Prior transient ischemic attack is independently associated with lesser in-hospital case fatality in acute stroke


*Judit Zsuga, MD, PhD, Department of Neurology, Faculty of Medicine, University of Debrecen, H-4012 Debrecen, PO Box 31, Hungary. Email:


Aim:  Ischemic preconditioning has been well established in healthy human hearts, but limited information is available about its occurrence or its integrity in the brain. The aim of the present study was therefore to investigate whether a prior cerebral ischemic episode (stroke or transient ischemic attack [TIA]) is able to confer protection against ischemic stroke, reflected by in-hospital case fatality.

Methods:  A total of 2874 acute stroke patients included in the prospective, hospital-based Debrecen Stroke Database were studied, of whom 673 had previous stroke and 195 had prior TIA.

Results:  Following adjustment for active confounders, TIA but not stroke in the history was associated with decreased odds for in-hospital case fatality (odds ratio, 0.53; 95% confidence interval: 0.29–0.98; P = 0.041). The fitness of the final multiple regression model was good (Hosmer–Lemeshow goodness-of-fit χ2 statistic (P = 0.328).

Conclusion:  TIA may have an ischemic preconditioning effect in the human brain.

THERE IS ONGOING debate concerning cerebral ischemic preconditioning (IP) in humans conferred by a transient ischemic attack (TIA). Although first described by Dahl and Balfour in 1964,1 attempts to define its significance started only after promising results concerning myocardial preconditioning were reported. IP is an adaptive response in which a brief exposure to ischemia markedly enhances the organ's (brain, heart) ability to withstand a successive sustained potentially injurious ischemic episode.2 While angina pectoris antecedent of myocardial infarction is perceived to be a clinical correlate of experimental preconditioning stimuli,3,4 whether TIA, defined analogously to angina as ischemia-induced functional deterioration in the absence of structural deficit, is a risk or a protecting factor is yet to be determined. Some uncertainty may be attributed to confounders inherent to strokes following earlier TIA because large vessel atherosclerosis associated with better outcome is frequently overrepresented, while the potentially more severe, frequently unheralded cardioembolic strokes are underrepresented in patient cohorts5,6. To the best of our knowledge only six studies have dealt with the question of cerebral ischemic preconditioning provided by prior TIA in humans.5–10

Out of these considerations we set out to answer the question of whether there is an association between in-hospital case fatality of ischemic stroke patients and the history of TIA at an undefined time in the past, by analyzing a substantial cohort of patients included in the large-scale, single-center Debrecen Stroke Database.


The Debrecen Stroke Database is a prospective, hospital-based registry, covering the patient population of Debrecen city (estimated population 230 000). Approximately 60% of all stroke patients in the catchment area are referred to the Stroke Unit at the Department of Neurology, University of Debrecen.11 All consecutive patients admitted to the Stroke Unit between 1996 and 2000 were entered into the registry, totaling 3755 records.12


The present investigations were conducted in accordance with the Declaration of Helsinki (as revised in Edinburgh 2000). In order to provide further support for or against the theory of cerebral IP conferred by prior TIA we selected in-hospital case fatality (some authors refer to in-hospital case fatality as in-hospital mortality) as the primary end-point rather than the commonly used global assessment variables, for example the National Institutes of Health (NIH) stroke scale or the Barthel index, which encompass subjective elements, thus introducing observational bias and inter-rater variability.

We defined TIA as an acute focal loss of cerebral or ocular function presumed to be of ischemic origin that resolved within 24 h.13 In order to minimize reporting bias (because patients with mild stroke more readily recall previous TIA) information concerning the presence or absence of TIA in the case history was verified using previous medical records.

On admission risk factors including hypertension, diabetes, peripheral artery disease, atrial fibrillation, and smoking status were recorded. Next, each patient was examined by a stroke neurologist, and Mathew score (a global assessment variable characterizing stroke severity) was determined.14 Blood pressure readings, 12-lead electrocardiogram, and routine laboratory tests, including serum glucose, cholesterol and triglyceride level were also obtained. B-mode or duplex ultrasound examination of the internal carotid artery (ICA) was also performed in most patients during their hospital stay. The diameter of the carotid artery was measured at the site of maximal stenosis in the extracranial ICA according to the European Carotid Surgery Trial method.15 The status of the ICA was characterized as follows: normal, no abnormality; intimasclerosis, <30% stenosis; carotid artery stenosis, stenosis between 30% and 99%; and carotid artery occlusion, stenosis >99%.

