Duration of Diabetes and Risk of Ischemic Stroke: The Northern Manhattan Study

Abstract

Background and Purpose

Diabetes increases stroke risk, but whether diabetes status immediately prior to stroke improves prediction, and whether duration is important, are less clear. We hypothesized that diabetes duration independently predicts ischemic stroke.

Methods

Among 3,298 stroke-free participants in the Northern Manhattan Study (NOMAS), baseline diabetes and age at diagnosis were determined. Incident diabetes was assessed annually (median=9 years). Cox proportional hazard models were used to estimate hazard ratios and 95% confidence intervals (HR, 95% CI) for incident ischemic stroke using baseline diabetes, diabetes as a time-dependent covariate, and duration of diabetes as a time-varying covariate; models were adjusted for demographic and cardiovascular risk factors.

Results

Mean age was 69±10 years (52% Hispanic, 21% white, and 24% black); 22% were diabetic at baseline and 10% developed diabetes. There were 244 ischemic strokes, and both baseline diabetes (HR 2.5, 95% CI 1.9-3.3) and diabetes considered as a time-dependent covariate (HR 2.4, 95% CI 1.8-3.2) were similarly associated with stroke risk. Duration of diabetes was associated with ischemic stroke (adjusted HR=1.03 per year with diabetes, 95% CI=1.02-1.04). Compared to non-diabetic participants, those with diabetes for 0-5 years (adjusted HR=1.7, 95% CI=1.1-2.7), 5-10 years (adjusted HR=1.8, 95% CI=1.1-3.0), and ≥10 years (adjusted HR=3.2, 95% CI=2.4-4.5) were at increased risk.

Conclusion

Duration of diabetes is independently associated with ischemic stroke risk adjusting for risk factors. The risk increases 3% each year, and triples with diabetes ≥10 years.

INTRODUCTION

Diabetes mellitus (DM) is a major public health burden. In the US, more than 23.6 million people have diabetes.¹ Prevalence is increasing as the population ages, and certain populations, such as minority groups, are more vulnerable. The number of people diagnosed with diabetes is estimated to increase 165% between 2000 and 2050². The independent association of diabetes with stroke is now well documented^3-10. Notably, almost all prospective cohort studies have evaluated diabetes as an exposure at baseline, or the time of participant enrollment, and measured its effect on stroke outcomes during follow-up. In large cohorts with long periods of follow-up, a significant proportion of participants who were non-diabetic at baseline become diabetic over the period of follow up¹¹. Taking this change in status into consideration may be expected to provide a more precise estimate of stroke risk.

While some studies have assessed the relationship of duration of diabetes to risk of cardiovascular outcomes such as coronary heart disease, cardiovascular mortality, peripheral arterial disease, carotid wall thickness, and thin cap fibroatheroma^12-18, the association between duration of diabetes and stroke risk is less well studied.

The Northern Manhattan Study (NOMAS) prospective cohort, with its annual evaluation of diabetes diagnosis, provides an opportunity to investigate the utility of incorporating serial interval evaluations of diabetes into risk prediction estimates. We investigated the effect of updated diabetes status on stroke by incorporating diabetes as a time-dependent covariate. We further examined the effect of duration of diabetes on ischemic stroke risk.

METHODS

NOMAS is a prospective population-based cohort study, designed to determine stroke incidence, risk factors, and prognosis in an urban multiethnic population. Northern Manhattan is defined as the area in New York City north of 145th Street, south of 218th Street, bound on the west by the Hudson River and separated from the Bronx on the east by the Harlem River. The cohort has a racial/ethnic mixture consisting of 52.3% Hispanic, 24.3% non-Hispanic black, and 20.9% non-Hispanic white residents.

Selection of Prospective Cohort

The study has been previously described in detail.^19-22 Briefly, community participants were eligible for enrollment if they (1) had never been diagnosed with a stroke, (2) were ≥40 years of age, and (3) resided for ≥3 months in a household with a telephone in northern Manhattan. Subjects were identified with random digit dialing employing dual-frame sampling to identify both published and unpublished phone numbers. The protocol was approved by the Institutional Review Board at Columbia University Medical Center and the Miller School of Medicine, University of Miami, and participants provided informed consent.

Baseline Evaluation

Baseline data were collected via interviews by trained research assistants, medical record review, physical and neurologic examination by study investigators, in-person measurements, and collection of fasting blood specimens for glucose and lipid measurements. A standardized questionnaire was adapted from the Behavioral Risk Factor Surveillance System²³ developed by the Centers for Disease Control and Prevention regarding the following conditions: diabetes, hypertension, hypercholesterolemia, smoking, peripheral vascular disease, transient ischemic attack, and cardiac disease (including angina, myocardial infarction, coronary artery disease, atrial fibrillation, and valvular heart disease).

