Risk Factors for Stroke in Type 2 Diabetes Mellitus:  United Kingdom Prospective Diabetes Study (UKPDS) 29 FREE

THE INCIDENCE of cardiovascular disease is increased 2- to 3-fold in patients with type 2 diabetes mellitus, and this increase cannot be explained by the presence of classic risk factors for atherosclerosis, such as smoking, hypertension, and dyslipidemia.¹ In patients with cerebrovascular disease, the presence of type 2 diabetes increases the risk of ischemic cerebral infarction, which accounts for more than three quarters of all strokes, but is not associated with an increased risk of cerebral hemorrhage.^2- 3 The crude incidence of stroke among patients with type 2 diabetes can be more than 3 times that in the general population,^4- 14 with particularly high rates reported in Sweden¹¹ and the southeastern United States.^8,15 The relative risk of stroke in patients with type 2 diabetes reaches a maximum in the 40- to 60-year-old group, and women comprise a greater proportion of patients with stroke than in the nondiabetic population.¹⁶

Nonmodifiable risk factors such as age, sex, race, and heredity are associated with stroke in subjects with¹⁶ and without¹⁷ diabetes. Hypertension, cardiac disease (including atrial fibrillation), and cigarette and alcohol use are modifiable risk factors in patients without diabetes,¹⁷ but an association with dyslipidemia is less clear.¹⁸ Studies¹⁹ of diabetic patients have consistently identified hypertension as the major risk factor. Associations have been found between stroke and other manifestations of atherosclerosis, cardiac failure, and nonrheumatic atrial fibrillation,^20- 23 but in diabetic patients, inadequate glycemic control, dyslipidemia, obesity, smoking, and microvascular disease have not been identified as independent risk factors.¹⁶ Many studies, however, have used relatively small numbers of selected patients from cross-sectional studies and a limited number of risk factors in the assessment of stroke in patients with type 2 diabetes.

We have investigated risk factors for stroke in patients with newly diagnosed type 2 diabetes mellitus recruited to the United Kingdom Prospective Diabetes Study (UKPDS).²⁴ White patients without known atheromatous disease were observed for a median of 7.9 years. The results indicate that appropriate management of hypertension and atrial fibrillation may prove the most effective primary prevention strategies in patients with type 2 diabetes.

The UKPDS recruited 5102 patients aged 25 to 65 years with newly diagnosed type 2 diabetes (fasting plasma glucose levels, >6 mmol/L [>108 mg/dL] on 2 occasions) between December 1, 1977, and March 31, 1991. An additional 2006 patients of similar age, sex, and fasting plasma glucose levels were excluded because they had severe vascular disease (myocardial infarction in the past year, current angina, or heart failure), accelerated hypertension, proliferative or preproliferative retinopathy, renal failure with plasma creatinine levels of greater than 175 µmol/L (>2.0 mg/dL), other life-threatening disease such as cancer, an illness requiring parenteral steroid therapy, an occupation precluding insulin treatment, language difficulties, or the presence of ketonuria (urine ketone bodies, >3 mmol/L), suggestive of type 1 diabetes mellitus.

There were 4178 white patients, of whom 381 had known cardiovascular disease (previous myocardial infarction or electrocardiographic [ECG] Q-wave abnormality [202 patients], angina [7 patients], heart failure [1 patient], intermittent claudication [120 patients], and a previous stroke or transient ischemic attack [51 patients]) and were therefore excluded. Biochemical measurements were not performed until 1981, and some patients had no valid data for 1 or more of the other variables, leaving 2704 patients with a complete set of risk factor information at baseline. The final analysis was done in 3776 patients who had data relating to age, sex, and hypertension or normotension categorization. The study protocol was approved by the institutional Ethics Committee in each of the 23 centers. All recruited patients gave informed consent to participation.

During an initial 3-month period in which patients received dietary therapy alone, contraceptive or hormone replacement therapy was stopped and a loop diuretic (furosemide [frusemide]) was substituted for benzothiadiazide treatment unless these changes were considered inappropriate on clinical grounds. After the initial dietary therapy, patients were randomly allocated to different hypoglycemic treatments according to the UKPDS protocol.²⁴ Those allocated to diet formed a conventional-therapy group, and those allocated to receive sulfonylurea, insulin, or metformin hydrochloride comprised an intensive-therapy group.

