Keywords
Retinal artery occlusion - Prediction - Risk scores - CHA
2DS
2-VASc score - ESSEN Stroke Risk score
Introduction
Retinal artery occlusion is an ocular emergency, often resulting in irreversible vision
loss.[1] Retinal artery occlusion is strongly associated with multiple cardiovascular diseases,
including stroke.[2]
[3]
[4]
[5]
[6]
[7] Stroke is the second leading cause of mortality worldwide, only exceeded by heart
disease.[8]
The association between retinal artery occlusion and stroke has been investigated
in previous studies, in which the risk of stroke in retinal artery occlusion patients
has been determined to range between two and nine-fold higher compared with randomly
selected controls, depending on time after event, with the highest incidence up to
1 month after the retinal artery occlusion event.[9]
Retinal artery occlusion and stroke have been described as equivalent.[10]
[11]
[12]
[13] The retina is a component of the nervous system, during embryonic development the
retina arises from the neural tube and is an extension of the diencephalon.[1]
[14]
[15] Furthermore, the blood–brain barrier shares common structural and functional features
with the blood–retina barrier.[16]
[17] The retina and the brain share blood supply. The central retinal artery is the first
branch of the internal carotid artery, which also supplies the brain with blood.[1]
[18] In addition, the pathogenesis of retinal artery occlusion and stroke is in both
cases thromboembolic.[1]
[5]
[19]
[20]
All these shared features support a strong association between retinal artery occlusion
and stroke. The prevalence of retinal artery occlusion will increase with the globally
ageing population,[21] making optimized management and clear guidelines essential to reduce the disease
burden associated with retinal artery occlusion, especially for an important disease
such as stroke. Guidelines recommend that patients with retinal artery occlusion are
immediately referred for stroke evaluation to reduce the risk of stroke, the evaluation
should include brain imaging such as ultrasound (UL), computed tomography (CT), or
magnetic resonance (MR) scans.[3]
[10]
[11]
[22]
[23]
[24]
[25]
[26]
[27]
[28] However, among ophthalmologists, not all retinal artery occlusion patients are referred
to immediate stroke risk evaluation[25]
[26] and no specific guidelines specify which patients should be referred.
This study will investigate whether existing risk assessment models can be utilized
in patients with retinal artery occlusion to evaluate the 1-year risk of stroke and
the need for referral for stroke evaluation.
Methods
Data Sources
For this cohort study, data were obtained using different Danish nationwide registries.
All residents in Denmark are provided a unique personal identification number at birth
or immigration. Using this identification number, data from the included registers
can be cross-linked. Three registers were used in this study: first, The Danish Civil
Registration System, where information about the date of birth, sex, vital status,
and migration can be obtained[29]; second, The Danish National Patient Register, which includes all hospital admissions
along with diagnoses, procedures, and outpatient services,[30] since 1994, the Danish National Patient Register has classified data using the International
Classification of Diseases version 10 (ICD-10); and finally, The Danish National Prescription
Registry, holding information on individual-level data about all prescription drugs
sold in Danish pharmacies. The Danish National Prescription Registry uses the global
Anatomical Therapeutic Chemical classification codes.[31]
Study Population and Risk Stratification
The study population included all subjects with a discharge code of retinal artery
occlusion (ICD-10: H340, H341, and H342, including subcodes) between January 1, 1995
and December 31, 2018. The date of the diagnosis was defined as the index date. Subjects
were excluded from the population if they were <18 years, died on the index date,
immigrated to Denmark within the last year, or had inconsistent demographic information.
Baseline characteristics were obtained from the Danish National Patient Register and
the Danish National Prescription Registry. Baseline characteristics concerning medication
were based on claimed prescriptions within 1 year prior to the index date. Furthermore,
the prevalence of scanning was retrieved from the Danish National Patient Registry
using procedure codes for CT, MR, and UL scanning of the head or neck.
