Association of poor sleep and HbA1c in metformin-treated patients with type 2 diabetes: Findings from the UK Biobank cohort study

The American Diabetes Association recommends a glycated haemoglobin target of less than 7% for treating type 2 diabetes mellitus. However, it is still being determined if poor sleep affects this therapeutic goal, despite being treated with the blood-glucose-lowering medication metformin. Thus, we used data from 5703 patients on metformin monotherapy participating in the UK Biobank baseline investigation between 2006 and 2010. We combined self-reported chronotype, daily sleep duration, insomnia, daytime sleepiness and snoring into a multidimensional poor sleep score ranging from 0 to 5, with higher scores indicating a less healthy sleep pattern. With each point increase on the poor sleep score scale, the odds of patients having an glycated haemoglobin of ≥ 7% increased by 6% (odds ratio [95% confidence interval], 1.06 [1.01, 1.11], p = 0.021). When examining the components of the poor sleep score separately, snoring was specifically associated with a glycated haemoglobin of ≥ 7% (1.12 [1.01, 1.25] versus no snoring, p = 0.038). However, adjusting for health and lifestyle conditions, such as body mass index, weekly physical activity level and hypertension status, eliminated the significant associations between the poor sleep score and snoring with glycated haemoglobin of ≥ 7%. Our findings suggest that poor sleep, specifically snoring, a symptom of obstructive sleep apnea, may interfere with the therapeutic goal of achieving a glycated haemoglobin below 7%. However, other factors known to be promoted by poor sleep, such as high body mass index, low physical activity and hypertension, may also contribute to the link between poor sleep and higher glycated haemoglobin levels.

level below 7% is a recommended goal in the pharmacotherapy of T2DM (American Diabetes Association, 2009).
In this study, we aimed to explore the relationship between poor sleep patterns and HbA1c levels in patients with T2DM on metformin monotherapy.We combined self-reported sleep characteristics, such as chronotype, daily sleep duration, insomnia, daytime sleepiness and snoring, into a multidimensional poor sleep score.We hypothesized that the odds of patients having an HbA1c level of ≥ 7% increase as the number of poor sleep characteristics reported by patients increases.If our study confirms this hypothesis, it could suggest that improving sleep may be a viable strategy for keeping HbA1c levels below 7% in patients undergoing metformin monotherapy for T2DM.

| Design and participants
The UK Biobank is a prospective study that recruited over 500,000 participants aged 40-73 years between 2006 and 2010 from across the UK.To identify patients with T2DM who participated in the UK Biobank baseline investigation, we applied a validated algorithm based on self-reported disease, medication and T2DM diagnosis in medical history (Eastwood et al., 2016).In addition, all participants who used glucose-lowering drugs and had an HbA1c ≥ 6.5% (48 mmol mol À1 ) but did not have gestational diabetes or type 1 diabetes according to the algorithm (Eastwood et al., 2016) were counted as patients with T2DM.Following exclusions specified in Table 1, for example, patients with comorbidities like renal failure that can interfere with the interpretation of the HbA1c (Jacobs et al., 2022;Jung et al., 2018;Radin, 2014)

| Assessment of patients' sleep patterns
Based on patients' answers to an electronic questionnaire, the following five criteria were used to determine the likelihood of suffering from an unhealthy sleep pattern in the present study, as each of them has been linked to HbA1c or T2DM (Li et al., 2021;Liu et al., 2022;Reutrakul et al., 2013;Shan et al., 2015;Wei et al., 2020): • late chronotype (herein defined as present if patients reported "evening" or "more evening than morning" circadian preference); • short sleep duration (herein defined as present if patients reported a daily sleep duration of less than 7 hr) or long sleep duration (defined as ≥ 9 hr per day); • insomnia (herein defined as present if patients reported to suffer "sometimes" or "usually" from insomnia symptoms); • reports of snoring (yes-answer); • and daytime sleepiness (herein defined as present if patients reported to suffer "sometimes", "often" or "all the time" from daytime sleepiness).
We scored each unhealthy sleep behaviour with 1 point.Thus, the poor sleep scale ranged from 0 (referring to healthy sleep) to 5 (unhealthy sleep).

| Ascertainment of HbA1c
The UK Biobank centrally determined blood HbA1c levels with highperformance liquid chromatography using the Bio-Rad VARIANT II TURBO HbA1c analyser.An HbA1c of less than 7% is recommended as a therapeutic goal by the American Diabetes Association (American Diabetes Association, 2009).For the analysis, an HbA1c of ≥ 7% was used as an outcome.

| Potential confounders
The following potential confounders were included in our logistic regression analysis: age; sex; ethnicity; Townsend index reflecting socioeconomic status; body mass index (BMI); region of the assessment centre; hypertension (defined as the use of antihypertensive drugs, systolic blood pressure ≥ 140 mmHg and diastolic blood pressure ≥ 90 mmHg, respectively); T2DM duration (the time difference between age at T2DM diagnosis and age at UK Biobank investigation); smoking (never versus previous/current); alcohol intake frequency (never versus occasional/sometimes versus usually); and physical activity level (low versus moderate versus high), which divided according to the short-form International Physical Activity Questionnaire based on the total metabolic equivalent minutes per week (Craig et al., 2003).

