Glycaemic control and excess risk of ischaemic and haemorrhagic stroke in patients with type 1 diabetes: a cohort study of 33 453 patients

Authors


  • Prior presentation: Parts of this study were presented in abstract form at the 51st European Association for the Study of Diabetes annual meeting in Stockholm, 14–18 September 2015.

Abstract

Objective

To estimate the excess risk of stroke in relation to glycaemic control in patients with type 1 diabetes.

Methods

In this prospective, matched cohort study, we identified patients with type 1 diabetes, aged ≥18 years, who were registered in the Swedish National Diabetes Register from 1998–2011 and five control subjects for each case from the general population, matched for age, sex and county of residence. The risks of all strokes, ischaemic stroke and haemorrhagic stroke were estimated using Cox hazard regression.

Results

Of 33 453 type 1 diabetes patients [mean age, 35.5 (SD 14.4) years; mean follow-up, 7.9 (SD 4.3) years; and mean diabetes duration, 20.2 years (SD 14.6)], 762 (2.3%) were diagnosed with stroke compared with 1122 (0.7%) of 159 924 control subjects [mean follow-up, 8.2 (SD 4.3) years]. The overall multiple-adjusted hazard ratios (HRs) for type 1 diabetes patients versus control subjects were 3.29 (95% CI: 2.96–3.66) and 2.49 (95% CI: 1.96–3.16) for ischaemic and haemorrhagic stroke, respectively. The risk of ischaemic and haemorrhagic stroke incrementally increased with increasing HbA1c; the risk of ischaemic stroke was significantly increased with HbA1c within target [≤6.9% (≤52 mmol mol-1)] [multiple-adjusted HR 1.89 (95% CI: 1.44–2.47)]. For HbA1c ≥9.7% (≥83 mmol mol-1), there was a markedly increased risk of both ischaemic and haemorrhagic stroke, with multiple-adjusted HRs of 7.94 (95% CI: 6.29–10.03) and 8.17 (95% CI 5.00–13.35), respectively.

Conclusions

Individuals with type 1 diabetes have an increased risk of ischaemic and haemorrhagic stroke, increasing markedly with poor glycaemic control.

Introduction

Diabetes mellitus is an established risk factor for cardiovascular disease [1]. Even though type 1 diabetes accounts for only a minor proportion of diabetes cases globally, it is important because it often affects children and young adults living with the disease for many years. The results from a growing number of studies in the last decade have indicated that type 1 diabetes is a risk factor for stroke [2-4]. Age and blood pressure level are examples of factors that predict stroke in individuals with type 1 diabetes [5, 6] as well as in those without diabetes [7]. However, chronic hyperglycaemia is a risk factor specific amongst individuals with diabetes but has not been widely investigated as a risk factor for stroke in those with type 1 diabetes.

High HbA1c, the most common measure of hyperglycaemia in diabetes care, is strongly associated with an increased risk of microvascular complications in individuals with type 1 diabetes [8, 9] and has also been shown to affect the risk of stroke and other macrovascular complications [10, 11]. However, most studies have not been adequately powered to provide reliable results, particularly regarding haemorrhagic stroke, which comprises a small subgroup of all stroke cases in Western populations [2, 4, 5, 10, 12]. Using the Swedish National Diabetes Register (NDR), we conducted a large, observational, cohort study to evaluate the risk of stroke in individuals with type 1 diabetes at different HbA1c levels compared with the risk in the general population.

Materials and methods

Study design and participants

The NDR is a quality assurance instrument in diabetes care for patients 18 years of age or older and was started in 1996 [13]. In 2013, the register was estimated to include over 90% of all individuals with type 1 diabetes in Sweden [14]. Type 1 diabetes is defined in the register as treatment with insulin and a diagnosis at ≤30 years of age. This definition has been validated as accurate in 97% of the cases recorded in the register [15]. Information on risk factors, complications of diabetes and medication are reported to the register from the local care provider on a yearly basis.