Patients were seen every 3 months in UKPDS clinics, and any possibly diabetes-related clinical events were recorded. The administrator requested full information from the center, general practitioner, or other health care professionals. A file without details of randomized or actual therapies was evaluated by 2 independent clinical assessors to ascertain whether predetermined criteria for such end points were met. If the 2 assessments did not agree, the information was presented to a panel of 3 other independent senior physicians for a final decision. Each end point was classified according to the International Classification of Diseases, Ninth Revision,²⁵ the codes for stroke being 430 to 434.9 and 436.

The height and waist and hip circumferences were measured at the time of the diagnosis of type 2 diabetes mellitus, and tobacco smoking and amount of exercise taken were assessed by questionnaire. Blood pressure was recorded as the mean of measurements taken 2 and 9 months after diagnosis using an electronic sphygmomanometer. Hypertension was diagnosed if the patient had a systolic blood pressure of 160 mm Hg or greater, a diastolic blood pressure of 90 mm Hg or greater, or both, or if antihypertensive therapy had already been prescribed. At the visit 3 months after initial dietary therapy, patients attended a UKPDS clinic in the morning after a 10-hour overnight fast to have a blood specimen drawn for standard biochemical tests and to provide a spot urine specimen for the assessment of albuminuria. Retinopathy was assessed by 4 horizontal 30° color photographs per eye with modified Wisconsin grading.²⁴ Electrocardiographic tracings were evaluated by 2 trained coders, and a Minnesota code was assigned.²⁴

Fasting plasma glucose levels were measured in each center, with a monthly quality assurance scheme showing a coefficient of variation of less than 4%. Glycosylated hemoglobin (A_1c) levels were measured by a high-performance liquid chromatography analyzer (Bio-Rad Diamat; Bio-Rad Laboratories, Hemel Hempstead, England) with a reference range of 4.5%-6.2%. Plasma triglyceride levels were measured using a commercial kit (GPO-PAP [glycerol phosphate oxidase-p-aminophenazone]; Boehringer Mannheim, Lewes, East Sussex, England), without correction for free glycerol, on a centrifugal analyzer (Cobas FARA; Roche Diagnostica, Welwyn Garden City, Herts, England). Plasma cholesterol levels were measured using a high-performance cholesterol oxidase-p-aminophenazone (CHOD-PAP) method with Preciset cholesterol standard (Kit C system; Boehringer Mannheim) on the centrifugal analyzer, with low-density lipoprotein and high-density lipoprotein cholesterol levels after precipitation.²⁶ Plasma insulin levels were measured with a radioimmunoassay that cross-reacted 100% with proinsulin. The urine albumin concentration was expressed relative to the mean urine creatinine concentration of 11 mmol/L in men and 8 mmol/L in women to allow for urine dilution.²⁶ The urine albumin concentration was measured by radioimmunoassay or immunoturbidimetry.

Age was categorized as younger than 50 years, 50 to 54 years, 55 to 59 years, or 60 years or older. Other continuous variables were categorized by tertiles calculated from 3797 persons with no prior indication of atherosclerosis. Cox proportional hazards models,²⁷ with censoring at 10 years' follow-up, were used to assess the effect of potential risk factors on fatal or nonfatal stroke. The dependent variable was the time to the first end point or to censoring. Relationships of single risk factors with events, after adjusting for age and sex, were assessed in 2704 patients with data on all risk factors. A multivariate selection of risk factors was made by a stepwise procedure, after adjusting for age and sex. The final multivariate analysis was done in the 3776 patients who had data for all the selected variables. Estimated hazard ratios are represented graphically with 95% confidence intervals estimated for each group by treating the relative risks as floating absolute risks.²⁸

The association of urine albumin concentration (categorized as <50 or ≥50 mg/L for microalbuminuria and <300 or ≥300 mg/L for proteinuria) with strokes was determined by adding urine albumin concentrations to the multivariate model obtained by the stepwise selection procedure in 2973 patients who had a valid measurement. The same method was used for the presence of atrial fibrillation in 3728 patients who had a baseline ECG. In 2161 patients who had retinal photographs assessed by a modified Wisconsin 191 grading of 4 color photographs per eye,²⁴ 3 categories—no retinopathy, the presence of microaneurysms in 1 or both eyes, and more severe retinopathy (eg, hard exudates, hemorrhages, or intraretinal microvascular abnormalities)—were used in analysis.

