Association of wearable device-measured physical activity with full-course diabetic retinopathy and retinal traits: insights from the middle-aged and older adult cohort

Abstract

Purpose

To assess accelerometer-measured physical activity (PA) in patients with all stages of diabetic retinopathy (DR) and investigate its association with specific retinal structural metrics.

Methods

This extensive cohort study included 13,600 participants with an average age of 56.39 years. These subjects were divided into four groups: non-diabetes mellitus, prediabetes mellitus (Pre-DM), diabetes mellitus (DM) without DR, and DR. We evaluated multivariate-adjusted associations of PA with DR progression using logistic regression and with retinal sublayer thickness using hierarchical linear model (HLM). The mediating role of body mass index (BMI) was tested to investigate the true association between PA and the full spectrum DR.

Results

As DR progressed, the durations of moderate-intensity PA (MPA) and moderate-vigorous PA (MVPA) decreased significantly by 29% (odds ratio (OR) = 0.71, 95% CI = 0.57–0.90) to 78% (OR = 0.22, 95% CI = 0.14–0.35) and 21% (OR = 0.79, 95% CI = 0.71–0.89) to 55% (OR = 0.45, 95% CI = 0.30–0.67), respectively. Morning MPA and MVPA (6:00–12:00) were protective factors against DR, whereas late-night PA (0:00–5:59) heightened DR risk. The multivariate-adjusted linear interaction model revealed that the positive effect of MPA and MVPA on the thickness of ganglion cell-inner plexiform layer (GCIPL), macular thickness (MT), and inner nuclear layer-external limiting membrane was significantly associated with DR disease status (interaction P < 0.05). Higher MPA and MVPA were correlated with accelerated thickening rates of the GCIPL and MT sublayers, ranging from Pre-DM to those with established DR. 35.7% and 58.7% of the associations between MPA, MVPA, and the full spectrum DR were mediated by lower BMI, respectively.

Conclusions

The diminution of PA is associated with the progression of DR and the attenuation of retinal sublayer thickness, and our findings support current PA recommendations promoting interventions to decelerate DR progression and preserve retinal health.

Supplementary Information

The online version contains supplementary material available at 10.1186/s13098-025-01745-1.

Background

Diabetic retinopathy (DR) is a common microvascular complication of diabetes mellitus (DM) and a leading cause of preventable blindness among the working-age population worldwide [1]. It is characterized by chronic progressive damage to the retinal microvasculature and associated structural changes, ultimately leading to vision impairment or loss if not treated promptly [2]. As the duration of the disease progresses, poor glycemic control, hypertension, and oxidative stress are considered risk factors for the progression of DR. Effective management of hyperglycemia and blood pressure can help reduce this risk [3], [4]. Physical activity (PA) effectively regulates glycemic levels by improving insulin sensitivity, lowering hemoglobin A1c (HbA1c), and reducing oxidative stress in the retina [5]. PA is an modifiable risk factor in the management of diabetes and the prevention of DR. Nevertheless, the specific impact of PA on the prevention and progression of DR remains inadequately studied and its effects remain uncertain.

It should be noted that most studies on PA and diabetes rely on self-reported questionnaires, which are prone to recall bias [6], social desirability bias, and misclassification bias. Accelerometry is now the gold standard for objectively measuring PA [7] and is increasingly important in evaluating the relationship between PA and ophthalmic diseases. Recent methods analyze triaxial acceleration data at subsecond resolutions to more accurately predict energy expenditure, summarized as daily minutes of light, moderate, or vigorous PA [8], [9]. The intensity of specific activities can be expressed in metabolic equivalents—the ratio of the working metabolic rate to the standard resting metabolic rate— which estimate energy expenditure [10]. To our knowledge, the relationship between PA and the progression from preclinical diabetes to the vision-threatening stage of DR has not been documented.

The retina, a key target of diabetic complications,is highly vulnerable to diabetes-induced metabolic stress [11]. Even before the clinical manifestations of diabetes appear, subtle changes will occur in the retina [12], [13]. Further research indicates that the retinal thickness of diabetic patients decreases, regardless of whether they exhibit clinical symptoms of DR [14], [15], [16]. The research conducted by Van Dijk et al. demonstrated that the retinal thickness within the macula decreases in patients with diabetes or mild DR [17], [18]. It is critical to examine how PA influences retinal structure across all stages of DR to elucidate its mechanistic role.

Therefore, this study aims to explore the association between accelerometer-measured PA and the DR progression, including retinal structural changes in middle-aged and elderly population. Utilizing data from the UK Biobank (UKB), we assessed PA types, encompassing total, light, moderate, vigorous, and moderate-to-vigorous PA, across individuals with different stages of DR, spanning from non-diabetes mellitus (Non-DM) to severe DR. Additionally, since weight loss plays a critical role in diabetes prevention, we assessed the contribution of weight control to the PA-DR association.