Interval Evaluation

All participants were prospectively followed annually through telephone interviews, and mean duration of follow-up at time of analysis was 9.0 ± 3.7 years. The yearly contact rate was 99%. Subjects were interviewed to determine changes in vital status, detect cardiac and neurologic symptoms and events, and review any hospitalizations. The phone assessment served as a screen for vascular events. The simple stroke question (“Since your last contact have you been diagnosed with a stroke?”) during telephone interview was 92% sensitive and 95% specific with in-person assessment, physician interviews, medical records, and neuroimaging data used as the gold standards²¹. Participants with affirmative responses to neurological symptoms underwent examination and review by a study neurologist or had medical records reviewed. Hospital surveillance of admission and discharge were performed to provide data that may have been missed during the annual telephone follow-up.

Measure of Exposure

Diabetes at baseline was defined if a participant reported a history of medical diagnosis of diabetes mellitus or treatment with oral hypoglycemic agents or insulin. In addition, fasting blood glucose (FBG) ≥126 mg/dl (6.5mmol/L) was used among those who did not self-report diabetes to adjudicate diabetic status and find “unaware” cases. FBG was measured using a Hitachi 747 automated spectrometer (Boehringer, Indianapolis, Indiana). Age at time of diagnosis was also recorded for those with self-report of diabetes at baseline, and diabetes duration was calculated. During follow-up evaluations among non-diabetic participants, the first follow-up contact at which there was self-report of new diagnosis of diabetes, treatment with oral hypoglycemic drugs, or insulin therapy was used to define conversion to diabetes during follow-up.

Duration of diabetes was calculated from the onset of diabetes up to the date of ischemic stroke or censoring.

Validation of diabetes status determined during follow-up

There were 74 instances in which a participant reported treatment with anti-diabetic drugs or insulin without ever reporting the diagnosis of diabetes. Medical records were reviewed to validate these responses. This self-report of medications or insulin without report of the diagnosis of diabetes was consistent with a diagnosis of diabetes in 92% of cases. We reflected these changes in the analyses.

Similarly, there were 87 participants who self-reported the diagnosis of diabetes once, but never reported any subsequent treatment. All these cases were confirmed to be diabetic on medical record review.

To avoid potential bias caused by intensive medical record review selectively among those with possible diabetes conversion during follow-up, and to validate the non-report of diabetes diagnosis or treatment among the remainder of the cohort, medical record review of a random computer generated list of 50 participants who never reported diabetes during follow-up was done. Two participants carried the diagnosis in the medical record but had not been detected in interview that year. On further review, it was determined that these participants did not report the diagnosis because it was “diet controlled diabetes” at that time, and they reported the diagnosis on subsequent follow-up once they started taking medication.

Measurement of Outcome

Stroke was defined during follow-up by the first symptomatic occurrence of any type of stroke including intracerebral hemorrhage, subarachnoid hemorrhage, and cerebral infarction, as previously described¹⁹. Medical records were reviewed to verify details ofsuspected events. At least two neurologists reviewed data independently and classified strokes. Any disagreements were adjudicated by the principal investigators (RLS/ME). We used only ischemic strokes for current analyses.

STATISTICAL ANALYSES

All analyses were performed using SAS version 9.1 (SAS Institute, Cary, North Carolina). The distributions of diabetes at baseline were calculated, both overall and by subject characteristics, including demographics and risk factors. Cox proportional hazard regression models were fitted to calculate hazard ratios and 95% CI (HR, 95% CI) for ischemic stroke as the outcome. Main predictors were 1) baseline diabetes, 2) diabetes as time-varying covariate (incorporating new-onset diabetes during follow-up), and 3) duration of diabetes as time-varying covariate. We used duration of diabetes as a continuous measure as well as categorized at 5 and 10 years in order to examine for threshold effects.

Models unadjusted and adjusted for demographic factors (age, sex, race-ethnicity, insurance status, and educational level), and behavioral and medical risk factors (hypertension, cardiac disease, high-density lipoprotein (HDL), low-density lipoprotein (LDL), current smoking, past smoking, alcohol consumption, waist circumference, and physical activity) were constructed. Assessment for two-way interactions was conducted. We compared the Akaike Information Criterion (AIC) among the baseline, and the time-varying diabetes models.

