Twenty-Year Time Trends in Long-Term Case-Fatality and Recurrence Rates After Ischemic Stroke Stratified by Etiology

Introduction

Based on routine mortality statistics, ischemic stroke mortality decreased between 1998 and 2015 about 50% in Germany.¹ It is unclear, to what extent this decline is attributable to changes in stroke incidence or to changes in case fatality and recurrence rates. Previous analysis from the Erlanger Stroke Project showed that ischemic stroke incidence in Germany decreased between 1995 and 2010 in men but not in women with substantial differences in etiological subtypes of ischemic stroke according to the TOAST (Trial of ORG 10172 in Acute Stroke Treatment) classification.² However, the burden of stroke in the population is not only influenced by incidence but also by case-fatality rates (CFR) as decreasing CFR accompanied by constant incidence rates leads to increasing prevalence.³ Another important factor that drives the burden of stroke in the population are recurrence rates, because a second stroke is associated with a longer in-hospital stay, higher mortality, and a higher degree of disability.^4,5

Epidemiological data on the long-term time trends in CFR and recurrence rates in patients with ischemic stroke are scarce. Recent publications from the population-based South London Stroke Register showed improved survival rates in all pathological subtypes of stroke.^6,7 A systematic review showed that the 5-year stroke recurrence rates decreased over time across the 13 hospital- or community-based stroke registers included.⁸ However, in the population of Perth in Australia, no significant time trends in 5-year stroke recurrence rates were found.⁹ CFR and recurrence rates differ substantially between etiological subtypes of ischemic stroke based on TOAST criteria.^10–12 To the best of our knowledge, long-term data on patterns of CFR and recurrence rate from a population-based register stratified by ischemic stroke cause are lacking.

Therefore, by updating a prior analysis from the Erlangen Stroke Project from 1994 to 1996, we investigated long-term time trends in CFR and stroke recurrence rates by ischemic stroke cause over a 20-year time period based on data from the population-based Erlangen Stroke Project.¹¹

Methods

The data that support the findings of this study are available from the corresponding author upon reasonable request.

Erlanger Stroke Register

The Erlanger Stroke Project is an ongoing prospective population-based stroke register in Germany, covering a source population of 105 164 inhabitants (2010). Within this source population, incidence and outcome of stroke is continuously monitored since 1994. The methodology of the study has been described in detail elsewhere.¹³ All hospitalized and nonhospitalized patients with stroke with registered residence in Erlangen were identified by regular checks of hospital admission, discharge records, nursing homes, and general practices using standardized criteria to ensure completeness of case ascertainment.¹³ Every patient is followed-up 3 months, 12 months, and annually until death. Patients are contacted for follow-up by study nurses and research assistants. In case the patient could not be contacted for follow-up, the Population Register of the City of Erlangen is checked for a possible change of address or death. If the patient died during the follow-up period, the death certificates is reviewed, and the cause of death is ascertained from all available medical records.

Stroke was determined according to the World Health Organization criteria.¹⁴ Classification of pathological subtypes of stroke was determined based on brain CT or MRI scan, with 95% imaging rate.¹³ For the present analysis, patients with first-ever ischemic stroke from 1996 to 2015 were included.

Etiological subtype of ischemic stroke was defined according to TOAST classification.¹⁵ We used the 5 major categories of TOAST: large-artery atherosclerosis (LAA) including large-artery thrombosis and artery-to-artery embolism, cardioembolism, small-artery occlusion (SAO), stroke of other determined cause, and stroke of undetermined cause (UND) comprising stroke of undefined and concurrent cause. Interrater reliability of the classification of TOAST for the present analysis was good (κ=0.65 [95% CI, 0.35–0.96] between 1995 and 1998 and 0.63 [95% CI, 0.43–0.83] between 1999 and 2010) as previously published.^2,11 Recurrent stroke was defined as new neurological deficit at least 24 hours after the incident stroke according to the World Health Organization definition excluding edema, mass effect, brain shift syndrome or hemorrhagic transformation, and procedure-related strokes.¹⁶