Methods

UK biobank study population

The UKB is a nationwide, community-based, prospective multicenter cohort study involving over 500,000 UK residents aged between 40 and 69 years from 2006 to 2010. All participants underwent baseline assessments at one of the 22 UK National Health Assessment Centers in Scotland, England, and Wales. Data were collected using questionnaires, physical measures, and biological sample collection at baseline. Ophthalmic assessments, including optical coherence tomography (OCT) imaging, were introduced to the baseline assessment in 2009 for six assessment centres. Long-term longitudinal follow-up was also conducted to assess various health-related outcomes. The UKB study was approved by the UK National Health Service National Research Ethics Service (Ref 11/NW/0382), and informed consent was obtained from all participants. Details of the rationale, design, and assessments used in the UKB Study have been described elsewhere. The UKB provides all data analyzed herein under project reference and data transfer agreement #86091.

The current study initially recruited 57,473 participants with full-course DR data and accelerometer-measured PA records. After quality control, we excluded 40,080 participants (69.7%): 384 with visual impairment (VI), 39,696 with missing VA data and 3,793 with low-quality accelerometer data. The final analytical cohort comprised 13,600 participants, stratified as follows: 10,751 normal subjects, 1,480 prediabetes mellitus (Pre-DM) subjects, 1,212 diabetes mellitus-no diabetic retinopathy (DM-NDR) patients, and 157 with DR. In exploring the associations between PA and retinal sublayers, 10,833 participants with OCT data exhibiting low signal-to-noise ratios or poor segmentation were excluded, resulting in 2,767 participants with high-quality OCT data. These participants were further subdivided into the following groups: normal subjects (2,223 participants), Pre-DM (282 participants), DM-NDR (241 participants), and DR (21 participants). Participants with preexisting ocular or neurological conditions that could confound the results of OCT were excluded based on validated International Classification of Diseases, Ninth Revision/Tenth Revision (ICD-9/10) codes. These conditions included glaucoma, high myopia, trauma-induced low vision, and neurodegenerative disorders such as Alzheimer’s disease, Parkinson’s disease, or dementia. The detailed participant flow diagram, including eligibility criteria and subgroup classifications, is presented in Supplementary Figure 1.

Accelerometer physical activity

The 7-day PA data were collected using raw accelerometer recordings (Axivity AX3, Open Lab, Newcastle University) with a sampling frequency of 100 Hz and a dynamic range of ± 8 g. Participants were excluded if they: (1) wore the accelerometer for less than 72 h; (2) lacked data points for any 1 h within the 24-hour cycle during the study period; or (3) exhibited implausibly high activity values (average vector magnitude > 100 mg). Data were also excluded when sensor clipping occurred (exceeding the ±8 g dynamic range). To account for potential variations in local gravitational acceleration, PA data were adjusted for local gravity per participant. Participants with insufficient stationary data for calibration purposes were excluded from the analysis. A full description of data collection and processing, as well as additional information on accelerometer data processing, analysis, and behaviour classification, was provided elsewhere. The average vector magnitude in milligravities (mg) was used to estimate the total volume of PA. The minutes per week (min/week) of light, moderate, and vigorous PA were determined from the time spent in acceleration ranges of 30–125 mg, 125–400 mg, and > 400 mg, respectively. Additional details regarding the collection of accelerometer data and quality control are provided in Supplementary Table 1.

The classification of Moderate PA (MPA) (< 150 min/week, ≥ 150 min/week) and vigorous PA (VPA) (< 75 min/week, ≥ 75 min/week) is derived from the current World Health Organization (WHO) recommendations for PA [19]. Total Physical Activity (TPA) is calculated as the sum of light, moderate, and vigorous PA, expressed in weekly minutes. The Metabolic Equivalent of Task (MET) is an objective measure of energy expenditure, considering individual body weight. Moderate-vigorous PA (MVPA) is determined by the sum of MPA (with an average of 4 METs) and VPA (with an average of 8 METs) and is represented in weekly MET minutes.

Ascertainment of visual impairment

VA was measured using the logarithm of the minimum angle of resolution (LogMAR) VA chart (Precision Vision, La Salle, IL, USA). Testing distance was 4 m (with habitual correction). If participants failed at 4 m, distance was reduced to 1 m. Participants were asked to read the letters from the top downward, and the total number of correctly identified letters was converted into a LogMAR value. VI was defined as presenting VA worse than 0.3 LogMAR (Snellen equivalent 20/40). Detailed methods for VA assessment in the UKB study can be obtained from prior publications [20].

Ascertainment of full-course DR status

Non-DM was determined by self-report undiagnosed diabetes and HbA1c concentration < 39 mmol/mol (<5.7%) at baseline. Pre-DM was ascertained from HbA1c concentrations ranging from 39 to 48 mmol/mol (5.7%-6.5%) without physician-diagnosed or hypoglycemic medicationsuse. DM was defined by meeting any of the following criteria: (1) self-reported physician diagnosis of DM; (2) documented DM diagnosis per ICD-10 classification; (3) current use of glucose-lowering medications (oral hypoglycemic agents or insulin); or (4) HbA1c level ≥48 mmol/mol (≥6.5%). Participants with DM were subsequently stratified into two subgroups: DM-NDR and DR, based on ophthalmologist-confirmed DR diagnosis.