In this study, we included 33 965 type 1 diabetes patients registered at least once in the NDR from 1 January 1998 until 31 December 2011. From the Swedish Total Population Register, we randomly selected five control subjects matched for age, sex and county of residence for each type 1 diabetes patient. This procedure, used in other studies [16], provided 169 825 controls from the general population. We excluded (i) control subjects who were registered with either type 1 or type 2 diabetes in the NDR (n = 6967), (ii) type 1 diabetes patients and controls with a diagnosis of stroke registered before starting the study (n = 506 and 2715, respectively), (iii) type 1 diabetes patients and controls who died before starting the study [(n = 3 and 205, respectively (i.e. usually control subjects who died between the date of random selection and the date of registration of the index case in the NDR)] and (iv) type 1 diabetes patients and their controls with missing vital status data in the NDR (n = 3 and 14, respectively). A total of 33 453 type 1 diabetes patients and 159 924 control subjects remained for analysis. All type 1 diabetes cases and their matched controls were followed from baseline registration in the NDR until first hospitalization for stroke, death or 31 December 2011, whichever came first. Informed oral consent was obtained from each patient included in the NDR.

Procedures

Information on comorbidities and stroke end-points for all participants was linked to our study from the Swedish National Patient Register (NPR) and Cause of Death Register by the personal identification number (PIN), which is unique for every citizen in Sweden. Information concerning place of birth and educational level was linked by the PIN from the Longitudinal Integration Database for Health Insurance and Labour Market Studies register. For patients with type 1 diabetes, data on risk factors, medication and lifestyle were retrieved from the NDR. After matching of data files, all personal identifiers were replaced by codes. The study was approved by the regional ethics review board at the University of Gothenburg.

Variables

Registration of all principal and contributory discharge diagnoses using the International Classification of Disease (ICD) codes is mandatory in Sweden, with the NPR operating on a national basis since 1987. The positive predictive value for diagnosis in the NPR varies, but is generally 85–95% for major cardiovascular diagnosis categories [17]. Diagnosis of stroke was validated in the NPR and Cause of Death Register for 2004 by measuring the diagnosis of stroke in the two registers combined and compared against a well-validated population-based epidemiological stroke register (the Northern Sweden MONICA). The positive predictive value was estimated to be 94%, and the sensitivity was 92% for the two registers combined [18].

The following codes were used to identify stroke prior to the start of the study: haemorrhagic stroke, 431, 432X (ICD-9), I61 and I62.9 (ICD-10); and ischaemic stroke, 433, 434, 436, 437X (ICD-9), I63, I64 and I67.9 (ICD-10). To identify comorbidities at study entry, the following codes were used: acute myocardial infarction (AMI), 410 (ICD-9) and I21 (ICD-10); coronary heart disease (CHD), 410–414 (ICD-9) and I20–I25 (ICD-10); atrial fibrillation (AF), 427D (ICD-9) and I48 (ICD-10); valve disease, 394–397, 424 (ICD-9), I05–I09 and I34–I36 (ICD-10); heart failure (HF), 428 (ICD-9) and I50 (ICD-10); and cancer 140–208 (ICD-9) and C00–C97 (ICD-10).

The following ICD-10 codes as a primary diagnosis were used to define stroke end-points: all strokes (I61, I62.9, I63, I64 and I67.9), ischaemic stroke (I63, I64 and I67.9) and haemorrhagic stroke (I61 and I62.9).

Educational level was categorized as low (compulsory only), intermediate and high (university or similar). Blood pressure was the mean value (mmHg) of two supine readings after rest for at least 5 min. Only active smokers were considered to be smokers for the present analysis. Analysis of microalbuminuria and HbA1c was performed at the local laboratory. A quality assessment organization regularly validates all healthcare laboratories in Sweden. Renal impairment was categorized as normoalbuminuria, microalbuminuria, macroalbuminuria or end-stage renal disease (ESRD). Microalbuminuria was defined as two of three urine samples obtained within 1 year with either an albumin:creatinine ratio of 3–30 mg mmol-1 (approximately 30–300 mg g-1) or a urinary albumin clearance of 20–200 μg min-1 (20–300 mg L-1). Urinary albumin excretion was defined as macroalbuminuria if the albumin:creatinine ratio was >30 mg mmol-1 (approximately ≥300 mg g-1) or urinary albumin clearance was >200 μg min-1 (>300 mg L-1). ESRD was defined as an estimated glomerular filtration rate of <15 mL min-1 or the need for renal dialysis or renal transplantation. Originally, healthcare units in Sweden used the HbA1c method calibrated to the high-performance liquid chromatography mono-S method. In September 2010, there was a nationwide change to the calibration recommended by the International Federation of Clinical Chemistry and Laboratory Medicine (IFCC) and the National Glycohemoglobin Standardization Program (NGSP). HbA1C values were converted according to the NGSP and are reported in percentages and in mmol mol-1 [19].

