Trends Over Time in the Risk of Stroke After an Incident Transient Ischemic Attack
Background and Purpose—Long-term population trends in the early risk of stroke after transient ischemic attack (TIA) are unknown. We hypothesized that there has been an appreciable decline in the risk of stroke after TIA for the last decade.
Methods—Population-level cohort study from Victoria, Australia (population 5.6 million), using linked data from hospitals, emergency departments, and death records (2001–2011), with a 2-year clearance period to define incident TIAs. Age-specific rates/1000, yearly incident rate ratios, and age–sex-adjusted risk of stroke after TIA were computed.
Results—The mean age of 46 971 patients with TIA was 71 (SD=15), 52% women. In patients ≥65 years, annual TIA rates declined between 2001 and 2011 from 5.8 to 4.8/1000 (men) and from 5.3 to 4.2/1000 (women). Yearly incident rate ratios were 0.97 (95% confidence interval, 0.96–0.98) in men and 0.97 (95% confidence interval, 0.97–0.98) in women. Overall, the 90-day stroke risk was 3.1%. Age–sex-adjusted risk of stroke at 90 days after a TIA decreased by 3% per year (odds ratio for the effect of year, 0.97; 95% confidence interval, 0.95–0.99). Male sex, direct discharge from emergency departments, public hospital care, stroke unit care, and absence of vascular risk factors were associated with a downward yearly trend of stroke within 90 days of TIA.
Conclusions—Over the last 10 years, there has been a measurable decline in the 90-day risk of stroke after an incident TIA and overall decline in rates of TIA in Victoria, Australia. These trends may reflect improved primary and secondary prevention efforts for the last decade.
Stroke and transient ischemic attack (TIA) present a significant public health problem worldwide. TIA may account for 30% of acute cerebrovascular disease occurring in a population.1 In earlier settings of suboptimal secondary prevention, TIA was reported to herald the stroke onset in 10% of patients within 90 days.2
Several preventative interventions for cardiovascular disease have gained ground for the last 2 decades, including the use of antiplatelet, blood pressure, and lipid-lowering agents and lifestyle programs for smoking and obesity. There is evidence that the incidence,3 hospitalization, and mortality rates for stroke overall4 are declining in Belgium (1984–1999), Rochester (1985–1989), and more recently in France (2002–2006).5–7 It is unknown whether these trends extend to the risk of stroke after TIA, as people with TIA represent a high-risk group. Such a study is logistically challenging because it requires data capture of all TIA cases and subsequent ischemic strokes and for this to be repeated within a defined population over time. Routinely collected hospitalization and death registration data for TIA and stroke may provide an alternative source of information on this issue at a population level. Such data would assist in informing further planning of health services for TIA.
Our aim was to determine the trends over time in the risk of ischemic stroke after an incident TIA. We hypothesized that there would have been an appreciable decline in the risk of ischemic stroke 90 days after TIA for the last decade. Secondarily, we aimed to examine the trends in rates of incident TIA for the same period, hypothesizing that there would be a continuing decline in these rates.
Design and Methods
We undertook a population-based cohort study of patients presenting with TIA for acute care in the state of Victoria, Australia, using linked data from emergency departments (EDs), hospital discharges, and death certificates.
Victoria is the second most populous State in Australia with a population of 5.6 million.8 All residents have universal access to publicly funded hospital medical care. In addition, ≈45% also have private health insurance, allowing access to private hospitals.9 Stroke units are present predominantly in public hospitals because they require sufficient numbers of junior and senior physicians to enable organized 24-hour stroke care.10
Hospital discharge and ED data are compiled by >300 hospitals.11,12 Although both public and private hospitals contribute to the hospital discharge data, ED data are based mostly on public EDs because these settings provide the majority of emergency care in Victoria. Diagnostic information is coded according to the International Statistical Classification of Diseases and Related Health Problems, 10th Revision, Australian Modification (ICD-10-AM)13 and procedural information according to the Australian Classification of Health Interventions.14 For hospital data, coding occurs at discharge when experienced clinical coders review the entire medical record and record the diagnoses and procedures relevant to the admission. For ED visits, the attending doctor records the diagnoses and the procedures undertaken. In the event of uncertainty, coders liaise with clinicians for clarification.