Statistical analysis

For the sake of comparing the baseline characteristics of patients without any previous ischemic event or with prior TIA and stroke, the normality of all continuous variables was checked using the Shapiro–Wilk test. In the case of normality, analysis of variance (anova) was used. Categorical variables were compared using Pearson χ2 test.

Exploratory logistic regression was done to obtain the best multiple model identifying the factors linked to in-hospital case fatality (primary end-point concerning stroke outcome) associated with IP and estimating the effects thereof, expressed as odds ratios (OR). The initial multiple model was based on variables identified a priori on the basis of critical evaluation of available scientific literature and by simple modeling (change of OR at significance <0.05). Accordingly the following variables were included: history of prior TIA or stroke, age, sex, arterial hypertension, diabetes, smoking, peripheral artery disease, atrial fibrillation, risk factors modifiable on admission (systemic blood pressure, and serum glucose level) and serum cholesterol, and triglyceride level determined on admission. Variables with many missing values (>100) were indicator coded. The fit of the final model, including only the explanatory variable and the active adjustment factors, was assessed using Hosmer–Lemeshow goodness-of-fit χ2 test, and models were regarded significant at P ≥ 0.05. Statistical analysis was performed using Stata 8.2 (Stata, College Station, Texas, USA).


The patients' fate was as follows: of the 3755 patients we excluded those who were admitted because of TIA (n = 332) according to the aforementioned definition or who had intracerebral or subarachnoidal hemorrhage verified on computed tomography (n = 475, n = 71, respectively). All ischemic stroke patients were included in the analysis regardless of stroke severity on admission characterized by Mathew score. (For three patients included in the database the information concerning the type of stroke was missing, therefore these patients were excluded from further analysis.)

Of the 2874 patients 2006 had no prior ischemic episode (‘first ever’ group), 673 had prior stroke (prior stroke group) and 195 had prior TIA (prior TIA group). Patients with prior TIA had significantly higher Mathew scores, indicating less pronounced severity on admission (first-ever group, 70.60 ± 17.84; prior TIA, 64.66 ± 20.69; prior stroke, 75.13 ± 16.38; P < 0.001). Additional baseline characteristics of patients with or without antecedent ischemic events are shown in Table 1.

Table 1.  Baseline characteristics of ischemic stroke patients from the Debrecen Prospective Stroke Database
FeatureFirst ever (n = 2006)Prior stroke (n = 673)Prior TIA (n = 195)P
  1. All continuous data passed the equal variance test, thus comparisons were performed using one-way analysis of variance.

  2. ICA, internal carotid artery; NS, not significant; TIA, transient ischemic attack.

In-hospital mortality: deceased (%)13.3620.957.69<0.001
Age (years) (mean ± SD)67.70 ± 13.0669.87 ± 11.1568.02 ± 12.430.006
% male51.3057.2155.900.020
Risk factors (%)
Treated arterial hypertension41.0857.3650.77<0.001
Untreated arterial hypertension15.0511.8916.92 
Peripheral arterial disease14.2116.0711.79NS
Atrial fibrillation11.4316.2018.56<0.001
Status of ICA (%)   <0.001
Carotid artery stenosis15.1117.6614.53 
Carotid artery occlusion6.0112.275.81 
Smoking status (%)   NS
1–10 cigarettes/day10.7510.639.94 
11–20 cigarettes/day13.2210.8310.56 
21–40 cigarettes/day6.675.632.48 
>40 cigarettes/day1.851.252.48 
On-admission parameters (mean ± SD)
Mathew score70.60 ± 17.8464.66 ± 20.6975.13 ± 16.38<0.001
Systolic blood pressure (mmHg)156.29 ± 25.89155.08 ± 27.06155.69 ± 25.76NS
Diastolic blood pressure (mmHg)89.32 ± 12.1789.30 ± 11.9089.28 ± 11.28NS
Serum glucose (mmol/L)7.02 ± 3.046.82 ± 2.876.63 ± 2.610.011
Serum cholesterol (mmol/L)6.09 ± 1.365.88 ± 1.335.85 ± 1.24NS
Serum triglyceride (mmol/L)1.76 ± 1.261.67 ± 1.011.64 ± 0.82<0.001

A possible impact of IP on in-hospital case fatality is shown in Table 2. We found that ischemic stroke patients experiencing previous TIA had a lower proportion of in-hospital case fatality than patients with prior stroke or no heralding ischemic episode (P < 0.001).

