Figure 4A. Heterogeneity among studies was moderate with I2= 59.95%. The association between short sleep and moderate DR or greater also was not significant, OR 1.35 (95%CI 0.82, 2.33), I2=0.00%, Figure S7. Subgroup analysis based on methods of sleep duration assessment (subjective vs. objective) revealed no significant associations between short sleep and DR (Figure S11). Exploration of the source of heterogeneity was performed and revealed that age (<60 vs. ≥60 years), and BMI (≤25 vs. >25 kg/m2) contributed to the heterogeneity. Association between short sleep and DR was significant in those ≥60 years, OR 1.50 (95%CI 1.27, 1.77), I2=0.0%, but not in those <60 years, OR 0.88 (95%CI 0.74, 1.04), I2=0.0% (Figure S8). However, no significant association between short sleep and DR was found in BMI subgroup analyses; heterogeneity was reduced to 0.0% in studies with participants’ BMI >25 kg/m2 but remained moderate in those with BMI ≤25 kg/m2, I2=62.64% (Figure S9).

Long sleep (>8 hours) was found to be significantly associated with DR (OR 1.41, 95%CI 1.21, 1.64), I2=0.0%, Figure 4B. When focusing on those with at least moderate DR, long sleep was not significantly associated with at least moderate DR, OR 1.73 (95%CI 0.92, 3.27) with low heterogeneity, I2=0.0%, Figure S10. When analyzing by methods of sleep duration assessment, it was found that long sleep duration as assessed by questionnaires was associated with DR, OR 1.39 (95%CI 1.15, 1.68) but not when assessed by actigraphy, OR 1.00 (95%CI 0.08, 12.34), Figure S12.

Mean differences (MD) in sleep duration between those with DR vs. no-DR were not found to be significant (MD −0.08 hours, 95%CI −0.33, 0.22), I2=65.21%, Figure S13. Similarly, the differences between those with moderate DR or greater vs. no-DR were not significant, MD 0.19 hours (95%CI −0.12, 0.49), I2=0.0%, Figure S14. Methods of sleep duration assessment (objective vs. questionnaire) were explored as potential sources of heterogeneity; however, these were not found to be contributing (Figure S15).

Eight studies20–22,31,62,66,68,69 (n=2,208,550) were included in the analyses, with seven utilizing PSQI and one utilizing insomnia diagnosis code. The results revealed that poor sleep satisfaction was significantly associated with DR, OR 2.04 (95%CI 1.41, 2.94), I2= 77.67, Figure 5A. The association remained significant with greater effect size when focusing on patients with moderate DR or greater, OR 3.94 (95%CI 1.49, 10.40), I2=78.32%, Figure S16.

Sleep satisfaction was poorer in those with DR vs no-DR, and those with moderate DR or greater vs. no-DR as reflected by higher PSQI scores, MD 2.81 (95%CI 1.89, 3.72), Figure 5B, and 2.67 (1.72, 3.62), Figure S17, respectively. Heterogeneity was nonexistent in both analyses (I2=0.0%).

Four studies were included in the analysis (n=378)20–22,54. The results revealed that there were no differences in sleep efficiency between those with DR vs. no-DR, MD −1.63% (95%CI −4.87, 1.62), I2 44.34%, Figure 6A; nor in those with moderate DR or greater compared to no-DR, MD −1.85% (95%CI −5.63, 1.92), I2 48.16%, Figure S18.

Five studies (n=1,792) were included 20–22,54,61 (Table S4).There was no association between poor alertness (ESS>10) and DR, OR 0.85 (95%CI 0.40, 1.81), I2=32.56%, Figure 6B. Subjective alertness as assessed by ESS was not significantly different between those with DR vs. no-DR, MD 0.49 (95% CI −1.16, 2.14), Figure S19, and in those with at least moderate DR vs. no DR, MD 0.69 (95%CI −1.30, 2.69), Figure S20. The degrees of heterogeneity were moderate, I2 55.96% and 58.84%, respectively.

