Risks of stroke and myocardial infarction after retinal artery occlusion and their time dependence: a systematic review and meta-analysis

Abstract

Background

Retinal artery occlusion (RAO) is an ophthalmic emergency that may signal underlying cardiovascular disease. Both stroke and myocardial infarction (MI) share pathophysiological mechanisms and risk factors with RAO. This systematic review aimed to assess the risks of stroke and MI following the onset of RAO, with a focus on their time-dependent changes.

Methods

PubMed, Embase, and Web of Science were searched for articles reporting  incidence of stroke and MI after RAO. Newcastle–Ottawa Scale was employed to evaluate the quality of study. Pooled relative risks for stroke and MI risks following RAO were calculated by random-effects models, and the time-dependent change of these risks was analyzed.

Results

Twelve studies involving 319 748 people were included. Pooled analysis showed that RAO was associated with a significantly increased risk of stroke (RR = 3.64, 95% CI: 2.53–5.25, p < 0.0001), particularly within the first month after RAO. MI risk showed an upward trend (RR = 1.38, 95% CI: 0.80–2.40, p = 0.2478), but did not reach statistical significance. Time-stratified analyses revealed the highest incidence of stroke and MI within 30 days post-RAO.

Conclusion

RAO is strongly associated with an elevated short-term risk of stroke, underscoring the need for prompt cerebrovascular evaluation and monitoring within the first month. Although the association between RAO and MI was less conclusive, cardiovascular assessment remains warranted. These findings highlight RAO as an important marker of systemic vascular disease requiring timely intervention.

Supplementary Information

The online version contains supplementary material available at 10.1007/s00415-025-13445-x.

Introduction

Cardiovascular diseases (CVD) now account for approximately one-third of all global deaths [1]. The global mortality rate due to CVD has increased by 12.5% over the past decade and the economic burden of CVD on individuals and society is progressively increasing, positioning it as a major public health issue [2, 3]. According to the World Health Organization, stroke and myocardial infarction (MI) are responsible for over 45% of cardiovascular disease-related fatalities [4].

Stroke and MI are types of cardiovascular diseases caused by the sudden obstruction of arteries due to atherosclerosis or thrombosis, leading to irreversible cell death and tissue injury [5, 6]. Recent treatments, including stent implantation, intravenous thrombolysis, and endovascular thrombolysis, have yielded favorable outcomes in stroke and MI patients. However, current treatments still face significant temporal and spatial limitations. Thrombolytic therapy for stroke patients must be administered within a 4.5-h window, while the optimal time window for MI patients is only 120 min [7, 8]. Studies indicate that only 9.2–29.3% of patients receive effective treatment within the prescribed time window [9]. Therefore, the prompt identification of individuals at risk for stroke or MI, coupled with timely intervention, may offer significant clinical benefits.

Retinal artery occlusion (RAO) is an ophthalmic emergency caused by exogenous embolism, endarteritis, or atherosclerosis. RAO can lead to a significant decline in visual acuity, visual field loss, or even blindness, severely affecting the life quality of patients [10]. RAO and stroke share similar pathophysiology, involving arterial obstruction caused by thromboembolism, atherosclerosis, and vasospasm [11]. So, RAO is considered a “retinal stroke” [12]. The European Group for the Evaluation of Ocular Disease Regression study suggests that the cardiovascular risk factors contributing to the increased incidence of RAO include obesity, hypertension, hypercholesterolemia, and diabetes, among others. Overall, at least one cardiovascular risk factor is identified in 67% of patients [13]. Recent recommendations from the American Heart Association and the American Stroke Association emphasize the rapid referral of individuals with acute RAO to the nearest stroke center for urgent treatment [14]. Recent studies suggest that fundus microvascular abnormalities may serve as significant biomarkers for systemic vascular diseases, with RAO potentially acting as an early indicator of CVD. Its association with stroke and MI has emerged as a clinical concern [9, 15].

This study systematically synthesizes relevant literature on the risks of stroke and MI following RAO, aiming to examine the correlation between RAO and flowing CVDs and evaluate their temporal trends in these risks post-first RAO.