Spectral-domain OCT imaging

Spectral-domain OCT imaging was performed using the Topcon 3D OCT1000 Mark II under mesopic conditions without pupillary dilation. Images were acquired in 3-dimensional 6 × 6 mm² [2] macular volume scan mode (512 A-scans per B-scan, 128 horizontal B-scans in a raster pattern). Both eyes were imaged (right eye first). Eyes with a low signal-to-noise ratio or failed segmentation were excluded from analysis [21]. Details acquisition protocols and quality control criteria are described in Supplementary Figure 1. We used the average thickness parameters obtained from the macular 6-grid. Participant-level thickness parameters were derived by averaging measurements from both eyes, using only high-quality images [22].

Covariates assessment

This study comprehensively adjusted for potential confounding factors across multiple domains. Demographic covariates included age (calculated from birth date to baseline assessment date), sex, self-reported ethnicity (categorized as White or non-White), Townsend index (an area-based measure derived from postal codes and Townsend deprivation scores), educational attainment (classified as university or non-university level). Lifestyle factors incorporated smoking status (non-smoker or current/former smoker), alcohol consumption (non-drinker or current/former drinker), body mass index (BMI, calculated as weight [kg] /height [m]²), Diet quality was assessed using a healthy diet score (range: 0-5), with 1 point awarded for meeting each of the following criteria: ≥4 tablespoons of vegetables intake per day, ≥ 3 servings of fruits daily, ≥ 2 servings of fish weekly, ≤ 2 weekly unprocessed red meat servings, and ≤ 2 weekly processed meat servings. Season of accelerometer wear was determined by accelerometer start date: spring (March-May), summer (June-August), autumn (September-November), winter (December-February)), family history of major depressive disorder (self-reported), and comorbidities such as hyperlipidemia (defined by self-report, antihyperlipidemic medication use, or serum cholesterol ≥ 6.21 mmol/L), hypertension (self-reported antihypertensive medication use, mean systolic blood pressure > 130 mmHg, or mean diastolic blood pressure  > 80 mmHg), and cancer (self-reported). Additional details about these measurements can be found in the UKB online protocol. All models are mutually adjusted for MPA and VPA to account for intensity interactions. Details variable definitions are listed in Supplementary Table 1.

Statistical analysis

Demographic characteristics stratified by TPA were determined using analysis of variance (ANOVA) for continuous variables and chi-square tests for categorical variables. Subsequent model adjusted for covariates significantly associated with PA.

Logistic regression models utilized healthy individuals as the reference group to investigate independent associations between DR progression stages and specific PA intensity levels. To investigate whether diurnal variation in activity patterns affect the progression of DR, we divided daily activity into four segments: morning (6:00–11:59), afternoon (12:00–17:59), evening (18:00–23:59), and late night (0:00–5:59). To address potential selection bias, we performed propensity score matching (PSM) at a 1:10 ratio using nearest neighbor matching with a caliper width of 0.05. The DR group served as cases, while controls, Pre-DM, and DM groups formed the reference population. Matching covariates included age, sex, ethnicity, Townsend deprivation index, education level, smoking status, alcohol consumption, healthy diet score, family history of depression, hypertension, hyperlipidemia, accelerometer wear season, and cancer history. The PA-DR association was subsequently reevaluated in the PSM cohort.

Hierarchical line models (HLM) analyzed interactions between MPA, MVPA and retinal sublayers, adjusting for all confounders. Retinal multilayer measurements include the ganglion cell inner plexiform layer (GCIPL), macular thickness (MT), retinal nerve fibre layer (RNFL), inner nuclear layer-external limiting membrane (INL-ELM), inner segment-outer segment to retinal pigment epithelium (ISOS-RPE), external limiting membrane to inner segment-outer segment (ELM-ISOS), and retinal epithelium (RPE).

This study employed a nonparametric bootstrap approach with 1,000 iterations to assess BMI’s mediating role in the MVPA-DR association [23]. We specified two regression models: (1) an outcome model examining joint effects of PA and BMI on DR progression and (2) a mediator model evaluating PA’s influence on BMI. Continuous variables were standardized as z-scores. Participant-level natural indirect effects (NIE) and total effects (TE) were calculated and then aggregated, with bias-corrected bootstrap percentile confidence intervals with mediation significance determined by 95% confidence intervals (CIs) excluding zero and mediation proportion (NIE/TE). This analytical framework accounts for the sampling distribution’s asymmetry while preserving an appropriate causal interpretation of the mediated pathway.

The statistical software packages used for the analysis were STATA 17 (StataCorp LP) and R version 4.3.2. All P values were two-tailed, and a significance level of P < 0.05 was used.