| Statistical analysis
Mean and standard deviation (SD) are presented unless stated otherwise.Statistical analyses were performed using IBM SPSS Statistics version 28.0 (IBM, Armonk, NY, USA).Comparisons of group characteristics were performed by Chi-square testing for categorical variables and generalized linear models for continuous variables.Furthermore, we conducted logistic regression analyses to investigate the association between unhealthy sleep patterns and the risk of having an HbA1c level above the target range (≥ 7%).We also performed separate logistic regression analyses for each of the five sleep components used to calculate the poor sleep score to determine their associations with the odds of patients having an HbA1c level above the target range.The results were reported as odds ratios (ORs) and 95% confidence intervals (CIs).
Three statistical models were used in our analyses.The crude model used either the sleep score or any of the five sleep characteristics as an exposure.In model A, we added patients' age, sex, ethnicity, Townsend index, smoking status, alcohol intake frequency, assessment centre, and diabetes duration to the regression as potential confounders.Poor sleep patterns have been linked to higher BMI, greater risk of hypertension, and lower levels of physical activity (Bromley et al., 2012;Kecklund & Axelsson, 2016;Lv et al., 2022;Schmid et al., 2015).Therefore, in model B, we additionally adjusted for BMI, hypertension status and weekly physical activity level.We considered p-values less than 0.05 to be statistically significant.

| Characteristics of the study population
The study included 5703 patients with T2DM who were treated with metformin.The mean age of the patients was 60 years (SD 6.8, range 40-70 years).Of the patients, 41.5% had an HbA1c level of ≥ 7%.Furthermore, 5.3% of the patients had a poor sleep score of 0 (indicating the healthiest sleep pattern), while 1.0% had a poor sleep score of 5 (indicating the least healthy sleep pattern).
Furthermore, Chi-square testing revealed that the patients with higher poor sleep scores were less physically active (Pearson Chi 2 = 73.05,p < 0.001).Table 2 presents more characteristics based on the poor sleep score.

| Association of sleep with the odds of having an HbA1c above the range
With each point increase on the poor sleep score scale, the odds of patients having an HbA1c above the target range (i.e.≥ 7%) significantly increased by 7% in the crude model (p = 0.006) and 6% in model A (p = 0.021; Table 3).However, when additionally adjusting for patients' BMI, hypertension status and weekly physical activity level, the poor sleep score was no longer associated with patients having an HbA1c of ≥ 7% (p = 0.186; Table 3).
When entering each sleep component separately into the logistic regression models as an exposure of interest, snoring was associated with 15% higher odds of patients having an HbA1c above the recommended target range in the unadjusted analysis (p = 0.012) and 12% in model A ( p = 0.038; Table 3), respectively.However, the association between snoring and HbA1c of ≥ 7% failed to reach significance in model B, that is, when additionally adjusting for patients' BMI, hypertension status and weekly physical activity level (p = 0.186; Table 3).Finally, in the crude model, daytime sleepiness was associated with higher odds of patients having an HbA1c of ≥ 7% (p = 0.042); however, this association was no longer present in model A (p = 0.051) or B ( p = 0.155; Table 3).