Method

Search strategy

Two researchers conducted searches on PubMed, Embase, and Web of Science using the expanded medical subject headings “retinal artery occlusion,” “central retinal artery occlusion,” “branch retinal artery occlusion,” “stroke,” and “myocardial infarction,” while also manually reviewing the reference lists of selected articles (see Supplemental Material for the detailed search strategy). The search strategy was confined to human studies and literature published in English. The research was conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines, and the study protocol was pre-registered in PROSPERO (CRD420251007733).

Inclusion and exclusion criteria

The inclusion criteria for studies were as follows: (1) case–control or cohort studies, (2) published in English language, (3) included patients who developed a stroke or MI following RAO, (3) incidence of stroke or MI post-RAO was recorded, (4) the definite time of follow-up duration was documented. Case reports, commentaries, review articles, and conference abstracts were not considered. Studies were excluded if the follow-up duration was incompatible with that of other studies or the statistical results were presented in person-years. Articles with fewer than five patients in the outcome group were also excluded.

Article screening and data extraction

Following the removal of duplicates, titles, abstracts, and full texts was screened by X.C. and R.Z. The resulting list of studies was reviewed by a third reviewer (C.W.), with any disagreements regarding study eligibility being reconciled in consultation. Data from each eligible report study were independently extracted by two authors (C.W. and X.C.) using a predefined extraction template, with any discrepancies resolved through discussion with all authors. The extracted data included—(1) study details: title, authors, year, sample size, and follow-up duration; (2) patient baseline characteristics: age and gender; (3) outcome measures: stroke and MI events.

Article quality assessment

The Newcastle–Ottawa Scale was employed to evaluate the quality of each included study, with scores of 1–3, 4–6, and 7–9 categorized as low, moderate, and high quality, respectively. Studies featuring a meta-analysis generally required a score of ≥ 5. The assessment was conducted independently by two researchers (C.W. and X.C.), with any discrepancies resolved through third-party mediation.

Statistical analysis

Data analysis was conducted following the procedures outlined previously. R 4.3.0 software was used for statistical analysis and graphing, with the R package meta employed to generate forest plots and publication bias analysis. A significance threshold of 0.05 was established for statistical testing, and p values were used to assess heterogeneity. To evaluate the risk of stroke and MI following RAO, pooled relative risks for stroke and MI were calculated for patients with and without RAO using fixed effects analysis, unless heterogeneity was detected (p < 0.05), in which case random-effects analysis was applied. When substantial differences exist among the included studies in terms of population characteristics, study design, or research background, a random-effects model should be applied. A one-by-one exclusion method was used to perform sensitivity analysis, while publication bias was assessed using both Begg’s test and Egger’s test.

Results

Results of the literature search

The electronic literature search yielded 5 236 references (including 1 629 duplicates), of which 12 publications were deemed appropriate for inclusion (Fig. 1). Following removal of duplicates, the titles and abstracts of 3 607 publications were screened. The full texts of 113 articles were reviewed, with 12 studies meeting the inclusion criteria. Non-eligible studies encompassed replies, comments, reviews, case reports, papers lacking comparisons, studies irrelevant to the issue, studies having inaccessible full texts, and studies from which data could not be extracted.

Fig. 1.

Selection of articles for this systematic review. Preferred Reporting Items for Systematic Reviews and Meta-Analysis flowchart of study selection

Basic characteristics and quality assessment of the literature

Twelve selected papers were retrospective cohort studies and case-control study, primarily originating from the United States (n = 4), Korea (n = 3), China (n = 2), Poland (n = 1), and Hungary (n = 1). In addition, one study adopted the data from the TriNetX research network. (Table 1). These studies collectively reported data on a total of 73,624 RAO patients with either stroke or MI. Sample sizes ranged from 140 to 191 377 (median, 2 404 patients). The average age of RAO patients, reported in ten studies, ranged from 49.72 to 82.75 years. Among 12 studies reviewed in full text, 9 studies investigated the difference of stroke risk between RAO patients and controls, with 7 studies examining the risk at various intervals following RAO. Five studies assessed MI risk following RAO and non-RAO controls, with three studies analyzing MI risk at different intervals post-RAO. Stroke was more common compared with MI. Overall, 5792 (7.87%) patients had a stroke, while 1979 (2.69%) patients had a MI. The total follow-up duration ranged from 1 to 12 years post-RAO.

Table 1.