Table 2.  In-hospital case-fatality in relation to prior TIA or stroke or the absence of any previous ischemic event
In-hospital mortalityFirst everPrior strokePrior TIAP
  1. TIA, transient ischemic attack.

Deceased n (%)268 (13.36)141 (20.95)15 (7.69)<0.001
Survived n (%)1738 (86.64)532 (79.05)180 (92.31)

The unadjusted OR for in-hospital case fatality derived from simple logistic regression was 0.54 (95% confidence interval [CI]: 0.31–0.93, P = 0.026) for the prior TIA group, and 1.708 (95%CI: 1.37–2.15, P < 0.001) for the prior stroke group. Further details of the simple models are shown in Table 3.

Table 3.  Factors associated with in-hospital case fatality: Simple logistic regression model
Lower limitUpper limit
  1. CI, confidence interval; ICA, internal carotid artery; OR, odds ratio; TIA, transient ischemic attack.

Prior stroke1.721.372.15<0.001
Prior TIA0.540.310.930.026
Age (years)1.0451.0351.055<0.001
Risk factors
Treated arterial hypertension1.140.911.420.243
Untreated arterial hypertension0.580.400.840.004
Peripheral arterial disease1.110.841.480.479
Atrial fibrillation2.021.552.63<0.001
Status of ICA   <0.001
Carotid artery stenosis2.681.245.800.012
Carotid artery occlusion5.492.4912.12<0.001
Smoking status   0.13
1–10 cigarettes/day0.930.611.420.753
11–20 cigarettes/day0.720.481.110.134
21–40 cigarettes/day0.730.411.320.304
>40 cigarettes/day2.060.974.400.061
On-admission parameters
Mathew score0.920.910.93<0.001
Systolic blood pressure (mmHg)1.000.991.000.78
Diastolic blood pressure (mmHg)0.990.981.000.519
Serum glucose (mmol/L)<0.001
Serum cholesterol (mmol/L)0.840.750.940.002
Serum triglyceride (mmol/L)0.710.580.871.001

Following the adjustment for confounding factors the OR remained practically unchanged (OR, 0.53; 95%CI: 0.29–0.98, P = 0.041) for the prior TIA group, but it decreased and lost its statistical significance in the prior stroke group (OR, 1.26; 95%CI: 0.95–1.67, P = 0.104). The final model proved to be significant when assessed using Hosmer–Lemeshow goodness-of-fit χ2 statistic (P = 0.328).

On basis of the final model presented in Table 4, we found that older age involves elevated in-hospital case fatality odds, thus patients are characterized by an extra 3.2% of odds with each year increase. Male gender also incurs increased odds in this final model. Additionally the OR for the elevation of serum glucose level by 1 mmol/L was 1.42 (95%CI: 1.29–1.55; P < 0.001). Finally, stroke evolution on basis of cardiac embolization seemed to be more fatal than that formed on the basis of large vessel atherosclerosis.

Table 4.  Factors associated with in-hospital case fatality: Multiple logistic regression model
Lower limitUpper limit
  • Hosmer–Lemeshow goodness of fit χ2 test statistic.

  • CI, confidence interval; OR, odds ratio; TIA, transient ischemic attack.

