Introduction

Chronic obstructive pulmonary disease (COPD) is characterized by persistent respiratory symptoms and airflow limitation due to airway and/or alveolar abnormalities usually caused by significant exposure to noxious particles or gases1. Although the diagnosis of COPD is based on spirometry, it is well recognized that COPD is associated with a range of systemic extrapulmonary effects1. Indeed, comorbidities are frequently present and contribute to the severity of disease2,3,4. Cardiovascular disease (including hypertension, coronary artery disease, congestive heart failure, stroke and peripheral arterial disease) is a common comorbidity in COPD2,5.

Smoking is the most important risk factor for the development of COPD, but it is also a major risk factor for the development of cardiovascular disease6,7. Indeed, cigarette smoke not only affects the airway epithelium, but also causes vascular endothelial damage and loss of microvascular integrity in several organs, including the heart, brain, and kidney8,9. In addition, systemic inflammatory processes, oxidative stress, and hypoxia may contribute to microvascular complications in COPD10,11. Since it has been suggested that changes in the microvasculature may be associated with cardiovascular disease12,13, monitoring the properties of the microvascular system in COPD could be valuable for additional phenotyping with possible added-value in clinical practice for cardiovascular risk stratification14.

Retinal microvasculature can be easily visualized using fundus photography and can serve as a non-invasive proxy to assess the microvascular health status15. Findings from large epidemiological studies demonstrated that structural assessment of retinal microvasculature can provide useful information for cardiovascular risk prediction16,17,18,19. Indeed, it has been recognized that changes in the retinal microvasculature may precede the occurrence of hypertension, obesity and diabetes mellitus (DM)20. Earlier population-based studies demonstrated that more narrow retinal arteriolar vessels are associated with hypertension, while wider retinal venular vessels are associated with systemic inflammatory markers, hyperglycaemia, obesity and dyslipidaemia18,21. Similar findings were observed in patients with type 2 DM, heart failure or rheumatoid arthritis22,23,24,25.

To date, systematic studies characterizing the retinal microvasculature in patients with COPD are scarce. Data from a population study suggest that the width of retinal venules is inversely associated with forced expiratory volume in one second (FEV1) and FEV1/forced vital capacity (FVC)26,27. This would indicate that vascular comorbidities associated with obstructive decrements in lung function may be characterized by defects in the microvascular circulation, as assessed with fundus imaging26.

The primary aim of this study was to measure retinal vessel widths in patients with COPD and to determine the association between these metrics and lung function parameters, functional outcomes and cardiovascular risk factors. A priori, we expected an inverse association between retinal vessel diameter and lung function parameters in patients with COPD26,27. Moreover, based on similarities with type 2 DM and heart failure, we postulated that the microcirculation, as assessed by retinal vessel diameters, is inversely associated with systemic cardiometabolic parameters and functional outcomes in patients with COPD22,23,24.

In addition, we aimed to determine the effects of an exercise-based pulmonary rehabilitation program on retinal vessel diameters in patients with COPD. Based on earlier work, demonstrating that regular exercise can induce retinal arteriolar dilatation and venular constriction in healthy adults28,29, we hypothesized that regular exercise training can change the dimensions of retinal vessel diameters in these patients and hence provide support for improved microcirculatory status as a result of regular exercise.

Results

Four hundred sixty-two patients with COPD referred for pulmonary rehabilitation at Ciro were screened between July 2016 and June 2017. One hundred forty-six patients were ineligible to participate due to various reasons (Fig. 1). In addition, 65 eligible subjects declined participation because of disinterest, and in two patients it was not possible to obtain retinal images due to small pupils. Two hundred forty-nine patients underwent fundus photography, of which three patients had ungradable retinal images. So finally, 246 patients were included in the analyses.

Figure 1
figure 1

Study flow chart: screening and enrollment of study participants.

Full size image

Characteristics

Patient characteristics are listed in Table 1. Slightly more than half of the patients were male and mean age was 64.4 ± 8.5 years. Patients had mild to very severe COPD and were frequently diagnosed with risk factors of cardiovascular diseases, including hypertension, (pre-) DM, and dyslipidemia. Furthermore, 47.6% of the patients had a history of cardiovascular diseases, including coronary artery disease (20.8%), peripheral arterial disease (10.4%), stroke (8.0%) and congestive heart failure (6.8%). Patients showed impaired aerobic capacity, as assessed by cardiopulmonary exercise test (CPET) and six-minute walk test (6MWT). Long-term oxygen therapy was used by 18.3% of the patients.