Statistical analysis

The unadjusted incidence rates for stroke end-points are presented as events per 1000 patient-years of follow-up with 95% confidence intervals (CIs) estimated by exact Poisson confidence limits. To investigate the risk of stroke for type 1 diabetes patients with different updated mean HbA1C levels, three Cox regression models were constructed with controls as the reference. Updated mean HbA1c represents the mean value of all preceding measurements and was updated for each new measurement (e.g. at registration of the third HbA1c value, the updated mean HbA1c at that time was the mean of three HbA1c values). Time at risk was calculated from study entry to first admission to hospital with a principal diagnosis of stroke, death, or 31 December 2011, whichever came first. Model 1 was an unadjusted model matched for age and sex. Model 2 was adjusted for time-updated age and sex and stratified for the duration of diabetes; model 3 was additionally adjusted for maximum educational level and comorbidities (AF and CHD) at baseline (other comorbidities were not significant and not included). Three stratified analyses were performed on the main model (model 3) to evaluate the effect of sex, diabetes duration and renal impairment on the relationship between HbA1c and stroke.

To study the relationship between HbA1c levels and the risk of stroke exclusively in type 1 diabetes patients, Cox regression models were constructed with different adjustments. The lowest HbA1c category [≤6.9% (≤52 mmol mol-1)] was used as a reference. In all Cox regression models, hazard ratios (HRs) and associated 95% CIs for stroke were estimated. All P-values are two-sided, and values <0.05 were considered to indicate statistical significance. The assumption of the proportional hazard was tested and found to hold for all Cox regression analyses. All analyses were performed using sas software 9.4 (SAS Institute Inc. ,Cary, NC, USA).

Results

For control subjects, only data on education, country of birth and comorbidities were available. Type 1 diabetes patients were more often born in Sweden than controls (Table 1 and Table S1). Baseline comorbidities of AMI, CHD, HF and cancer were more common in type 1 diabetes patients compared with controls, and the presence of these baseline comorbidities increased with higher HbA1c category. Diabetes duration, BMI and average systolic blood pressure in diabetes patients initially increased with each higher baseline HbA1c category and then slightly decreased in the very highest HbA1c categories. The difference in average blood pressure between the highest and lowest HbA1c categories was only 2.6 mmHg for systolic and 3.0 mmHg for diastolic blood pressures. The proportion of smokers amongst type 1 diabetes patients increased with higher HbA1c categories.

Table 1. Characteristics of type 1 diabetes patients, by baseline HbA1c category, and population-based controls at first registration in the NDR (1998–2011)
 HbA1c categories at baselinea
 Controls n = 159 924All type I diabetes n = 33 453≤6.9% (≤52 mmol mol-1) n = 60717.0–7.8% (53–62 mmol mol-1) n = 76437.9–8.7% (63–72 mmol mol-1) n = 88248.8–9.6% (73–82 mmol mol-1) n = 5377≥9.7% (≥83 mmol mol-1) n = 3940Missing value n = 1598
  1. NPR, National Patient Register; NDR, National Diabetes Register; NGSP, National Glycohemoglobin Standardization Program; IFCC, International Federation of Clinical Chemistry and Laboratory Medicine.

  2. All cases and controls were free of previous stroke. Data for categorical variables are shown as n (%); data for continuous variables are shown as mean (SD). aHbA1c expressed as % (mmol mol-1), in accordance with the NGSP and IFCC, respectively. bPercentages of these variables represent proportion of participants without missing data on each variable. A number of type 1 diabetes patients without missing data were education category 33 190, antihypertensive treatment 33 732 and smoking 31 019.

