The high risk of stroke immediately after transient ischemic attack

Methods.

All residents of the Province of Alberta (population 3,009,900)19 are eligible for coverage under a publicly funded and universally available health-care system. The only exceptions are members of the Canadian Military and the Royal Canadian Mounted Police (less than 1% of the Alberta population) who receive health care coverage through the federal government. Each resident covered by the plan has a unique lifetime personal health number which can be used to link to a variety of data sources. We used four administrative data systems in this study. Hospital inpatient data were acquired from a nationally standardized collection of hospital morbidity data. Emergency department data were collected from the Ambulatory Care Classification System, which contains information on emergency department services, day surgeries, outpatient treatment programs/clinics, and other procedures that do not require overnight stay. These two data systems provide the information for the definition of cases, and the follow-up of cases through the health service system. Summary mortality data were assembled from the Alberta vital statistics death registration file. Our final source of data, a fee-for-service practitioner claims database, was used to differentiate first-ever compared to recurrent stroke events. Virtually all Alberta physicians bill the provincial government through this fee-for-service system.

TIA was defined by ICD-9-CM code 435.x. We isolated the first presentation to an Alberta emergency department by an Alberta resident in the 1999/2000 fiscal year. Since hospital admission is preceded by entry through the emergency department, these data include hospitalized patients. TIA diagnosed in a physician’s office without presentation to an emergency department was not included in this study due to the expected poor diagnostic accuracy of billing code data. However, based upon data from Texas, this could be expected to constitute 5% or less of all TIA.18 TIA patients with past stroke or TIA were defined by a past emergency department visit for cerebrovascular disease up to 3 years previously, as well as cases which had been hospitalized for cerebrovascular disease or for which any physician had billed a diagnosis of cerebrovascular disease in the previous 7 years. All other cases were defined as first-ever cases. Estimated age-adjusted rates of TIA per population were standardized using the direct method to the 1996 Canadian population.

Death was designated for both emergency department (which includes deaths that occur in contact with the ambulance service) and in-hospital events according to a death code contained within the data systems themselves. Additional mortality information was obtained from the provincial vital statistics registry. Discharge included discharges to home, nursing homes, and other long-term care facilities. Persons who were transferred from the initial site of inpatient care were further followed through the hospital inpatient or hospital outpatient system and categorized as discharged or dead according to their final separation status. Thus, an episode of service for all cases terminates upon discharge or death.

Stroke, TIA, and myocardial infarction after stroke were defined using hospital discharge data and the ambulatory care classification system database. Recurrent stroke is used to mean ischemic stroke, subarachnoid hemorrhage (SAH), or intracerebral hemorrhage (ICH) occurring after the index TIA. In diagnosing stroke and TIA we considered only ICD-9 codes 430, 431, 433–436; we included hemorrhage stroke types because intracranial hemorrhage is equally important to patients as an outcome event as ischemic stroke. Carotid endarterectomy was defined using procedure codes in the physician billing data. Individuals were characterized by age, sex, and rural versus urban place of residence according to postal code.20 We derived a proxy for socioeconomic status using information within a population health care registry which groups the population based on the payment of a health care premium (a quarterly fee associated with health care coverage). We defined persons of low socioeconomic status as those who had their health care premiums paid in part or in full by the government (persons on social assistance, persons of low income, and persons with treaty [aboriginal] status) and the remainder of the cases as non-low socioeconomic status. Diabetes mellitus was defined by a combination of claims and inpatient hospitalization data.21,22⇓ A person was considered to have diabetes if he or she had a minimum of two physician visits for diabetes over a 2-year period or a single hospitalization for diabetes in a single year. A person was identified as hypertensive if he or she had two or more diagnoses of hypertension in the past 7 years.

The authors had complete control of the design and analysis of this study independent of the funding sources. The study was conducted under the ethical and legal auspices of the Health Information Act of Alberta.

Results.

A total of 2,285 residents of Alberta had at least one TIA diagnosis in an Alberta emergency department, 1,475 of which were identified as first-ever cerebrovascular events (table). The emergency room, age-adjusted incidence was 68.2 per 100,000 (95% CI 65.3 to 70.9) for all TIA and 44.1 per 100,000 (95% CI 41.8 to 46.4) for first-ever TIA. The mean age of all cases was 71.4 (SD 13.8) and 48.9% were men. A minority of patients (n = 601, 26.3%) were admitted to hospital. Patients admitted to hospital were more likely to live in a rural locale (RR = 1.9, 95% CI 1.6 to 2.2) and be older than 65 years (RR = 1.5, 95% CI 1.2 to 1.8). Sex, hypertension, diabetes mellitus, and socioeconomic status were not different between admitted and non-admitted patients.