Table 1 Baseline characteristics of COPD patients (n = 246).
Full size table

Retinal findings

Patients had a mean central retinal arteriolar equivalent (CRAE) of 152.5 ± 12.7 µm and mean central retinal venular equivalent (CRVE) of 233.7 ± 21.1 µm. Retinal abnormalities were present in 146 (59%) patients, of which 47 (19%) had multiple abnormalities. Most common abnormalities were increased vessel tortuosity (n = 44), peri-papillary atrophy (n = 42), (suspicion of) cataract (n = 34), and (suspicion of) glaucoma (n = 22) (See Supplementary Table S1).

Associations with retinal vessel diameters

Gender and age-adjusted univariate associations between retinal vessel diameter and lung function parameters, clinical variables, cardiovascular risk factors and functional outcomes of pulmonary rehabilitation are listed in Table 2. Older patients with COPD had a significantly smaller CRAE and CRVE. Several other variables were inversely associated with CRAE, including transfer factor for carbon monoxide (TLCO %predicted), arterial oxygen saturation at rest (SaO2), systolic blood pressure, diastolic blood pressure, and triglycerides levels (Table 2). In addition, SaO2 and systolic blood pressure were inversely associated with CRVE, whilst low-density lipoprotein cholesterol levels and level of systemic inflammation were positively associated with CRVE (Table 2). Results did not demonstrate an association between retinal vessel diameters and FEV1 or FEV1/FVC. Furthermore, no association was found between retinal vessel widths and baseline exercise capacity (Table 2).

Table 2 Age- and sex-adjusted association between CRAE and CRVE and clinical, cardiovascular and functional measures.
Full size table

Multivariate regression models for CRAE and CRVE are listed in Table 3 (See Supplementary Table S2 for the complete model with all covariates). Model 1 identifies an association between older age and higher systolic blood pressure and narrower retinal arterioles (β: −0.224 (95% confidence interval (CI) −0.439, −0.009), p = 0.042; and β: −0.136 (95%CI −0.210, −0.062), p < 0.001) when adjusting for relevant confounders, including sex, body mass index (BMI), fat free mass index (FFMI), FEV1/FVC, TLCO, residual volume (RV), total lung capacity (TLC), diastolic blood pressure, arterial partial pressure of oxygen (PaO2), long-term oxygen therapy, smoking status, smoking pack-years, blood glucose, dyslipidemia, pulmonary and cardiovascular medication, cardiovascular comorbidities and fellow vessel diameter. Higher systolic blood pressure, low level of systemic inflammation and older age were associated with narrower retinal venules; β: −0.229 (95%CI −0.379, −0.120), p = 0.041; −13.767 (95%CI −24.556, −2.978), p < 0.001; and −0.654 (95%CI −1.089, −0.381), p = 0.042; respectively. These associations remained significant after further adjustment for fellow retinal vessel diameter (model 2; Table 3).

Table 3 Multiple regression model of CRAE and CRVE.
Full size table

Response to pulmonary rehabilitation

One hundred and ninety-six patients with COPD completed pulmonary rehabilitation and assessment after rehabilitation. Characteristics of these patients are listed in Table 4. Baseline characteristics of these patients were not significantly different compared to patients who did not complete pulmonary rehabilitation and assessment after rehabilitation.

Table 4 Baseline characteristics of COPD patients stratified according to response to pulmonary rehabilitation.
Full size table

After pulmonary rehabilitation, patients had a significantly improved six-minute walk distance (14 ± 50 m, p = 0.001) and cycle endurance time on constant work rate test (CWRT) (224 ± 358 s, p < 0.001). 33.5% of the patients achieved a clinically relevant improvement in 6MWT, whilst 51.5% of the patients achieved a clinically relevant improvement in cycle endurance time. No significant differences were found in CRAE (0.02 ± 7.2 µm, p = 0.97) and CRVE (0.61 ± 11.1 µm, p = 0.52) following pulmonary rehabilitation.

