Data was collected from the following Danish nationwide registers: The Danish National Patient Register contains information regarding all hospital admissions since 1974 classified by diagnoses according to International Classification of Diseases (ICD). ICD-8 codes were used from 1974 to 1995, and ICD-10 codes from 1996 onwards. [28] In the Register of Medicinal Product Statistics, all data on dispensed prescription drugs have been collected since 1994 coded according to The Anatomical Therapeutic Chemical (ATC) Classification System. [29, 30] Information regarding death and cause of death was accessed via the Danish Cause of Death Register. [31] Finally, linkage of data between registers is possible due to the unique identification number afforded all Danish residents.

RAO was identified according to ICD-8 code 37702 and ICD-10 code H340, H341 and H342. RVO was identified according to ICD-8 37703 and ICD-10 H348. Unspecified retinal vascular occlusion was defined by ICD-10 code H34 without specification of RAO or RVO. AF was identified by ICD-8 code 42793,42794 and ICD-10 code I48.

Other comorbidities established from discharge diagnosis were heart failure (ICD-8: 425, 4270, 4271, ICD-10:I110,J819, I42 and I50). Prior stroke, transitory ischemic attack or systemic thromboembolism (ICD-8: 433,434,435,436,444, ICD-10: I63, I64DG458, G459 and I74) and vascular disease (myocardial infarction and peripheral vascular disease (ICD-8: 410, 440, ICD-10: I21, I22, I700, I702, I703, I704, I705, I706, I707, I708 and I709).

Preceding use of anticoagulation was identified as redeemed prescriptions of warfarin (ATC: B01AA) prior to index date i.e. the date of the qualifying diagnosis of retinal vascular occlusions for cases. Index date for controls were the same date as for the corresponding cases. Medication at baseline was defined as redeemed prescriptions 180 days prior to admission. Prescriptions of any glucose-lowering drug (ATC: A10) defined diabetes mellitus. Specific drugs identified included; clopidogrel (ATC: B01AC04), dipyridamole (ATC: B01AC07), non-steroidal anti-inflammatory drugs (ATC: M01A except M01AX05), loop diuretics (ATC: C03C) and acetylsalicylic acid (ATC: B01AC06 and N02BA01). Notably, although acetylsalicylic acid can be sold over-the-counter, most use continues to be registered due to reimbursement policies.

Hypertension was defined as concurrent treatment with two antihypertensive drugs of different classes: antiadrenergic (ATC: C02A, C02B, C02C), non-loop diuretics (ATC: C02L, C03A, C03B, C03D, C03E, C03X, C07B, C07C, C07D, C08G, C02DA, C09BA, C09DA and C09XA52), vasodilators (C02DB, C02DD and C02DG), beta blockers (C07A, C07B, C07C, C07D and C07F), calcium channel blockers (C08, C07F, C09BB and C09DB), and renin angiotensin system inhibitors (ATC: C09AA, C09BA, C09BB, C09CA, C09DA, C09DB, C09XA02 and C09XA52). This method has previously been validated. [26]

All patients with discharge diagnosis of retinal vascular occlusions (i.e. RAO and RVO) from January 1st 1997 to December 31st 2011 were identified. Patients with previous AF or previous use of warfarin were excluded. Inclusion in the study happened on the day of discharge of retinal vascular occlusion. Each patient with retinal vascular occlusion was matched according to sex and age with five controls from the Danish National Patient Register using the Greedy Matching Macro (http://​www.​mayo.​edu/​research/​documents/​gmatchsas/​doc-10027248). Controls were free from previous AF or prior use of warfarin. Patients were followed until incident of AF, death, or end of study period.

Rates of AF were calculated as new cases of AF per 100 person-years. Cumulative incidence was calculated taking into account the competing risk of death. Risk of AF according to retinal vascular occlusion was determined by Cox regression analysis and presented unadjusted and adjusted for hypertension, diabetes, vascular disease, and prior stroke/systemic thromboembolism/transitory ischemic attack. Independent subgroup analyses were performed for RAO, RVO, and unspecified occlusions, using the corresponding controls as reference. Fulfillment of all model assumptions including proportional hazards, goodness of fit chi-square test, and significance of interactions between all covariates adjusted for in the model were tested and found valid.. Data management and statistical analyses were performed using SAS 9.2 (SAS Institute Inc., Gary, NC, USA) and R 3.0.2. (R Core Team (2013). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. URL https://​www.​R-project.​org).

Retrospective register based studies do not require ethical approval in Denmark and the study was approved by the Danish Data Protection Agency (Ref. no. 2007–58-0015 /Internal ref. no. GEH-2014-013, I-Suite number: 02731).