The fundamental characteristics of the literature included

Study	Year	Country	Study type	Follow-up (years)	Sample size	Sex (female, n, %)	Age (mean ± sd/range)
							RAO	Control
Bruno et al. [16]	1995	USA	Retrospective cohort study	3.4	140	0	69	68.6
Chang et al. [17]	2012	Taiwan, China	Retrospective cohort study	3	3 248	1 393, 42.89%	60.14 ± 14.71	60.16 ± 14.66
Park et al. [18]	2015	Korea	Case–control study	1	1 585	653, 41.20%	61.60 ± 15.00	–
Rim et al. [19]	2016	Korea	Retrospective cohort study	12	2 404	1 096, 45.59%	–	–
Hong et al. [20]	2017	Korea	Retrospective cohort study	1	151	49, 32.50%	60.80 ± 15.30	–
French et al. [21]	2018	USA	Retrospective cohort study	1	191 337	135 636, 70.89%	–	–
Xiao et al. [22]	2020	China	Case–control study	5	270	39, 14.40%	59.98 ± 10.26	60.01 ± 10.14
Scoles et al. [23]	2021	USA	Retrospective cohort study	1	36 426	18 333, 50.30%	74.17 ± 8.58	74.24 ± 9.31
Suri et al. [24]	2022	USA	Retrospective cohort study	1	14,527	6 995, 48.00%	69.00 ± 13.00	–
Walek et al. [25]	2022	Poland	Retrospective cohort study	12	278	102, 36.69%	70.09 ± 11.13	70.17 ± 11.49
Wai et al. [26]	2023	TriNetX	Retrospective cohort study	10	69,044	17 728, 51.00%	66.00 ± 15.20	66.00 ± 15.20
Balla et al.[27]	2023	Hungary	Retrospective cohort study	10	338	144, 42.60%	64.18 ± 10.00	63.88 ± 10.43

RAO patients		Control patients		Effect size: HR, RR, IRR, or OR (95% CI)	Comments
Size	Even t(n, %)	Size	Event
70	Stroke: 17, 8.5% MI: 6, 2.7%	70	Stroke: 2, 0.80% MI: 7, 2.9%	Stroke: ARR 9.90(2.30–43.10) MI: ARR 1.40(0.70–2.90)	Positive association
464	Stroke: 91, 19.61%	2784	Stroke: 280, 10.05%	Stroke: HR 2.07(1.64–2.63)	Positive association
1585	Stroke: 152, 9.59% MI: 15, 0.90%	–	–	–	–
401	Stroke: 60, 15.00%	2003	Stroke: 160, 8.00%	HR 1.78 ( 1.32–2.41)	Positive association
151	Stroke: 13,  8.6%	–	–	–	–
3338	Stroke: 141,  4.22%	187,999	Stroke: 1394,  0.74%	Stroke: IRR 33.10(9.80–84.60)	Positive association
45	Stroke: 17, 37.80% MI: 3, 6.7%	225	Stroke: 1, 0.40% MI: 3, 1.3%	Stroke: OR 23.43(12.75–43.04)	Positive association
18,213	Stroke: 1807, 9.90%	18,213	Stroke: 606, 3.30%	Stroke: HR 2.97(2.71–3.26)	Positive association
14,527	Stroke: 1 577, 10.90% MI: 615, 4.20%	–	–	–	–
139	IS: 14, 10.07% MI: 3, 2.16%	139	IS: 7, 5.04% MI: 2, 1.44%	–	Positive association
34,522	Stroke: 1841, 14.59% MI: 1331, 10.55%	34,522	Stroke: 1160, 9.18% MI: 1190, 9.43%	Stroke: RR 5.20(4.67–5.79) MI: RR 1.72(1.51–1.97)	Positive association
169	IS: 12, 7.1% MI: 6, 3.55%	169	IS: 0, 0% MI: 0, 0%	IS: OR 8.18(3.09–21.64) MI: OR 3.10(1.36–7.08)	Positive association

RAO retinal artery occlusion, MI myocardial infarction, IS ischemic stroke, HR hazard ratio, RR relative risk, IRR incidence rate ratio, OR odds ratio

Quality assessment

Eight studies were evaluated as high quality, while four studies were evaluated as moderate quality (Table S1). All studies assessed the incidence of stroke or MI using standard and reliable methods. Only nine studies included a control group to compare the incidence of stroke or MI between RAO patients and non-RAO populations. Furthermore, five studies had short follow-up durations, with none exceeding 1 year.