RESULTS

Baseline characteristics

Baseline characteristics of the cohort are shown in Table 1. The mean age was 69±10 years; 62.8% of the cohort were women, 52.3% Hispanic, 20.9% non-Hispanic white, and 24.3% non-Hispanic black.

Table 1.

Baseline socio-demographic and cardiovascular risk factors in the Northern Manhattan Study

Sociodemographic and cardiovascular risk factors (N (%) or Mean ± SD)		Total	Diabetic at enrolment	Non-diabetic at enrolment	Persistently non-diabetic	Newly diabetic during follow-up	p
N		3298	716 (21.7%)	2582 (78.2%)	2244 (68.0%)	338 (10.2%)
Age (years)		69 ± 10	69 ± 8	69 ± 11	70 ± 11	66 ± 9	0.43
Sex	Women	2071 (62.8%)	438 (61.2%)	1633 (63.2%)	1425 (63.5%)	208 (61.5%)
	Men	1227 (37.2%)	278 (38.8%)	949 (36.8%)	819 (36.5%)	130 (38.5%)	0.31
Race/Ethnicity	White	690 (20.9%)	100 (14.0%)	590 (22.9%)	556 (24.8%)	34 (10.1%)	Ref
	Black	803 (24.3%)	196 (27.4%)	607 (23.5%)	552 (24.6%)	55 (16.3%)	< .0001
	Hispanic	1726 (52.3%)	408 (57.0%)	1318 (51.0%)	1076 (48.0%)	242 (71.6%)	< .0001
Completed High School Education		1511 (45.8%)	282 (39.4%)	1229 (47.6%)	1107 (49.3%)	122 (36.1%)	< .0001
Insurance	Medicaid/No insurance	1435 (43.5%)	375 (52.4%)	1060 (41.1%)	870 (38.8%)	190 (56.2%)	< .0001
	Medicare/Private Insurance	1841 (55.8%)	335 (46.8%)	1506 (58.3%)	1358 (60.5%)	148 (43.8%)	Ref
Smoking Status	None	1548 (46.9%)	317 (44.3%)	1231 (47.7%)	1076 (48.0%)	155 (45.9%)	Ref
	Past	1179 (35.7%)	273 (38.1%)	906 (35.1%)	787 (35.1%)	119 (35.2%)	0.09
	Current	569 (17.3%)	126 (17.6%)	443 (17.2%)	379 (16.9%)	64 (18.9%)	0.42
Moderate Alcohol Consumption		1075 (32.6%)	181 (25.3%)	894 (34.6%)	796 (35.5%)	98 (29.0%)	< .0001
Any Physical Activity		1909 (57.9%)	380 (53.1%)	1529 (59.2%)	1352 (60.2%)	177 (52.4%)	0.0027
Cardiac Disease		792 (24.0%)	210 (29.3%)	582 (22.5%)	500 (22.3%)	82 (24.3%)	0.0002
Waist circumference (inches)		36.8 ± 5.0	38.4 ± 5.0	36.3 ± 4.9	36 ± 5.0	38.4 ± 4.6	< .0001
High density lipoprotein (mg/dL)		46.8 ± 14.6	43.7 ± 14.0	47.6 ± 14.6	48.4 ± 14.7	42.7 ± 13.3	< .0001
Low density lipoprotein (mg/dL)		129.1 ± 36.1	126.3 ± 39.1	130.0 ± 35.3	129.8 ± 35.3	130.8 ± 35.0	0.02
Systolic Blood Pressure (mm Hg)		143.7 ± 21.0	146.7 ± 20.0	142.9 ± 21.3	142.7 ± 21.2	144.2 ± 22.0	< .0001

SD=Standard deviation

* Baseline diabetics versus baseline non-diabetics

At baseline, 574 participants (17.4%) self-reported diabetes and 142 subjects had FBG>126 mg/dl, for a total of 716 (21.8%) with diabetes at baseline. Approximately 93-96% of participants visited their primary care physician at least once during the prior year for each year of follow up. Among those who were non-diabetic at baseline (n=2582), 338 subjects (13.1%) reported new-onset diabetes during a mean 9.0 years of follow-up.

Diabetes at baseline and as a time-dependent covariate

There were 244 incident ischemic strokes. Baseline diabetes was associated with risk of stroke (unadjusted HR 2.6, 95% CI 2.0-3.3). In the fully adjusted model, adjusted for demographic and other cardiovascular risk factors including smoking, alcohol consumption, LDL, HDL, blood pressure, waist circumference, history of cardiac disease, and physical activity, the association was unchanged (adjusted HR 2.5, 95% CI 1.9-3.3) (Table 2). When new-onset diabetes was taken into account as a time-varying covariate, diabetes was still associated with risk of ischemic stroke (adjusted HR 2.4, 95% CI 1.8-3.2). The AICs for the models were similar (3333.1 for the model using baseline diabetes and 3335.2 in the model using diabetes as a time-dependent covariate).