Death registrations are maintained by the Registry of Births, Deaths, and Marriages.15
Data from EDs, hospitals, and deaths are linked to one another at the Victorian Department of Health where linkage quality is assessed by a series of internal logic checks and manual review.16 These data do not contain identifying variables when released for research use.
Case Definition of Incident TIA Episodes
All TIA episodes from hospital and ED data were extracted between July 1, 2001, and June 30, 2011, based on the 0.4-digit TIA ICD-10 codes G450-G459 (Table I in the online-only Data Supplement).
These TIA episodes were then restricted to incident (first) episodes of TIA by ensuring that the case had not presented to hospital or ED with a TIA or stroke (ICD-10-AM codes 1630–1639) during the previous 2 years. This is termed as a 2-year clearance period.
Quality of Coding
Data standards are maintained with regular, independent coding audits using a selected random sample17 where an expert coder recodes the entire hospitalization blinded to the original codes assigned for that discharge. Both must rely on the TIA and stroke diagnosis available in the hospital chart. Using the expert coder as the reference standard, the coding quality for TIA and stroke in Victorian public hospital data is found to be high (sensitivity of 89% (95% confidence interval [CI], 84%–94%) and a positive predictive value of 93% (95% CI, 89%–97%)). When assessing reliability, the kappa is found to be 0.91, indicating high agreement beyond chance.
The primary measure was the occurrence of nonfatal and fatal ischemic stroke within 90 days after an incident TIA. Fatal strokes were identified by a manual review of the primary cause of death by 2 stroke experts (V.K. Srikanth and T.G. Phan) and then automated for complete capture. Cases of index TIA with lengths of stay ≥90 days were excluded from further analysis because it was thought unlikely they represented cases of TIA only.
Risk Factors and Procedures
Cardiovascular risk factors included hypertension, smoking, elevated lipids, obesity, atrial fibrillation, and carotid stenosis; relevant investigations and procedures were computed tomography brain with/without angiography, and MRI with/without angiography (Table I in the online-only Data Supplement).
Rurality of residence and private hospital care were defined using standard classifications.12 The presence of a stroke unit within a hospital was based on a review of the lists of hospitals by T.G. Phan and V.K. Srikanth with reference to a list published by the National Stroke Foundation of Australia.10
We calculated age- and sex-specific rates of TIA per 1000 from 2001 to 2011 with Victorian population data as the denominator.8 A Poisson regression, stratified by age group (18–39, 40–64, and 65+) and sex and further adjusted for age as a continuous variable, was fitted to check for changes in these rates over time.
For the primary outcome analysis, multivariable logistic regression models were fitted to assess whether there was an age–sex-adjusted year effect for stroke after TIA at 90 days. Of note, we initially attempted to fit a Cox proportional hazards model, but our key exposure variable (year) violated the proportional hazards assumption. As the at-risk period was of relatively short duration (90 days), logistic regression was chosen as a suitable alternative. To assess whether the primary outcome was confounded by a secular trend in diagnostic accuracy of TIA, we restricted the analysis to a subsample of those who had brain imaging during the TIA episode because this may signify a higher suspicion of TIA. A further analysis was conducted to adjust for MR brain imaging to assess whether misclassification of stroke as TIA early in the time period could explain our results.
Secondary analyses were also conducted to examine the influence of other factors, including risk factors and process factors, on any observed trends.
A sensitivity analysis was conducted restricting the sample to cases of TIA who were hospitalized with a diagnosis of TIA in the principal diagnostic position and who had not had a TIA or stroke in the previous 5 years (a 5-year clearance); this case definition is likely to overlap with published analyses of cases of TIA presenting to hospital5,18 and is likely to remove ≈97% of prevalent TIAs and strokes in comparison to the 2-year clearance which removes 93%.19 In Australia, the principal diagnosis is defined as the diagnosis chiefly responsible for occasioning a service event or episode.20
Stata12 and 13 (Stata Corporation, College Station, TX, 2006) were used for all statistical analyses.