Results

The demographic characteristics of participants stratified by TPA tertiles are shown in Table 1. Overall, compared to individuals in the lowest tertile of TPA, those in the highest tertile were younger, had similar educational levels, lower BMI, higher healthy diet scores, were less likely to have a family history of depression, and had a lower prevalence of hyperlipidemia, hypertension, and cancer ( all P < 0.05).

Table 1.

Demographic characteristics of participants stratified by TPA tertiles

Baseline Characteristics	Lowest	Middle	Highest
Number of subjects	3,470	4,793	5,337
The Full-course DR, No. (%)
Non-DM	2,454 (70.7)	3,838 (80.1)	4,459 (83.6)
Pre-DM	389 (11.3)	524 (10.9)	567 (10.6)
DM-NDR	553 (15.9)	381 (8.0)	278 (5.2)
DR	74 (2.1)	50 (1.0)	33 (0.6)
Age, mean (SD), yrs	57.67 (7.69)	56.51 (7.72)	55.44 (7.81)
Gender, No. (%)
Female	3,067 (88.4)	4,530 (94.5)	5,181 (97.1)
Male	403 (11.6)	263 (5.5)	156 (2.9)
Ethnicity, No. (%)
White	3,243 (93.5)	4,515 (94.2)	4,972 (93.2)
Non-white	227 (6.5)	278 (5.8)	365 (6.8)
Townsend index, mean (SD)	-1.00 (2.86)	-1.32 (2.76)	-1.32 (2.73)
Education level, No. (%)
College or university degree	1,434 (41.3)	2,085 (43.5)	2,310 (43.3)
Others	2,036 (58.7)	2,708 (56.5)	3,027 (56.7)
Smoking status, No. (%)
Never	1,980 (57.2)	2,873 (60.1)	3,209 (60.3)
Former/current	1,482 (42.8)	1,907 (39.9)	2,113 (39.7)
Alcohol intake, No. (%)
Never	147 (4.2)	168 (3.5)	238 (4.5)
Former/current	3,322 (95.8)	4,622 (96.5)	5,097 (95.5)
Body mass index, mean (SD)	28.43 ± 5.77	26.61 ± 4.76	25.37 ± 4.32
Health diet score, mean (SD)	2.70 ± 1.03	2.81 ± 1.03	2.88 ± 1.04
Family history of severe depression, No. (%)
No	2,961 (85.3)	4,059 (84.7)	4,521 (84.7)
Yes	509 (14.7)	734 (15.3)	816 (15.3)
History of hypertension, No. (%)
No	890 (25.7)	1,518 (31.7)	2,028 (38.0)
Yes	2,580 (74.3)	3,275 (68.3)	3,309 (62.0)
History of hyperlipidemia, No. (%)
No	1,612 (46.5)	2,522 (52.6)	3,118 (58.4)
Yes	1,858 (53.5)	2,271 (47.4)	2,219 (41.6)
History of cancer, No. (%)
No	3,137 (90.7)	4,366 (91.3)	4,895 (91.9)
Yes	322 (9.3)	417 (8.7)	430 (8.1)

Continuous variables are described as mean ± standard deviation, and categorical variables are described as numbers and percentages. Abbreviations: PA, physical activity; Non-DM, no diabetes mellitus; Pre-DM, prediabetes mellitus; DM-NDR, diabetes mellitus without diabetic retinopathy; DR, diabetic retinopathy; SD, standard deviation

Associations of PA with full-course DR

Accelerometer-measured PA significantly declined with DR progression (Supplementary Table 3). Among individuals with Pre-DM, fewer met WHO-recommended PA thresholds: 29% fewer achieved MPA guidelines (odds ratio (OR) = 0.71, 95% CI = 0.57–0.90, P = 0.004) and 21% fewer met MVPA targets (OR = 0.79, 95% CI = 0.71–0.89, P < 0.001). Only 22–45% of DR patients maintained compliance with WHO recommendations. A significant inverse correlation was observed between MVPA and the progression of DR in the PSM cohort. Notably, in the Pre-DM group, only 58% achieved MVPA guidelines, a significantly lower proportion than controls (OR = 0.58, 95% CI = 0.37–0.92; P = 0.019) (Supplementary Table 7).

Associations of daily activity rhythms with full-course DR

Subsequent time-stratified analyses (Table 2) demonstrated that morning PA provided the most potent protective effect against DR progression, whereas nighttime activity was associated with an elevated risk. Compared to healthy controls, each incremental unit of morning MPA was associated with dose-dependent reductions in the likelihoods of developing Pre-DM (OR = 0.81, 95% CI = 0.71–0.91, P = 0.001), DM-NDR (OR = 0.56, 95% CI = 0.45–0.69, P < 0.001), and DR (OR = 0.38, 95% CI = 0.24–0.58, P < 0.001). A similar pattern was observed for morning MVPA (06:00–11:59), with each additional unit corresponding to 22% (OR = 0.78, 95% CI = 0.70–0.88, P < 0.001), 56% (OR = 0.44, 95% CI = 0.36–0.53, P < 0.001), and 68% (OR = 0.32, 95% CI = 0.21–0.48, P < 0.001) risk reductions for Pre-DM, DM-NDR, and DR, respectively.