Only two studies explored sleep timing and regularity in T2D patients with and without DR who were non-nightshift workers. The first cross-sectional study was conducted in Thailand and included 10 healthy controls, 10 individuals with T2D without DR and 15 individuals with T2D with at least moderate DR21. Sleep timing was assessed by actigraphy. Sleep midpoint (time point between sleep onset and offset) was not significantly different among the three groups. Sleep irregularity (as assessed by variability of sleep duration across the night) was higher in T2D with DR as compared to control participants (75.1(23.9) vs. 47.8 (21.2) minutes) but not with T2D and no-DR. Another cross-sectional study was conducted in the U.S. and included 45 participants (15 healthy controls, 15 T2D with no-DR, and 15 T2D with at least moderate DR)20. Sleep midpoint did not differ among groups, but sleep duration variability was significantly higher in patients with DR vs. no-DR (89.3 (42.8) vs. 55.4 (21.9) minutes.

For melatonin, seven studies were included20–22,24,28,68,70, in which four utilized overnight urinary 6-sulfatoxymelatonin, one utilized plasma melatonin at 6 AM68, one utilized plasma melatonin at 8 AM71, and one utilized salivary melatonin every 4 hours with peak and average nocturnal values (peak values were included in the analysis)22. Pooling these studies revealed that those with DR had significantly lower melatonin/melatonin metabolites than no-DR group (standardized mean difference, SMD, −0.94, 95%CI −1.44, −0.44), I2=82.13%, Figure 7A. When analyzing only the studies utilizing overnight urinary 6-sulfatoxymelatonin/creatinine ratio, the level was significantly lower in those with DR vs. no-DR, MD −7.12 ng/mg (95%CI −11.46, −2.77), I2=77.71%, Figure 7B. Meta-regression was conducted to explore the source of heterogeneity. Sensitivity analysis considering sex distribution of participants (male <50% vs. ≥50%) was performed and this reduced heterogeneity in both groups. SMD of melatonin between those with DR vs. no-DR in studies with <50% male participants was −1.45 (95%CI −1.90, −0.99), I2=48.33%, and studies with ≥50% male participants was −0.54 (95%CI −0.82, −0.26), I2=0.0%, Figure S21.

One study explored DLMO using 9-hour salivary melatonin collection in the evening and revealed that some participants did not have detectable DLMO20. The result revealed that a lower proportion of patients with DR had detectable DLMO as compared to no-DR or control groups (33.3% vs. 85.7% vs. 83.3%, p=0.049).

Five studies (n=562) qualified for systematic review17,39,72–74, Table 3. Three studies were prospective cohorts17,39,72, one was a retrospective cross-sectional study73 and one was an RCT74. All studies were conducted in North America or the U.K. Due to the varying nature of the studies and their outcomes, meta-analysis could not be performed. Therefore, narrative review was provided.

Altaf et al. studied 38 adults with T2D with moderate to severe OSA who were referred for CPAP treatment, with an average follow up of 43 months39. Of the 38 individuals, 15 were CPAP compliant (no definition provided). Progression from no DR or background DR to advanced DR, defined as pre-proliferative and proliferative DR, was significantly less in CPAP compliant vs. non-compliant groups (0.0% vs. 39.1%). The study also explored the progression to maculopathy, defined as exudates, microaneurysm, hemorrhage and/or thickening within one disk diameter of the fovea. No differences in progression to maculopathy was found between the two groups.

Kaba et al. conducted a prospective cohort (12-month follow up) in 44 eyes of T1D and T2D with DR and moderate to severe OSA, as determined by AHI≥15 from PSG17. Thirty eyes had co-existing diabetic macular edema (DME) and 14 eyes did not have DME. For those with DME, 6 of 30 individuals were CPAP compliant (no definition provided). CPAP compliant participants had significantly lower DR severity score as compared to CPAP non-compliant participants at 12-months [1.00(0.0) vs. 1.36 (0.80), p=0.042]. For those with DME, 5 out of 14 participants were CPAP compliant. No significant differences between the CPAP compliant vs. non-compliant groups in DR severity score at 6- and 12-months were found.