Non-responders to pulmonary rehabilitation had a significantly worse baseline lung function, and lower BMI and FFMI compared to responders. Baseline retinal vessel diameters and changes in retinal vessel diameters after pulmonary rehabilitation were not significantly different between responders and non-responders (Table 4). In addition, baseline CRAE or CRVE were not significantly associated with change in functional outcomes of pulmonary rehabilitation, including six-minute walk distance (β for CRAE: −0.002 (95%CI −0.035, 0.032); p = 0.930 and β for CRVE: −0.007 (95%CI −0.062, 0.047); p = 0.792, respectively) and cycle endurance time on CWRT (β for CRAE: 0.001 (95%CI −0.003, 0.006); p = 0.562 and β for CRVE: 0.002 (95%CI −0.005, 0.010); p = 0.520, respectively). Moreover, no associations were found between change in CRAE or CRVE following pulmonary rehabilitation and change in six-minute walk distance (β for CRAE: −0.004 (95%CI −0.030, 0.021); p = 0.754 and β for CRVE: 0.003 (95%CI −0.036, 0.043); p = 0.868, respectively) and cycle endurance time on CWRT (β for CRAE: −0.004 (95%CI −0.007, 0.003); p = 0.517 and β for CRVE: −0.001 (95%CI −0.006, 0.004); p = 0.768, respectively).

Discussion

This is the first study measuring retinal vessel diameters in a large group of well-characterized COPD patients. We showed that retinal arteriolar and venular diameters were independently associated with systolic blood pressure, whereas retinal venular diameter was also related to systemic inflammatory status and age. Retinal vessel diameters, as a proxy for microvascular health, did not change after an exercise-based pulmonary rehabilitation program. Retinal vessel diameters were not associated with important functional outcomes of pulmonary rehabilitation. Additionally, we found retinal abnormalities in more than half of the patients with COPD.

To date, only few studies examined the relation between retinal vascular parameters and lung function26,27,30. Data from one large population-based study demonstrated that changes in retinal venular diameter were associated with decrements in lung function in healthy individuals (28 mL decrement in FEV1 per SD unit of CRVE), especially in current smokers (93 mL decrement in FEV1 per SD unit of CRVE)26. This association between retinal vessel diameter and lung function was not confirmed by our data from a large group of COPD patients. This might be explained by the compromised lung function with a narrower range of FEV1 values in our population compared to the study of Harris et al.26. In addition, our FEV1 values are at the lower bound of the distribution of values measured in the latter study. Similar to our findings, Chew et al. and Ugurlu et al. did not demonstrate an association between retinal vessel diameter and severity of COPD27,30.

Structural assessment of retinal microvasculature can provide useful information for cardiovascular risk prediction16,17,18. To date, no studies have evaluated retinal vessel parameters and cardiovascular risk parameters in COPD. Although earlier population-based studies demonstrated that a lower FEV1 is an independent risk factor for cardiovascular morbidity and mortality31,32, we could not establish an association between lung function parameters and retinal vessel diameter in patients with COPD. Then again, we did find an association between retinal vessel diameter and traditional cardiovascular risk factors, including systolic blood pressure and level of inflammation. Considering the high prevalence and incidence of cardiovascular comorbidity in COPD patients2,33, and the fact that cardiovascular comorbidities are often undiagnosed34,35, an easy to obtain fundus image may be an interesting metric for further cardiovascular phenotyping of patients with COPD. However, we emphasize the need for determining the clinical relevance of retinal imaging in cardiovascular risk prediction in patients with COPD.

It is increasingly recognized that systemic inflammatory markers are involved in the pathogenesis of cardiovascular diseases and can predict cardiovascular events36,37, and systemic inflammatory processes may have a role in the association between retinal vessel diameter and cardiovascular diseases, including coronary heart disease, DM, hypertension and stroke38. Large population-based studies already demonstrated that systemic inflammation is associated with retinal venular diameter. The Rotterdam study, the multi-ethnic study of atherosclerosis and the Beaver Dam Eye Study reported associations of nonspecific and specific systemic inflammatory markers (i.e. white blood cell count, erythrocyte sedimentation rate, high-sensitive C-reactive protein (hs-CRP), interleukin (IL)−6) with larger venular diameter18,38,39. Our study is the first to establish an association between systemic inflammation and retinal venular diameter in patients with COPD. Moreover, this association was consistently seen regardless of smoking status, and other cardiovascular risk factors. In addition, we identified an independent association between retinal vessel diameter and systolic blood pressure. The inverse relation between retinal vessel diameter and elevated systolic blood pressure has already been reported in several large population-based studies18,21,40, but, to the best of our knowledge, never in patients with COPD. Future longitudinal studies will be needed to demonstrate the usefulness of retinal blood vessel analysis as part of cardiovascular risk screening in patients with COPD.