Prior TIA0.530.290.970.041
Prior stroke1.260.951.670.104
Atrial fibrillation1.441.042.000.03
Treated arterial hypertension1.050.801.380.71
Untreated arterial hypertension0.570.370.890.013
Carotid artery stenosis0.180.0540.620.006
Carotid artery occlusion0.330.0961.140.081
1–10 cigarettes/day1.130.691.840.62
11–20 cigarettes/day1.090.661.820.728
21–40 cigarettes/day1.000.502.000.985
>40 cigarettes/day2.140.795.800.132
Serum glucose mmol/L1.421.291.56<0.001
Serum triglyceride mmol/L0.750.600.940.011


The major finding of the present study is that TIA in the past might induce IP, resulting in decreased case fatality in acute stroke. We found a robust association between TIA before stroke and decreased in-hospital case fatality.

The present finding that a prior TIA, by functioning as an ischemic challenge, improves outcome after a subsequent ischemic stroke is in agreement with the majority of previous studies.6–9 It should be noted that we determined the OR for the unfavorable outcome (in-hospital case fatality), whereas others generally investigated IP by determining the OR for the complementary favorable outcome event (e.g. presence of IP was indicated by OR >1). In order to enable comparison of results, we calculated the reciprocal of the OR for favorable outcome (when appropriate). Accordingly, the adjusted OR for unfavorable outcome ranged from 0.65 (95%CI: 0.49–0.91) to 0.28 (95%CI: 0.11–0.79).5,8,9 Our adjusted OR for in-hospital case fatality of 0.53 (95%CI: 0.29–0.98) fits well with these previously reported data.

Lack of IP was reported, however, on the basis of the Northern California TIA trial.10 Here the authors attempted to determine whether strokes occurring within 1 week of an initial TIA are less severe than those occurring later, when IP is not expected. The rationale for that study design was to control for the differences in the pathophysiology of atherosclerotic strokes heralded by TIA and cardioembolic strokes. Thus of the 1707 TIA patients, 180 patients who developed ischemic stroke within 90 days of the initial TIA were included in the analysis. These 180 patients were then investigated on the basis of the timing of the prior TIA (within 1 day, between days 2 and 7, and between days 8 and 90). Disabling or non-disabling stroke was defined as the measure of outcome. The adjusted OR for non-disabling stroke was 0.28 (95%CI: 0.07–1.11), well in agreement with that of previous studies reporting the presence of IP. The level of significance, however, was P = 0.08. The authors interpreted these results as the absence of IP, but admitted that there was a possibility of missing the clinically significant effect of IP due to the restricted power of their study.

The present study has a number of limitations inherent of observational studies. One such limitation was the lack of detailed information concerning the antecedent TIA such as the duration, the vascular territory in which the TIA occurred, and the time that elapsed between the TIA and the subsequent ischemic stroke. Based on experimental evidence in animals it takes around 24 h to develop ischemic tolerance following ischemic challenge, which lasts for approximately 7 days, and is completely resolved by the end of the second week. The situation, however, is less unequivocal in humans. There are contradictions concerning the significance of temporal associations because some authors report a lack of importance,8 while others were able to identify a window of protection similar to that observed in animal studies.5 A possible explanation for the protective effect witnessed in the prior TIA group may be as follows. With the wider accessibility of more sophisticated imaging methods, evidence is starting to accumulate that a TIA is not merely a transient loss of function in a specified vascular area, but a marker for ongoing cerebrovascular ischemia in general. For example a diffusion-weighted imaging study assessing patients with TIA showed that TIA increases the risk of silent ischemia. The authors stated that new events on magnetic resonance imaging are more frequent than clinical events.16 Starting from this we propose that a patient with a previous TIA continuously receives clinically silent transient ischemic preconditioning' stimuli. This ongoing subtoxic ischemia is then able to trigger the endogenous protective signaling mechanism that will prove beneficial in the event of a subsequent stroke.

Nevertheless, considering the magnitude of this prospective database, we presume that the results are reliable and thus make it unlikely that this positive association is due to any unidentified bias. Another potential source of bias toward the TIA group may be introduced by not questioning the patients or their relatives about previous undocumented TIA. A large nationwide survey, including 10 112 participants that assessed the public knowledge of TIA and prevalence of TIA diagnosis concluded that relying solely on patient recall for estimating past TIA may lead to both overestimation and underestimation of the true prevalence of prior TIA.17

A further study limitation was that we lacked data concerning the use of anti-platelet (AP) medication or chronic anticoagulation (AC) in secondary prevention. Although such agents are likely to decrease the risk of in-hospital case fatality, they presumably exert this effect in both case scenarios given after an initial stroke (prior stroke group) and TIA (prior TIA group). Starting from this the beneficial effects of AP and AC drugs should be observable in both the prior TIA and the prior stroke group as decreased OR. Nonetheless only the prior TIA group manifested this lower OR (after correcting for the confounders), proposing that a prior TIA per se exerts beneficial effects of concurrent AP/AC use.

