Retinal Artery Occlusion and the Risk of Stroke Development
Twelve-Year Nationwide Cohort Study
Background and Purpose—Our aim was to evaluate the risk of subsequent stroke development after retinal artery occlusion (RAO).
Methods—National registry data were collected from the Korean National Health Insurance Service, comprised 1 025 340 random subjects. Patients diagnosed with RAO in 2002 and 2003 were excluded. The RAO group was composed of patients with an initial diagnosis of either central or other RAO between January 2004 and December 2013 (n=401). The comparison group was composed of randomly selected patients (5 per RAO patient; n=2003) who were matched to the RAO group according to sociodemographic factors and year of RAO diagnosis. Each sampled patient was tracked until 2013. Cox proportional hazard regression was used.
Results—Stroke occurred in 15.0% of the RAO group and in 8.0% of the comparison group (P < 0.001). RAO was associated with an increased risk of stroke occurrence (hazard ratio, 1.78; 95% confidence interval, 1.32–2.41). The magnitude of the RAO effect for stroke was larger among younger adults aged <65 years (hazard ratio, 3.11) than older adults aged ≥65 years (hazard ratio, 1.26). However, the risk of subsequent stroke was significantly increased in older adults aged ≥65 years at the 4-year follow-up (hazard ratio, 1.58; 95% confidence interval, 1.01–2.48).
Conclusions—RAO was significantly associated with subsequent stroke after adjusting for comorbidities and sociodemographic factors. These findings are limited by uncontrolled confounding factors and need to be replicated by other observational studies.
Retinal artery occlusion (RAO) is considered an ocular emergency because it often leads to permanent vision loss.1 Anatomically, the retina and brain share the same blood supply, the internal carotid artery, and the retina is a component of the nervous system, representing an extension of the diencephalon. Developmentally, the retina originates from the neural tube, which is a part of the brain.2 RAO and stroke share a common pathophysiological mechanism of thromboembolism3 and share common risk factors, including hypertension, diabetes mellitus, and hyperlipidemia.4 Recent studies suggest that RAO is associated with stroke and mortality caused by stroke.5–15 In the present study, we therefore investigated the possible association between RAO and the subsequent risk of stroke occurrence based on RAO cases and their controls obtained from a nationwide representative sampling of 1 025 340 adults.
The study protocol was approved by the Institutional Review Board of Severance Hospital, Yonsei University College of Medicine, Seoul, Korea.
All individuals in South Korea are obligated to enroll in the single-payer, Korean National Health Insurance Service, and nearly all of the medical data in the health system are centralized in a large database. Claims are accompanied by data on diagnostic codes, procedures, prescription drugs used, and personal information. The Korean government uses a unique identification number for each Korean resident to identify every person in the health care system. Furthermore, the Korean National Health Insurance Service uses diagnostic codes based on the Korean Classification of Diseases (KCD), which is a similar system to the International Classification of Diseases (ICD). This study used the National Health Insurance Service National Sample Cohort 2002–2013 (NHIS-NSC 2002–2013), which was released by the Korean National Health Insurance Service. The data comprised 1 025 340 nationally representative random subjects, amounting to ≈2.2% of the entire population in the Korean National Health Insurance Service in 2002. Proportionate stratified random sampling was used based on a total of 1476 strata (2 categories for sex, 18 categories for age group, and 41 categories for income).
A retrospective nationwide propensity score–matched cohort study was performed. The RAO group included all patients who received inpatient and outpatient care between January 2004 and December 2013 for an initial diagnosis of central RAO or other RAO (KCD code H34.1, corresponding to ICD-9-CM code 362.31 Central RAO; KCD code H34.2, other RAO, corresponding to ICD-9-CM code 362.32/362.33 arterial branch occlusion/partial arterial occlusion). Because of a higher RAO incidence in 2002 to 2003 compared with other periods of study, patients in 2002 to 2003 were excluded as a chronic condition to ensure that the RAO group included only subjects with new episodes. Patients previously diagnosed with stroke before 2004 (KCD codes I60–I69, corresponding to ICD-9-CM codes 430–438, cerebrovascular disease) were also excluded. We included patients who were diagnosed with RAO before their stroke, based on the visitation date. Finally, 401 eligible RAO patients in 2004 to 2013 were identified after excluding potential preexisting cases of both RAO and stroke. These cases were regarded as new incident cases of RAO. We selected 2003 patients (5 per RAO patient) from the database, who were matched to the RAO group in terms of age, sex, residential area, household income, and year of RAO diagnosis. Each patient was tracked for ≤10 years from the time of (1) diagnosis of RAO or (2) randomly selected visitation date in matched-year of enrollment, to the last follow-up to identify those patients who developed stroke (KCD codes I60–I63, corresponding to ICD-9-CM codes 430–434, cerebrovascular disease except transient cerebral ischemia, ill-defined disease, or late effects).