Table 2.

Association of daily activity rhythms of MPA and MVPA within 24 hours and the full-course DR

	Non-DM	Pre-DM		DM-NDR		DR
		OR (95%CI)	P-value	OR (95%CI)	P-value	OR (95%CI)	P-value
MPA
0:00–5:59	1 [reference]	1.07 (0.95, 1.20)	0.263	1.42 (1.17, 1.72)	< 0.001*	1.59 (1.09, 2.31)	0.016*
6:00–11:59	1 [reference]	0.81 (0.71, 0.91)	0.001*	0.56 (0.45, 0.69)	< 0.001*	0.38 (0.24, 0.58)	< 0.001*
12:00–17:59	1 [reference]	0.84 (0.74, 0.95)	0.005*	0.65 (0.52, 0.81)	< 0.001*	0.65 (0.43, 0.98)	0.040*
19:00–23:59	1 [reference]	0.87 (0.77, 0.99)	0.030*	0.68 (0.55, 0.84)	< 0.001*	0.70 (0.46, 1.06)	0.093
MVPA
0:00–5:59	1 [reference]	1.07 (0.95, 1.20)	0.252	1.33 (1.10, 1.62)	0.004*	1.33 (0.92, 1.93)	0.133
6:00–11:59	1 [reference]	0.78 (0.70, 0.88)	< 0.001*	0.44 (0.36, 0.53)	< 0.001*	0.32 (0.21, 0.48)	< 0.001*
12:00–17:59	1 [reference]	0.84 (0.74, 0.94)	0.003*	0.50 (0.41, 0.61)	< 0.001*	0.49 (0.33, 0.73)	< 0.001*
19:00–23:59	1 [reference]	0.88 (0.78, 1.00)	0.046*	0.53 (0.43, 0.64)	< 0.001*	0.51 (0.34, 0.77)	0.001*

MPA and VPA were mutually adjusted. Multivariable models were adjusted for age at baseline, baseline age, gender, ethnicity, smoking status, alcohol intake, education level, Townsend index, family history of severe depression, seasons of accelerometer wear, health diet score and comorbidities (hypertension, hyperlipidemia and cancers). Abbreviations: DR, diabetic retinopathy; MPA, moderate physical activity; MVPA, moderate-vigorous physical activity; Non-DM, no diabetes mellitus; Pre-DM, prediabetes mellitus; DM-NDR, diabetes mellitus without diabetic retinopathy; DR, diabetic retinopathy; OR, odd ratio; CI, confidence interval. *Values with p < 0.05 are considered statistically significant

Analysis of afternoon activity (12:00–17:59) revealed that increased MVPA was associated with a 16-51% reduction in DR progression risk across different subgroups (Table 2). Notably, in the balanced matched cohort, MVPA performed between 6:00 AM and 12:00 PM continued to mitigate the risk of DR development at all disease stages (Supplementary Table 8).

Associations of PA and retinal sublayer thickness with full-course DR

We included 2,767 participants with qualified OCT data to investigate the association between device-measured PA and specific retinal thickness parameters. There was a significant interaction between MPA, MVPA, and overall DR, with MVPA affecting the thicknesses of retinal sublayers GCIPL (interaction P value = 0.005), MT (interaction P value = 0.002), and INL-ELM (interaction P value = 0.014) at different stages of DR. Specifically, in the normal population, the average thickness of various retinal layers increased at a steady rate with the increment of MPA and MVPA. For each unit increase in MPA, the thickness of MT and INL-ELMincreased by (276 + 0.00412) µm and (80.1 + 0.000473) µm, respectively. Similarly, for a one-unit increase in MVPA, the thickness of MT and INL-ELM increased by (276 + 0.000927) µm and (80.2 + 0.000515) µm, respectively (Figure 1).

Fig. 1.

Hierarchical subgroup analysis for the association between MPA and MVPA with different retinal layer thickness in the full-course DR. a. association between GCIPL thickness and MPA the full-course DR; b. association between MT and MPA the full-course DR; c. association between INL-ELM thickness and MPA the full-course DR; d. association between GCIPL thickness and MVPA in the full-course DR; e. association between MT and MVPA the full-course DR; f. association between INL-ELM thickness and MVPA in the full-course DR. Abbreviations: MPA, moderate physical activity; MVPA, moderate-vigorous physical activity; GCIPL, ganglion cell-internal plexiform layer; MT, macular thickness; INL-ELM, inner nuclear layer-external limiting membrane; Non-DM, no diabetes mellitus; Pre-DM, prediabetes mellitus; DM-NDR, diabetes mellitus without diabetic retinopathy; DR, diabetic retinopathy. *Values with p < 0.05 are considered statistically significant

As the disease progresses to the Pre-DM, the increase in MPA leads to a faster rate of thickening in the GCIPL, MT, and INL-ELM sublayers, with respective rates of 0.005 (β = 0.005, 95%CI = 0.002–0.007, P-value = 0.002), 0.014 (β = 0.014, 95%CI = 0.004–0.017, P-value = 0.001), and 0.004 (β = 0.004, 95% CI = 0.000-0.007, P-value = 0.028). (Table 3).