Female72 846 (46%)15 120 (45%)2689 (44%)3426 (45%)3940 (45%)2420 (45%)1913 (49%)732 (46%)
Baseline age, years35 (14) n = 159 92436 (14) n = 33 45334 (14) n = 607137 (15) n = 764337 (15) n = 882436 (14) n = 537733 (13) n = 394032 (14) n = 1598
Born in Sweden138 571 (87%)31 415 (94%)5669 (93%)7224 (95%)8342 (95%)5053 (94%)3673 (93%)1454 (91%)
Education categoryb
High53 716 (34%)10 156 (31%)2496 (41%)2601 (34%)2603 (30%)1279 (24%)725 (19%)452 (29%)
Registration in the NPR prior to baseline
Coronary heart disease1107 (<1%)1351 (4%)162 (3%)315 (4%)408 (5%)244 (5%)166 (4%)56 (4%)
Heart failure281 (<1%)464 (1%)57 (1%)90 (1%)143 (2%)73 (1%)69 (2%)32 (2%)
Variables in the NDR only
HbA1c, NGSP% (SD) IFCC mmol mol-1 (SD) 8.2 (1.5) 65.8 (15.8) n = 31 8556.3 (0.5) 45.6 (5.5) n = 60717.4 (0.2) 57.3 (2.6) n = 76438.3 (0.3) 67.2 (2.8) n = 88249.2 (0.3) 76.9 (2.8) n = 537710.8 (1.0) 95.0 (11.3) n = 3940 
Duration of diabetes, years 20.2 (14.6) n = 33 45316.1 (15.7) n = 607121.7 (14.9) n = 764322.6 (14.2) n = 882421.2 (13.4) n = 537718.1 (12.7) n = 394016.5 (14.7) n = 1598
BMI, kg m-2 25.0 (4.0) n = 29 20624.6 (4.0) n = 538525.0 (3.8) n = 694525.3 (3.8) n = 796525.4 (4.1) n = 483625.0 (4.6) n = 343624.6 (5.0) n = 639
LDL, mmol L-1 2.66 (0.83) n = 11 2602.53 (0.76) n = 23412.60 (0.79) n = 27152.69 (0.82) n = 29892.74 (0.86) n = 17602.87 (0.94) n = 13142.58 (0.82) n = 141
Systolic blood pressure, mmHg 126.7 (16.9) n = 31 102124.2 (15.7) n = 5749126.6 (16.5) n = 7318128.3 (17.0) n = 8448127.6 (17.1) n = 5150126.8 (18.0) n = 3692123.5 (16.6) n = 745
Diastolic blood pressure, mmHg 73.5 (9.1) n = 31 10272.1 (8.9) n = 574973.0 (8.9) n = 731873.8 (9.1) n = 844874.3 (9.3) n = 515075.1 (9.5) n = 369272.9 (9.4) n = 745
Antihypertensive treatment, n (%)b 6895 (22%)869 (15%)1566 (21%)2097 (25%)1250 (24%)902 (24%)211 (17%)
Smoking, n (%)b 4206 (14%)505 (9%)764 (11%)1099 (13%)849 (17%)837 (23%)152 (13%)

The mean follow-up period was 7.9 years for type 1 diabetes patients and 8.2 years for control subjects. Amongst 33 453 type 1 diabetes patients, 762 (2.3%) received a diagnosis of stroke compared with 1122 of 159 924 (0.7%) controls. The incidence rates as cases per 1000 patient-years were 2.87 (95% CI: 2.67–3.09) for all strokes, 2.47 (95% CI: 2.29–2.67) for ischaemic stroke and 0.40 (95% CI: 0.33–0.49) for haemorrhagic stroke (Table S2). The corresponding incidence rates for controls were 0.86 (95% CI: 0.81–0.91), 0.70 (95% CI: 0.65–0.74) and 0.16 (95% CI: 0.14–0.18) cases per 1000 patient-years, respectively.

The following results were obtained from the final Cox regression model that compared individuals with type 1 diabetes with controls [adjusted for time-updated age, sex, stratified by diabetes duration, maximum education category and baseline comorbidities (CHD and AF)]. Type 1 diabetes patients as a group had an increased risk of all strokes and ischaemic stroke with HRs of 3.14 (95% CI: 2.85–3.46) and 3.29 (95% CI: 2.96–3.66), respectively. Table 2 shows the estimates from the Cox regression models where the risk of all strokes and ischaemic stroke was increased for type 1 diabetes patients compared with controls in every HbA1c category. The risk gradually increased with higher HbA1c levels to a maximum HR of 7.98 (95% CI: 6.46–9.85) for all strokes and 7.94 (95% CI: 6.29–10.03) for ischaemic stroke in the highest HbA1c category. The HR for haemorrhagic stroke for type 1 diabetes patients as a group was 2.49 (95% CI: 1.96–3.16) compared with controls. The risk of haemorrhagic stroke was not increased for type 1 diabetes patients in the lowest HbA1c category, but in all higher HbA1C categories, the risk gradually rose to a maximum HR of 8.17 (95% CI: 5.00–13.35) in the highest category. HR values from different adjustment models for all stroke categories are shown in Table S3.

