Background and aim There is uncertainty about the long-term prognosis after spontaneous intracerebral haemorrhage (ICH). Therefore, we systematically reviewed the literature for studies reporting long-term survival and ICH recurrence, and their predictors.
Methods We searched Ovid Medline 1946–2011 inclusive for cohort studies of ≥50 patients reporting long-term (>30 days) outcome after ICH. Two reviewers independently extracted data from each study. We meta-analysed 1-year and 5-year survival data from population-based studies using a random effects model (and quantified inconsistency using the I2 statistic).
Results We identified 122 eligible studies. The pooled estimate of 1-year survival was 46% (95% CI 43% to 49%; nine population-based studies (n=2408); I2=27%) and 5-year survival was 29% (95% CI 26% to 33%; three population-based studies (n=699); I2=6%). In 27 cohort studies, predictors most consistently associated with death were increasing age, decreasing Glasgow Coma Scale score, increasing ICH volume, presence of intraventricular haemorrhage, and deep/infratentorial ICH location. The annual risk of recurrent ICH varied from 1.3% to 7.4% in nine studies and this risk was higher after lobar ICH than non-lobar ICH in two of three hospital-based studies. Four studies reporting the risks of recurrent ICH and ischaemic stroke after ICH found no significant differences between these risks.
Conclusions Less than a half of patients with ICH survive 1 year and less than a third survive 5 years. Risks of recurrent ICH and ischaemic stroke after ICH appear similar after ICH, provoking uncertainties about the use of antithrombotic drugs.
Statistics from Altmetric.com
Spontaneous (non-traumatic) intracerebral haemorrhage (ICH) is an important cause of stroke, with an annual incidence of 24.6 per 100 000.1 ICH accounts for 10% of stroke in high-income countries and 20% in low-income and middle-income countries.2 A recent systematic review found that the 1 month case fatality after ICH has remained unchanged for several decades at ∼40%,1 but outcome in the longer term is less clear. Survivors seem to be at considerable risk of serious vascular events: recurrent ICH—which may be more frequent after lobar ICH—has been the principal concern for stroke physicians,3 but thrombotic (vaso-occlusive) diseases may be at least as frequent, raising important questions about the potential benefits and hazards of antithrombotic (antiplatelet and anticoagulant) drugs.4
The last systematic review of the long-term outcome after ICH focused on the risks of recurrent ICH and ischaemic stroke, which were pooled from patients in heterogeneous studies who may have had different baseline risks of these two events; that review reported a higher annual risk of recurrent ICH (2.3%) than ischaemic stroke (1.1%) after ICH.3 Since the review, several cohort studies have been published, which would incorporate the temporal trends that have been noted in some recent community-based cohorts5 ,6 and provide the opportunity to directly compare the risks of recurrent ICH and ischaemic stroke after ICH in cohorts quantifying both of these risks.
Therefore, we systematically reviewed published cohort studies reporting all major outcomes after ICH to characterise long-term (>30 days) survival and its predictors, change in survival over time, rates of recurrent ICH and other serious vascular events (stratifying by index ICH location where possible), and functional outcome.
We included longitudinal cohort studies of ≥50 patients with non-traumatic (spontaneous) ICH, whether or not they were explicit about the absence of an identified underlying cause (so-called ‘primary’ ICH), which reported long-term (>30 days) clinical outcomes (death, recurrent ICH, vaso-occlusive events, epileptic seizure(s) or functional outcome) and/or tested for predictors of outcome. We excluded studies if: <90% of ICH had been confirmed by imaging or pathological examination; they had included patients with pure intraventricular haemorrhage (IVH), pure subarachnoid haemorrhage or ICH secondary to an underlying cause (such as an arteriovenous malformation or intracranial venous thrombosis); or they had used International Classification of Diseases (ICD) codes that include pure IVH (ICD-9 code 431 and ICD-10 code161.5) to determine eligibility.