Stroke occurred in 346 (15.1%) patients within 1 year of TIA. Of these strokes, 342 (98.8%) were AIS, 3 (0.9%) were ICH, 1 (0.3%) was SAH, and 28 (8.1%) were deceased within 30 days of their stroke. The modeled risk of stroke was 1.4% (95% CI 1.0 to 1.8) at 2 days, 6.7% (95% CI 5.7 to 7.7) at 30 days, 9.5% (95% CI 8.3 to 10.7) at 90 days, and 14.5% (95% CI 12.8 to 16.2) at 1 year (figure 1). Half of the risk at 1 year was accrued at 38 days. Adjusted risk of stroke at 1 year was greatest among patients with hypertension (HR 1.54, 95% CI 1.22 to 1.94) or diabetes mellitus (HR 1.35, 95% CI 1.06 to 1.75) or who were older than 65 years (HR 1.33, 95% CI 1.04 to 1.71) but this was not evident at 2 days or 30 days. Hypertension showed the strongest quantitative relationship with stroke (see the table, figure 2). First-ever TIA patients were half as likely to have a stroke. Rural residents were less likely to have a stroke (HR 0.77, 95% CI 0.60 to 0.99). Admission to hospital was associated with a reduced risk of stroke (HR 0.73, 95% CI 0.57 to 0.95) (figure 3). Overall, the differences in risk among stratified groups according to comorbid risk factors were small.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1. Stroke-free survival. Survival function estimate = estimated proportion of the patients who are outcome free.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2. Covariate adjusted survival curves showing difference between stroke-free survival in diabetics, hypertensives, age > 65, rural-urban place of residence, first-ever cases, socioeconomic status (SES).

• Download figure
• Open in new tab
• Download powerpoint

Figure 3. Stroke-free survival after admission to hospital.

Recurrent TIA occurred in 2.5% (95% CI 2.0 to 3.0) at 2 days, 6.2% (95% CI 5.2 to 7.2) at 30 days, 7.9% (95% CI 6.8 to 9.0) at 90 days, and 12.3% (95% CI 10.9 to 13.7) at 1 year. The risks of recurrent TIA were not related to any of the risk factors studied (HR ∼ 1). Myocardial infarction occurred in 2.1% (95% CI 1.6 to 2.8) of patients. The risk of the combined outcome of stroke, myocardial infarction, or death was 21.8% (95% CI 20.0 to 23.6) at 1 year. Overall mortality at 1 year after TIA was 5.9%. Mortality was predicted only by age greater than 65 years (HR 7.99, 95% CI 4.28 to 14.93). Carotid endarterectomy was performed in 1.5% (95% CI 1.1 to 2.1) of patients. The median time from the TIA event to carotid endarterectomy was 78 days (interquartile range 15 to 130 days).

A total of 2,199 patients (96.2%) saw a second physician in Alberta within 1 year of their TIA and 86 (3.8%) did not. Of those who saw a physician, 272 of 1,927 (12.4%) had a stroke whereas among those who did not seek a physician, 74 of 86 (86.0%) had a stroke (RR 7.0, 95% CI 6.0 to 8.0). When time to review by a second physician was modeled as a time dependent covariate within the Cox proportional hazards model, stroke free survival dropped off rapidly. Examination of the risk of stroke beginning 1 week after TIA, when any events that occurred in the first week were ignored, showed that the late risk is similar between those who saw their personal physician early or late (figure 4). Time is therefore a confounder of the relationship between seeing a physician and stroke recurrence. This observation emphasizes the importance of time.

• Download figure
• Open in new tab
• Download powerpoint

Figure 4. Time to physician follow-up after TIA and risk of stroke. The top panel shows stroke-free survival from the date of TIA. The bottom panel shows stroke-free survival beginning 1 week after the TIA when events in the first week are excluded from the analysis. GP = general practitioner.

Discussion.

The unique aspect of our study is that it is a population-based study. Our analysis provides contemporary data to confirm and expand upon past population based estimates of TIA prognosis. The early risk of stroke was 6.7% at 30 days. This rate may be an underestimate because our study excluded patients who had recurrent events within the first day. For example, patients who came to the ER with a TIA were discharged and had a stroke later that same day would not have been identified by our methods. Such patients were common in a northern California study accounting for a 24-hour early stroke rate of 4.2% (personal communication, S. Claiborne Johnston, 2003). Some of these patients may represent what has been previously called fluctuating or progressing stroke. Accounting for this rate, our results are comparable to rates of recurrent stroke observed in northern California10 and to rates reported from the Oxfordshire Community Stroke Study.9

Early stroke occurs commonly after TIA but the 30-day risk in our study was not predicted by hypertension, diabetes, or age. Hypertension was a predictor of recurrent stroke at 90 days (p = 0.004) but this was not true of diabetes or age. These risk factors emerged only when the longer term 1-year risk was considered. Patients with a past TIA or stroke and patients who were unable to see a second physician after the emergency room encounter were more likely to have a stroke. However, this relationship is confounded by time; since the risk of stroke after TIA is related to time, a relationship must emerge between time-to-physician and stroke after TIA. The confounding effect emphasizes the importance of time in predicting early stroke but does not indicate how to prevent future stroke.