To assess the effect of regression to the mean on systolic blood pressures, a second measurement was taken in a subset of 477 patients who were randomly allocated to dietary therapy and continued this treatment alone for a further 12 months after randomization. Patients were categorized according to their baseline value into the tertiles calculated from the 3797 persons with no prior indication of coronary artery disease. In the highest and lowest blood pressure groups, the difference in the means of the 2 samples taken 12 months apart gave an estimate of regression to the mean. These differences were then applied to the respective baseline mean values of the upper and lower tertiles in the 3776 patients, to correct for a regression to the mean. To estimate the effect of a 10-mm Hg increment of systolic blood pressure on the risk of stroke, the systolic blood pressure was fitted in the stepwise selected Cox model as a continuous variable, leaving other risk factors as categorical variables. This estimate was also adjusted for a regression to the mean.

Statistical analyses were performed using commercial software (SAS Institute, Inc, Cary, NC). They did not include any reference to allocated or actual therapy during the 10-year follow-up period. Data are reported as mean±1 SD, geometric mean (1 SD interval), or percentages.

Of 3776 patients in the study, 99 (2.6%) had a stroke during follow-up. The median duration of follow-up was 7.9 years with an interquartile range of 6.0 to 10.0 years (the data of 27.1% of patients were censored at 10 years). Baseline data aggregated by sex for the 2704 patients with complete risk factor data are shown in Table 1. After adjusting for age and sex, hypertension and systolic blood pressure were significantly and independently associated with stroke (Figure 1), whereas the body mass index (BMI or Quetelet index; calculated as weight in kilograms divided by the square of the height in meters), waist-hip ratio, smoking, exercise level, fasting plasma glucose levels, hemoglobin A_1c level, serum insulin level, serum lipid variables, and diastolic blood pressure were not (Table 2).

In a multivariate analysis of 3776 patients, age, sex, and the presence of hypertension were the major risk factors. When hypertension was replaced by the systolic blood pressure, this was also a significant risk factor, with estimated hazard ratios of 1.96 (95% confidence interval, 1.01-3.81) for the middle tertile and 2.99 (1.57-5.69) for the top tertile. When fitted as a continuous variable, after adjusting for a regression to the mean, for a 10-mm Hg rise in blood pressure, the hazard ratio increased by 0.54.

Of the 2973 patients with a valid urine albumin measurement at baseline, 80 patients (2.7%) subsequently had a nonfatal or fatal stroke. After adjusting for age and sex, a urine microalbumin excretion of greater than 50 mg/L was associated with stroke, with an estimated hazard ratio of 2.3 (1.4-4.0) compared with a urine albumin excretion of 50 mg/L or less. When urine albumin excretion was added to the model with age, sex, and the presence of hypertension, neither it nor retinopathy was a significant predictor of stroke (P>.10).

Of 3728 patients with a baseline ECG, 98 (2.6%) had a stroke during follow-up. Atrial fibrillation was present in 28 patients initially, and 6 (21.4%) of these had a stroke within 10 years. Thus, of all patients with stroke (n=99), atrial fibrillation had been documented in 6 (6.1%). Atrial fibrillation was a significant risk factor, with an estimated hazard ratio of 8.65 (3.5-18.4), when added to the model with age, sex, and the presence of hypertension (P<.001) (Table 3).

Our data confirm the importance of nonmodifiable and modifiable risk factors for stroke in diabetes mellitus. Increasing age and male sex were the nonmodifiable risk factors present in our cohort, and hypertension and atrial fibrillation were the only modifiable factors present among a broad range of clinical and biochemical variables entered in the model.

The risk of stroke increased progressively for the 4 age categories in the present study; a patient older than 60 years at diagnosis had almost 5 times the risk of a patient younger than 50 years at baseline. Male sex was also associated with stroke in our diabetic patients, with men having a relative risk of more than 1.5 times that of women. Nevertheless, a woman with diabetes probably has less cerebrovascular "protection" associated with sex because the relative risk of stroke in women with diabetes compared with those without diabetes (between 2.6 and 13) is generally higher than that of diabetic vs nondiabetic men.^12,14

Consistent with our findings, hypertension, especially systolic hypertension, has been recognized previously¹⁶ as a major risk factor for stroke in patients with diabetes. An evaluation of the effect of changes in blood pressure and antihypertensive treatment from the time of diagnosis on stroke incidence was beyond the scope of the present study and has been reported recently elsewhere.²⁹ Nevertheless, given the difficulty in achieving acceptable blood pressure control in diabetic patients with hypertension,³⁰ it is not surprising that the blood pressure at the time of diagnosis was a strong predictor of the occurrence of stroke. Other features of the "metabolic syndrome," including obesity, hyperinsulinemia, and dyslipidemia, were not associated with stroke in our patients, consistent with previous studies of type 2 diabetes.¹⁶ A recent meta-analysis¹⁸ has indicated that hypertension, but not elevated blood cholesterol levels, is related to the incidence of stroke in the general population.