Risk stratification was performed based on the CHA2DS2-VASc score and the ESSEN Stroke Risk score used on the retinal artery occlusion population
evaluated at the time of diagnosis. The CHA2DS2-VASc score was developed for risk stratification of stroke in patients with atrial
fibrillation.[32] The CHA2DS2-VASc score ranges from 0 to 9 and is calculated based on points accumulated from
prevalent congestive heart failure (1 point), hypertension (1 point), diabetes mellitus
(1 point), prior ischemic stroke/systemic embolism/transient ischemic attack (2 points),
vascular disease (1 point), age 65 to 74 years (1 point), age ≥75 years (2 points),
and female sex (1 point).[33] The ESSEN Stroke Risk score was developed for risk stratification of recurrence
in patients with stroke. The ESSEN Stroke Risk score ranges from 0 to 9 and is calculated
based on points accumulated from prevalent hypertension (1 point), diabetes mellitus
(1 point), myocardial infarction (1 point), peripheral arterial disease (1 point),
ischemic stroke or transient ischemic attack (1 point), other cardiovascular events
(1 point), age 65 to 75 years (1 point), age >75 years (2 points), and current smoking
(1 point).[34] Patients were divided into groups based on the sum of the specified points. Scores
of ≥4 points were collapsed due to the small sample size. See [Supplementary Table S1] for details on score definitions.
Outcomes
In this study, the outcome of interest was stroke (ICD-10: I63 and I64). Patients
were followed until stroke, the competing event of death or censoring due to emigration,
or administrative end of follow-up at December 31, 2018. The follow-up time was 1
year.
Statistics
Survival analyses were used to assess the risk of stroke in patients with retinal
artery occlusion. Baseline characteristics were summarized using descriptive analysis
with proportions for categorical covariates and means for continuous variables. In
addition, the proportion of retinal artery occlusion patients using platelet aggregation
inhibitors or getting a CT, MR, or UL scanning within 48 hours was identified. The
Aalen–Johansen estimator was used to estimate the cumulative incidence of medical
use 3 months and 1 year after the index date. Additionally, an Aalen–Johansen estimator
was used to estimate the cumulative incidence of stroke according to score strata
for both risk scores considering death as a competing event.
The crude event rate for stroke was calculated as events per 100 person-years. Subsequently,
a Cox proportional hazard ratio model and a Fine and Gray model ([Supplementary material]) were used to calculate the hazard rate ratio and subdistributional hazard rate
ratio of stroke according to the risk score level,[35]
[36] the latter with death as competing event. Crude models were used since the performance
of the risk score was investigated. A likelihood ratio test was performed to assess
the overall effect of the risk scores compared with the null model.
C-statistics or the area under the receiver operating characteristics-curve provided
a measure of the discrimination of the investigated model.[37] To investigate the accuracy of the risk prediction in terms of both discrimination
and calibration, the Brier score was calculated. The Brier score is applicable for
survival data and is the mean squared difference between the observed status and the
predicted risk. A Brier score of 0 would indicate a perfectly calibrated prediction
model.[38]
[39] Afterward, the index of prediction accuracy was determined, which is 1 minus the
ratio between the null model and Brier score of the investigated model.[39]
[40]
Results
A total of 7,906 retinal artery occlusion patients were identified with a mean age
of 69.2 years and of which 46.9% were female. Of the included patients, 73.3% had
a CHA2DS2-VASc score of 2 or above and 54.8% had an ESSEN Stroke Risk score of 2 or above,
suggesting experiencing cardiovascular events prior to their retinal artery occlusion
event ([Table 1]). At baseline 40.7% of the study population had a recent prescription of platelet
aggregation inhibitors and after 1 year this percentage reached 65.9% (95% confidence
interval [CI]: 64.8–66.9%). Within the first 48 hours of their diagnosis, only 5.9%
of the RAO patients had a hospital record of a procedure for scanning of the carotid
artery to check for stenosis ([Table 2]).