Mason et al. studied 28 T2D patients with OSA and clinically significant macular edema (CSME) in a 6-month prospective study72. CPAP compliance was defined as average use ≥2.5 h/night. No significant differences were observed in fundus photograph or macular edema between the CPAP compliant (n=13) and the noncompliant (n=15) groups, but visual acuity improved in the CPAP compliant group as compared to the non-compliant group; adjusted difference 0.11, p=0.047.

Smith et al. performed a retrospective cross-sectional review of 321 T2D participants who were being treated with CPAP (63 with DR and 258 with DR)73. CPAP compliance was defined as average use ≥4 h/night, ≥70% of nights. CPAP compliance significantly predicted lower prevalence of any DR in a multivariate analysis, OR 0.49 (95%CI 0.26, 0.94).

One RCT was conducted by West et al in 131 T2D subjects with DME and visual impairment with severe OSA (AHI≥15 or 4% oxygen deoxygenation index (ODI) ≥20)74. Participants were randomized to CPAP plus ophthalmology care vs. routine ophthalmology care only for 12 months. CPAP compliance was defined as average use ≥2 h/night. In this study, CPAP reports at 3 months showed that 19% use ≥ 4h/night between 60–100% of the time, 27% at 3 months and 22% at 12 months. There were no differences between CPAP vs. routine care groups in progression of retinopathy or central macular thickness. In addition, there were no differences in visual acuity between CPAP vs. control groups; nor between CPAP compliant vs. non-compliant groups.

Overall, these studies yielded mixed results with potential beneficial effects of CPAP in patients with diabetes and OSA on DR or visual acuity. In addition, CPAP compliant criteria varied by studies and compliance rate was generally low.

Descriptions of the quality of the case-control and cohort studies are presented in Figure S22A and Figure S22B, respectively. All cohort studies had a low risk of bias in domains D3-D7 and reported a low risk of bias in domains D1, D2, and D8 with percentages of 85.7% and domain D9 with a percentage of 71.4%. In case-control studies, domains D4, D5, D7, and D8 similarly all had a low risk of bias while domains D1, D2, D3, D6 and D9 had a 75% low risk of bias. A large majority of the analyzed studies were cross-sectional in nature as shown in Figure S22C. Most cross-sectional studies reported a low risk of bias in domains D1-D7 with respective percentages: 78.4%, 86.5%, 91.9%, 89.2%, 91.9%, 83.8%, and 89.2%. The one RCT was rated low risk of bias in all domains, suggesting good quality. Overall, quality assessment on a domain basis indicated a higher quality with respect to each study type.

For OSA, a small study effect was suggested from the Egger test (P-value = 0.036) and asymmetrical funnel plot. Contour enhanced was further explored. Some studies were missing in the significant area indicating that asymmetrical funnel plot might be resulted from heterogeneity between studies rather than publication bias, Figure S23. For sleep duration, results from Egger test (P-value = 0.213) and symmetrical funnel plot suggest no publication bias for pooling the effect size of sleep duration, Figure S24.

For sleep satisfaction, publication bias was analyzed for the studies using PSQI. Small study effect was indicated by both the Egger test (P-value = 0.007) and an asymmetrical funnel plot. Contour enhanced funnel plot revealed that certain studies were missing in both significant and non-significant areas, suggesting that the asymmetrical funnel plot may be a result of publication bias, Figure S25. For melatonin, the results of the Egger test (P-value = 0.996) and the symmetrical funnel plot both indicate no evidence of publication bias, Figure S26.

