In our study of 1,268 adults with type 1 diabetes and ETDRS grading of diabetic retinopathy severity, we show that non-proliferative retinopathy increases the risk of stroke. Interestingly, this main observation emerged regardless of the presence of diabetic kidney disease. Furthermore, the risk associated with non-proliferative retinopathy was of the same magnitude as that of PDR. We were also able to show that the incidence of both ischemic and hemorrhagic stroke increased with the severity of diabetic retinopathy. And, as another novelty, we observed a higher incidence of stroke in participants with CSME.

Our findings are in line with previous data from the FinnDiane Study, reporting that severe diabetic retinopathy, indicated by a cruder evaluation, i.e., retinal photocoagulation, increases the risk of stroke in type 1 diabetes [13]. We now, for the first time, show that this holds true for a broader spectrum of diabetic retinopathy determined from fundus images. Furthermore, in the present study, the association was independent of diabetic kidney disease, which is somewhat surprising as diabetic kidney disease is currently one the strongest known risk factors for stroke in type 1 diabetes [3, 6].

Outside the FinnDiane Study, there is relatively limited research evaluating stroke in individuals with type 1 diabetes [2] and previous studies assessing retinopathy and stroke have shown inconclusive results [23], although retinopathy severity seems to be a risk marker for stroke in type 2 diabetes [7, 23]. The Action to Control Cardiovascular Risk in Diabetes (ACCORD) Study [7] reported that both mild and moderate to severe retinopathy are associated with stroke in type 2 diabetes and that the stroke risk increases by retinopathy severity. Their cut-off values for retinopathy severity grouping differed slightly from our study, and, furthermore, the ACCORD study did not mention diabetic kidney disease in their report. Nonetheless, we obtained similar effect sizes and, as the ACCORD Study [7], we also found the HR for stroke to increase by retinopathy severity in a Cox model not including diabetic kidney disease. In our Cox model including kidney disease, however, the HR for stroke was similar in non-proliferative retinopathy and PDR.

In our Cox model adjusting for diabetic kidney disease, however, non-proliferative retinopathy was as strongly associated with stroke as PDR. It is possible that non-proliferative retinopathy is an earlier marker of stroke in type 1 diabetes, compared to PDR, and thus might have a larger impact on the cause-specific hazard when the presence of other risk factors is low. In individuals displaying several risk factors for stroke, such as in those with PDR who also had a high prevalence of diabetic kidney disease, the impact of diabetic retinopathy might become lesser. Regardless, the significant effect of diabetic retinopathy on stroke risk was observed in both non-proliferative retinopathy and PDR in the Cox model.

In the cumulative incidence function analyzing the competing risk of stroke and death, the probability of death was higher than the probability of stroke in both non-proliferative retinopathy and PDR. Interestingly, whereas PDR was associated with death in the Fine-Gray model, non-proliferative retinopathy was not. The subdistribution hazards for stroke in the Fine-Gray model were largely unchanged compared to the Cox model, except that the association between stroke and PDR weakened slightly, so that it was non-significant, but borderline, for PDR.

We further assessed ischemic and hemorrhagic stroke, as well as subtypes of ischemic stroke, as separate outcomes in marginal analysis. For ischemic strokes, the cumulative incidence in non-proliferative retinopathy resembled that of PDR. This was particularly prominent for lacunar strokes. Moreover, in the Cox analysis not adjusted for diabetic kidney disease, the HRs in non-proliferative retinopathy and PDR were similar.

In our study, the ETDRS score indicates the mildest possible diabetic retinopathy for a participant during follow-up, as we have not assessed the progression of diabetic retinopathy. Similarities between the groups could be due to the progression of retinopathy, which the Cox model does not show. Whether or not progression was present to some degree, having non-proliferative retinopathy, i.e., more than only microaneurysms, at baseline was significantly associated with an increased risk of ischemic and lacunar stroke when diabetic kidney disease was not regarded. Non-lacunar stroke, on the other hand, was not associated with diabetic retinopathy in a corresponding model.