Table 2. Adjusted hazard ratios for stroke (and 95% confidence intervals) for time-updated mean HbA1c categories versus population-based controls as examined by Cox regression
 HR (95% CI)a P-value
HbA1c categories at baselineb
Controls (reference)≤6.9% (≤52 mmol mol-1)7.0–7.8% (53–62 mmol mol-1)7.9–8.7% (63–72 mmol mol-1)8.8–9.6% (73–82 mmol mol-1)≥9.7% (≥83 mmol mol-1)
  1. aHRs adjusted for time-updated age, sex, stratified by diabetes duration category, maximum educational level and baseline comorbidities (coronary heart disease and atrial fibrillation). bHbA1c categories expressed as % (mmol mol-1), in accordance with the National Glycohemoglobin Standardization Program and International Federation of Clinical Chemistry and Laboratory Medicine, respectively.

Stroke (events/individuals) 1855/191 0041.001.74 (1.35–2.25) <0.00012.16 (1.83–2.55) <0.00013.52 (3.08–4.03) <0.00014.38 (3.67–5.21) <.00017.98 (6.46–9.85) <0.0001
Ischaemic stroke (events/individuals) 1544/191 0041.001.89 (1.44–2.47) <0.00012.20 (1.84–2.64) <0.00013.82 (3.31–4.41) <0.00014.57 (3.78–5.53) <0.00017.94 (6.29–10.03) <0.0001
Haemorrhagic stroke (events/individuals) 311/191 0041.001.10 (0.52–2.33) 0.812.00 (1.34–3.00) 0.00072.19 (1.49–3.22) <0.00013.52 (2.23–5.54) <0.00018.17 (5.00–13.35) <0.0001

The results from the analysis stratified by sex, diabetes duration and renal impairment category for all strokes are shown in Fig. 1 and Figs S1 and S2. The corresponding analysis for the subtypes of stroke is shown in Tables S4–6. The effect of HbA1c on the risk of stroke and its subtypes was similar for men and women and for all three categories of diabetes duration. An increased risk of stroke with higher HbA1c category was also observed in all renal impairment categories. The highest risk of all strokes compared with controls was found in type 1 diabetes patients with severe ESRD and HbA1c ≥9.7% (≥83 mmol mol-1): HR 27.66 (95% CI: 14.22–53.78) (Fig. S2).

Figure 1.

Adjusted hazard ratios (HRs) for stroke (and 95% confidence intervals) for time-updated mean HbA1c categories stratified by sex versus population-based controls as examined by Cox regression, adjusted for time-updated age, stratified by diabetes duration, maximum educational level and baseline comorbidities (coronary heart disease and atrial fibrillation). HbA1c categories expressed as % in accordance with the National Glycohemoglobin Standardization Program.

Estimates of risk of stroke with different adjustments within the diabetes patient group are shown in Table 3 [with HbA1c ≤6.9% (≤52 mmol mol-1) as the reference]. The risk of stroke gradually increased in the higher HbA1c categories compared with the reference, from HbA1c ≥7.9% (≥63 mmol mol-1) for all and ischaemic strokes and from HbA1c ≥8.8% (≥73 mmol mol-1) for haemorrhagic stroke. For ischaemic stroke, the risk was most attenuated by adjustment for lipids and lipid-lowering medication [HR 2.81 (95% CI: 1.78–4.43) in the HbA1c category ≥9.7% (≥83 mmol mol-1)]. For haemorrhagic stroke, adjustment for renal impairment had the largest attenuating effect [HR 4.13 (95% CI: 1.67–10.22) in the HbA1c category ≥9.7% (≥83 mmol mol-1)].

Table 3. Adjusted hazard ratios for stroke (and 95% confidence intervals) by time-updated HbA1c categories amongst type 1 diabetes patients as examined by Cox regression
 Hazard ratio (95% CI) P-value
Model 1Model 2Model 3Model 4Model 5Model 6
  1. HbA1c categories expressed as % (mmol mol-1), in accordance with the National Glycohemoglobin Standardization Program and International Federation of Clinical Chemistry and Laboratory Medicine, respectively.