DISCUSSION

This systematic review and meta-analysis comprehensively examined multidimensional sleep health in diabetic retinopathy, and whether treatment of OSA, a common comorbid sleep disorder in those with diabetes, could improve visual health and DR. We found that having OSA was associated with ~34% increase in the risk of having DR, and the findings were confirmed in the studies utilizing objective assessments of OSA. According to American Academy of Sleep Medicine Clinical Practice Guideline, objective assessment is mandatory for OSA diagnosis since questionnaire may not be accurate and could be associated with false positive result and low specificity75. In addition, subjects with at least moderate DR had higher mean AHI, by 1.86 events/h, than those without DR which support the association between OSA and DR. In our study, significant association between OSA and DR was only observed in those<60 years of age. These findings were consistent with the recent publication by Vgontzas AN, et al which demonstrated increased mortality risk in mild and moderate OSA only in those <60 years of age76. This could partly be due to referral bias, in which healthier older people were referred to a sleep laboratory; also, a survivor bias, in which older people remaining alive tend to be heathier therefore OSA may have less effects on DR development 77. Furthermore, older age may be protective from mild chronic intermittent hypoxia associated with apnea-related activation of cardioprotective adaptive pathways77,78. In addition, three of five studies exploring the use of CPAP showed potential benefits in decreasing DR progression or its severity, although the only RCT did not show positive results. For sleep duration, long sleep was associated with 41% increase in the risk of having DR, while short sleep was associated with DR in those ≥60 yr. Poor sleep satisfaction (insomnia, poor self-reported sleep quality) doubled the risk of having DR, the association of which was even stronger with at least moderate DR. However, in this review, alertness and sleep efficiency were not associated with DR, but the number of studies was small. The evidence regarding sleep timing was limited but suggestive of increased sleep variability in those with DR. Lastly, melatonin production (saliva, serum, overnight urinary 6-sulfatoxymelatonin) was found to be significantly lower in those with DR. The results of this review highlight multiple abnormalities in sleep health in people with DR.

The pathogenesis of DR involves changes in the neurovascular unit (blood vessels, neurons, glia cells)79. Impaired autoregulation of the capillary bed and capillary wall dysfunction alter blood flow to the retina and can lead to the leakiness of the capillaries80. Hyperglycemia, inflammation and hypoxia leading to overexpression of vascular endothelial growth factor (VEGF) can contribute to this process79,81. OSA, associated with insulin resistance and hyperglycemia, can contribute to this pathogenesis as increased inflammatory markers and VEGF levels were shown in those with OSA82,83, and further that CPAP treatment led to a reduction in VEGF levels84. Further, OSA was shown to be associated with thinning of the retinal nerve fiber layer (RNFL)85 and CPAP treatment improved RNFL thickness86. In a separate study, 36 patients with OSA syndrome who had treatment with uvulopalatopharyngoplasty or CPAP showed improvement in retinochoroidal parameters (e.g. choroidal and retinal thickness, capillary density index) at 6-months while these parameters worsened in no-treatment group.87 Our meta-analysis results, although limited by most studies being cross-sectional in design, supported the association between OSA and DR, and were in agreement with previous meta-analysis results14 but with a larger number of included studies. In addition, favorable results were seen on some studies exploring OSA treatment and DR progression, but these were limited by a small number of participants, non-RCT design, and varied CPAP compliance criteria among studies. Thus, larger and longer follow up RCTs are needed to evaluate the benefits of CPAP or other OSA treatment on DR incidence/progression.

Both long and short sleep durations have been shown to predict diabetes, exhibiting a similar effect size to that of physical inactivity88. Likewise, a recent meta-analysis has demonstrated that both long and short sleep durations increase the risk of DR15. In contrast, our results only illustrate this relationship between long sleep and increased DR risk, likely partly due to additional studies being included in the current review. Upon subgroup analysis for individuals aged 60 and older, the connection between short sleep duration and DR risk emerged in our current analysis, suggesting that short sleep could be more detrimental to ocular health in this group than the younger population. Long sleep could possibly serve as a marker of poor health or poor sleep quality, often associated with insomnia and OSA, thus an individual might spend a longer time in bed. Additionally, the link between long sleep and DR may also be explained through factors like decreased physical activity, increased insulin resistance, and obesity89. Furthermore, systemic inflammation, integral to DR pathogenesis, is associated with both long and short sleep, showing heightened circulating cytokines (CRP and IL-6)90, which could exacerbate the development of DR91. Note that in our analysis, only self-reported, not objectively measured, long sleep was associated with DR. This could be partly due to a small number of included studies utilizing objectively measured sleep duration. It is also possible that self-reported long sleep duration reflected time spent in bed, rather than the actual sleep duration. Individuals with DR generally have other medical comorbidities and possibly spend longer time in bed than those without DR.

