The role of blood pressure in risk of ischemic and hemorrhagic stroke in type 1 diabetes

We defined type 1 diabetes as diabetes diagnosis before 40 years of age and insulin medication commenced within 1 year after diagnosis. The mean age at baseline was 37.4 ± (standard deviation) 11.9 years, the median duration of diabetes was 20.9 (interquartile range 11.5–30.4) years, and 52% of the participants were men. Each participant collected timed urine samples for the measurement of urinary albumin excretion rate (UAER). Diabetic nephropathy (DN) was defined as having a UAER of ≥ 200 µg/min or ≥ 300 mg/24 h or having end-stage renal disease (ESRD). ESRD was defined as ongoing dialysis treatment or kidney transplantation. Severe diabetic retinopathy (SDR) was defined as history of retinal photocoagulation. Coronary heart disease (CHD) was defined as a history of myocardial infarction or coronary artery revascularization, or treatment with long-acting nitroglycerin.

Blood pressure was measured by a trained nurse during the exam visit twice in the sitting position with a 10 min rest before the first measurement, and the mean values of these two measurements were calculated for both the systolic blood pressure (SBP) and the diastolic blood pressure (DBP). Pulse pressure (PP) was calculated as SBP–DBP. Mean arterial pressure (MAP) was calculated as \({\text{DBP}} + {\raise0.7ex\hbox{$1$} \!\mathord{\left/ {\vphantom {1 3}}\right.\kern-0pt} \!\lower0.7ex\hbox{$3$}}{\text{PP}}\). Antihypertensive medication was defined as the use of any anti-hypertensive agent; angiotensin-converting-enzyme inhibitors, angiotensin receptor blockers, calcium channel blockers, β-blockers, diuretics, or other antihypertensive agents (mainly moxonidine). Hypertension was defined as SBP > 140 mmHg and/or DBP > 90 mmHg. 24-h Na and 24-h urinary potassium (24-h K) excretion were measured from a 24-h urine collection. Sodium/potassium ratio (Na/K ratio) was defined as 24-h Na divided by 24-h K. Information on sodium and potassium urinary excretion was available for 2402 (59%) individuals during follow-up.

A detailed description of the identification and classification of the strokes has previously been described in detail [13]. For this study, we had additional follow-up data available until the end of 2012. Follow-up data was available in all participants. In short, individuals with an incident stroke were identified from the FinnDiane questionnaires, death certificates, and the National Care Register of Health Care. Medical records from the attending hospitals, brain computed tomography and magnetic resonance images, as well as death certificates were ordered on these individuals. Based on these data, all the strokes were classified into ischemic or hemorrhagic stroke by two stroke neurologist (J.P and R.L), with the help from an experienced neuroradiologist when needed. Follow-up time was calculated from the baseline visit until the last date the participants were known to be free of stroke, or until the date of the first stroke for those who suffered a stroke, resulting in 45,050 person-years of follow-up.

All variables included were tested for normal distribution by visual inspection. Parametric continuous variables were analyzed with Student’s t-test, and the results are presented as means with standard deviation. Non-parametric variables were analyzed with Mann–Whitney U-test, the results are presented as medians with interquartile range. The difference in categorical variables between groups was tested with the χ²-test. Cox proportional hazard models were performed to study how SBP, DBP, MAP, and PP affected the risk of any stroke, ischemic stroke, and hemorrhagic stroke. The results are presented as hazard ratio (HR) with 95% confidence interval (CI). Blood pressure variables were included in different models because of collinearity, with only one blood pressure variable in each model. The multivariate models were adjusted for the variables that statistically differed in the univariate analyses (Additional file 2: Table S2), which resulted in adjustments for sex, diabetes duration, waist circumference, DN, SDR, HbA_1c, LDL-cholesterol, current or history of smoking, use of antihypertensive medication, presence of CHD, and year of the FinnDiane study visit.

Since participant mortality due to other causes than stroke is a competing event to stroke, we performed competing risk analyses to take this into account. We examined the association between SBP and stroke using the proportional subdistribution hazards regression model by Fine and Gray [14], with death as a competing event. The results are presented as subhazard ratio (SHR) with 95% CI, and is defined as the hazard of a given cause in the presence of the competing event.

To assess the shapes of the association between the continuous variables SBP, DBP, MAP, and PP and the risk of stroke, we fitted unadjusted Cox proportional hazard models with each variable included as a restricted cubic spline with three knots. The number of knots was selected based on the Akaike information criteria [15], and located at the 10th, 50th, and 90th percentiles, corresponding to the values listed in Additional file 1: Table S1. The median values were used as the reference point in the analyses. HR was then estimated based on the restricted cubic spline models and used to create plots for graphical assessment of the relationships. We implemented the Wald test to determine whether the relationship between the blood pressure variables and stroke significantly deviated from linearity. P < 0.05 was considered statistically significant. The restricted cubic spline analyses were performed with Harrell’s Regression Modelling Strategies (rms) package [16], while the competing risk analyses were performed with the ‘cmprsk’ package [17] in the R software [18]. All other analyses were performed with the SPSS Statistical software 25.0 (IBM Corporation, Armonk, NY).

As in the general population [19, 20], SBP seems to be a very strong risk factor for stroke also in type 1 diabetes. SBP remained significant as a risk factor for total stroke and ischemic stroke when compared with the other variables in the Cox regression models. PP is the pulsatile variable of blood pressure and, as such, a marker for large artery stiffness. PP has been shown to increase the risk of stroke in the general population, and the association seems to be linear [2]. The association of DBP and stroke is, however, less clear since this blood pressure variable does not increase with age, instead, it decreases [21], leading to a weaker association with stroke. This also explains the non-linear association with DBP and stroke in the restrictive cubic spline regression models. Rawshani et al. [7] also showed that the association with DBP and stroke is rather weak. MAP, on the other hand, is considered to be the perfusion pressure in organs in the body and is composed of both SBP and DBP [22]. The significant association reflects this for any stroke as well as for both subtypes in the present study. We are not aware of any previous studies on MAP and stroke in type 1 diabetes.