  2. Model 1 (99.4% of individuals without missing data): adjusted for updated age and sex; model 2 (99.4% of individuals without missing data): as model 1 plus stratified by diabetes duration; model 3 (98.7% of individuals without missing data): as model 2 plus adjusted for education category and updated comorbidities (coronary heart disease and atrial fibrillation); model 4 (94.5% of individuals without missing data): as model 3 plus adjusted for updated mean systolic blood pressure, updated mean BMI and smoking; model 5 (86.1% of individuals without missing data): as model 3 plus adjusted for updated mean HDL, updated mean LDL and lipid-lowering medication; model 6 (94.8% of individuals without missing data): as model 3 plus adjusted for updated renal impairment categories.

Stroke (events/individuals)754/33 263754/33 263750/33 007706/31 608493/28 809727/31 697
≤6.9% (≤52 mmol mol-1)1.001.001.001.001.001.00
7.0–7.8% (53–62 mmol mol-1)1.28 (0.96–1.70) 0.0961.23 (0.92–1.64) 0.171.21 (0.91–1.62) 0.191.30 (0.96–1.76) 0.0921.02 (0.73–1.43) 0.901.23 (0.92–1.66) 0.17
7.9–8.7% (63–72 mmol mol-1)2.07 (1.58–2.72) <0.00011.97 (1.50–2.59) <0.00011.91 (1.45–2.51) <0.00011.96 (1.47–2.63) <0.00011.66 (1.21–2.29) 0.00181.86 (1.40–2.48) <0.0001
8.8–9.6% (73–82 mmol mol-1)2.53 (1.89–3.40) <0.00012.41 (1.79–3.23) <0.00012.25 (1.67–3.03) <0.00012.25 (1.64–3.08) <0.00011.91 (1.33–2.72) 0.00042.06 (1.51–2.80) <0.0001
≥9.7% (≥83 mmol mol-1)4.29 (3.12–5.90) <0.00014.21 (3.06–5.79) <0.00013.89 (2.82–5.37) <0.00013.61 (2.56–5.08) <0.00013.31 (2.20–4.97) <0.00013.19 (2.28–4.47) <.0001
Ischaemic stroke (events/individuals)650/33 263650/33 263647/33 007614/31 608424/28 809626/31 697
≤6.9% (≤52 mmol mol-1)1.001.001.001.001.001.00
7.0–7.8% (53–62 mmol mol-1)1.20 (0.88–1.63) 0.241.16 (0.85–1.58) 0.341.15 (0.84–1.57) 0.381.20 (0.87–1.66) 0.271.02 (0.71–1.48) 0.901.17 (0.85–1.61) 0.33
7.9–8.7% (63–72 mmol mol-1)2.08 (1.56–2.77) <0.00011.99 (1.49–2.65) <0.00011.92 (1.43–2.57) <0.00011.92 (1.41–2.60) <0.00011.74 (1.23–2.45) 0.00171.90 (1.40–2.57) <0.0001
8.8–9.6% (73–82 mmol mol-1)2.43 (1.78–3.33) <0.00012.32 (1.70–3.18) <0.00012.17 (1.58–2.98) <0.00012.09 (1.50–2.92) <0.00011.85 (1.26–2.73) 0.00172.05 (1.47–2.85) <0.0001
≥9.7% (≥83 mmol mol-1)3.95 (2.80–5.57) <0.00013.90 (2.76–5.50) <0.00013.58 (2.53–5.06) <0.00013.27 (2.27–4.71) <0.00012.81 (1.78–4.43) <0.00013.03 (2.11–4.36) <0.0001
Haemorrhagic stroke (events/individuals)104/33 263104/33 263103/33 00792/31 60869/28 809101/31 697
≤6.9% (≤52 mmol mol-1)1.001.001.001.001.001.00
7.0–7.8% (53–62 mmol mol-1)1.90 (0.83–4.36) 0.131.74 (0.76–4.01) 0.191.72 (0.75–3.96) 0.202.37 (0.90–6.19) 0.0801.01 (0.41–2.46) 0.981.71 (0.74–3.94) 0.21
7.9–8.7% (63–72 mmol mol-1)2.07 (0.91–4.71) 0.0831.86 (0.82–4.25) 0.141.83 (0.80–4.17) 0.152.45 (0.94–6.40) 0.0671.20 (0.50–2.87) 0.691.63 (0.71–3.76) 0.25
8.8–9.6% (73–82 mmol mol-1)3.36 (1.43–7.88) 0.00533.03 (1.29–7.13) 0.0112.83 (1.20–6.71) 0.0183.90 (1.45–10.52) 0.00712.22 (0.88–5.60) 0.0912.16 (0.90–5.19) 0.084
≥9.7% (≥83 mmol mol-1)6.98 (2.90–16.83) <0.00016.50 (2.69–15.71) <0.00016.32 (2.60–15.38) <0.00017.03 (2.49–19.82) 0.00026.65 (2.55–17.35) 0.00014.13 (1.67–10.22) 0.0021