This study was approved by the Monash University Human Research Ethics Committee.
We identified 46 971 cases of incident TIA in the 10-year study period (Table 1 and Table II in the online-only Data Supplement). Almost 80% were aged ≥60 years, 52% were women, 33% were married, 18% resided in rural areas, and 76% were born in an English-speaking country. The majority of patients (86%) were treated in public hospitals, with only 14% receiving any private hospital care and 51% treated in a hospital with a stroke unit. The proportion of ED-only cases increased from 38% to 45% between 2001 and 2011, whereas those receiving any brain imaging and any MRI increased from 52% to 53% and from 3% to 5%, respectively.
Age-specific rates of TIA declined for patients aged ≥65 years, but increased for those aged <65 years in the 10-year period (Table II in the online-only Data Supplement). The yearly incidence rate ratios confirmed these trends (Table 2).
Either during the presentation or in the preceding 2-year-period, 41% of people with TIA had ≥1 vascular risk factor (Table IV in the online-only Data Supplement). This proportion was higher (56%) in the sensitivity analysis (5 year clearance and restricted inclusion to only those with a principal diagnosis of TIA during an admission [n=13 091]).
Overall, 1263 (2.9%; range, 2.6%–3.6%) patients had a stroke within 90 days of TIA (Table V in the online-only Data Supplement). After adjusting for age and sex, we observed a 3% yearly decrease in the odds of stroke at 90 days for all TIAs (odds ratio [OR] for the effect of year, 0.969; 95% CI, 0.950–0.988; P=0.004; Table 3). In a restricted sample of patients with TIA who received brain imaging, the proportion of patients suffering stroke within 90 days was slightly higher for the time period (3.1%–4.7%), but the yearly OR for stroke after TIA sample (OR, 0.959; 95% CI, 0.930–0.988) was similar to that for the whole. Further, adjusting for the use of MRI did not alter the declining yearly trend of stroke after TIA (OR, 0.959; 95% CI 0.931–0.988). The results were similar in a sensitivity analysis that included only incident TIAs with a 5-year clearance.
In prespecified secondary analyses, a downward yearly trend in 90-day stroke risk was observed for patients directly discharged from ED (P≤0.001), in men (P≤0.05), among those receiving public hospital care (P≤0.001), having access to stroke units (P≤0.05), receiving brain imaging (P=0.001), and in the absence of vascular risk factors (P≤0.001; Table 3).
These are the first data to demonstrate a yearly decline in the risk of stroke within 90 days after an incident TIA for the last decade. This declining trend was more likely among men, in those treated in public hospitals, with access to stroke unit care, requiring ED-only care, receiving brain imaging, and without a concurrent traditional cardiovascular risk factor. Consistent with previous studies showing a decline in stroke incidence,5–7 we also found a decline in rates of hospitalization for TIA in people aged ≥65 years for the last decade, but a paradoxical increase in those aged <65 years.
This study has several strengths. It is the first population-based study of trends over time in the risk of ischemic stroke after an incident TIA. Our sample size was large, allowing for the detection of small effects. We carefully defined incident TIA using a 2-year clearance period and found consistent results in a sensitivity analysis using a 5-year clearance period. We used well-audited and standard methods of diagnostic coding to identify cases of TIA and stroke and well-established data linkage procedures. The uptake of MRI was low and, therefore, is unlikely to affect our case definition of TIA over time and thus confound the estimates of decreasing stroke rate after TIA. We also explored the effect of several indicators of care to provide some clues to explain observed trends in the risk of ischemic stroke after TIA. On an additional note, with respect to TIA overall, only 1 other study has provided data on temporal trends in TIA incidence.5 In contrast to our study, the previous study included only hospitalized patients, whereas we were able to also include those who presented to EDs but were not hospitalized (a substantial proportion), thus enhancing the generalizability of our results.