Statistical Analysis

Incidence rates were age-standardized according to the European standard population.¹⁷ Recurrent stroke was defined as first recurrent stroke after the index event (ischemic stroke) regardless of the pathological subtype (ischemic, hemorrhagic stroke, or other) of recurrent stroke. Crude 3 months, 1 year, and 5 years CFR and recurrence rates were estimated with Kaplan-Meier and stratified by sex and according to TOAST classification. Age-adjusted CFR and recurrence rates were estimated using Cox regression. Sex-specific time trends for CFR and recurrence rates were estimated Cox proportional hazard regression models. The Cox regression models were adjusted for age, sex, and year of event and stratified by etiological subtypes according to the TOAST classification. Due to the small sample size, stroke of other determined cause was excluded in survival analysis. Competing risk analysis were performed as sensitivity analysis for time trends in recurrence rates¹⁸ with death from any cause being used as a competing event and recurrent stroke as an event of interest. Competing risk calculates the probability of getting a recurrent stroke under the condition that no recurrent event occurred before or that the person might have died. We used Aalen Johansen estimator to estimate the cumulative incidence as a function of recurrent stroke.¹⁹ The subdistribution hazard regression from Fine and Gray was used to investigate time trends in the cumulative risk of recurrent stroke under competing risks adjusted for age, sex, and stratified for etiological subtype according to TOAST classification.²⁰

Ethics

Written informed consent to participate was given by patients or their legal representatives. The study was approved by the Ethics Committee of the Medical Faculty of Friedrich-Alexander-University Erlangen-Nürnberg (Reference number: 249_15 Bc).

Results

Overall, 3346 patients with ischemic stroke were registered in the Erlanger Stroke Project between January 1996 to December 2015. Mean age was 74±13 years, 53.2% were women. Distribution of subtypes was as follows: LAA, 8.4%; cardioembolism, 23.0%; SAO, 28.0%; OC, 2.1%; and UND, 38.5%. Mean age in years was 71±12 in LAA, 77±11 in cardioembolism, 71±12 in SAO, 58±17 in OC, and 75±13 in UND; 41.5% of LAA, 56.7% of cardioembolism, 50.1% of SAO, 55.5% of SAO, 55.9% of OC, and 55.5% of UND were women. The average annual age-standardized incidence rate of total ischemic stroke from 1996 to 2015 was 109.7 per 100 000 overall, 75.8 in women and 131.6 in men; average annual age-standardized incidence rate per stroke subtypes per 100 000 was as follows: 11.2 in LAA, 22.7 in cardioembolism, 32.2 in SAO, 4.4 in OC, and 39.5 in UND.

Case-Fatality Rates

Overall, 3-month, 1-year, and 5-year survival probabilities by sex and TOAST subtype are shown in Table 1 and Figure 1; 5-year survival probability was highest in patients with stroke due to SAO (73.8% [95% CI, 70.4–77.3]) and lowest in cardioembolic stroke (40.9% [95% CI, 37.2–45.0]). Table 2 shows time trends determined by hazard ratios for the year of event of the adjusted Cox proportional hazard regression for death for the entire observation period as well as censored after 3 months, 1 year, and 5 years. The same linear trends were observed in Kaplan-Meier survival estimates stratified for the year of event in 5-year groups with 1996 to 2000 having the lowest and 2011 to 2015 having the highest survival rates (Figure I in the Data Supplement). Overall, Cox regression showed a trend for declining stroke mortality during the entire time period (hazard ratio for year of event, 0.98 [95% CI, 0.97–0.99], P<0.001). Stratified Cox regression revealed a significant decrease in stroke mortality over time in men as well as in women. Further significant decreases over time in stroke mortality were observed in cardioembolism in total as well as for men. In SAO, we observed statistically significant declining stroke mortality rates in all for all and in women. In UND, stroke mortality decreased significantly over time in total and in men.