Data analysis

Comparison of the incidence of stroke and MI after RAO

A total of 57,361 patients were included in the RAO group. Among them, 4,192 patients had complications after the onset of the disease, with an pooled incidence rate of 12.84%, 95% CI (7.44–19.42%). There were 246,088 cases in the control group, among which 2,840 patients developed stroke, and the incidence of stroke was 2.72%, 95% CI (0.92–5.37%). The studies exhibited significant heterogeneity (p < 0.0001), necessitating the use of a random-effects model. The result showed that the incidence of following stroke was significantly higher in the RAO group [RR = 3.64, 95% CI (2.53, 5.25)] (Fig. 2A).

Fig. 2.

Forest plots of the incidence of stroke and MI after RAO. A Comparison of the incidence of stroke after RAO; B comparison of the incidence of MI after RAO

A total of 13,046 patients were included in the RAO group, among whom 1350 patients developed MI after the onset of the disease, with an incidence rate of 5.96%, 95% CI (2.84–10.01%). The control group included 13,226 patients, among whom 1,202 cases developed MI, with an incidence rate of 3.16%, 95% CI (0.30–8.37%). Heterogeneity between studies was substantial (p = 0.1554), and a random-effects model was applied bacause of the substantial differences in study populations. The results showed that there was no statistically significant difference in the incidence of MI after the onset between the RAO group and the control group, but there was a clear trend [RR = 1.38, 95% CI (0.80, 2.40)] (Fig. 2B).

Sensitivity analysis

Sensitivity analyses were performed using a one-by-one exclusion procedure. After sequentially removing individual studies, the effect sizes of the remaining studies were summed to observe changes in the total effect size. The results show that the conclusions remained stable after the exclusion of any individual study (Fig. 3A). The results were not significantly affected by the removal of any original studies, suggesting that the findings of the meta-analysis can be considered relatively reliable.

Fig. 3.

Sensitivity analysis of random-effects model for stroke and MI. A Sensitivity analysis of random-effects model for stroke; B sensitivity analysis of random-effects model for MI

The sensitivity analysis results for post-RAO MI studies are shown in Fig. 3B, demonstrating that the combined effect size decreased from [RR = 1.38, 95% CI (0.80, 2.40)] to [RR = 1.12, 95% CI (1.04, 1.21)] following the exclusion of the Xiao et al. 2020  or the Balla et al. 2023 study. Conversely, exclusion of the Wai et al. 2023 study resulted in an increase from [RR = 1.38, 95% CI (0.80, 2.40)] to [RR = 2.16, 95% CI (0.71, 6.51)], indicating a substantial influence of these two studies on the outcome.

Publication bias analysis

In studies examining stroke risk after RAO, we performed Begg’s test, which yielded Z = 0.63, p = 0.5361, and Egger’s test, which revealed t = 0.11, p = 0.9163, suggesting the absence of significant publication bias. Similarly, for post-RAO MI studies, Begg’s test showed no significant bias in which Z = 1.47, p = 0.1416, and Egger’s test also indicated no significant bias which yielded t = 1.53, p = 0.2245. These results indicate that the reliability of the conclusions drawn from this meta-analysis is unlikely to be affected by publication bias.

A contour-enhanced funnel plot was generated using the shear-and-complement method to assess publication bias. Figure S1A illustrates publication bias in studies examining stroke risk after RAO, while Fig. S1B depicts publication bias in studies on MI. The asymmetry observed in the funnel plot was corrected by including two dummy studies. One of the ancillary studies was located in the non-significant zone, suggesting the potential existence of unpublished, non-statistically significant studies. This highlights the possibility of publication bias in this analysis.

Incidence of stroke at different intervals following RAO

The occurrence of stroke at various intervals following RAO is detailed in Figs. 4A, S2. The risk of stroke was highest within 30 days of RAO, with an incidence of 3.56% (95% CI 1.60–6.25%). This risk gradually decreased, with an incidence of 0.48% (95% CI 0.30–0.71%) between 31–90 days and 0.36% (95% CI 0.04–0.96%) between 91–180 days.

Fig. 4.