Table 1
Baseline characteristics of the population comprising patients with retinal artery
occlusion
|
Characteristics
|
Population
|
|
n
|
7,906
|
|
Female % (n)*
|
46.9 (3711)
|
|
Age mean (SD)*
|
69.2 (13.4)
|
|
Arterial hypertension % (n)
|
40.4 (3196)
|
|
Diabetes % (n)
|
14.6 (1157)
|
|
Ischemic heart disease % (n)
|
18.3 (1449)
|
|
Peripheral artery disease % (n)
|
10.8 (856)
|
|
Stroke % (n)
|
11.8 (933)
|
|
Atrial fibrillation % (n)
|
9.0 (713)
|
|
Heart disease % (n)
|
12.8 (1009)
|
|
Renal disease % (n)
|
4.0 (313)
|
|
Cataract % (n)
|
22.8 (1806)
|
|
Glaucoma % (n)
|
5.9 (467)
|
|
CHA2DS2-VASc % (n)
|
|
|
0
|
9.6 (758)
|
|
1
|
17.0 (1343)
|
|
2
|
20.2 (1600)
|
|
3
|
21.3 (1687)
|
|
≥4
|
31.8 (2518)
|
|
ESSEN % (n)
|
|
|
0
|
15.1 (1192)
|
|
1
|
30.2 (2386)
|
|
2
|
24.2 (1912)
|
|
3
|
16.1 (1271)
|
|
≥4
|
14.5 (1145)
|
Abbreviations: SD, standard deviation.
Table 2
Treatment characteristics after the index date for retinal artery occlusion diagnosis
|
Characteristics
|
% (95% CI)
|
|
CT, MR, or UL scan of head or neck within 48 h
|
5.9 (exact)
|
|
Platelet aggregation inhibitors
|
|
|
Baseline
|
40.7 (exact)
|
|
3 mo after index date
|
55.6 (54.4–56.7)
|
|
1 y after index date
|
65.9 (64.8–66.9)
|
Abbreviations: CI, confidence interval; CT, computed tomography; MR, magnetic resonance;
UL, ultrasound.
During the 1-year follow-up, a total of 565 stroke events were observed. The overall
rate per 100 person-years was 7.99 (95% CI: 7.36–8.68). This incidence rate increased
correspondingly with the level of both scores (p < 0.00). An incidence rate of 3.62 (95% CI: 2.46–5.31) per 100 person-years was determined
for patients with a CHA2DS2-VASc score of 0, reaching 13.25 (95% CI: 11.78–14.89) per 100 person-years for patients
with a score of 4 or above. For the ESSEN Stroke Risk score, the incidence rate was
3.97 (95% CI: 2.97–5.32) per 100 person-years in patients with a score of 0 up to
16.43 (95% CI: 14.01–19.27) per 100 person-years in patients with a score of 4 or
above. All incidence rates depending on both the CHA2DS2-VASc score and the ESSEN Stroke Risk score are summarized in [Table 3].
Table 3
Stroke in patients with retinal artery occlusion
|
Events
n
|
Person-time
100 PY
|
Rate
/100 PY (95% CI)
|
Risk
% (95% CI)
|
Effect
HR (95% CI)
|
p-Value
|
|
CHA2DS2-VASc
|
|
|
|
|
|
|
|
0
|
26
|
7.19
|
3.62 (2.46–5.31)
|
3.4 (2.3–4.9)
|
Ref
|
|
|
1
|
71
|
12.54
|
5.66 (4.49–7.15)
|
5.3 (4.2–6.6)
|
1.56 (0.99–2.44)
|
|
|
2
|
92
|
14.63
|
6.29 (5.13–7.72)
|
5.8 (4.7–7.0)
|
1.71 (1.11–2.65)
|
|
|
3
|
97
|
15.27
|
6.35 (5.21–7.75)
|
5.8 (4.7–7.0)
|
1.72 (1.12–2.65)
|
|
|
≥4
|
279
|
21.06
|
13.25 (11.78–14.89)
|
11.2 (10.0–12.5)
|
3.48 (2.33–5.20)
|
0.000
|
|
ESSEN
|
|
|
|
|
|
|
|
0
|
45
|
11.32
|
3.97 (2.97–5.32)
|
3.8 (2.8–5.0)
|
Ref
|
|
|
1
|
130
|
22.08
|
5.89 (4.96–6.99)
|
5.5 (4.6–6.5)
|
1.47 (1.04–2.06)
|
|
|
2
|
127
|
17.09
|
7.43 (6.24–8.84)
|
6.7 (5.6–7.9)
|
1.82 (1.30–2.56)
|
|
|
3
|
112
|
11.00
|
10.18 (8.46–12.25)
|
8.9 (7.4–10.6)
|
2.47 (1.74–3.48)
|
|
|
≥4
|
151
|
9.19
|
16.43 (14.01–19.27)
|
13.4 (11.5–15.4)
|
3.85 (2.76–5.38)
|
0.000
|
Abbreviations: CI, confidence interval; HR, hazard rate ratio; PY, person years.