• Download figure
• Download PowerPoint

Stroke Recurrence

Pathological subtypes of second strokes were as follows: 84.5% ischemic stroke; 5.7% intracerebral hemorrhage; 0.2% subarachnoid hemorrhage; and 9.6% stroke pathology not specified or missing. Overall, 3-month, 1-year, and 5-year recurrence rates by sex and TOAST subtype were shown in Table 3 and Figure 2. Undefined strokes had the highest (21.9% [CI, 18.7–25.0]) and SAO in women (16.0% [95% CI, 11.7–20.1) and LAA in men (16.6% [95% CI, 8.7–23.9]) the lowest rate of stroke recurrence after 5 years. There were no significant trends for decline in stroke recurrence estimated with Cox regression (Table 4). In the sensitivity analysis taking death as competing risk into account, the UND had the highest risk of recurrent stroke followed by SAO (Table I in the Data Supplement). After considering death as competing risks, there were significant trends for a decline in stroke recurrence in all the patients. Stratified competing risk analysis showed significant decreases in risk of recurrent stroke in women and in women with UND (Table II in the Data Supplement). There was no linear trend in recurrence rates stratified for 5-year groups of year of event (Figure II in the Data Supplement).

• Download figure
• Download PowerPoint

Discussion

In our population-based sample, nearly every second patient died 5 years after the first event. There were substantial variations in survival rates across etiological subtypes observed with highest rates for cardioembolism and lowest for SAO, respectively. CFR decreased over the 20 years of observation, showing similar trends across stroke etiological subtypes except for LAA. About every fifth patient suffered from a recurrent stroke within 5 years after the first event with the highest recurrence rates observed in cardioembolism and UND. Cox regression revealed no significant decrease in long-term recurrence rates over time. Competing risks analysis showed similar patterns with additional decreasing patterns in stroke recurrence overall and in women.

Our 3-month CFR of 13% are lower than the mean weighted CFR of 21.8% reported from the European Registers of Stroke collaboration of 6 population-based stroke registers in Europe in 2004 to 2006, but similar to the lowest rate across the participating registers observed in Dijon.²¹ On the contrary, the 1-year (63.7%) and 5-year survival rates (42.8%.) for ischemic and hemorrhagic stroke in the South London Stroke Register were lower compared with our study.²² This might be caused by the fact that hemorrhagic strokes were included in the South London Stroke Register which had lower survival rates than ischemic strokes. Only few studies investigated time trends in long-term CFR, most of them showing decreasing CFR. For example, in a Danish population-based cohort found a decrease in 30-day and 5-year mortality rates after ischemic stroke between 1994 and 2011.²³ In addition, in the Netherlands in a linkage study of national registries, age-and sex-specific 30-day and 1-year (excluding the first 30 days) CFR decreased between 1997 and 2005 in most age-groups.²⁴ Controversially, 90-day CFR increased in the Swedish Stroke Registry between 1995 and 2010, which the authors referred to the higher proportion of more severe strokes.²⁵

To the best of our knowledge, our study is the first to investigate long-term time trends in CFR stratified by etiological stroke subtypes according to standardized mechanism-based classification scheme such as the TOAST classification. The patterns of the stratified 5-year CFR by TOAST are in line with previously published data from our group reporting survival up to 2 years.¹¹ In the Nanjing Stroke Registry, highest 1-year survival rates were reported for patients with SAO (92.7%), followed by UND (89.4%), cardioembolism (88.1%), and LAA (84.2%).²⁶ Most CFRs in our study were lower compared with this data, and we found the lowest 1-year survival rates in cardioembolism (67.9%). Furthermore, the Dijon Stroke Registry analyzed 28-day CFR stratified by cause (defined as macroartheromics, microatheroma, and cardioembolic). They reported significantly improvements CFR in stroke due to microatheroma from 1985 to 2004, similar to our findings.²⁷

The decrease in CFR might be caused by improvements in stroke management and treatment. For example, after the introduction of stroke units in 1994/1995, the number of stroke units, one of the most effective options in stroke management,²⁸ increased in Germany up to 255 in 2012.²⁹ Also, an increased uptake in acute therapies, such as thrombolytic therapy and improved overall care, might have led to the observed reduction in CFR in Germany. For example, based on German administrative hospital data proportion of patients with stroke being admitted to a stroke unit increased between 2005 and 2010 from 15% to 52%.³⁰ The decrease in CFR might also be attributed to changes over time in patient characteristics affecting outcome such as stroke severity, comorbidities, or frequency of early complications. For example, in the Austrian Stroke Unit Registry, the stroke severity significantly decreased by 1 point in the National Institutes of Health Stroke Scale between 4 and 3 in men and between 5 and 4 in women.³¹