| DISCUSSION
Patients with T2DM often experience sleep disturbances (Tan et al., 2018).As suggested by previous studies (Brouwer et al., 2020;Liu et al., 2022;Mokhlesi et al., 2021;Reutrakul et al., 2013), sleep patterns hallmarked by poor sleep characteristics like short sleep, insomnia and obstructive sleep apnea (OSA) are associated with higher HbA1c, a widely used marker for assessing long-term glucose control in patients with T2DM.For example, a Mendelian Randomization Study using data from the UK Biobank found that frequent insomnia symptoms appear to cause higher HbA1c levels in the general population (Liu et al., 2022).In a separate study of 172 patients with T2DM using 7-day wrist-actigraphy and sleep questionnaires,  Note: Model A: controlling for age, sex, ethnicity, Townsend index, smoking status, alcohol intake frequency, assessment centre and diabetes duration.Model B: controlling for Model A variables, and BMI, hypertension and physical activity level.Abbreviations: CI, confidence interval; HbA1c, glycated haemoglobin; OR, odds ratio; ref, reference group in the logistic regression.
While these studies indicate that features of poor sleep patterns are associated with higher HbA1c levels, it is still being determined if poor sleep patterns interfere with common goals in the pharmacotherapy of T2DM, like having an HbA1c of lower than 7% (American Diabetes Association, 2009).
In this study, we therefore aimed to assess the impact of poor sleep on glycaemic control among metformin-treated patients with T2DM, using a poor sleep score.Our findings suggest that an increased number of reported unhealthy sleep characteristics were associated with higher odds of having an HbA1c of ≥ 7%.Specifically, snoring emerged as the most significant risk factor for poor glycaemic control, although this association was no longer significant after accounting for known health and lifestyle factors related to sleep, such as BMI, physical activity level and hypertension status (Bromley et al., 2012;Kecklund & Axelsson, 2016;Lv et al., 2022;Schmid et al., 2015).These results suggest that poor sleep may interfere with glycaemic control among patients with T2DM undergoing metformin therapy, by promoting a sedentary lifestyle, contributing to high BMI and increasing the risk of hypertension.Notably, we found that patients with worse sleep patterns had higher BMI levels, with a difference of up to 4.2 units compared with those with healthier sleep patterns.Additionally, patients with poor sleep patterns were less likely to engage in high levels of physical activity and more likely to have a low weekly physical activity level.
In a previous clinical trial that involved 221 overweight or obese adults with prediabetes or treatment-naive T2DM, a higher apnea-hypopnea index (AHI), a marker of OSA severity, was associated with higher HbA1c.Specifically, it was observed that for every 40-unit increase in AHI, there was a corresponding 0.12% increase in HbA1c (Mokhlesi et al., 2021).In our study, we found that both snoring and daytime sleepiness (the latter only in the crude analysis) were associated with an increased risk of patients having an HbA1c level of ≥ 7%.These are common symptoms of OSA (Chung et al., 2008), suggesting that the observed association between poor sleep patterns and elevated HbA1c levels in our study may primarily be due to OSA.This hypothesis is further supported by our finding that the association between poor sleep score and snoring with HbA1c ≥ 7% was no longer significant after controlling for additional risk factors for OSA, such as high BMI and presence of hypertension (Chung et al., 2008).With these findings in mind, treatment of OSA with continuous positive airway pressure (CPAP) therapy, for example, may hold promise in achieving an HbA1c level of less than 7%.In line with this assumption, a proof-of-concept study has demonstrated that after just 1 week of whole-night CPAP treatment, patients with OSA experienced a decrease of approximately 0.8 mmol L À1 in their 24-hr mean plasma glucose levels (Mokhlesi et al., 2016).However, meta-analytic evidence paints a rather heterogeneous picture of the effects of CPAP on glycaemic control in patients with T2DM and comorbid OSA.For example, a meta-analysis using data from 228 adults with T2DM and coexisting OSA found that CPAP treatment did not significantly impact HbA1c compared with usual care or sham CPAP (Zhu et al., 2018).In contrast, in a recent meta-analysis including 587 patients with T2DM and comorbid OSA, CPAP was associated with a 0.32% lower HbA1c level than sham CPAP or no CPAP (Shang et al., 2021).Factors such as patients' CPAP compliance, CPAP adherence (i.e.whether nightly CPAP treatment is limited to the first 4 hr of sleep versus the whole sleep period) and the duration of CPAP therapy may explain the variable effects of CPAP on glycaemic control in patients with T2DM and comorbid OSA.
Several limitations to our study should be acknowledged.First, our analysis was limited to patients receiving metformin monotherapy, the most commonly prescribed first-line medication for T2DM (Vigneri & Goldfine, 1987).As the pharmacotherapy of T2DM continues to advance (DeMarsilis et al., 2022), it is unclear whether the observed association between poor sleep and HbA1c extends to patient groups receiving other antidiabetic therapies.Additionally, although we excluded patients with comorbidities that could interfere with HbA1c interpretation, such as leukaemia, anaemia and renal failure (Jacobs et al., 2022;Jung et al., 2018;Radin, 2014), residual confounding cannot be entirely ruled out despite extensive adjustments.Moreover, the sleep data used in this study were based on self-reporting and may be susceptible to reporting bias.Furthermore, no information regarding the dosing of metformin was collected during the UK Biobank baseline investigation.Finally, our study design was cross-sectional, which makes it challenging to draw causal inferences.Future longitudinal studies incorporating objective measures of sleep (e.g.sleep wearables) may help gain a more comprehensive understanding of the complex interplay between sleep patterns and long-term glycaemic control in patients with T2DM.

| CONCLUSIONS
In conclusion, this study provides evidence that poor sleep, mainly snoring, may be associated with difficulty achieving the recommended HbA1c target of less than 7% in patients with T2DM on metformin monotherapy and, as such, adds further evidence that screening for sleep health may be a helpful add-on in the standard care of T2DM, as recently proposed (ElSayed et al., 2023).However, it is essential to note that other factors, such as high BMI, low physical activity and hypertension, likely contribute to this link.Therefore, addressing these underlying health conditions and improving sleep may help optimize glycaemic control in patients undergoing metformin therapy for T2DM.
, 5703 metformin-treated patients with T2DM remained for analysis.Records on antihyperglycaemic medication derived from the UK Biobank interview.The UK Biobank study was approved by the National Health Service National Research Ethics Service (ref.11/NW/0382).All participants provided written informed consent to participate in the UK Biobank study.Information about ethics oversight in the UK Biobank can be found at https://www.ukbiobank.ac.uk/ethics/.
short and long sleep duration and subjective sleep quality were T A B L E 2 Baseline characteristics of metformin-treated patients with T2DM, stratified by multidimensional sleep status Tan and Christian Benedict designed the study.Pei Xue performed the statistical analysis, interpreted the data, and wrote the initial draft with supervision from Christian Benedict.Pei Xue and Christian Benedict obtained financial support.All authors reviewed and approved the final version of the article submitted for publication.