The 1-year cumulative incidence for the groups with no points in the CHA2DS2-VASc score and the ESSEN Stroke Risk score were, 3.4% (95% CI: 2.3–4.9%) and 3.8%
(95% CI: 2.8–5.0%), respectively. The cumulative incidence ranged from 5.3 to 5.8%
for CHA2DS2-VASc scores of 1 to 3 and reached 11.2% (95% CI: 10.0–12.5%) for a CHA2DS2-VASc score of 4 or above. For the ESSEN Stroke Risk score, the cumulative incidence
ranged from 5.5%-8.9% for scores of 1-3 and reached 13.4% (95% CI: 11.5–15.4%) for
a score of 4 or above ([Table 3]). The cumulative incidence for CHA2DS2-VASc scores of 1, 2, and 3 was similar, where the cumulative incidence for the ESSEN
Stroke Risk score increased for each stratum.
For the CHA2DS2-VASc score, the hazard rate ratio for score 1 was 1.56 (95% CI: 0.99–2.44) and 3.48
(95% CI: 2.33–5.20) for the score 4 or above, respectively, with score 0 used as reference.
The hazard rate ratio for the ESSEN Stroke Risk score range between 1.47 (95% CI:
1.04–2.06) for score 1 and 3.85 (95% CI: 2.76–5.38) for score 4 or above. In a Fine
and Gray model conducted to evaluate the score on the risk scale, the effect estimates
were comparable between the two models. Estimates of the Fine and Gray model are summarized
in [Supplementary Table S2].
The predictive abilities of the two risk scores were investigated using C-statistics,
Brier score, and index prediction accuracy score. These were all comparable for both
scores. The C-statistics for the CHA2DS2-VASc score and the ESSEN Stroke Risk score, respectively, were 61% (LB-UB: 58–63%)
and 62% (LB-UB: 59–64%). The Brier score for the CHA2DS2-VASc score was 2.715 (LB-UB: 2.375–3.051), which yielded an index of prediction accuracy
of 0.00. Comparable results were identified for the ESSEN Stroke Risk score with a
Brier score of 2.713 (LB-UB: 2.372–3.044) and an index of prediction accuracy of 0.00
([Table 4]).
Table 4
Predictive performance of stroke in patients with retinal artery occlusion using the
CHA2DS2-VASc score and the ESSEN Stroke Risk score
|
C-statistics % (LB–UB)
|
Brier score % (LB–UB)
|
Index of prediction accuracy score
|
|
Null model
|
|
2.716 (2.375–3.052)
|
|
|
CHA2DS2-VASc
|
61 (58–63)
|
2.715 (2.375–3.051)
|
0.0002 (−0.0009–0.0021)
|
|
ESSEN
|
62 (59–64)
|
2.713 (2.372–3.044)
|
0.0007 (−0.0004–0.0030)
|
Abbreviations: LB, lower bound; UB, upper bound.
Discussion
In this nationwide cohort study of patients with retinal artery occlusion, the risk
of a subsequent stroke increased significantly with increasing points in both the
CHA2DS2-VASc score and the ESSEN Stroke Risk score. However, the discrimination of the risk
assessment models was poor due to the relatively low prevalence of stroke in the population.