Discussion

In this nationwide cohort study, individuals with type 1 diabetes had a threefold increased risk of being diagnosed with stroke compared with their matched controls from the general population. The risk of stroke was raised in all HbA1c categories, ranging from a 75% increase in the lowest HbA1c category to almost an eightfold increase in risk in the highest category. An effect of HbA1c on the risk of stroke was observed in men and women, for type 1 diabetes patients with a short and long duration of disease and in all three categories of renal impairment. These findings indicate an independent effect of glycaemic control on the risk of stroke. Even type 1 diabetes patients maintaining HbA1c levels within the recommended target [≤6.9% (≤52 mmol mol-1)] and with normoalbuminuria had an increased risk of all strokes because of an increased risk of ischaemic stroke events in this group.

Several previous studies have estimated the risk of stroke in type 1 diabetes patients compared with the general population [2-4, 20]. In a study that included participants with a similar age range as in the present study, a comparable incidence rate was observed [3]. The risk of stroke for type 1 diabetes patients as a group has been estimated to be between twofold and fourfold higher than in the general population [3, 20]. To the best of our knowledge, no previous study has estimated the risk of stroke for type 1 diabetes patients at different HbA1c levels and compared this risk with that of the general population.

HbA1c is a well-established risk factor for microvascular complications in type 1 diabetes [9, 21]. In the last decade, an increasing number of studies have shown that HbA1c affects the risk of cardiovascular disease and stroke in type 1 diabetes patients, even though the number of stroke events were often limited [10-12, 22, 23]. The finding in our study of an increased risk of all strokes and ischaemic stroke in type 1 diabetes patients with higher HbA1c levels is consistent with these previous results. The association between glycaemic level and the risk of haemorrhagic stroke is less clear. Recently, Hagg et al. [11] found that HbA1c was not a risk factor for haemorrhagic stroke in type 1 diabetes patients, whereas Secrest et al. [5] reported that HbA1c predicted haemorrhagic but not ischaemic stroke. In present study, there was a 2.5-fold excess risk of haemorrhagic stroke for type 1 diabetes patients as a group compared with controls, and this risk was significantly higher in all but the lowest HbA1c category. Therefore, we found an increased risk of ischaemic and haemorrhagic stroke with higher HbA1c levels in type 1 diabetes patients.

Blood pressure is an important risk factor for stroke in type 1 diabetes patients [11], but the difference in average blood pressure between the highest and lowest HbA1c category was comparatively minor (2.4 mmHg for systolic and 3.0 mmHg for diastolic blood pressures). The proportion of type 1 diabetes patients receiving antihypertensive treatment was 15% in the lowest HbA1c category, but approximately the same (21–25%) in all higher HbA1c categories. Additionally, in the analysis within the type 1 diabetes group, we were able to adjust for systolic blood pressure and still found an effect of HbA1c level on the risk of stroke (Table 3, model 4). Therefore, we believe that blood pressure level might explain some, but not all, of the increased risk of stroke with higher HbA1c category.

Previous studies have demonstrated that diabetic nephropathy is a predictor of stroke [5, 6]. In the present study, the risk of all strokes, and stroke subtypes, within the same HbA1c category gradually increased with worse renal impairment category (Fig. S2 and Table S6), which is in line with previous findings [5, 6]. The effect of HbA1c on the risk of haemorrhagic stroke for type 1 diabetes patients with normoalbuminuria was weaker than that for ischaemic stroke. Only type 1 diabetes patients with normoalbuminuria and HbA1c in the highest category ≥9.7% (≥83 mmol mol-1) had a significantly increased risk of haemorrhagic stroke compared with control subjects. For ischaemic stroke, the risk was significantly increased in all HbA1c categories for type 1 diabetes patients with normoalbuminuria. Further studies are needed to elucidate the relationships between diabetes, glycaemic control and subtypes of stroke in individuals with type 1 diabetes.