The regression models were adjusted for patient age (<50, 50–59, 60–69, 70–79, or ≥80 years), sex (men/women), household income (<30%, 30–70%, and ≥70%), and geographic location according to 4 regions (Seoul, a metropolitan area in Korea; the second area included the largest province, the third area included the second largest city, and the 2 second to third largest provinces; and the fourth area included other areas). Comorbidities, such as hypertension, ischemic heart disease, atrial fibrillation, diabetes mellitus, chronic renal failure, and hyperlipidemia diagnoses based on the KCD, may be associated with an increased risk of stroke.16–18 Therefore, we defined these comorbidities as any diagnoses between 2002 and 2013.
Descriptive statistics of the study population were performed. After accounting for sociodemographic differences between baseline characteristic for RAO patients and the general population, propensity score matching was performed. We calculated propensity scores by estimating a logistic regression to predict RAO occurrence using—and controlling for—sociodemographic factors (eg, age, sex, residential area, and household income) and year of RAO diagnosis. Matching was performed using a greedy macro with the estimated propensity score. An 8-to-1 digit greedy matching algorithm was then used to identify a unique matched control for each RAO patient according to the propensity score.19 Under this algorithm, if this match could not be found, the algorithm then proceeded sequentially to the next highest digit match (a 7-, 6-, 5-, 4-, 3-, 2- or 1-digit match; 8→1 digit match) of the propensity score to determine the next-best matches in a hierarchical sequence until no more matches could be found. When a match was found, it was not considered again. From the preliminary analysis, the stroke occurrence (exposure) for the comparison group was determined to be ≈10%. We were able to achieve >80% power in detecting a hazard ratio (HR) of ≥1.7 with a sample size of 2400 (1:5 ratio of cases:controls) based on a conservative assumption of the effectiveness of our matching (correlation of exposure, 0.3).20 Univariate and multivariate Cox proportional hazard regression analyses were performed to identify hazards associated with stroke using HRs and 95% confidence intervals (CIs).
The overall stroke-free rate was estimated using the Kaplan–Meier method, and the log-rank test was used to determine differences in stroke occurrence. The proportional hazards assumption was assessed by a Cox model with Schoenfeld residuals, and the assumption was not violated in overall or subgroup analyses except for analyses of the older age group (≥65 years of age). A time-dependent effect of RAO for stroke (χ2=4.36; P=0.037) in the older age group was observed; therefore, we treated RAO incidence as a time-dependent variable in a piecewise model for further analysis.21 The total follow-up duration was divided into 3 periods: 0–4, 4–7, and 7–10 years. When we evaluated the effect of RAO on incidence of stroke, the RAO group had a higher risk of stroke (HR, 1.85) until the fourth year of follow-up and a lower risk of stroke during the 4- to 7-year and 7- to 10-year follow-up periods (HR, 0.45 at 4–7 years and 1.01 at 7–10 years). Therefore, we chose 0- to 4-year stroke-free survival for the older age group as a study exit such that the proportional-hazards assumption was not violated. Even though the proportional hazards hypothesis is violated for the ≥65-year-old age group during the 10-year follow-up, we hypothesized a constant proportional hazards for 10 years and presented the HR and HR at 0 to 4 years (Figure 1). A threshold of P=0.05 was selected for statistical significance. The statistical software SAS System for Windows, version 9.4 (SAS Institute Inc, Cary, NC) and Stata/MP, version 14.0 (StataCorp, College Station, TX) were used to perform the statistical analyses.