The recurrence rates observed in our study are in line with the pooled estimates of a recent meta-analysis based on population-based studies.⁸ There are only a few studies reporting recurrence rates stratified by stroke cause. We found that patients with LAA have a high risk to get a second stroke within the first year, whereas after 5 years the risk in the LAA subgroup was relatively small in comparison to the other stroke subtypes. This higher risk might be because the CFR after 5 years in patients with LAA is high, and therefore, a substantial proportion of the patients died before they can get a second stroke. The rates in SAO are relatively low over the whole observation period. A previous systematic review found lower recurrence rates up to 1 month in lacunar versus nonlacunar infarctions, however, these differences were not statistically significant in time period up to 12 months.³² The risk of a recurrent stroke in cardioembolism and UND remains high after the first year after the event. The high risk in cardioembolism might be attributable to patients with atrial fibrillation having a higher risk of recurrence.³³ The high risk of recurrent stroke in the UND group might be caused by a substantial proportion of patients with undetected atrial fibrillation in this group as recent studies revealed that intensified monitoring for atrial fibrillation in this group yield to a proportion of 16.1% patients with potential cardioembolic stroke.³⁴

We found no clear time trends in overall stroke recurrence rates as well after stratification for sex or stroke subtype in our main analysis using Cox regression, whereas we found a few significant time trends in stroke recurrence rates after taking competing risks into account. Our main analysis is in line with a systematic review from 2018 comprising 34 randomized controlled trials, hospital-based and community-based studies published before December 11, 2016, which also found no statistically significant time trends in stroke recurrence rates after ischemic stroke or transient ischemic attack.³⁵ The authors concluded that this might partly be due to heterogeneity in stroke cause as no stratification for stroke subtype was performed.³⁵ However, after stratification for ischemic stroke cause in our study, we did not find any clear time trend in recurrence rate in our main analysis despite substantial heterogeneity in stroke recurrence rates by etiological subgroups. We only found a decrease in stroke recurrence rates over time in patients in the UND group, when we considered competing risks. One possible reason might be, that, although new medication for stroke prevention are available such as nonvitamin K oral anticoagulants, the control of cardiovascular risk factors (eg, hypertension, atrial fibrillation) after a stroke is still not optimal in Germany.³⁶ Besides, adherence to secondary preventive measures might decrease during follow-up, leading to a decrease in the effect of secondary prevention. An alternative explanation could be that our study was underpowered to detect a small decrease in stroke recurrence rates due to the small sample size, especially in the later years. Another reason might be that people die before they can get a second stroke. Therefore, we included as a sensitivity analysis an analysis accounting for competing risks, where we observed a significant decrease in stroke recurrence in total and in women.

Our study has several strength and limitations. One of the strengths is the population-based design and the long follow-up over 20 years with constant standardized data collection over the entire time period. In addition, our data is able to report data on the TOAST classification of the first event over the whole time period which allows stratifying time trends in CFR and recurrence rates by etiological subtypes of the first event. Our study also has several limitations. First, we have a high rate of undefined causes according to TOAST classification. However, the proportion of patient with undefined stroke is within the range of other population-based registers.^37,38 Second, no information on the etiological subtype of recurrent stroke was available, which might influence time trends in stroke recurrence. Third, no adjustment for stroke severity was possible, as stroke severity is not collected in the register by a standardized scale such as the National Institutes of Health Stroke Scale. Fourth, the source population of our registry is relatively small, which might limit the statistical power of the study. However, the Erlanger Stroke Project is the largest and longest running population-based stroke registry in Germany, and only a limited number of other registers worldwide comprise such a long observation period.

Conclusions

Decreasing CFR over the past 20 years were observed across all stroke subtypes that might be associated by improved acute treatment options. However, there is still room for improvement secondary prevention measures in ischemic stroke survivors, as we did not observe a clear reduction in stroke recurrence rates.

Acknowledgments

We thank their fellow participants of the Erlangen Stroke Project: Universitätsklinikum Erlangen, Waldkrankenhaus St. Marien, Klinikum am Europakanal, the General Practitioners Association Erlangen, the Regional Public Health Office of Erlangen and the City Council of Erlangen. We would like to express their gratitude to the 100 Erlangen general practitioners, their staff, the community nurses, and the patients and their family members, without whose cooperation and help this study would not have been possible.

Footnotes