Table 3.

The association between PA and different retinal thickness stratified by DR status

	Non-DM		Pre-DM		DM-NDR		DR
	β (95%CI)	P-value	β (95%CI)	P-value	β (95%CI)	P-value	β (95%CI)	P-value	P-interaction
MPA
GCIPL	0.001 (-0.000, 0.002)	0.190	0.005 (0.002, 0.007)	0.002*	0.005 (0.000, 0.008)	0.034*	0.005 (-0.055, 0.088)	0.517	0.005*
MT	0.004 (0.001, 0.005)	0.010*	0.014 (0.004, 0.017)	0.001*	0.009 (0.000, 0.017)	0.047*	0.026 (-0.145, 0.212)	0.592	0.003*
INL-ELM	0.000 (-0.000, 0.002)	0.162	0.004 (0.000, 0.007)	0.028*	0.002 (-0.002, 0.006)	0.345	0.012 (-0.039, 0.037)	0.948	0.018*
MVPA
GCIPL	0.000 (-0.000, 0.000)	0.269	0.001 (0.000, 0.001)	0.005*	0.001 (-0.000, 0.002)	0.035*	0.001 (-0.013, 0.021)	0.542	0.005*
MT	0.001 (0.000, 0.001)	0.010*	0.003 (0.001, 0.004)	0.002*	0.002 (0.000, 0.004)	0.044*	0.006 (-0.035, 0.050)	0.620	0.002*
INL-ELM	0.000 (0.000, 0.000)	0.185	0.001 (0.000,0.001)	0.025*	0.001 (-0.000, 0.001)	0.343	0.003 (-0.009, 0.009)	0.921	0.014*

The heretical line model (HLM) was adjusted for age at baseline, gender, ethnicity, smoking status, alcohol intake, education level, Townsend index, family history of severe depression, seasons of accelerometer wear, health diet score and comorbidities (hypertension, hyperlipidemia and cancer). Abbreviations: MPA, moderate physical activity; MVPA, moderate-vigorous physical activity; Non-DM, no diabetes mellitus; Pre-DM, prediabetes mellitus; DM-NDR, diabetes mellitus without diabetic retinopathy; DR, diabetic retinopathy; GCIPL, ganglion cell-internal plexiform layer; MT, macular thickness; INL-ELM, inner nuclear layer-external limiting membrane;

*Values with p < 0.05 are considered statistically significant.

In the diabetic stage, MPA exhibited a significant and positive correlation with the GCIPL and MT sublayers, with incremental rates of 0.005 and 0.009, respectively. MVPA exhibited a similar positive trend, but with an increment rate ranging from 0.001 to 0.002.

Additionally, it is noteworthy that the impact of MPA and MVPA on the thickness of the GCIPL and MT sublayers is more pronounced in the DR stage. When the disease progresses into DR, for each unit increase in MPA, the incremental rates for GCIPL, MT are 0.005 and 0.026, respectively. Similarly, for every unit increase in equivalent MVPA, the thickness of the GCIPL and MT sublayers increases by (72.2 + 0.000132) µm and (268 + 0.00603) µm, respectively. (Figure. 1) After excluding 93 individuals with conditions independently affecting OCT outcomes, such as glaucoma, high myopia, low vision caused by severe trauma, or neurological disorders including Alzheimer’s disease, Parkinson’s disease, or dementia, sensitivity analyses demonstrated that the primary outcomes remained robust (Supplementary Table 1).

The impact of BMI on PA and full-course DR

Mediation analysis revealed significant inverse associations between PA and the occurrence of DR with MPA (β = -0.033, p = 0.004), MVPA(β = -0.07, p < 0.001), and TPA (β = -0.017, p = 0.004) all showing protective effects. Specifically, BMI mediated 35.7% of the MPA-DR association and 58.7% of the MVPA-DR association across the full-course DR, respectively. (Supplementary Table 5).

Discussion

The present study investigated accelerometer-based PA, full-course DR, and retinal structural parameters in a large cohort. Objectively measured PA, whether moderate or moderate-vigorous intensity, was associated with delayed onset and advancement of DR. Our findings highlight the importance of meeting current aerobic PA guidelines (150–300 min of MVPA per weekly), specifically engaging in daytime MVPA (06:00-18:00) decreases DR risk by 38–84% and improves retinal structural characteristics.

Lower PA levels were significantly associated with DR progression. Notably, more than half of patients with established DM or vision-threatening DR fail to meet the WHO-recommended PA guidelines. These findings align with Zhao et al.'s Behavioural Risk Factor Surveillance System (BRFSS) analysis of 99,172 individuals, where only 18.5% of type 2 diabetes patients met PA guidelines[24]. Consistent with prior research, an additional study also found that those who participated in lifestyle PA were less likely to be diagnosed with DR [25]. Multiple lines of evidence suggest that higher PA correlates with reduced DR incidence [26] and may ameliorate DR severity [27]. However, the absence of comprehensive retinal fundus imaging data in the UKB database precluded a detailed investigation of severity-specific associations across the spectrum of DR. Future studies should incorporate graded clinical data to further validate these findings.