Table 1 shows the characteristics of the study population for the 2 cohorts: the RAO group and the comparison group. We examined 401 RAO patients and 2003 sociodemographically matched comparison subjects. Stroke occurred in 15.0% of the RAO group and in 8.0% of the comparison group (P<0.001). Patients with RAO were more likely to experience hypertension, ischemic heart disease, atrial fibrillation, diabetes mellitus, chronic renal disease, and hyperlipidemia relative to the comparison group. No significant difference in year of enrollment, age, sex, residential area, or household income was detected between the 2 groups because these variables were used for sample matching.
Table 2 shows the HRs for stroke during the ≤10-year follow-up period using univariate and multivariate Cox regressions. RAO patients had an increased risk of subsequent stroke (HR, 1.78; 95% CI, 1.32–2.41). Comorbidities, such as hypertension (HR, 1.95; 95% CI, 1.30–2.91), ischemic heart disease (HR, 1.89; 95% CI, 1.42–2.53), and atrial fibrillation (HR, 1.78; 95% CI, 1.19–2.68) were significantly associated with subsequent stroke. In terms of sociodemographic characteristics, increasing age was significantly associated with subsequent stroke.
A total of 9681 person-years, including 1543 person-years for RAO patients and 8138 person-years for matched controls, were examined. Stroke occurred in 38.9 per 1000 person-years for the RAO patients and 19.7 per 1000 person-years for the comparison group (Figure 1). Subgroup analysis for age, stratified into 2 subgroups: <65 and ≥65 years, and additional analyses for the age group ≥65 years, based on a 4-year follow-up as an study exit, were provided. RAO patients aged <65 years were significantly associated with a greater risk of stroke (HR, 3.11; 95% CI, 1.89–5.10). When we assumed the proportional hazard risk for the 10-year follow-up, RAO patients aged ≥65 years exhibited an increased risk for stroke (HR, 1.26; 95% CI, 0.84–1.90); however, this result was not statistically significant. Additional analyses of the ≥65-year age group at the 0- to 4-year follow-up as a study exit, which satisfied the proportional hazard risk during this period, showed that RAO patients aged ≥65 years were associated with an increased risk of stroke (HR, 1.58; 95% CI, 1.01–2.48). RAO was associated with elevated stroke risk in both men and women (HR, 2.11; 95% CI, 1.41–3.15 for men; HR, 1.53; 95% CI, 0.95–2.45 for women).
Figure 2 shows the results of the Kaplan–Meier survival curve. In Figure 2A, the log-rank test indicates that RAO patients developed stroke significantly more frequently than the comparison group (P<0.001, log-rank test). In Figure 1B, the survival curve shows the difference of stroke incidence between age groups; stroke occurred more frequently for the older age group (≥65 years) than for the younger age group (<65 years). The trends of stroke-free survival rates for older adults (≥65 years) were time dependent, from the beginning of the study to approximately the 4-year follow-up; the stroke-free rate of the RAO group showed a more abrupt decrease than the comparison group, but the gaps in stroke-free incidences between the 2 groups were smaller after the 4-year follow-up. In Figure 2C, the survival curve represents the difference of stroke incidences for RAO in men and women; overall, stroke occurred more frequently in men than in women. Despite the different incidences of stroke by sex, the gap between the comparison group and the RAO group in Figure 2C for each sex was large, suggesting that RAO was associated with the risk of stroke in both sexes, even though the results for women were marginally significant. There was no difference in the incidence of stroke between the central RAO and the other RAO (P=0.810, log-rank test; Figure 2D).
In this study, we examined 401 RAO patients and 2003 sociodemographically matched comparison subjects whose data were obtained from a national database of 1 025 340 randomly selected subjects in Korea. We found that RAO patients exhibited a higher subsequent incidence of stroke during the 10-year follow-up period after adjusting for hypertension, ischemic heart disease, atrial fibrillation, diabetes mellitus, chronic renal failure, and hyperlipidemia.