Time-dependent change of disease risk following RAO. A Time-dependent change of stroke risk following RAO; B time-dependent change of MI risk following RAO

The incidence of MI at various intervals following RAO is shown in Figs. 4B, S3. Similar to stroke risk, the risk of MI was higher within 30 days following RAO, with an incidence of 0.88% (95% CI 0.00–3.90%). This risk gradually decreased, with an incidence of 0.09% (95% CI 0.05–0.15%) between 31–90 days and 0.17% (95% CI 0.11%-0.24%) between 91–180 days.

Discussion

The primary aim of this study was to investigate the risk of stroke and MI in patients with RAO and its time-dependent changes. Our results demonstrated that the stroke risk in RAO patients was significantly elevated compared to controls, with RR of 3.64. The incidence of MI in RAO patients after the onset of the disease tended to be higher than that in the control group, but there was no statistically significant difference. Our results also showed that the risk of stroke was greatest within the first month after the onset of RAO, with an increased trend of risk of MI observed within the first month following RAO onset. Collectively, these results underscore the clinical significance of prompt cardiovascular and cerebrovascular evaluation for patients with RAO.

Research on cerebral small-vessel lesions, such as transient ischemic attack (TIA), suggests that the occurrence of small-vessel infarctions, like TIA, may significantly influence stroke risk at an early stage [28, 29]. Therefore, RAO, as a small-vessel lesion, may provide valuable insights into CVD risk evaluation. Taken together, our data suggest that prompt neurovascular evaluation of patients with RAO may have important clinical implications. However, it should be noted that our findings of a significantly elevated stroke risk following RAO must be interpreted within the existing literature on ischemic events. Research on cerebral small-vessel lesions suggests that the occurrence of small-vessel infarctions, like TIA and RAO, may indicate a significantly increased risk of stroke. A clear consensus indicates that the risk of stroke following RAO is generally lower than TIA. The risk of stroke after TIA has been reported to reach 5.2% within 7 days and to range between 4.03% and 39.2% within 90 days [30, 31]. Medium- to long-term follow-up indicates that the 1-year risk of stroke after TIA ranges from 14.5% to 21.4%, with the 10-year risk approaching 20% [32, 33]. The main reason may lie in the differing etiologies of the two conditions. RAO is most often caused by small emboli originating from fragmented carotid atherosclerotic plaques that obstruct the retinal artery, whereas TIA is commonly attributed to large-artery atherosclerosis, cardioembolic sources, or small-vessel disease. The etiology of TIA largely overlaps with that of ischemia stroke, making it a stronger predictor of subsequent stroke [34]. Nevertheless, the comparatively lower stroke risk following RAO should not be regarded as benign. The overall risk of major vascular events (stroke, MI, and vascular death) remains substantially elevated compared to the general population.

French et al. [21] conducted a follow-up study on 3 338 patients with RAO using the U.S. National Health Insurance database, revealing that the incidence of stroke in RAO patients was 33.1 times higher than in the control group (95% CI 9.8–84.6), with the highest stroke risk occurring within 2 weeks of onset. Rim et al. [19] found that the risk of ischemic stroke was 1.78 times higher than that in the control group post-RAO (95% CI 1.32–2.41). Both studies demonstrate an elevated stroke risk within 2 weeks following RAO, consistent with our study’s conclusion that the stroke risk is most pronounced during the first month post-RAO. Because the time scales selected for statistics are different, our scope is broader. Therefore, it is crucial to regularly assess the cerebrovascular condition of patients within the first month following RAO to ensure timely intervention.

Regarding the risk of MI after the onset of RAO, there are contradictory research conclusions. Bruno et al. [16] found no statistically significant difference in the risk of MI after RAO (RR = 1.40, 95% CI 0.70–2.90). However, Wai et al. [26] reported a significantly higher risk of MI at 1 year, with a RR of 1.72 (95% CI 1.51-1.97). This discrepancy may stem from variations in temporal stratification. In this study, the relative risk of MI was 3.00 (95% CI 1.79–5.04) at 2 weeks post-RAO in a subgroup analysis, while the risk decreased to 1.72 (95% CI 1.57–1.97) at 1 year. This suggests that short-term risk may be obscured in long-term studies lacking sufficient temporal stratification. Park et al. [18].analyzed 1,655 patients with central retinal artery occlusion using a self-controlled case-series design, finding that the risks of hemorrhagic stroke and MI had no temporal correlation with the occurrence of RAO. However, our study identified a non-significant trend suggesting an increased risk of MI at 30 days post-RAO, but this finding contrasts with prior studies. Unfortunately, we were unable to include their data in our study due to discrepancies in statistical methodologies. The pathogenesis of RAO differs in part from that of MI. Although both conditions share common risk factors such as hypertension and hyperlipidemia, RAO is primarily caused by emboli originating from unstable atherosclerotic plaque rupture, whereas MI more often results from the chronic progression of coronary atherosclerosis [35]. This distinction may help explain the discrepancy between the significantly increased risk of stroke and the relatively insignificant risk of MI observed in RAO patients. Stroke is mainly attributable to large-artery atherosclerosis, cardioembolic sources, or small-vessel disease, which overlap considerably with the mechanisms underlying RAO [36]. Nevertheless, cardiovascular basic assessment of RAO patients remains essential.