The long median delay from event to carotid endarterectomy and the relatively low proportion of patients who undergo endarterectomy are intriguing. Rapid endarterectomy after TIA may prevent stroke but may carry higher perioperative morbidity. Because one would expect 10 to 15% of TIA patients to have carotid stenosis as the mechanism of TIA, the low proportion of patients undergoing endarterectomy suggests potential underutilization of this procedure. These observations are less reliable because of the small numbers of carotid endarterectomies performed.

We were unable to confirm that diabetes mellitus and older age were predictors of early recurrent stroke, as was suggested in a previous study.10 Our study design prevented the assessment of TIA symptoms and the duration of those symptoms as risk factors for recurrence. However, as expected, the risk of stroke at 1 year was predicted by age, diabetes mellitus, and hypertensive status. These were small differences that may not be clinically significant at an individual patient level, but assume increasing importance when considering whole populations.

Admission to hospital after TIA was minimally protective in preventing stroke. Caution is warranted in interpreting this observation. The risk of stroke within 24 hours of TIA onset was not captured by our study and admission to hospital may have been highly effective in preventing some very early strokes. The late risk is likely to be the same between admitted and non-admitted patients because risk factors are similar.

Interestingly, approximately 1% of recurrent strokes were hemorrhages. The emergence of data on microbleeds observed on T2*-weighted MRI among stroke patients24–27⇓⇓⇓ leads to speculation that at least some patients with TIA have symptoms because of hemorrhage and not ischemia. Microbleeds may not be asymptomatic. While it is known that survivors of symptomatic ICH may be at least as likely to have recurrent ischemia as recurrent hemorrhage,28 an imaging study of TIA and stroke recurrence is needed to determine the role microbleeds may play in symptomatic TIA. Antithrombotic therapy, initiated after a TIA, might also be a contributing cause to subsequent hemorrhage. Intriguingly, there is an association between hemorrhage, antiplatelet therapy, and microbleeds.29

The age-adjusted incidence of TIA estimated from our case definition ranges between 44 and 68 per 100,000 population, estimates that match recently published rates from Belgium30 and Cincinnati, OH.31 However, these results reflect only TIA diagnosed in the emergency department and likely underestimate the true incidence. Nevertheless, these rates support the validity of our approach despite known limitations of ICD-9-CM coding such as suboptimal specificity, and the restriction of our case definition to emergency room diagnoses only. Unfortunately, these limitations may not be improved upon by the ICD-10.32 However, the apparent accuracy of our data may have resulted from counterbalancing biases. Although it is unknown in Alberta how many patients with TIA are seen by their family physician and not in the emergency room, we may have underestimated TIA by not capturing TIA diagnosed in a physician’s office and may have overestimated true TIA by using administrative data. Importantly, the key clinical issue remains the prognosis of TIA after the patient leaves the emergency room, regardless of whether the diagnosis is technically correct. Because timeliness is so critical, the urgent management of TIA will always depend upon the diagnostic impression of the emergency physician.

Limitations of our approach include those inherent in using passive surveillance with administrative data.18 However, a unique aspect of administrative data in Alberta is the Ambulatory Care Classification System, which captures emergency department visits. Previous studies of administrative data have not assessed such a system.18 It is likely therefore that only cases occurring in the community with presentation to a family physician were missed by our case definition. Most importantly, a previous study has confirmed that only 5.6% of ICD-9 diagnoses of TIA could be refuted based upon chart review and among these patients three individuals had a stroke. The risk of stroke was not different after exclusion of patients with a TIA diagnosis refuted by neurologist chart review.10 Furthermore, further evidence has confirmed that ICD-9 codes for stroke using the first diagnosis position have an 88% positive predictive value compared to chart review,33 suggesting that our recorded outcome events were true events.

The Alberta population is an open population. Approximately 3.7% of the population emigrated from the Province of Alberta during the study period. A majority of these people emigrated within Canada.19 Patients who had recurrent events, myocardial infarction, or death or who underwent carotid endarterectomy in another province would not have been captured by our methods. Similarly, patients who emigrated to Alberta and had a TIA in Alberta may have been misclassified as first-ever cases when they had previously had events in another jurisdiction. Because emigration and immigration made up a small proportion of the total population and because there is no reason to believe that migration patterns are different among patients with and without TIA, we believe it unlikely that this effect significantly biased our results.

Importantly, we could not assess the severity of recurrent stroke from administrative data. However, mortality at 30 days among the recurrent strokes was 8.1%. Stroke can be devastating even in its apparently minor forms. Our results confirm that the risk of stroke after TIA is high but contradict previous estimates of important risk factors for early stroke occurrence. We support the recent emphasis on TIA as an emergency.34 Rapid investigation and treatment of patients with TIA as a strategy to prevent stroke is current priority for stroke research. Models to identify those at particular risk are needed to rationalize resource allocation to those patients at highest risk.