The retinal vessel exhibits unique characteristics that can be directly and noninvasively examined using ophthalmoscopy.22 Because of potentially serious disabilities after stroke, study of the retinal vasculature, including the RAO, has gained increasing attention. However, few previous clinical or epidemiologic studies have been performed because of the rarity of RAO. From a cohort study that included 98 RAO patients over a period of 5 years, the average annual incidence of stroke was 2.5%, the first 1-year risk of stroke was 7.4%, and the Kaplan–Meier survival curve of stroke-free incidences was similar to that of the present study (Figure 2A), showing a relatively abruptly decreasing stroke-free incidence at the beginning of the study.7 The most common cause of RAO is retinal embolism, which can be classified as 1 of 3 types: (1) calcific, (2) cholesterol, and (3) platelet-fibrin.12 From a cohort study including 70 subjects with asymptomatic retinal cholesterol embolism and 70 controls, stroke occurred at an annual incidence of 8.5% in RAO patients and 0.8% in controls.8 The large epidemiologic study of the Beaver Dam Eye Study and the pooled study of the Beaver Dam Eye Study and Blue Mountains Eye Study reported associations between retinal arteriolar emboli and incident stroke mortality.9,13,14 From a hospital-based study, including 439 RAO patients, transient ischemic attack/cerebrovascular accidents occurred frequently in both 234 central RAO patients (19%) and 141 branch RAO patients (17%) when compared with the US population (4.3% in the matched US population for central RAO and 3.4% in the matched US population for branch RAO). These results are similar to the present study in terms of similar incidences of stroke comparing central RAO and other RAO patients (Figure 2D). Taiwan has a nationwide health insurance system similar to that of our country. A study from Taiwan reported a significant association between RAO and stroke including 464 RAO patients and 2784 matched controls: (1) RAO patients (19.6%) were more likely to experience a stroke during the 3-year follow-up period when compared with controls (10.1%); (2) the risk of stroke occurrence was particularly high at the beginning of the study; and (3) the incidence rate ratio of 3.34 was highest in a younger age group of ≤60-year patients when compared with other age groups.15 These results were consistent with those of our study. However, stroke in the present study did not include transient ischemic attack; therefore, the incidence rate was lower in our results than in the report from Taiwan.
In Figure 2B, a relatively abrupt decrease in stroke-free incidence for the RAO group from the beginning of the study was observed until the 4-year follow-up period, and our additional analyses showed that older adults with RAO experienced an increased risk of stroke in this period (HR, 1.58; Figure 1). Even if there was no difference in the cumulative incidence of stroke between the RAO group and the comparison group for 10 years in the ≥65 years of age group, it is important to be aware that strokes often occur in the early period of observation in both age groups. In our previous study, retinal vein occlusion was associated with a 1.5-fold increased risk of stroke.23 Finally, we confirmed that both the retinal vein and the artery occlusion were associated with a subsequently increased stroke risk. However, the different time-dependent characteristics for stroke development between the retinal vein occlusion and RAO should be emphasized. The risk of subsequent stroke is particularly high immediately after RAO development (Figure 2A) although an abrupt decrease in the stroke-free rate was not observed in retinal vein occlusion patients in our previous study.23 Diffusion-weighted magnetic resonance imaging within 7 days of the onset of visual loss detected acute ischemic stroke in 8 patients (34.2%) of the total of 33 RAO patients in an earlier study at our institute24; however, selection bias exists because patients who did not have neuroimaging were excluded. The American Heart Association/American Stroke Association recommends that all patients with suspected brain or retinal ischemia undergo immediate brain imaging and etiological work-ups25; however, only 35% of ophthalmologists refer patients with central RAO to the emergency room.26 Our findings confirm that a systemic evaluation to prevent stroke among patients with RAO should be performed as soon as possible because of the elevated early stroke risk, but it is debatable whether RAO patients without other neurological deficits should be referred to emergency departments equipped with a stroke unit or referred to outpatient clinics.27
Based on Figure 2C, even though stroke occurred more frequently in men than in women, the stroke-free rate between the RAO group and the comparison group in each sex showed differences early in the study period. However, the differences between RAO and the comparison group were smaller in women than in men; therefore, the effect size for incident stroke in the RAO group was larger among men (HR, 2.11) than among women (HR, 1.53) after adjustment (Figure 1). In addition, subjects with central RAO were more likely to have major underlying causes, including large artery atherosclerosis, cardioembolism,24 or stroke, as well as a transient ischemic attack,15 compared with patients with branch RAO. However, the present study, which was based on a long-term follow-up period, showed no difference in occurrence for stroke (not including transient ischemic attacks) between the central RAO and the other RAO patients (Figure 2D). RAO is, therefore, associated with an increased risk of stroke, regardless of RAO subtype, sex, or age.