When comparing on-admission stroke severity among the first ever, prior stroke and prior TIA groups, we found that patients with prior TIA had a significantly higher Mathew score. One might argue that the reason for detecting lower in-hospital case fatality is that the investigation was biased because patients with prior TIA had less-severe strokes. Upon entering the Mathew score into the starter multiple logistic regression model, we found that the model became highly redundant (data not shown). Given the fact that both in-hospital case fatality and stroke severity are different definitions of outcome after a cerebral ischemic insult, we think that decreased stroke severity on admission is in fact another manifestation of the IP phenomenon.

The present study has considerable advantages when compared with other studies. First, instead of using a global assessment variable, consisting of subjective elements, commonly used by others (the NIH stroke scale, Barthel index), we chose a hard outcome measure, in-hospital case fatality, as the primary outcome variable, the judgment of which is unambiguous. Second, we included every patient suffering from ischemic stroke, regardless of the severity of stroke on admission. In one study showing an independent association between TIA and favorable outcome, patients presenting with stupor or coma were excluded.9 Similar bias was introduced into another study when the investigators excluded patients who died from the final analysis, with patients with prior TIA dying more frequently.8

Stroke studies evaluating whether the occurrence of prior TIA is associated with the severity of stroke outcome implicitly presume that other aspects of pathophysiology are similar to the group lacking any prior ischemic episode.5,6 Generally this is not the case because cardioembolic strokes are less likely to be heralded by TIA, and are associated with greater severity, potentially confounding the comparison between patients having or not having previous TIA. In contrast, large artery atherosclerosis per se may be associated with a more favorable stroke outcome because it is associated with smaller emboli giving rise to more distal vessel occlusion, and different composition that enhances spontaneous thrombolysis and earlier recanalization. Together with the opening of collateral intracerebral vessels, these effects all contribute to smaller infarcts and preserved brain tissue.18 A bias common to some studies was the overrepresentation of atherosclerotic strokes and the underrepresentation of cardioembolic ones.5,6 In the present sample we found that a lower percentage of patients had carotid artery atherosclerosis and a higher frequency of atrial fibrillation in the prior TIA group. Furthermore, the final multiple model is unique because it contains detailed information about carotid artery atherosclerosis (presence of intimasclerosis, carotid artery stenosis or carotid artery occlusion) and of the presence or absence of atrial fibrillation, the primary source of cerebral emboli, thus their effect on in-hospital case fatality is controlled for.

Because hyperglycemia is able to worsen stroke outcome (EUSI guidelines recommend that serum glucose level be lowered below 10 mmol/L),19 and because it is able to prevent IP per se20,21 this parameter was controlled for by our multiple logistic model.

We found that the OR for untreated hypertension was lower than that for treated arterial hypertension. This is probably a type I or alpha error.

Summarizing, we suggest that prior TIA but not prior stroke offers benefit, because preceding TIA was significantly associated with a lower probability (OR: 0.53) for in-hospital case fatality following adjustment for confounders. Therefore we conclude that there is an independent association between prior TIA and lower in-hospital case fatality following a subsequent stroke, and we have provided additional evidence for the beneficial effect of IP in the clinical setting. Starting from this, the effect of overly aggressive therapeutic interventions against mild ischemic episodes should be re-evaluated.22 Additionally, efforts should be made to identify the cellular and molecular mechanisms underlying this phenomenon to exploit these mechanisms yielding safer and more effective stroke therapies.


The authors thank Dr László Kardos, biostatistician, for his advice on data analysis. This work was supported by grant No. ETT 130/2003 and by the Öveges program of the National Office for Research and Technology (NKTH No.OMFB-01458/2006).