Etiologies for ischemic strokes can be classified as large-artery atherosclerosis, small-vessel occlusion, other determined etiology, undetermined etiology or cardioembolism according to the TOAST (Trial of Org 10,172 in Acute Stroke Treatment) criteria. [32] Cardio-embolism is estimated to be the source of stroke in 16–30% of all ischemic strokes, and 61% in patients with AF.

In stroke patients, guidelines recommend timely diagnosis of AF is prioritized to ensure prescription of oral anticoagulation to prevent future ischemic strokes. As AF in many cases is paroxysmal, detection rate of AF after ischemic stroke depends on the applied diagnostic method. Using a 7-day event-recorder, AF can be detected in 12.5–15% of stroke patients free of previous – known – AF. [33, 34] The detection rate increases further with continuous implantable and external cardiac monitoring. [35, 36]

The association of AF and RAO is not investigated to the same extent as the association of AF and ischemic stroke. Whereas one study found a similar detection rate of AF in RAO patients AF (12.5% of patients using 24-h ECG (at least)), other studies indicate that AF is more common in patients with ischemic stroke and find a higher number needed to screen in patients with RAO than in patients with ischemic stroke [14, 37, 38]. Patients with AF have a higher ratio of hemispheric to retinal occlusions compared to patients with AF, which may indicate that other embolic sources than the atria such as a carotid stenosis play a larger role in RAO patients. [39]

No studies have however investigated the proportion of cardio-embolic strokes in RAO. [40‐42] In our population, less than 10% of the RAO patients were diagnosed with AF during follow-up. Several studies suggest assessment of cardiovascular risk in RAO could reasonably comprise Holter monitoring and trans-esophageal echocardiography. [21, 23, 33, 43] A clinical trial of patients with implanted devices capable of detecting silent episodes of AF would contribute to furthering knowledge on the association of AF with subsequent risk of retinal vascular occlusion.

We have previously shown how retinal vascular occlusion (either RAO or RVO) is an independent and substantial risk factor for stroke and thromboembolism in patients with known AF. In this study of 87,000 patients with AF, HR of RAO and RVO was 1.39 (1.08–1.79) and 1.26 (1.02–1.54), respectively, after adjustment for cardiovascular comorbidities and medication. [27]

In contrast to results from the present study, findings from another study found an increased risk of AF in patients with RVO, which may be explained by difference between the study designs or ethnicity of the populations. [44] Because RVO share risk factors with ischemic stroke, RVO may be interpreted as a stroke surrogate marker. [16] As such, AF detection in RVO may still be relevant especially since AF commonly develops asymptomatically, and the silent development of incident AF in patients with prior retinal vascular occlusions would infer greater risk of stroke in the absence of anticoagulation.

Since this is a registry-based cohort study, the results represent an association between the two conditions rather than a causal relation. The hypothesis that subclinical AF may be an underlying cause of RAO is supported by the significant increase in cumulative incidence of AF amongst RAO patients compared to controls. Comparably, no significant difference was found between RVO and controls. This observed difference is also relevant when addressing the issue of detection bias: Because this is an observational study, patients with retinal vascular occlusion conceivably receive greater medical attention after study inclusion compared with controls (i.e. surveillance bias), but the cumulative incidence of AF did not differ between RVO patients and controls, thereby countering the premise of surveillance bias. Thus, the observed results support a hypothesis of RAO, and not RVO, being caused by – undiagnosed – silent AF.

As we did not have access to information about risk factors such as obesity, excessive use of alcohol, the increase in risk of incident AF in patients with RAO may reflect a higher frequency of unmeasured risk factors associated with cardiovascular disease compared to controls and patients with RVO. However, cumulative incidence of death was similar in RVO and RAO patients, thereby implying that the increased risk of AF in RAO – and not RVO – reflects a genuine association independent of differences in health status of cases and controls.

This study substantiates the importance of a careful work-up required in patients with retinal vascular occlusion, which is in accordance with other studies. [37, 45] Detection of AF leads to indication for oral anticoagulation (in the presence of at least one CHA₂DS₂-VASc point) to efficiently reduce risk of ischemic stroke. [46, 47] Further studies are needed to estimate the effect of oral anticoagulation on the risk of RAO.

The primary strength of the study is the substantial size of the cohort of patients diagnosed with retinal vascular occlusion in Denmark between 1997 and 2011, and the 1:5 matched population without retinal vascular occlusion.

However, only retinal vascular occlusions diagnosed in hospitals were included, as we did not have access to data regarding patients treated outside of hospitals. As such, it is possible that the identified retinal vascular occlusions treated in hospitals are more severe compared to cases treated outside of hospitals. Consequently, if the association between retinal vascular occlusion and AF is more pronounced for more severe cases, our study may overestimate this association.