The need for specific and effective guidelines in relation to the management of retinal
artery occlusion patients is especially important regarding stroke. Worldwide stroke
has been ranked the second most common cause of death and the third most common cause
of disability. A study of the global disease burden of stroke estimated that the global
absolute number of first-time stroke event in 2010 was 16.9 million, with a total
of 5.9 million stroke-related death the same year.[41] The burden of stroke-related deaths has similarly been identified in patients with
retinal artery occlusion. The mortality rate was investigated in a population of 8,580
participants of older white persons and determined to be higher in patients with retinal
artery occlusion compared with controls, with an all-cause mortality of 56% in retinal
artery occlusion compared to 30% in controls.[42] The hazard rate ratio of stroke-related mortality was 2.0 (95% CI: 1.1–3.8), which
was statistically significant compared to controls.[42] A population of 4,926 participants aged 43–84 years was investigated in both 1999
and 2003. In 1999, it was determined that patients with retinal artery occlusion are
three times more likely to experience a fatal stroke compared to controls during an
8-year follow-up period.[43] In 2003, the same research group described the risk of a mention of stroke on the
death certificate of patients with retinal artery occlusion to be 2.4 times higher
compared with controls without a retinal artery occlusion event during an 11-year
follow-up period.[44]
Based on these elevated risks, the need to identify patients with retinal artery occlusion
at increased risk of stroke is highly relevant. It is important to emphasize that
this study focuses on the stroke risk and whether the included risk scores are able
to select patients with retinal artery occlusion who would benefit from early instigation
of treatment to prevent subsequent stroke. We found that only 5.9% of the included
patients were scanned within 48 hours after their retinal artery occlusion diagnosis
([Table 2]). Even though it is recommended. For all risk strata, irrespective of score, the
risk of stroke is high even within the first month after the occlusion ([Fig. 1]). This could indicate that a subgroup of patients is at acute risk of stroke and
would benefit from an immediate referral for stroke evaluation and prophylaxis treatment.
The physician may need a predictive tool to identify the patients at high risk of
stroke to refer them for immediate stroke evaluation. Thereby, giving the physician
a tool to ensure acute referral for stroke assessment for the more critical patients
following their retinal artery occlusion event and when adequate a subacute referral
for the less critical patients.
Fig. 1 Cumulative incidence of stroke during the 1–year follow-up.
In this study, the CHA2DS2-VASc score was chosen as one of the risk stratification options. Originally, the
score was developed as a stroke risk stratification schema for patients with atrial
fibrillation.[32] The related CHADS2 score and the CHA2DS2-VASc score show comparable clinical utility when predicting stroke. However, the
CHADS2 score categorizes the largest proportions of patients into the intermediate-risk
strata. The CHA2DS2-VASc score on the contrary distributes patients more equally in the different groups,
which enables a more specific evaluation of the individual patient. In addition, the
CHA2DS2-VASc score identified extremely low-risk patients more precisely compared to the
CHADS2 score.[32]
[45]
The second risk stratification option used was the ESSEN Stroke Risk score. The ESSEN
Stroke Risk score was developed in specific populations constituted of stroke patients
in randomized controlled trials and mainly to predict recurrent stroke.[34] Therefore, it would be beneficial to investigate the usages of this risk score in
different populations. Retinal artery occlusion and stroke have been described as
equivalent diseases,[11] making a retinal artery occlusion population an obvious choice to investigate the
predictive properties of the ESSEN Stroke Risk score.
The two risk scores are conceptually similar, with the addition of points corresponding
to the presence of risk factors. Furthermore, both risk scores are recognized as useful
for the prediction of stroke risk in the predefined populations, which suggests that
a comparison of the use of the two risk scores in the retinal artery occlusion population
would be relevant. The tendency observed in the Cox model suggested that these risk
scores could hold the potential to predict stroke in patients with retinal artery
occlusion, supported by the statistically significant likelihood ratio test.
The C-statistics and the individual Brier score for both the risk scores showed acceptable
values for prediction models. The C-statistics for the risk scores included in this
study were similar to comparable prediction models with C-statistics ranging between
57 and 75%.[32]
[46]
[47] However, the index of prediction accuracy showed the prediction performance of the
risk scores were close to zero when compared to the null model representing the population
prevalence of stroke. The C-statistics and the Brier score estimates the accuracy
of a prediction model, whereas the index of prediction accuracy yields a measure considering
the discrimination and calibration adjusted for the null model. In this study, the
Brier score of the null model was undistinguishable from the Brier scores of the risk
stratification models, suggesting that the discrimination of the subgroup of patients
experiencing stroke into risk strata with risks between 3% and 14% was quantitatively
not enough to yield improvement of the Brier score. Combined, the results suggest
that the risk scores stratify patients with clinically relevant increasing risk of
stroke; however, even in the strata with the highest risk, the proportion of patients
who do not develop stroke is high.