In contrast to population-based studies, our findings did not demonstrate an association between retinal venular diameter and blood glucose level or DM. To date, little is known on the exact mechanisms for the association of wider retinal venular diameter with hyperglycemia and DM,. It has been speculated that venular widening may reflect systemic inflammatory processes involved in the pathogenesis of impaired glucose metabolism41, which is further supported by findings from epidemiological studies, showing an association between wider retinal venules and elevated systemic inflammatory markers18,38,39. It is therefore possible that the high proportion of patients with increased levels of systemic inflammation in this study might attenuate the association between retinal venular diameter and DM. This is confirmed by the study of Heitmar et al., in which no association was found between retinal vessel diameter and DM in different patients groups with increased levels of inflammation42.

It has been suggested that hypoxia may be involved in the etiology of some retinal abnormalities in COPD43,44. Indeed, a constant oxygen supply is crucial for adequate organ function44., andeven small changes in oxygen supply to the retina can result in tissue hypoxia and retinal changes44. In addition, exposure to corticosteroids may contribute to an increased risk of glaucoma and cataract, though results in patients with COPD are contradictory45,46,47. An earlier review demonstrated that structural and functional changes in the retinal microvasculature are more common and severe in patients with COPD compared to non-COPD controls, including increased retinal vessel diameter, lower retinal arterial oxygen saturation, impaired haemodynamics and increased resistive index of the orbital vessels43. Chew et al. also demonstrated that retinal abnormalities were more frequently present in hospital patients with COPD compared to other hospital patients (80% vs. 50%)27. Furthermore, patients with COPD were more likely to report visual impairment compared to subjects without COPD (14.0% vs. 9.6%)48. In line with this, ophthalmological screening in this study identified a high prevalence of retinal abnormalities in patients with COPD. More interestingly, these abnormalities were not recorded in their medical history. As retinal abnormalities can result in a reduced visual acuity49, which is related to a decreased quality of life and physical functioning, and increased risk of falling50, an ophthalmological screening might be valuable in patients with COPD.

Lower levels of physical activity are associated with wider retinal venular diameters51,52. In addition, it has been suggested that physical exercise has a positive effect on the structure (i.e. retinal arteriolar dilatation and venular constriction) and functionality of the microvasculature in subjects with microvascular diseases and physically inactive adults28,53,54. To date, only few studies evaluated the effects of different exercise modalities on microvasculature29,55,56. In patients with migraine and patients with unipolar depression it has been demonstrated that high-intensity interval training is slightly more effective in improving microvascular structure and function than moderate, continuous exercise55,56. It remains challenging to determine which exercise modalities are the most optimal for improving microvascular structure and function for different patient groups and disease severities. Indeed, this is the first study in patients with COPD. Our results demonstrate that an exercise-based pulmonary rehabilitation program did not affect static measurements of retinal vessel diameter in patients with COPD. Although exercise training (i.e. high-intensity interval and continuous) and duration of the pulmonary rehabilitation program were comparable, or even longer compared to earlier studies, it is possible that this program may not provide optimal stimuli to induce significant effects in retinal vessel diameters56. Moreover, the timing of pulmonary rehabilitation may not be optimal to improve microvascular function. Indeed, it has been recognized that microvascular dysfunction is an early event in the natural history of COPD57,58, and it is an open issue whether microvascular dysfunction is still reversible in patients with advanced COPD. It is possible that the response to an exercise program can be attenuated in patients with COPD, as endothelial dysfunction and impaired flow-mediated dilation are often present in these patients59. Moreover, the possible effects of exercise training might be influenced by the high prevalence of cardiovascular comorbidities and use of cardiovascular medication. Similar to our findings, it was shown that pulmonary rehabilitation does not appear to be effective in improving arterial stiffness and endothelial dysfunction in patients with COPD60,61. Future studies should investigate the most suitable timing and exercise regimes for improving microvascular structure and function. In this context, one should also consider a more integral assessment of the microvasculature and include the analysis of biomarkers of endothelial function (e.g. endothelin) and functional markers such as dynamic retinal imaging and Laser-Doppler flowmetry62.