The cardioembolic source or emboli from carotid artery are common causes of RAO.28 Interestingly, central RAO is associated with visible emboli in ≤20% of patients,29 but branch retinal artery occlusion has a higher rate of visible emboli, approaching 60% to 70%.5 Other mechanisms of retinal artery occlusion should be considered besides emboli such as local thrombosis, hypercoagulability, vasospasm, and hypoperfusion.30,31 After adjusting for atrial fibrillation in both groups, we found that the subsequent stroke rate is still higher in RAO patients than in controls. This results indicate that subsequent stroke is higher in RAO patients than in controls but not because atrial fibrillation is more common in RAO patients. However, other mechanisms of small-vessel disease, including local thrombosis, hypoperfusion, and vasospasm, might play a role in the development of RAO and subsequent stroke after RAO development.
Because of the rarity of RAO, few large population-based studies of RAO have been performed; thus, this study over an extended time period, using a relatively large sample size of RAO patients, provides a novel contribution to the field.
The limitations of this study include (1) the possibility of misclassification of diagnoses for RAO or comorbidities; (2) the possible under-reporting of asymptomatic RAO; (3) the possibility of delayed visits to the ophthalmologist and delayed diagnosis of RAO; (4) the possibility that chronic RAO patients were not fully excluded; (5) the inability to collect other important health-related information, such as smoking or body mass index, preventing control for these confounding factors; (6) the possibility that medical claims might have included biased controls when compared with general population-based controls, who neither received medical care nor had a specific diagnosis; and (7) the possibility that differences exist between South Korean and Western populations, limiting the generalizability of our findings. In addition, evaluation of stroke subtype using ICD and KCD is impractical32; therefore, further study is needed to evaluate what subtypes of stroke are associated with RAO.
RAO was associated with stroke development after adjusting for comorbidities and sociodemographic factors. This finding and the findings of our previous study suggest that retinal vessel occlusion, including the artery and vein, may play an important role in preventing strokes. Artery occlusion in the retina, regardless of age, sex, or branch versus central RAO, may therefore need more emergent systemic screening than retinal vein occlusion. These findings are limited by uncontrolled confounding factors, including behavioral risk factors such as smoking, and should be replicated by other observational studies.
We thank Peng Guan Ong, BSc (Hons, Biostatistician at Singapore Eye Research Institute with Specialization in Biostatistics, Singapore) for his help with statistical advice. This study used the National Health Insurance Service National Sample Cohort 2002–2013 (NHIS-NSC 2002–2013, NHIS-2015-2-070), which was released by the Korean National Health Insurance Service, and the authors alone are responsible for the content and writing of this article.
Sources of Funding
This research was supported by a grant from the Korea Health Technology R&D Project provided by the Korea Health Industry Development Institute, which is funded by the Ministry of Health & Welfare, Republic of Korea (HI13C1485).
- Received July 14, 2015.
- Revision received November 18, 2015.
- Accepted December 8, 2015.
- © 2016 American Heart Association, Inc.
- Von Graefe A.
- Hankey GJ,
- Slattery JM,
- Warlow CP.
- Wang JJ,
- Cugati S,
- Knudtson MD,
- Rochtchina E,
- Klein R,
- Klein BE,
- et al
- Goldstein LB,
- Adams R,
- Becker K,
- Furberg CD,
- Gorelick PB,
- Hademenos G,
- et al
- Parsons L.
- Patton N,
- Aslam T,
- Macgillivray T,
- Pattie A,
- Deary IJ,
- Dhillon B.
- Furie KL,
- Kasner SE,
- Adams RJ,
- Albers GW,
- Bush RL,
- Fagan SC,
- et al
- Merchut MP,
- Gupta SR,
- Naheedy MH.
- Goldstein LB.