Our research indicated that MVPA exerts a protective effect across all stages of the disease process, with PA performed in the morning potentially yielding the most pronounced benefits. Prior studies have demonstrated that engaging in PA during the daytime can reduce the risk of developing type 2 diabetes mellitus (T2DM) by 9% and 10%, even under varying sociodemographic factors, which aligns with our findings. The underlying mechanism by which different exercise rhythms delay the progression of DR might be closely linked to glycaemic metabolism, lipid metabolism, and skeletal muscle clocks within the body [28]. A clinical trial revealed that exercising at 10:30 a.m. was more effective in reducing hyperglycaemic levels than other time windows [29], whereas aerobic exercise performed in the afternoon or evening enhanced glycaemic control [30]. Additionally, engaging in high-intensity interval training (HIIT) exercises in the morning leads to a decrease in circulating lipid levels and an increase in carbohydrate concentrations, whereas exercising in the afternoon produces the opposite effect, with a reduction in circulating carbohydrate levels and an increase in lipid levels [31]. Another study has shown that different exercise rhythms have distinct effects on skeletal muscle clocks, and appropriate exercise timing can yield more significant benefits for metabolic disorders such as T2DM [32].

In conclusion, regardless of the 24-hour period in which the PA occurred, prolonged adherence to moderate-intensity aerobic exercise results in a significant decrease in fasting blood glucose levels and a reduction in the incidence of DR [33]. Additionally, it is interesting that both MPA and MVPA performed between 6:00 a.m. and 12:00 p.m. independently protect against DR, except PA occurring between 0:00 a.m. and 6:00 a.m., which was identified as an independent risk factor. Our findings suggest that choosing a more favourable time for MVPA may have a more beneficial effect on delaying the development of DR.

Structural retinal alterations precede the clinical diagnosis of DR. Establishing correlations between accelerometer-measured PA and spectral-domain optical coherence tomography (SD-OCT)-derived retinal thickness parameters may facilitate investigation into the potential neuroprotective benefits of exercise during the preclinical stages of DR. Our study demonstrates significant positive associations between increased MVPA duration and enhanced MT as well as RNFL thickness, independent of DR progression stages. These findings align with evidence from the PROGRESSA study, which reported positive correlations between self-reported and device-measured PA levels and retinal layer thickness. Specifically, participants in the highest quartiles of MPA and VPA exhibited significantly greater GCIPL thickness than those in the lowest quartiles (+ 0.57 μm, P < 0.001 and + 0.42 μm, P = 0.005, respectively) [34]. A prospective cohort study further substantiated these observations, showing a positive association between time spent in MVPA and reduced GCIPL thinning rate (β = 0.06 μm/y/SD; 95% CI: 0.01–0.105; P = 0.018)³⁵. Emerging evidence suggests PA may also exert protective effects on retinal microvasculature. Higher levels of PA are associated with a smaller foveal avascular zone (FAZ) area, and greater central vessel density and perfusion density [36].

This study demonstrates that BMI partially mediates the protective association between PA and DR, accounting for 35.7-58.7% of the total effect. Importantly, 41.3-64.3% of the protective effect was attributable to non-BMI metabolic pathways, including improved glycemic control, enhanced lipolysis and fat oxidation [37], and reduced systemic inflammation [38]. The observed association between the mechanisms through which PA improves metabolic regulation [39], [40]. These findings suggest that weight reduction alone provides limited benefits for DR prevention and that combined interventions incorporating exercise and weight management may yield optimal protective effects against DR development.

Furthermore, the beneficial effects of PA on the visual system may be even more pronounced in individuals with vision-threatening DR. PA has been recognized as a crucial factor for overall health. Numerous studies have identified that physical inactivity or sedentary behaviour is a significant risk factor for various complications of diabetes [41]. When engaging in PA for no less than 30 minutes, 5 days a week, the risk of DR can be reduced by 40% [42]. Therefore, drawing upon the Global Recommendations on PA for Health as well as the PA Guidelines for Adults issued by the United States Department of Health and Human Services, we recommend that individuals in the early and intermediate stages of chronic eye disease or at high risk, consider long-term adherence to MVPA (≥ 150 min per week) [43]. Future research should focus more cohorts with accelerometer-measured PA data to better explore lifestyle interventions for managing DR and compare the benefits of different exercise modes and durations.

Advantages and limitations

The cohort utilized in this study possesses several notable strengths. Our research leveraged a large, prospective, and diverse population sample from the UKB to acquire objective, reproducible, high-quality seven-day accelerometer-assessed PA data using Axivity AX3 devices, designed to capture individuals’ “typical activity patterns.” These devices offer open-source data processing capabilities, which assist in mitigating the self-reported bias commonly encountered in PA data derived from questionnaires. An extensive array of sociodemographic, lifestyle, and anthropometric covariates was included, allowing for a well-powered, broad investigation of the relationship with PA, while minimizing potential confounders.