Recent intervention studies have provided apparently conflicting data concerning the relationship between blood pressure control and the incidence of stroke in patients with type 2 diabetes. An analysis of UKPDS data on an intention-to-treat basis has revealed that hypertensive patients allocated to tight blood pressure control (target, <150/85 mm Hg; mean achieved blood pressure during a median of 8.4 years of follow-up, 144/82 mm Hg) had a 44% reduction in the incidence of fatal and nonfatal stroke compared with patients in a group with less tight control (target, <180/105 mm Hg; achieved, 154/87 mm Hg).²⁹ Preliminary data from the Systolic Hypertension in Europe Treated With Nitrendipine-based Antihypertensive Therapy Trial show that the excess risk of stroke associated with diabetes was abolished by antihypertensive treatment of older patients with type 2 diabetes mellitus and isolated systolic hypertension.³¹ In the Hypertension Optimal Treatment (HOT) trial,³² however, blood pressure reduction in a subgroup with diabetes was not associated with a reduced risk of stroke.

There are several possible explanations for the apparently discordant findings in the HOT study. First, fewer diabetic patients were entered in the HOT study than in the UKPDS, and the number of cardiovascular events was less than the investigators anticipated,³² thus reducing the power of the study to detect a beneficial effect of blood pressure reduction on stroke. Second, the average follow-up in the HOT study was only 3.8 years, and this may have been inadequate to show a treatment-related effect on the risk of stroke. The UKPDS data²⁹ show that the Kaplan-Meier plots for cumulative stroke in the 2 treatment groups were coincident for the first 3 years, with increasing separation thereafter. Third, the UKPDS used atenolol or captopril as primary therapy in the tight-control group, whereas the HOT study used a calcium channel blocker. The possible adverse effects of calcium channel blockers on cardiovascular outcomes³³ may also be a factor underlying the differences in the risk of stroke between hypertension intervention studies in type 2 diabetes.

The association between macroproteinuria and macrovascular disease that includes stroke is well recognized.^34- 35 Nevertheless, microalbuminuria was not a risk factor for stroke in the present study when included in a multivariate model with hypertension. This has also been reported for coronary artery disease in patients with type 2 diabetes³⁶ and suggests that microalbuminuria is, in part, a marker for hypertension rather than an independent risk factor for atherosclerosis. Although physical activity appears to have a beneficial effect on the incidence of stroke in middle-aged British men,³⁷ we found no such association in the present study.

Patients in our cohort who had atrial fibrillation on ECG at the time of diagnosis were more than 8 times as likely to suffer a stroke during the first 8 years of diabetes than those who were in sinus rhythm. This finding suggests that atrial fibrillation is the most important risk factor, either modifiable or nonmodifiable, in patients having newly diagnosed type 2 diabetes mellitus. The strength of association between atrial fibrillation and subsequent stroke has not been identified by previous studies and has significant implications for clinical management. Because atrial fibrillation can be intermittent and because it increases in prevalence with increasing age, more patients in the UKPDS than identified at the time of diagnosis likely had atrial fibrillation during the follow-up. Identification of these patients is clearly difficult without specific intensive ECG monitoring beyond that done every 3 years under the UKPDS protocol. Nevertheless, it is possible that, if this information were available, the association between atrial fibrillation and stroke would be even stronger than that evident from cross-sectional data alone.

In some previous reports of risk factors for stroke in diabetes, atrial fibrillation did not appear in logistic regression analyses. This includes longitudinal studies^35,38 of approximately 1000 Finnish patients aged 45 to 64 years with type 2 diabetes mellitus and a smaller-scale prospective study³⁹ of a cohort of 133 patients of similar age with newly diagnosed diabetes. Proteinuria, glycemic control, dyslipidemia, and autonomic neuropathy were among the independent risk factors for stroke identified in at least 1 of these 3 studies.^35,38- 39 The apparent differences between risk factor profiles in Finnish and UKPDS patients might reflect relative sample sizes and differences in the true underlying risk of stroke in the study populations¹¹ because, even allowing for differences in age and duration of diabetes, the incidence of stroke appeared higher in the Finnish patients (eg, 3.8% after 5 years and 14.7% after 10 years in those who were newly diagnosed³⁹) than in our cohort (2.6% after a median of 8 years). The selection of variables used in regression analysis, including the presence or absence of atrial fibrillation, may have been a key contributing factor.