Given that RAO is a rare condition, the use of national-level data may help mitigate selection bias arising from hierarchical disparities among healthcare institutions. A analysis by Park et al [18], using the Korean National Health Database, revealed a 21.5-fold increase in the 1-month risk of stroke in patients with RAO compared to controls. In contrast, a study by Chang et al. [17] did not identify a statistically significant association. This discrepancy may be attributed to differences in the case ascertainment methods; Park et al. validated diagnoses through ICD coding, while Chang et al. relied on the diagnosis made in the ambulatory and emergency room. Our clinical experience has revealed that the initial diagnoses in the emergency department may occasionally be inaccurate, leading to potential biases. Laczynski et al. [37] suggested large national database studies may be prone to issues such as misdiagnosis through ICD coding, resulting in the misclassification of some non-stroke patients. They initially screened patients with comprehensive ophthalmological examination and fluorescein angiographic verification and included 221 patients (5 with stroke, resulting in an overall incidence of 2.3%), with 4 patients presenting with RAO concomitant with stroke and 1 patient having a stroke following RAO. The authors concluded that the stroke risk in RAO patients was similar to that of the general population. It is important to note that this study was conducted in a university hospital, and the patient cohort may have experienced selection bias. Consequently, this data was excluded from the analysis due to discrepancies in statistical methodologies.

The American Academy of Ophthalmology guidelines for managing RAO emphasize that RAO is an indicator of potential future ischemic events and requires prompt evaluation and referral to a stroke center [14]. This guideline provides new evidence on post-RAO evaluation, highlighting the importance of stroke surveillance within the first month. The study recommends the following measures: (1) undergoing magnetic resonance imaging examinations during hospitalization; (2) paying attention to the possible symptoms of CVDs in patients after discharge. In the presence of cardiovascular risk factors such as diabetes and smoking, antiplatelet and anticoagulant therapy may be needed [38]. Furthermore, early MI detection can be achieved through electrocardiogram monitoring coupled with cardiac biomarker assessments [10].

This study faces a few limitations. First, some studies were excluded due to discrepancies in inclusion and exclusion criteria, as well as differences in statistical methodologies. Moreover, the sample sizes of the studies involving RAO patients and controls varied significantly (ranging from 31 to several thousand patients), and there were differences in confounding factors, including ethnicity, age, and lifestyle. Furthermore, due to the limitations in statistical methods, more precise temporal stratification was difficult, highlighting the need for further research to explore the temporal dynamics underlying this relationship.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Chuansen Wang and Xueying Chen were responsible for conducting the search, screening potentially eligible studies, writing the report, extracting and analyzing data, interpreting results. Ying Li was responsible for supervision. Runlang Zhu was responsible for extracting and analyzing data. Jiaqing Feng was responsible for visualization. Duan Chen was responsible for extracting and analyzing data. Ting Chen was responsible for designing the review protocol. Xuan Xiao was responsible for designing the review protocol and funding acquisition. Chuansen Wang and Xueying Chen contributed equally as co-first authors. Xuan Xiao and Ting Chen contributed equally as co-last authors.

Funding

This work was supported by the National Key R&D Program of China (2023YFC2308404), the Health Commission of Hubei Province Scientific Research Project (HBJG-250010), and the Technology Innovation Team Project of Hubei Province (220171677).

Data availability

The data and analysis code generated in this article can be obtained by contacting the author.

Declarations

Conflicts of interest

The authors declare that they have no competing interests.