The major strength of this study is the use of a wide array of clinical variables, cardiovascular risk factors, and functional outcomes, allowing a comprehensive analysis of possible variables related to retinal vessel diameter in a large group of well-characterized COPD patients. This study has the following limitations. First, this is an observational study and no causal relations can be established. Furthermore, it is possible that cardiovascular diseases preceded the development of COPD in some patients. Future studies are needed to determine the clinical relevance of retinal vessel parameters for the development of cardiovascular diseases in patients with COPD. Second, we did not include a healthy control group. Then again, it was already documented that patients with COPD have increased retinal arteriolar and venular diameters compared to non-COPD controls27,30. Furthermore, 20% of the patients in this study had retinal vessel calibers outside the earlier reported age- and sex-specific reference range (male: CRAE 129.15–202.49 µm; CRVE 170.52–241.99 µm and female: CRAE 145.92–217.07 µm; CRVE 178.21–259.72 µm)63. However, comparison with normative data should be interpreted with caution, as studies used different retinal vessel measurement software systems. It has been demonstrated that there is a poor agreement between different retinal measurement software, indicating that absolute measurements from the different software systems could not be interpreted interchangeably64,65. Third, static retinal vessel imaging only reflects a snapshot of the retinal microvasculature and provides limited information on functional alterations of the vasculature66,67. For future studies, we recognize the added-value of dynamic vessel analysis because this test has the potential to assess microvascular reactivity and autoregulation, which are suggested to have diagnostic potential in detecting subclinical endothelial dysfunction in the microvasculature and facilitate early diagnosis of cardiovascular diseases66,67. Fourth, we only quantified retinal vessel diameter, whilst other retinal vessel parameters, including vessel tortuosity and fractal dimension, may also provide valuable information regarding cardiovascular risk68. Fifth, proportion of patient in GOLD stage I and GOLD A were relatively small, and therefore caution is required when generalizing findings to the whole COPD population. Sixth, we only used hs-CRP as marker of systemic inflammation, and we did not measure other inflammatory markers that may be of interest in COPD, such as IL-6, tumor necrosis factor- α, and fibrinogen. Finally, a high proportion of patients had cardiovascular comorbidities, which may have influenced the findings of this study. However, multivariate analyses did not show an association between retinal vessel diameter and cardiovascular diseases or cardiovascular medication. Furthermore, additional analyses did not show significant differences in retinal vessel diameters between patients with and without cardiovascular diseases.

To conclude, we demonstrated that retinal arteriolar and venular diameters of patients with COPD were significantly associated with systolic blood pressure and level of systemic inflammation, but there was no evidence for an association with lung function parameters, functional outcomes or other cardiovascular risk factors. Furthermore, an exercise-based pulmonary rehabilitation program did not affect retinal vessel diameter. Future studies are needed to evaluate the association between physical activity level and retinal vessel diameters and the possible beneficial effect of increased physical activity level on microvascular health, as assessed by retinal vessel diameters, in patients with COPD.

Methods

Study design

This was a prospective observational study conducted in Ciro (Horn, the Netherlands)69. The study was approved by the Medical Research Ethics Committees United (MEC-U, Nieuwegein, the Netherlands; NL56813.100.16). The study was performed in accordance with the tenets of the Declaration of Helsinki and Good Clinical Practice and has been registered on www.trialregister.nl (NTR5896) before enrolment of the first volunteering participant. All subjects gave written informed consent.

Study population

Patients referred for clinical assessment and pulmonary rehabilitation to Ciro were screened69. Patients were eligible to participate when they were diagnosed with COPD and were clinically stable. Patients with a clinical diagnosis of abnormalities in the lens or cornea, which makes it impossible to image the retina with a fundus camera, were excluded. Additionally, patients with known retinal diseases were excluded. Furthermore, patients who were unable to provide informed consent due to cognitive problems or unable to speak and understand Dutch or English were excluded.

Exercise-based pulmonary rehabilitation program

All patients with COPD participated in a comprehensive pulmonary rehabilitation program at Ciro, as defined by the latest international ATS/ERS statement on pulmonary rehabilitation70. The program consists of 40 sessions and can be inpatient (8 weeks, 5 days/week) or outpatient (8 weeks, 3 half days/week, followed by 8 weeks 2 half days/week). In addition to progressive physical exercises, such as high-intensity interval and continuous training on a stationary cycle and treadmill, and strengthening exercises, the program also included non-exercising components such as occupational therapy, nutritional counselling, psychosocial counselling, exacerbation management, optimizing medication use, and/or educational sessions.