Despite the aforementioned strengths, our study still has several limitations. First, a significant limitation of this study is the reliance on single cross-sectional data to define PA. Although single-week PA has been validated as a reliable measurement tool in multiple large-scale cohort studies for exploring epidemiological associations between PA and chronic diseases [8], [44], [45], and strong correlations exist between PA levels measured at different time points across an individual’s life course [46], [47], [48], long-term PA inevitably fluctuates. Second, while we adhere to an established time-based segmentation method, varying definitions of the ‘morning’, ‘afternoon’, and ‘evening’ periods may lead to divergent outcomes. Thirdly, in our study cohort, the absence of comprehensive retinal fundus imaging data in the UKB database precluded a detailed investigation of severity-specific associations across the spectrum of DR. Future studies should incorporate graded clinical data to further validate these findings. Fourth, although the mediation analysis model in this study was theoretically grounded and large-scale prospective randomized controlled trials have demonstrated that increased PA directly reduces BMI and improves glycemic control [49], this analytical framework cannot definitively preclude the possibility of reverse or bidirectional causal pathways. Finally, while the prevalence of DR in our study cohort aligns with global epidemiological trends [50], the limited sample size reduces statistical power. Despite these limitations precluding causal inference between specific PA patterns and retinal thickness, the UKB cohort has strong population representativeness. Moreover, our findings are consistent with multiple prospective cohort studies, all demonstrating a significant positive correlation between MPA and retinal sublayer thickness [34]. Future research should utilize larger DR case samples and longitudinal designs to validate and extend our conclusions.

Conclusion

The current study provides evidence that long-term adherence to MVPA (≥ 150 min per week) may prevent or delay the progression of full-course DR and is positively associated with increased thickness in GCIPL, MT, and INL-ELM, suggesting a potential neuroprotective effect of PA. Our findings support current PA guidelines but suggest that they could be more stringent. Consequently, population-level interventions to increase PA should be combined with diet-induced weight loss strategies for optimal DR prevention.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Abbreviations

DR

Diabetic retinopathy

DM

Diabetes mellitus

PA

Physical activity

HbA1c

Haemoglobin A1c

UKB

UK Biobank

Non-DM

Non-diabetes mellitus

OCT

Optical coherence tomography

VA

Visual acuity

VI

Visual impairment

Pre-DM

Prediabetes Mellitus

DM-NDR

Diabetes mellitus without diabetic retinopathy

ICD-9/10

International Classification of Diseases, Ninth Revision/Tenth Revision

MPA

moderate physical activity

VPA

Vigorous physical activity

WHO

World Health Organization

TPA

Total physical activity

MET

Metabolic equivalent of task

MVPA

Moderate-vigorous physical activity

BMI

Body mass index

PSM

Propensity score matching

HLM

hierarchical line models

GCIPL

Ganglion cell inner plexiform layer

MT

Macular thickness

INL-ELM

Inner nuclear layer-external limiting membrane

RNFL

Retinal nerve fibre layer

ISOS-RPE

Inner segment-outer segment to retinal pigment epithelium

ELM-ISOS

External limiting membrane to inner segment-outer segment

RPE

Retinal epithelium

NIE

Natural indirect effects

TE

Total effects

CI

Confidence interval

OR

Odds ratio

T2DM

Type 2 diabetes mellitus

HIIT

High-intensity interval training

SD-OCT

Spectral-domain optical coherence tomography

FAZ

Foveal avascular zone

Author contributions

Study concept and design: Du ZJ, Zhang XY, Shen CX. Acquisition, analysis, or interpretation: All authors. Drafting of the manuscript: Shen CX, Du ZJ, Zhang XY. Critical revision of the manuscript for important intellectual content: Yu HH. Statistical analysis: Du ZJ, Shen CX. Obtained funding: Yu HH, Zhang XY, Chen YL. Administrative, technical, or material support: Yu HH. Study supervision: Zhang XY, Yu HH.

Funding

This study was supported by National Natural Science Foundation of China (U24A20707, 82171075, 82301260), Guang-dong Basic and Applied Basic Research Foundation (2023B1515120028), Scientific Research Project of Traditional Chinese Medicine Bureau of Guangdong Province (20231012), China Postdoctoral Science Foundation (2024T170185), Brolucizumab Efficacy and Safety Single-Arm Descriptive Trial in Patients with Persistent Diabetic Macular Edema (2024-29).

Data availability

No datasets were generated or analysed during the current study.

Declarations

Ethics approval

The UK Biobank study was approved by the UK National Health Service National Research Ethics Service (Ref 11/NW/0382), and informed consent was obtained from all participants.

Consent for publication

Written informed consent for publication was obtained from all participants.

Competing interests

The authors declare no competing interests.