In the general population, lacunar strokes have been associated with retinal vascular changes in a cross-sectional setting, and it has been contemplated that they share etiological features [24]. Follow-up data on lacunar strokes and retinal microvasculature are sparse in both the general population and type 1 diabetes and although our current observations, as well as our previous report, suggest that lacunar strokes and diabetic retinopathy are linked, this relationship, particularly the hypothesized common etiology, warrants further exploration.

When analyzing hemorrhagic stroke, the cumulative incidence increased stepwise with increasing retinopathy severity. To our surprise, we did not observe a significant association between hemorrhagic stroke and non-proliferative retinopathy. This was somewhat unexpected, as we have previously reported that asymptomatic cerebral microbleeds and moderate to severe non-proliferative retinopathy are associated [12]. PDR, on the other hand, increased the HR for hemorrhagic stroke in all models in this current study, which is in line with our previous observations [6]. PDR and diabetic kidney disease might largely represent the same participants in our current cohort, in which case the impact of milder retinopathy alone may be insignificant in comparison. Considering that we have previously reported that severe diabetic retinopathy and diabetic kidney disease combined increase the risk of hemorrhagic stroke more than either complication alone [13], this might be one explanation.

The cumulative incidence of stroke was increased in participants with CSME in similarity with diabetic retinopathy. By 20 years of follow-up, the cumulative incidence had reached 21% in participants with CSME, compared to 15% in non-proliferative retinopathy and 20% in PDR. In contrast to diabetic retinopathy, CSME did not increase the risk of stroke in a Cox model adjusted for diabetic kidney disease. In analysis of ischemic stroke, the cumulative incidence at 20 years was 17% in CSME, compared to 15% in PDR, and both variables were associated with an increased HR when diabetic kidney disease was not taken into account. For hemorrhagic stroke, the cumulative incidence was 6% in participants with CSME and 8% in PDR, but in contrast to PDR, CSME was not a risk factor for hemorrhagic stroke when kidney disease was regarded.

Possibly, we were unable to detect an existing association, as macular data was available only for a subset (85%) of our participants. The prevalence of any stroke, or stroke type, did not, however, differ between those with vs. without information on macular status. Although macular edema and severe diabetic retinopathy often coincide, their risk factors also differ somewhat, e.g., whereas high systolic blood pressure increases the risk of CSME, PDR develops independent of systolic blood pressure [25]. It should therefore not be ruled out that macular edema and diabetic retinopathy represent different risk profiles. In type 2 diabetes, cardiovascular disease seems more strongly linked to macular edema than to PDR, particularly when it comes to fatal coronary heart events. Yet, the link between macular edema and stroke is unknown in type 2 diabetes as well [8].

In this study, we were able to thoroughly classify all incident strokes from medical records. For those without stroke, we did not confirm stroke-free status further, and some might have asymptomatic cerebrovascular disease, such as silent lacunar strokes or a history of transient ischemic attacks. We have, however, excluded all individuals for whom the clinical and radiological observations were not in agreement, the date of stroke or stroke type could not be determined, or when information regarding stroke was otherwise insufficient.

While we consider our well characterized study cohort a strength of this study, it also comes with limitations. As the study cohort is derived from a Finnish population of adults with type 1 diabetes, our results may not be generalizable to other geographical and ethnical contexts, or to other age groups.

Furthermore, although diabetic retinopathy is well characterized in our participants and grading has been done by an experienced ophthalmologist, our data are limited to information available from fundus images and medical records and do not include optical coherence tomography (OCT), for example. More detailed fenotyping based on OCT or OCT Angiography might have provided us with additional information for further, or more precise, categorization of the participants. Regardless, as screening by fundus imaging is the gold standard for detecting diabetic retinopathy to date [26], diabetic retinopathy evaluation from retinal images can still be considered highly relevant. Retinal images could, thus, serve as an accessible tool for the assessment of cerebrovascular health.