Where data concerning atrial fibrillation have been available, only limited conclusions about their importance can be drawn. In a study⁴⁰ of 428 unselected hospital inpatients with cerebrovascular disease, 77 (18.0%) patients had diabetes mellitus and were also significantly more likely to have had a history of atrial fibrillation than the patients without diabetes. In an 8-year population-based study²¹ of subjects aged 35 to 74 years, 15.1% of diabetic patients with stroke had previously documented atrial fibrillation compared with 10.7% of those without diabetes. Although these studies confirm that atrial fibrillation increases the risk of stroke in patients with diabetes, our data suggest that it is the most important risk factor in younger patients with newly diagnosed type 2 diabetes mellitus who are relatively free of vascular disease.

Warfarin sodium therapy is associated with a two-thirds reduction in the risk of stroke when given for atrial fibrillation in the general population.⁴¹ Although there are no equivalent data for patients with diabetes, it is possible that the benefits may be even greater. Nevertheless, only 8 (30%) of our patients with atrial fibrillation on ECG at study entry were taking warfarin therapy at some stage during the first year of follow-up. The underuse of anticoagulant therapy for chronic atrial fibrillation is well recognized, with a similar proportion of patients with atrial fibrillation in the general population (about a third) receiving warfarin.^42- 43 These patients have an increased risk of hemorrhagic stroke of 3 such events per 1000 patient-years of treatment.⁴⁴ Because the risk of cerebral hemorrhage in diabetic patients may be less than that in persons without diabetes,^2- 3 warfarin therapy is not likely to markedly increase the incidence of this complication of anticoagulation therapy in type 2 diabetes mellitus.

Both aggressive treatment of hypertension and routine anticoagulation therapy for atrial fibrillation may reduce the risk of stroke in patients with newly diagnosed type 2 diabetes mellitus who are younger than 65 years. Although data from published intervention trials^29,45 in diabetic patients with hypertension have inconsistencies, the beneficial effects of antihypertensive treatment on the incidence of cerebrovascular disease that have been documented in subjects without diabetes appear even greater for diabetic patients,³¹ including those who are older than the patients in the present study, in whom the risk of stroke is increased. Other strategies may include aspirin therapy in any patient with atheromatous cardiovascular disease and prophylactic anticoagulation in those with cardiac failure or who have had a large anterior myocardial infarction.^21,46

Accepted for publication November 2, 1998.

This work has been supported in part by grants from the United Kingdom Medical Research Council, London; British Diabetic Association, London; the United Kingdom Department of Health, London; The National Eye Institute and The National Institute of Digestive, Diabetes, and Kidney Disease of the National Institutes of Health, Bethesda, Md; The British Heart Foundation, London; The Health Promotion Research Trust, London; Charles Wolfson Charitable Trust, London; The Alan and Babette Sainsbury Trust, London; The Clothworkers' Foundation, London, and The Oxford University Medical Research Fund Committee, Oxford. It has also been supported by grants from pharmaceutical companies, including Novo Nordisk A/S, Bagsværd, Denmark; Bayer plc, Newbury, England; Bristol-Myers Squibb Ltd, Hounslow, England; Hoechst Marion Roussel, Bridgewater, NJ; Eli Lilly and Co, Indianapolis, Ind; Lipha, Lyon, France; and Farmitalia Carlo Erba, St Albans, England. Additional assistance was provided by Boehringer Mannheim, Livingston, Scotland; Becton Dickinson, Oxford; Owen Mumford, Woodstock, England; Securicor, Sutton, England; Kodak, Hemel Hempstead, England; and Cortecs Diagnostics, Deeside, England.

The cooperation of the patients and many National Health Service staff and nonstaff at the centers is much appreciated.

Reprints: Philip Bassett, MA, University of Oxford, Diabetes Research Laboratories, Radcliffe Infirmary, Woodstock Road, Oxford OX2 6HE, England.