In individuals with type 2 diabetes, the risk of stroke increases after blood pressure exceeds 130/80 mmHg [23]. In our study, the risk of stroke starts to increase at similar DBP levels, while for SBP, linear increase is observed even earlier. In a recent study by Rawshani et al. [7], similar results were seen for SBP, although the risk started to increase for stroke at a slightly higher level compared to our study. The treatment recommendation for blood pressure levels in diabetes is, despite these facts, higher; according to the American Diabetes Association [24], medical treatment of high blood pressure in diabetes should be initiated at blood pressure levels of > 140/90 mmHg, and the treatment goal is a blood pressure of < 140/90 mmHg. However, a recently published guideline by the American Heart Association suggested that blood pressure levels in type 1 diabetes should be lower, recommending initiation of treatment already at blood pressure levels of > 130/80 mmHg, and that the blood pressure target should be < 130/80 mmHg in individuals with diabetes [22]. The results in our study support the new recommendations for blood pressure treatment targets in type 1 diabetes. Furthermore, despite the high prevalence of antihypertensive medication in the participants with stroke blood pressure levels were suboptimal, indicating that a more aggressive approach in treating blood pressure in these individuals should be considered.

To our surprise, there were no indications of a J-shaped phenomenon in our study for any of the studied blood pressure variables. The risk of stroke increased linearly for all subtypes for SBP, MAP, and PP, while the risk increment was non-linear for DBP. For SBP, this is in line with the study by Rawshani et al. [7], where the association with stroke was also linear. According to Zhao et al. [5], the J-shaped association was found mainly in individuals 30–59 years of age, and this association weakened in individuals older than 60 years. Even though the participants in our study were younger than 60 years, mean age around 50 years, their blood vessels could resemble those of non-diabetic individuals of older age due to the long diabetes duration and, thus, alternating blood glucose levels. In accordance with this, we have earlier shown that the pulse pressure in individuals with type 1 diabetes increases 15 to 20 years earlier than in non-diabetic individuals, suggesting early vascular aging in diabetes [25]. This can be explained by the fact that hyperglycemia causes endothelial dysfunction, leading to diabetic angiopathy and arterial stiffness, and all of these leading to hypertension [26‐28]. The same changes take place in normoglycemic persons during aging [29, 30]. On the other hand, apart from hypertension, the risk factors identified for stroke in type 1 diabetes are different compared to the general population [8, 31, 32]. The etiology of stroke is also different. In type 1 diabetes, the majority of strokes are caused by small-vessel disease or stroke of unclear etiology, while cardioembolism is less frequent in young individuals with type 1 diabetes compared with people without diabetes [33]. Carotid stenosis, a marker of large-artery atherosclerosis, is also less common, especially compared to individuals with type 2 diabetes [33]. Subclinical atherosclerosis is, however, known to associate with type 1 diabetes [34], so it might be that some of the strokes classified as unclear etiology could be due to atherosclerosis. In the present study we were unable to assess the etiology of strokes due to variation in and/or lack of etiological assessment performed at the different hospitals during the different time periods.

High salt intake, measured as 24-h urinary sodium excretion, increases the risk of cardiovascular disease, mainly stroke [9, 35]. The same can be seen in individuals with chronic kidney disease, for which the risk of stroke increases linearly when urinary sodium excretion increases [36]. A majority of our participants, 61%, with stroke had impaired kidney function, defined as diabetic nephropathy. Yet, we did not find any associations with urinary sodium and potassium excretion and stroke. One possibility for this result could be the low number of individuals tested; we only had information on urinary sodium and potassium excretion available for 115 participants with stroke. Many of these individuals also had micro- or macroalbuminuria, which in some part affect the urinary excretion. On the other hand, no trends towards any associations were seen in the analyses.

One of the weaknesses of this study was that we only had an office blood pressure measurement from a single time point available, and this could have been measured years before the incident stroke. Likewise, 24-h Na and 24-h K were also measured from a single urine collection. Ambulatory 24-h blood pressure or timed blood pressure measurements during a more extended time would be the best method to determine the impact of blood pressure on the risk of stroke. Unfortunately, in this large study population consisting of over 4000 participants, this was not feasible. Moreover, already with this single office blood pressure, we were able to show an increased risk of stroke in these participants, indicating that this single blood pressure measurement might be a sensitive enough tool to evaluate the risk of stroke years before an event.

Another limitation of this study is the number of stroke cases, particularly that of hemorrhagic strokes. This could affect the impact of sodium and potassium on the risk of stroke, as well as the J-shaped vs linear association between blood pressure levels and stroke risk. On the other hand, in the studies by Cederholm et al. [37, 38], no J-shaped relationships between low SBP and increased stroke risk was found even though the number of individuals was threefold larger in their second study, 13,000 vs 35,000 participants. Further studies on the potential J-shaped association between blood pressure and stroke in individuals with type 1 diabetes are required before any conclusions can be made.

The number of individuals studied is also one of the strengths of this study. Stroke is, fortunately, still a relatively rare complication in this age group. To this date, this is the largest study population of individuals with type 1 diabetes and stroke ascertained from medical files, which allowed us to also perform studies on the subgroups of stroke, ischemic and hemorrhagic stroke. Another strength in our study is the well-characterized population. All of the included strokes were identified from different sources, the information on strokes and their complications were verified from medical records, and all the included strokes were confirmed and classified with the same methods by expert stroke neurologists.