The most important strengths of the present study are the large number of individuals with type 1 diabetes and the presence of a control group from the general population. The size of the study population provided a larger number of strokes, both haemorrhagic and ischaemic, than in any previous study. Additionally, repeated measurements of several important characteristics in individuals with type 1 diabetes were available, as well as information on comorbidities and educational level for both control subjects and diabetes patients. However, there are also limitations that need to be considered. First, only stroke events leading to hospitalization or death outside hospital were captured. A recent study in Sweden showed that stroke awareness did not differ between patients with diabetes and individuals without regular healthcare contacts [24]. Therefore, we believe that the proportion of potential missed cases due to not seeking medical care should be the same between control subjects and individuals with diabetes. Secondly, no risk factor data other than demographic characteristics and comorbidities were available for the controls. HbA1c values were not available for the control subjects but those with diabetes were excluded. Given the relatively young age of the population, the numbers of undiagnosed cases of type 2 diabetes and of individuals in the prediabetic state with higher HbA1c should be very small amongst the controls, and accordingly the HbA1c level of the controls should be close to normal. Blood pressure, smoking and AF are important risk factors for stroke. However, mean blood pressure was not raised in individuals with type 1 diabetes and the proportion of participants with AF was similar to that of controls. The proportion of smokers amongst individuals with type 1 diabetes might be slightly lower than that in the general population. Accordingly, information on these variables is unlikely to have altered our results decisively. Thirdly, the diagnoses of haemorrhagic and ischaemic stroke were not formally validated, but computed tomography scans are routinely used in suspected stroke cases in Sweden, which should minimize misdiagnosis. In our study 14% of the strokes amongst type 1 diabetes patients were haemorrhagic. This is somewhat lower than was found for example by Hägg et al. [11]. However, we have not included subarachnoid haemorrhage in our study due to a different and more specific pathology. Additionally, the proportion of haemorrhagic strokes relative to all strokes differs between countries [25]. A previous study on the subtypes of stroke in Sweden demonstrated that 12% of the strokes were haemorrhagic (excluding subarachnoid haemorrhage) [26]. Therefore we believe our figures reflect the true proportion of haemorrhagic strokes amongst type 1 diabetes patients in Sweden. Fourthly, during the first years of the study not all hospitals in Sweden reported to the NDR. The coverage rose to approximately 50% of all type 1 diabetes patients in 2003 and to approximately 90% in 2013. However, hospitals report all their diabetes patients, both well and poorly controlled, to the NDR. Furthermore, a small number of patients with diabetes not registered in the NDR could be included amongst the controls during the first years of the study, potentially attenuating the effect of HbA1c level. Taken together, even though the coverage of the NDR was not complete during the first years of the study this should only have a minor effect on our results. Finally, as in all observational studies, we cannot definitively eliminate the possibility of residual confounding.

Conclusion

Individuals with type 1 diabetes have an increased risk of ischaemic and haemorrhagic stroke compared with the general population even when HbA1c levels are within target, and the risks are markedly increased with worse glycaemic control. Accordingly, continuous efforts to improve glycaemic control in individuals with type 1 diabetes are of major importance to protect this group against a disease with potentially devastating effects on daily life.

Conflict of interest statement

No conflict of interest to declare.

Acknowledgements

We thank all of the clinicians who were involved in the care of patients with diabetes for collecting data and staff at the NDR.

Funding

The study was financed by grants from the Swedish state, under the agreement between the Swedish government and county councils for economic support of research and education of doctors (the ALF agreement). This study was also funded by grants from the Novonordisk Foundation, the Swedish Society of Physicians, the Health & Medical Care Committee of the Regional Executive Board, Region Västra Götaland, the Swedish Heart and Lung Foundation, Diabetes Wellness, the Swedish Research Council (SIMSAM; grant numbers 2013-5187 and 2013-4236) and the Swedish Council for Working Life and Social Research (Epilife).

Authors’ contributions

C.H.S., M.L. and A.R. initiated the conception and design of the study, contributed to the analysis and interpretation of the data and drafted the article. A.-M.S. and S.G. contributed to the acquisition of data. A.M. performed the calculations. All authors contributed to revising the article critically for important intellectual content and final approval of the version to be submitted. A.R. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Ancillary