Measurements

Demographics and clinical characteristics, including BMI, FFMI, post-bronchodilator pulmonary function (FEV1, FVC, TLC, RV, and TLCO %predicted), SaO2, arterial partial pressure of oxygen (PaO2) and carbon dioxide (PaCO2), peak aerobic capacity (CPET), functional exercise capacity (6MWT and CWRT at 75% of peak work rate), blood pressure, and blood measurements (i.e. blood glucose, blood lipids (high-density lipoprotein (HDL), LDL, triglycerides), and hs-CRP), were obtained during routine assessment prior and after pulmonary rehabilitation, as described before71. Diagnosis of hypertension was based on the use of antihypertensive medication or a systolic blood pressure>140 mmHg and/or diastolic blood pressure>90 mmHg on three separate occasions72. The diagnosis of (pre-) DM was based on a fasting plasma glucose level ≥6.1 mmol/L. Dyslipidemia was defined as total cholesterol  ≥6.4 mmol/L, LDL cholesterol  ≥4.4 mmol/L, HDL cholesterol <0.9 mmol/L, triglycerides ≥1.94 mmol/L, or if receiving lipid-lowering medication. Hs-CRP was used as a marker of systemic inflammation and categorized using the established clinical cutoff points into low (<1 mg/L), borderline (1–3 mg/L), moderately high (3.01–10 mg/L) and markedly high (>10 mg/L)73. Medication use and cardiovascular comorbidities were obtained from patients’ medical records.

Retinal photography

A non-mydriatic Canon CR-2 fundus camera (Hospithera, Belgium) with a field of view of 45° was used to obtain high-resolution, optic disc centred images of the fundus of the right eye of each patient at start and end of pulmonary rehabilitation. Patients had to abstain from exercise in the hour prior to the measurement. A trained grader masked to patient characteristics, analyzed the retinal images using semi-automated vessel analysis software developed at VITO (Belgium; http:\\mona.health).

The six largest arterioles and venules coursing through a zone between 0.5 and 1 disc diameter from the optic disc margin were measured and summarized as CRAE and CRVE, representing the average diameter of arterioles and venules of the eye15,74. Images were not synchronized on the cardiac cycle, but instead retinal vessel diameters of two images were averaged to minimize random variation in retinal vessel diameter due to different stages of the cardiac cycle75.

In addition, all images were screened by two ophthalmologist for the following changes: retinal vascular changes (i.e. increased vessel tortuosity, increased venous diameter, retinal hemorrhage, pathological arterio-venous crossings), optic disc changes (i.e. glaucomatous changes/suspect of glaucoma, peripapillary atrophy, peripapillary bleeding, congenital anatomical optic nerve variation, tilted disc, pale optic disc), vitreoretinal or choroidal structural changes (i.e. macular glistening/epiretinal membrane, foveal/perifoveal pigmentary changes, hard exudates, drusen, choroidal nevi). Slightly blurry media was defined as (suspicious of) cataract.

Statistics

A priori, we have calculated the minimum sample size to achieve adequate power. Based on Chew et al.27, demonstrating that microvascular abnormalities were prevalent in 80% of the patients with COPD, we should include 246 participants (relative precision of 10% with a 95% confidence interval).

Data were tested for normality and are presented as mean and standard deviation (SD) unless noted otherwise. Age- and sex-adjusted stepwise multiple regression models were used to identify the association between retinal vessel diameter and lung function parameters, clinical variables, cardiovascular risk factors and functional outcomes of pulmonary rehabilitation (6MWT, cycle endurance time on CWRT). Variables were selected as possible confounders if they showed a significant association with retinal vascular width on univariate analysis (p ≤ 0.10), or if a variable was reported to be associated with retinal vessel width in previous literature. In case of high collinearity between two variables, the variable with greater clinical significance, based on the judgement of the authors, was chosen. Variables were included in the final model (Model 1) if they were significant (p ≤ 0.05) and/or they modified the estimates of the remaining parameters (>10% change). To account for potential confounding, models were additionally adjusted for fellow vessel diameter (i.e. for CRVE in models of CRAE, and vice versa76) (Model2). Benjamini-Hochberg corrections were applied to correct for multiple testing. General linear models for repeated measures were used to determine the effect of pulmonary rehabilitation on retinal vessel diameter for all patients with COPD and after stratification for response on pulmonary rehabilitation based on the minimal clinically important difference of the 6MWT and cycle endurance time (responder: +30 m on 6MWT77 and/or +100 s in cycle endurance time on CWRT78, non-responder: <30 m on 6MWT and <100 s in cycle endurance time on CWRT). All analyses were performed using SPSS 25.0 (SPSS Inc; Chicago, Illinois). Level of significance was set at ≤0.05.