Table 3.

Methods, limitations and results of relevant studies

First author	Methods	Results	Variables included in multivariate model	Limitations
Studies which used carotid disease as the dependent variable:
Cardoso [23]	Tertiary-care university hospital outpatients were consecutively recruited Required to have either any microvascular complication or macrovascular complication with at least 2 modifiable risk factors Excluded if > 80 years old, BMI ≥ 40 kg/m2, serum creatinine ≥ 2 mg/dl or poor life expectancy Participants followed up till first endpoint or end of study Cox regression used	No measure of carotid disease was associated with new or worsening DR in most adjusted model From personal communication from author, for highest versus lowest tertile of IMT, the HR (95% CI) was 0.99 (0.59–1.64) P = 0.95 for CCA, 1.23 (0.73–2.06) P = 0.44 for BIF and 1.17 (0.71–1.93) P = 0.53 for ICA Carotid plaque score ≥ 3 points was also not significant; 1.69 (0.88–3.24) P = 0.12	Age, sex, diabetes duration, BMI, smoking, physical activity, clinic SBP, number of antihypertensive drugs, use of insulin and statins, presence of macrovascular diseases and baseline DR, mean HbA1c, HDL and LDL during first year of follow up	Selection bias—recruited from tertiary hospital clinic so likely complex type 2 diabetes participants Potential measurement bias as single ophthalmologist Unclear risk of attrition bias
Hjelmgren [32]	Participants were recruited from the Western Region initiative to Gather Information of Atherosclerosis (WINGA) database Participants who were referred for ultrasound after suffering first ischaemic stroke or TIA were consecutively included Excluded if < 40 years old, no ultrasound within 6 months of event or if information on DR ambiguous Logistic regression used	Any DR did not increase the odds of carotid stenosis (OR: 0.79 (0.48–1.30), P = 0.35)	Age, CHD, HF, PAD and creatinine	Selection bias—only included those who had experienced an ischaemic stroke or TIA Potential detection bias as information from medical records may not be complete and detection may vary between hospitals/clinics Used stenosis > 50% as outcome so would have missed lower rates of occlusion/CIMT increase Potential measurement bias—unsure about who/how many conducted the carotid ultrasound High proportion of males Limited generalisability
Carbonell [7]	Recruited from two outpatient university hospital clinics belonging to the same health care organisation Participants were identified from electronic clinical records and included if > 18 years old and diabetes duration ≥ 1 year Excluded if history of CVD, diabetic foot disease, eGFR < 60 ml/min/1.73 m2 or uACR > 300 mg/g Logistic and multinomial logistic regression used	Advanced DR (OR: 2.66 (1.03–6.95) P = 0.044) but not mild DR (1.35 (0.66–2.76) P = 0.41) was independently associated with carotid plaque Advanced DR (OR 4.71 (1.48–15.04) P = 0.009) was also independently associated with increased odds of ≥ 2 carotid plaques The presence of any DR was not statistically significantly associated with any plaque (1.64 (0.85–3.17) P = 0.14) or ≥ 2 plaques (1.93 (0.83–4.47) P = 0.129)	Age, sex, diabetes duration, smoking, diastolic BP, dyslipidaemia, uACR, BMI, pulse pressure and LDL	Selection bias—only recruited from clinics Excluded those with CVD and in doing so may have excluded some with carotid disease Sample size was calculated on the presence of DR and not advanced DR Potential measurement bias as single ophthalmologist and sonographer at each site
Liu [34]	Participants were from the Diabetes Health Management Program, a community-based system of electronic health records recruited via free health check-up annually for residents and household survey at Meilong Town Excluded if < 40 years of age or history of CVD Linear and logistic multiple regressions used	Any DR was associated with CCA IMT (mm) (coefficient 0.015, P = 0.010, Standard error: 0.080) in linear Regression Any DR was associated with CCA IMT > 1 mm (OR:1.84 (1.02–3.31) P = 0.043), presence of plaque (1.87 (1.03–3.39) P = 0.039) and subclinical atherosclerosis (1.93 (1.03–3.60) P = 0.039) in most adjusted logistic regression models	Age, sex, alcohol use and LDL in all models. The logistic regressions also adjusted for smoking, hypertension, diabetes duration, HbA1c, use of antidiabetic drugs, insulin use, antihypertensive drugs, obesity, Triglycerides, total cholesterol, HDL, eGFR, uACR and GGT	Excluded those with CVD and in doing so may have excluded some with carotid disease Potential measurement bias as single ophthalmologist and sonographer Unclear risk of selection bias