The Brier score and the index of prediction accuracy seems contradictory to the rate,
risk, and effect estimated for stroke in patients with retinal artery occlusion, which
is due to the different focus for the analyses. The sensitivity of prediction models
for stroke are essential since stroke is a serious disease, where it is desirable
to exclude false negatives as much as possible. The Brier score does not consider
sensitivity or specificity of a model.[48] Therefore, it is important to consider the clinical aspect when analyzing the data.
The CHA2DS2-VASc score was developed to risk stratify for stroke in patients with atrial fibrillation.
However, atrial fibrillation mainly causes cardiogenic embolic strokes.[49]
[50] Retinal artery occlusion is associated with atherothrombotic embolization originating
from the carotid arteries, and cardiogenic embolization is not a primary risk factor.[5]
[51]
[52] This suggests a difference between the two diseases and may explain why the CHA2DS2-VASc score did not show the same results in patients with retinal artery occlusion
as in patients with atrial fibrillation. To accommodate the need for stroke risk stratification
in patients with retinal artery occlusion, it would be valuable to identify a prediction
model with accurate probabilistic abilities. The probabilistic abilities may be improved
if the risk score was constituted more specifically of atherosclerotic factors, since
retinal artery occlusion is predominantly embolic.[53] Furthermore, ophthalmologic factors could be considered as well, since retinal artery
occlusion have been associated with other ophthalmologic diseases.[52]
Limitations
This study is limited by its observational design and the use of ICD coding. However,
the components of the CHA2DS2-VASc score have been validated in the Danish registries with positive prediction
values of >90%.[54] Errors in the ICD coding result in random variation, which especially is problematic
in smaller populations. To reduce the possibility of random variation influencing
the analyses performed during this study, the highest CHA2DS2-VASc score group was collapsed to ensure larger groups.
Another limitation is the inclusion of a patients using antithrombotic treatment since
treatment will reduce the risk of stroke. This could result in an interpopulation
variation, including individuals with different risks for stroke, making the population
incomparable. However, due to the large proportion of individuals using antithrombotic
treatment, it was assessed that the interpopulation variation would not influence
the results significantly. No guidelines are available for the management of patients
with retinal artery occlusion; however, based on the included data, it was evident
that the majority of these patients use antithrombotic treatment, which may have an
impact on the risk score level compared to an untreated population. Furthermore, patients
with retinal artery occlusion may be managed differently in other countries, which
could influence the transferability of this study.
The population assessed at the index date can acquire comorbidities included in the
score over time, which would change their score and hence their risk of stroke.[55] This dynamic nature of stroke risk is not considered when conducting analyses during
this study.
Conclusion
The risk of stroke was associated with both increased CHA2DS2-VASc score and ESSEN Stroke Risk score. However, the association was stronger between
an increasing ESSEN Stroke Risk score and stroke in patients with retinal artery occlusion
compared to an increasing CHA2DS2-VASc score. The predictive probabilities of both the risk scores did not support
the use of either score for stroke risk stratification in retinal artery occlusion
patients.
What is Known on this Topic?
-
Retinal artery occlusion and stroke have been described as equivalents.
-
Referral to a stroke clinic of patients with retinal artery occlusion to undergo immediate
evaluation of stroke risk is recommended. However, ophthalmologists do not routinely
refer patients with retinal artery occlusion to specialized neurological investigation.
-
There are no available guidelines on which patients with retinal artery occlusion
that should be referred.
What Does this Paper add?
-
Increasing levels of the CHA2DS2-VASc score and the ESSEN Stroke Risk score were associated with increasing risk of
stroke in patients with retinal artery occlusion. The risk of stroke was high even
in strata with no or few risk factors.
-
The ESSEN Stroke Risk score was found qualitatively superior for risk stratification
compared to the CHA2DS2-VASc score.