Table 2.

Risk of ischemic stroke associated with baseline diabetes and diabetes as time-dependent covariate.

Models	Hazard ratio	95% Confidence interval
Baseline diabetes
Unadjusted (Diabetes only)	2.6	2.0, 3.3
Adjusted for demographic variables*	2.7	2.1, 3.5
Adjusted for demographic variables and cardiovascular risk factors†	2.5	1.9, 3.3
Diabetes as time-dependent covariate
Unadjusted (Diabetes only)	2.5	1.9, 3.2
Adjusted for demographic variables*	2.6	2.0, 3.4
Adjusted for demographic variables and cardiovascular risk factors†	2.4	1.8, 3.2

^*age, sex, education, race-ethnicity, and insurance.

^†age, sex, education, race-ethnicity, insurance, waist circumference, alcohol consumption, smoking status, physical activity, HDL, LDL, history of cardiac disease and systolic blood pressure.

There were no interactions between diabetes and age, race-ethnicity or sex.

Duration of Diabetes

The mean duration of diabetes among people who self-reported diabetes at baseline was 17.3 ± 11.6 years (median 13.7 years). Among the 338 subjects diagnosed with diabetes during follow-up, mean duration was 4.5 ± 3.2 years (median 4.2 years). With each year of diabetes, stroke risk rose by 3% (adjusted HR per year 1.03, 95% CI 1.02-1.04).

Duration was also categorized as <5 years, 5-10 years, and ≥10 years, with non-diabetic participants as a reference group. Without assuming linearity, the trichotomized duration of diabetes variable was fitted adjusting for other risk factors. The null hypothesis that all three groups had the same risk of stroke was rejected (Chi-square test with 2 degrees of freedom, p=0.01). Compared to non-diabetics, those with diabetes <5 years (adjusted HR 1.7, 95% CI 1.1-2.7), 5-10 years (adjusted HR 1.8, 95% CI 1.1-3.0), and ≥10 years (adjusted HR 3.2, 95% CI 2.4-4.5) had an increased risk of stroke (Table 3 and Figure).

Table 3.

Risk of ischemic stroke associated with duration of diabetes.

Diabetes duration	Hazard ratio*	95% Confidence interval	p value
Continuous model
per year	1.03	1.02, 1.04	<.0001
Categorical model†
≤5 years	1.72	1.09, 2.71	.02
5-10 years	1.83	1.13, 2.97	.01
≥ 10 years	3.23	2.36, 4.51	<.0001

^*Adjusted for age, sex, race-ethnicity, education, insurance, waist circumference, smoking status, alcohol consumption, physical activity, systolic blood pressure, history of cardiac disease, LDL cholesterol, HDL cholesterol.

^†Non-diabetic participants as reference group.

Figure.

Risk of ischemic stroke and duration of diabetes

DISCUSSION

In this prospective cohort, diabetes at time of enrolment was associated with ischemic stroke risk, consistent with estimates of association reported in other studies, varying from 1.3 to 4.0^{3-10, 24}. Contrary to our hypothesis, however, the magnitude of the association for diabetes with stroke risk was no different when we included diabetes as a time-dependent covariate. In traditional epidemiological analyses, the use of only baseline assessments of a risk factor, such as diabetes, could potentially bias study results towards the null. Our findings suggest that there is marginal incremental value to including further assessments of diabetes during follow-up in analyses of its effect on stroke risk. One practical implication for future epidemiological studies would be potential cost savings through avoiding additional lengthy interviews and assessments of risk factors during follow-up. Whether these findings are transferable to other risk factors and cardiovascular outcomes is not clear from these analyses.

There are several possible explanations for this absence of additional information from follow-up assessments in our analyses. The cardiovascular risk factor burden carried by the participants at enrolment (mean age 69 years) could already be high enough that development of diabetes during follow-up does not confer added information. Second, subjects newly diagnosed with diabetes may be more compliant with treatment, which has been shown to be beneficial for primary stroke prevention in our cohort²⁵. Third, the median duration of follow-up for baseline diabetics in our cohort was 13.7 years, but it was only 4.2 years for those who developed diabetes after baseline, which may not be sufficient to manifest cerebrovascular events. Lastly, we used both self-report and laboratory results to identify diabetes at baseline. However, self-report alone was used to define diabetes during telephone follow-up, which may have led us to miss cases, as nearly a third of people with diabetes may be undiagnosed²⁶.