Alonso [28]	Recruited based on medical records from an outpatient clinic and diabetic eye disease program Tried to match those with DR and those without on age and sex Excluded those with CVD or impaired renal function General linear models used for IMT and logistic regression for plaque presence	Any DR was associated with mean ICA IMT (P = 0.0176) but not CCA IMT or bifurcation IMT Any DR increased odds of any plaque (OR: 1.71 (1.03–2.85) P = 0.0366) and the odds of ≥ 2 plaques (3.17 (1.75–5.75)) P < 0.0001)	All models adjusted for age. Also, in general linear models CCA IMT adjusted for smoking; bifurcation IMT for hypertension; and ICA IMT for sex. In logistic regression for any plaque adjusted for hypertension and smoking; and for ≥ 2 plaques for sex and dyslipidaemia	Selection bias—only recruited from clinics Excluded those with CVD and in doing so may have excluded some with carotid disease Potential measurement bias as single ophthalmologist and sonographer
Jung [27]	Hospital patients’ notes were retrospectively reviewed Excluded if malignancy, hepatic failure, acute infection, acute metabolic complications, fatal arrhythmia or CVD Logistic regression used	Any DR was independently associated with CCA IMT > 1 mm (OR: 3.8 (1.4–10.2)) but not > 2 carotid plaques (OR: 5.7 (0.6–51.3))	Age, diabetes duration, smoking, hypertension, HbA1c, cardiac autonomic neuropathy, brachial-ankle pulse wave velocity, statin use, ACE-I/ARB use and eGFR	Potential detection bias -retrospectively analysed medical notes which relies on all data being available/recorded appropriately All tests would have been done as part of usual diabetes care so may reflect a higher risk population Excluded those with CVD and in doing so may have excluded some with carotid disease Potential measurement bias—single ophthalmologist and limited detail on who/how many performed ultrasounds Relatively small, young sample Limited generalisability
Cardoso [29]	Tertiary -care university hospital outpatients were consecutively recruited Required to have either any microvascular complication or macrovascular complication with at least 2 modifiable risk factors Excluded if > 80 years old, BMI ≥ 40 kg/m2, serum creatinine ≥ 2 mg/dl or poor life expectancy Generalised linear models were used with DR as a fixed factor to assess relationship with IMT Logistic regression used to assess relationship between DR and plaque score	Any DR was associated with increased odds of a plaque score > 2 (OR: 1.70 (1.02–2.84) P = 0.043) Any DR was not independently associated with IMT at ICA, BIF or CCA in either logistic or linear regression but effect size and P-values not given	The logistic regression for plaque score adjusted for age, sex, smoking, antihypertensive use and aortic pulse wave velocity. All IMT linear and logistic regressions adjusted for age and night-time pulse pressure. Additionally, CCA IMT adjusted for sex, smoking and antihypertensive use; bifurcation IMT for LDL and smoking; and ICA IMT for sex and C-reactive protein in logistic regression as well as smoking in linear regression.	Potential selection bias—recruited from tertiary hospital clinic so likely complex type 2 diabetes participants Potential measurement bias as single ophthalmologist
Son [10]	Consecutive patients of an outpatient diabetes centre diagnosed with diabetes during the study period were recruited Excluded those with longer duration of diabetes, CVD or cerebrovascular events Logistic regression used	Any DR increased the odds of plaque or increased CCA IMT > 0.9 mm (OR: 6.57 (1.68–25.71) P = 0.007) compared to those with CCA IMT < 0.9 mm and no plaque	Age, sex, smoking, hypertension, BMI, diabetic nephropathy, HbA1c, fasting glucose, HDL and LDL	High proportion of males Only comprised participants with newly diagnosed diabetes Small sample size Excluded those with CVD and in doing so may have excluded some with carotid disease Potential measurement bias – single ophthalmologist and limited detail on who/how many performed ultrasounds
Lacroix [33]	Patients with diabetes referred to a vascular laboratory were consecutively recruited Excluded those with life expectancy > 12 months, a recent (< 6 weeks) stroke or TIA, carotid surgery, cervical radiotherapy or symptoms of carotid disease Multiple logistic regression used	Any DR increased odds of any carotid stenosis (2.38 (1.06–5.33) P = 0.03) Any DR increased odds of carotid stenosis ≥ 60% (3.62 (1.12–11.73), P < 0.0001) compared to no or < 60% stenosis	Age > 70 years, hypertension, BMI, history of CHD and family history of diabetes in model for any stenosis. Sex, ABI and history of ischemic neurological disorder or cervical bruit in model for stenosis ≥ 60%	Selection bias—recruited from referrals to a specialist clinic and excluded those with symptoms of carotid disease Study was focussed on screening for carotid disease Potential measurement bias—limited detail on who/how many people performed ophthalmic exams or ultrasounds