There are also potential limitations to this study. Using diagnostic coding to retrospectively identify cases of TIA or stroke may be subject to a risk of misclassification. A substantial proportion of patients with suspected TIA presenting only to ED may in fact be a TIA mimic, such as migraine or syncope.21 However, diagnostic clarification may occur well after index presentation and only if reviewed by stroke specialists. Therefore, TIA diagnostic coding may be subject to problems with specificity of diagnosis. Diagnostic coding for stroke has a high positive predictive value when compared with gold-standard clinical neurologist diagnosis.22–24 Although diagnostic misclassification may therefore affect point incidence estimates of TIA, it is much less likely to influence our analysis of trends over time in stroke risk after TIA, or TIA rates, particularly because the same coding system was used for the study period. Notably, the yearly downward trend in the risk of stroke risk was still observed when restricted to patients receiving brain imaging during hospitalization. The use of brain imaging signifies a higher index of suspicion of a TIA as compared with conditions, such as syncope, thereby providing stronger support that such patients may have a genuine TIA.
Our study only deals with patients who were hospitalized, or received ED care in hospitals, and may miss those who did not attend hospital. However, the vast majority of Australian patients with transient neurological symptoms tend to directly attend hospital EDs rather than via primary care practitioners.25 This, and the large state-wide coverage of hospitals in our linkage data set, ensures generalizability to the majority of the population. Moreover, exclusion of those not attending hospital is unlikely to affect the validity of our longitudinal analysis, and any trend over time in hospitalization practices for TIA is likely to modify (and potentially explain) rather than confound our primary results.
Our primary results of a decline in 90-day ischemic stroke risk after TIA are novel and consistent with reports of a decline in overall stroke incidence for a similar time period in Australia and in other developed countries.3,4 Our subgroup analyses allow some speculation as to why this may be the case. The stronger decline in 90-day stroke risk seen in those receiving brain imaging, public hospital care, and having access to stroke units suggest an important effect of organized acute care post TIA and possibly better secondary prevention.26 Recent Australian and international data provide evidence for increased use of brain imaging over time and better organization of rapid TIA management resulting in higher rates of early use of antiplatelet therapy and carotid artery investigations.27,28 Rapid treatment pathways have resulted in low rates of stroke after TIA (<5%) compared with older estimates in settings of suboptimal prevention (≈10%).2 Such pathways are most likely to be instituted in hospitals with organized stroke units, which, in Australia, are predominantly in public hospitals.10 The greater decline in the risk of stroke among those managed only in the ED may signify an improved ability to better manage patients without admission, as well as a lower intrinsic risk of stroke in such patients compared with those who are admitted to hospital. Similarly, the greater decline among those without a concurrently diagnosed cardiovascular risk factor may be that such factors signify a greater intrinsic risk of stroke. The lack of a downward yearly trend for those with these cardiovascular risk factors may suggest that their treatment at a population level may not yet be optimal. However, we are cautious in our interpretations of these subgroup analyses, which are hypothesis-generating rather than conclusive.
We also found a population-level decline in the rates of acute presentations for TIA for the study period in those aged ≥65 years in both men and women. This decline in those ≥65 years is consistent with overall patterns of reduction in stroke incidence3,4 and incidence of TIA in the developed world.5–7 These results may indicate that cardiovascular prevention has improved at the population-level in older people. Consistent with this theory, the number of prescriptions for antiplatelet therapy (by 110%), BP lowering agents (by 45%), and cholesterol reduction (by 76%) markedly increased between 1997 and 2009 in Australia.29 Similar to the French study,5 we found an increase in TIA rates in those aged <65 years. In people <65 years, there may be an increasing awareness of cardiovascular risk of stroke, although diagnostic error in this age group cannot be excluded as a reason for the observed increase in rates. Younger people who present to hospital with transient neurological syndromes may in reality have other causes for their symptoms (eg, complex migraine, seizures), but be misdiagnosed and hence miscoded as TIA. However, it is important to note that the overall rate of TIA diagnosis in this age group was much less (0.43/1000) than in those aged >65 years (4.9/1000).