Among those with diabetes ≥10 years, risk of ischemic stroke is 3 times the risk among those without diabetes. Our study provides evidence that the risk of ischemic stroke increased continuously with duration of diabetes mellitus. The increase is not as much during the second half of the first decade, but rises steeply as the disease enters its second decade. This must however be interpreted keeping in mind that true onset of diabetes may be 4-7 years earlier than clinical diagnosis²⁷.

This is the first prospective cohort study to address the association of diabetes duration and ischemic stroke among both men and women. The Nurses’ Health study reported an association between diabetes duration and various stroke subtypes among women³, where the risk of ischemic stroke rose from 1.5 (0-4years) to 4.1 (>20 years). The maximum jump in the risk was seen at the 10 year mark, similar to our findings. Our cohort has men and women ≥40 years, and representation by Hispanic, white and black participants, as compared to the Nurses’ Health study cohort, comprised of predominantly white women ages 30-55 years at the time of enrollment.

Several potential mechanisms could explain the association of diabetes duration and stroke in our study. There is evidence of association between diabetes duration and atherosclerotic lesions, including intimal medial thickness and thin cap fibroatheromas^13,16. Carotid plaque thickness has been shown to predict ischemic stroke in our cohort²⁸. In addition, hypertension is twice as prevalent among diabetics as in people without diabetes²⁹, and long term hypertension causes accelerated microvascular and macrovascular complications among diabetics²⁹. The risk of microalbuminuria has been shown to increase with increasing duration of diabetes^{30, 31} and microalbuminuria has been reported as a strong and independent risk factor of stroke among patients with diabetes³². Other potential mediators may be endothelial dysfunction³³, and abnormalities in fibrinogen and clotting mechanisms^{34, 35}.

Our study has public health implications. Although stroke rates have been declining among diabetics³⁶, the rapid increase in diabetes incidence over the same period is leading to a higher overall stroke burden³⁶. In recent decades, the age of onset of type 2 diabetes has decreased, paralleling the obesity epidemic in young adults³⁷. As the population ages and the elderly live longer, more and more people will live with longer duration of the disease. It is thus important to better understand the dynamics between diabetes, time, and stroke, and emphasize the importance of interventions to prevent early diabetes. Minimising the number of years a patient has diabetes would help combat the rise in stroke risk with each year of the disease.

Our study has several strengths. NOMAS is designed to focus on risk factors for stroke in whites, blacks, and Hispanics living in the same community. The study has a large sample size, long duration of follow-up, minimal loss to follow-up, and detailed information on potential confounding factors. Our study design also allowed us to use diabetes as a time-dependent covariate to model risk of stroke, and study the relationship between diabetes duration and ischemic stroke among both men and women.

However, this study is not without limitations. First, because we were limited to using self-report to determine diabetes during follow up, we may have misclassified ‘unaware’ diabetics as non-diabetic, leading to a bias towards the null. Use of quantifiable measures of glycemic status such as FBG or hemoglobin A1c as time dependent covariates may have added additional prognostic information. On the other hand, our cohort may be atypical in that there was a high degree of follow up with primary physicians (93-96% annually) which may have led to a higher likelihood of diagnosis with diabetes. However, while we do have data collected during the annual follow-up interview on visits to primary care doctors, we do not have data on whether diabetes screening occurred during those visits. Second, the duration of follow-up may not have been long enough to bring out the difference between baseline and time-dependent models. Third, duration of diabetes at baseline was calculated based on participants’ self reported age of onset, which is vulnerable to inaccuracy, as there is a lag time between onset and diagnosis²⁷. The apparent threshold of ten years identified in this study, therefore, may be an underestimate. The analysis of duration and stroke is complicated by the fact that longer duration is associated with older age, and residual confounding cannot be ruled out. We did not have sufficient numbers to detect interaction by sex, age, and race/ethnicity, and thus cannot comment on the differential effect of diabetes or its duration in segments of the cohort. However, our findings are in agreement with other large population based cohorts.

In conclusion, use of diabetes as a time dependent covariate adds little incremental value to using diabetes at baseline as a risk factor for stroke. Duration of diabetes, however, increases the risk of ischemic stroke, independent of coexisting risk factors. As more people become diabetic earlier and live longer, this relationship assumes public health importance and warrants steps to institute long-standing and sustainable lifestyle changes for primary prevention, and appropriate long term management after diagnosis.