Distiller [31]	Patients with diabetes were recruited from the Centre for Diabetes and Endocrinology Included those with at least 10 measurements of Hba1c in last 5 years, normal renal function, no proteinuria Those on statins > 5 years, with an underlying autoimmune disease, nephropathy, on steroids or those with hypothyroidism with inadequate replacement were excluded Multiple logistic regression, linear regression and ordinal logistic regression used	Any DR increased the odds of plaque; OR: 3.65 (1.11–12.02) P = 0.033, but not IMT or IMT risk (effect for these not given)	In multiple regression for IMT adjusted for age, diabetes duration, BMI, hypertension and HDL. In ordinal regression for IMT risk adjusted for age, triglyceride:HDL ratio and HbA1c. In logistic regression for plaque adjusted for age, hypertension and smoking	Some selection bias as recruited from a diabetes centre and only Caucasians with long diabetes duration Potential measurement bias—limited information given on ascertainment of DR status

First author	Methods	Results	Results adjusted for	Limitations
Studies which used diabetic retinopathy as the dependent variable:
Ichinohasama [8]	Unclear how participants recruited, but underwent assessment at hospital Excluded those with HbA1c < 6.5 and if not on ongoing diabetes therapy, those on haemodialysis, or who had malignancy, inflammatory disease, chronic respiratory disease, macular degeneration, or glaucoma or other retinal disease Logistic regression analysis used	CCA IMT increased the odds of mild NPDR (OR: 8.65 (1.95–38.4) P = 0.005, per 1 mm increase)	Age, sex, duration of diabetes, HbA1c diastolic blood pressure, heart rate, creatinine, central macular thickness and mean blur rate in the overall optic nerve head	Potential selection bias—participant recruitment was not clear Only right sided IMT and DR assessed Participants had no or mild DR, none with more severe DR Excluded those with T2D with HbA1c < 6.5 and diet controlled Potential measurement bias as single ophthalmologist and sonographer
Yun [10]	Participants registered at a public health centre who had participated in another survey were recruited Excluded those with missing data including blood, urine, HbA1c, diabetes duration, CCA-IMT, carotid plaque, baPWV, DR outcome Logistic regression analysis used	CCA IMT was not associated with DR in most adjusted model (OR for tertile 2: 1.16 (0.67–2.02) and tertile 3: 1.06 (0.59–1.90) when compared to tertile 1, P = 0.844) Carotid plaque was not associated with DR (OR: 1.20 (0.75–1.91))	Age, sex, duration of diabetes, HbA1c, total cholesterol, triglycerides, HDL, eGFR, BMI and history of hypertension	High proportion of females Potential measurement bias—unclear about who graded DR and the reliability and validity of the ultrasounds
Araszkiewicz [30]	Hospital patients admitted for diabetes management recruited consecutively Excluded those > 50 years old, liver dysfunction, chronic kidney disease ≥ stage 3, anaemia, acute inflammation, CVD, CHD, PVD, DKA on admission or carotid stenosis > 50% Logistic regression analysis performed	CCA IMT not associated with DR in multivariate analysis (OR: 1.00 (0.99–1.01) P = 0.169 per 1 μm increase)	Age, sex, diabetes duration, albuminuria, BP, postprandial glucose, HbA1c, central augmentation index and peripheral augmentation index	Potential selection bias—recruited from a hospital Only right sided IMT measured Likely excluded those with higher IMT as excluded if stenosis > 50% and CVD Small numbers Potential measurement bias as single ophthalmologist and unclear who performed ultrasounds
Rema [9]	A random sample of 450 participants with known and 150 participants with newly diagnosed diabetes from a population-based study were assessed Multivariate regression models were used	Mean IMT increased the odds of DR (OR: 2.9 (1.17–7.33), P = 0.024 per 1 mm increase)	Age, HbA1c, duration of diabetes and microalbuminuria	Unclear selection bias risk Only used the right carotid ultrasound

DR diabetic retinopathy, VTDR vision threatening diabetic retinopathy, IMT intima-media thickness, CCA common carotid artery, ICA internal carotid artery, BIF bifurcation, CVD cardiovascular disease, uACR urinary albumin:creatinine ratio, CHD coronary heart disease, PVD peripheral vascular disease, DKA diabetic ketoacidosis, TIA transient ischaemic attack, HF heart failure, PAD peripheral arterial disease, BP blood pressure, BMI body mass index, LDL low density lipoprotein, HDL high density lipoprotein, eGFR estimated glomerular filtration rate, GGT gamma-glutamyl transferase, ACE-I angiotensin converting enzyme inhibitors, ARB angiotensin receptor blockers, ABIs ankle-brachial index