We have found a reduction over time in the risk of ischemic stroke after an incident TIA in a comprehensive population-level analysis of patients presenting for ED and hospital care. It is likely that improved emergency and hospital care and secondary and primary cardiovascular prevention play important roles in determining these trends.
Sources of Funding
The project received funding from a Pfizer Australia Cardiovascular Lipid Research Grant, 2010. V.K. Srikanth is the recipient of a National Health and Medical Research Foundation of Australia/National Heart Foundation Career Development Fellowship. A.G. Thrift was supported by a Senior Research Fellowship from the National Health and Medical Research Council (1042600).
Dr Phan received honoraria payments from Bayer, Genzyme and Boehringer Ingelheim. He is an Advisory Board Member, Genzyme (Fabry Disease). The other authors report no conflicts.
The online-only Data Supplement is available with this article at http://stroke.ahajournals.org/lookup/suppl/doi:10.1161/STROKEAHA.114.006575/-/DC1.
- Received June 30, 2014.
- Revision received August 21, 2014.
- Accepted September 2, 2014.
- © 2014 American Heart Association, Inc.
- Rothwell PM,
- Giles MF,
- Chandratheva A,
- Marquardt L,
- Geraghty O,
- Redgrave JN,
- et al
- Brown RD Jr,
- Petty GW,
- O’Fallon WM,
- Wiebers DO,
- Whisnant JP
- Buntinx F,
- Devroey D,
- Van Casteren V
- 8.↵Australian Bureau of Statistics. Australian Demographic Statistics 2012. http://www.abs.gov.au/ausstats/abs@.nsf/mf/3101.0. Accessed March 23, 2012.
- 9.↵Australian Bureau of Statistics. Private Health Insurance: A snapshot, 2004-05 2006. http://www.abs.gov.au/ausstats/abs@.nsf/mf/4815.0.55.001. Accessed March 23, 2012.
- 10.↵Stroke Foundation. Stroke Care Units 2014. http://strokefoundation.com.au/stroke-care-units/. Accessed March 25, 2014.
- 11.↵State Government of Victoria, Australia, Department of Health. Victorian Emergency Minimum Dataset (VEMD) 2012. http://www.health.vic.gov.au/hdss/vemd/index.htm. Accessed March 23, 2012
- 12.↵State Government of Victoria, Australia, Department of Health. Victorian Admitted Episodes Data Set (VAED) 2012. http://www.health.vic.gov.au/hdss/vaed/index.htm. Accessed March 23, 2012.
- Wollongong Uo
- 14.↵Australian Institute for Health and Welfare. Australian Classification of Health Interventions (ACHI) 7th edition 2013. http://meteor.aihw.gov.au/content/index.phtml/itemId/391343. Accessed March 10, 2013
- 15.↵State Government of Victoria, Australia, Department of Justice. Victorian Registry of Births, Deaths and Marriages 2012. http://www.bdm.vic.gov.au. Accessed March 23, 2012
- 16.↵State Government of Victoria, Australia, Department of Health. Victorian Data Linkages 2012. http://www.health.vic.gov.au/vdl/. Accessed March 23, 2012
- Gattellari M,
- Goumas C,
- Garden F,
- Worthington JM
- 20.↵Australian Institute for Health and Welfare. Episode of care—principal diagnosis 2012. http://meteor.aihw.gov.au/content/index.phtml/itemId/269654. Accessed March 10, 2013.
- Haesebaert J,
- Termoz A,
- Polazzi S,
- Mouchoux C,
- Mechtouff L,
- Derex L,
- et al
- Kokotailo RA,
- Hill MD
- Sacco S,
- Pistoia F,
- Carolei A
- Sanders LM,
- Srikanth VK,
- Jolley DJ,
- Sundararajan V,
- Psihogios H,
- Wong K,
- et al
- 29.↵Australian Institute for Health and Welfare. Stroke and its management in Australia: an update. 2013. http://www.aihw.gov.au/publication-detail/?id=60129543613. Accessed March 23, 2013