Poor sleep satisfaction was significantly associated with a 2-fold risk of having DR, and approximately 4-fold for moderate DR or greater. Poor sleep quality predicted poor glycemic control11, which is a risk factor for DR. In addition, insomnia was also reported to be associated with elevated inflammatory markers including C-reactive protein and interleukin-692, and reduced RNFL thickness93. A bidirectional association between sleep satisfaction and DR potentially exists, as patients with DR are likely to suffer from other diabetes complications and co-morbidities, causing sleep disruption. This could possibly become a vicious cycle in worsening DR severity. Whether treating insomnia or improving sleep quality will lessen DR severity remains to be elucidated.

Patients with DR had significantly lower melatonin/melatonin metabolites than those without DR. Evidence shows that lower melatonin secretion was independently associated with a higher risk of type 2 diabetes development94. Recent evidence suggested that individuals with DR may exhibit abnormal melatonin rhythm22. One possible explanation could be reduced intrinsically photosensitive retinal ganglion cell (ipRGC) function in DR. ipRGCs primarily control non-image-forming light responses, communicating to the suprachiasmatic nucleus within the hypothalamus and beyond. It has been found that reduced ipRGC function in DR is associated with circadian dysregulation and sleep disturbances95. The causal relationship between melatonin and DR has been explored in animal studies. Melatonin levels are decreased in the retina and could be rescued by insulin treatment 96. Melatonin treatment presents promise in DR intervention, seen in several animal studies. It reduces reactive oxygen species, which contributes to corrections of the amplitude of electroretinogram ‘a’ wave and ‘b’ wave97. Melatonin decreases inflammatory markers of DR, such as tumor necrosis factor-α, interleukin-1β, inducible nitric oxide synthase (iNOS)97, and matrix metalloproteinases-998. Melatonin also reduces the high glucose-induced increase in vascular endothelial growth factor in retinal Müller cells99. Disruption in melatonin regulation seen in DR could potentially alter circadian regulation and further disturb sleep patterns, creating a vicious cycle. Further research and clinical trials are crucial to establish the efficacy of melatonin supplementation in DR treatment.

Sleep efficiency and daytime alertness were not associated with DR in this analysis, although the number of included studies was small. The evidence regarding sleep timing and regularity was also limited, with two small studies suggesting an increase in sleep variability in DR. It was unclear if sleep variability, known to impact glycemic control100 and inflammation101, played a role in DR, or that the lower melatonin production in patients with DR contributed to difficulties maintaining regular sleep.

The main strength of this review is a comprehensive examination of multidimensional sleep health in DR. However, there are several limitations including small number of studies in some analyses, most studies being cross-sectional precluding a causal relationship assumption, high heterogeneity among studies and the presence of publication bias in certain analyses. Classifications of T1D were not uniform among studies included in this meta-analysis, thus some T1D participants could be insulin-treated T2D. However, this study aimed to include both types of diabetes. Certain diabetes treatments, such as glucagon like peptide 1 receptor agonists, were reported to worsen DR 102. However, we could not analyze the effects of diabetes medication use on DR as this information was not reported in the original studies. In addition, the definition and staging of DR varied among studies, but all definitions were based on the presence of vascular abnormalities and/or edema. Race/ethnicity could play a role in the relationship between sleep and DR, however, this could not be explored without individual patient level data.

In conclusion, this systematic review and meta-analysis demonstrated that poor sleep health, including poor satisfaction and long sleep duration, OSA, and low melatonin levels were associated with DR. Some evidence suggested that treating OSA may benefit visual health in individuals with diabetes, but results were mixed. Future studies should address whether sleep interventions to improve sleep health could benefit health outcomes in people with DR.