In February 2012, one investigator (MTCP) searched Ovid Medline (1946–2011) using a comprehensive electronic search strategy combining search terms for ICH that had been used in a prior Cochrane Library review7 and search terms for prognosis studies used in PubMed Clinical Queries8 and a search term for human studies (see online supplementary methods). We searched for eligible studies that had been missed by the electronic searches by screening the reference lists of included studies published in 2004–2011, screening the population-based studies included in a recent systematic review of short-term prognosis after ICH as well as rerunning its search strategy from December 2009 to December 2011.1
After removing duplicate entries, all titles and available abstracts were screened for eligibility by one investigator (MTCP). Studies that appeared to be eligible were read in full for the final assessment of eligibility, and uncertainties were resolved by discussion with a second reviewer (RA-SS). Included studies published in English were read by a third reviewer (AFF). We asked colleagues to assess studies published in languages other than English using a standardised assessment form.
Two reviewers (MTCP and AFF) independently extracted data from all the included studies published in English. Disagreements or uncertainties were resolved by arbitration or discussion with a third reviewer (RA-SS).
Summary measures and data analysis
If several publications arose from the same cohort, we only included the report with the largest sample size from each cohort. We included studies that had included patients with ‘primary’ or ‘secondary’ ICH if we could extract results for the group with ‘primary’ ICH alone. We sought to meta-analyse results from only the most generalisable studies, which had not selected participants according to a specific care setting or ICH location. We collected survival data as the proportion of patients with ICH surviving at particular time points (1 year and 5 years). When outcome measures were only quantifiable from a graph, we used Plot Digitiser (http://plotdigitizer.sourceforge.net/) to extract data. When not provided, we calculated 95% CIs for these data using the GraphPad QuickCals website (http://www.graphpad.com/quickcalcs/ConfInterval1.cfm, accessed August 2012). We meta-analysed survival data at 1 year and 5 years from population-based studies using StatsDirect statistical software V.2.7.2, using the DerSimonian-Laird random effects method, to generate their pooled estimates and their respective 95% CIs. We quantified the I2 statistic (a measure of inconsistency that describes the percentage of total variation across studies ie, due to heterogeneity rather than chance) in survival data from population-based studies and hospital-based studies separately. We assessed temporal trend in 1 year and 5 years survival data using weighted linear regression between survival and the midyear of the study periods. The regression model was weighted by the inverse of the standard error of the survival for each study. We used IBM SPSS Statistics V.20 (IBM Corp, Armonk, New York, USA) for regression analyses. We extracted data on outcome events as proportions at specific time points or as annual rates. We extracted outcome event rates stratified by ICH location when this was reported. We restricted the description of functional outcome to modified Rankin Scale (mRS) scores in population-based studies at 6 months and 1 year after ICH. We restricted our summaries of predictors of death or ICH recurrence to covariates that had been tested or adjusted for their association with outcome in at least five studies and restricted our summary to studies in which at least one of these covariates had been tested as a predictor. We have reported this systematic review in compliance with the PRISMA guidelines.9
We identified 304 studies that appeared to be eligible for this review and included 122 that described any of the prespecified outcomes, after excluding 182 studies for the following reasons (see online supplementary figure 1): sample size was <50,s1–17 cohort was not restricted to ‘primary’ ICH,s18–38 cohort included pure IVHs39–47 or pure subarachnoid haemorrhage,s48–53 ICD codes were used to detect ICH,s54–69 other report of an included cohort,s70–83 impossible to separate ICH from ischaemic stroke,s84–127 outcomes did not meet inclusion criteria,s128–144 outcome assessment within 30 days only,s145–166 <90% of ICH confirmed by imaging/pathology,s167–173 or the full text of the publication was unavailable.s174–182 Among the 122 included studies,10–54 s183–259 75 studies reported outcomes at time points that we were able to summarise in this review (see online supplementary table 1), but 47 studies did not.s187–193, s195, s197, s199, s201, s203, s205, s207, s211, s212, s216–219, s222, s225, s226, s228–233, s236, s237, s241–245, s247, s248, s250, s252–259
Long-term survival after ICH: summary and meta-analysis
Of 94 studies reporting estimates of survival, outcome was reported at 90 days (n=32), 6 months (n=18), 1 year (n=37), 2 years (n=7), 3 years (n=10), 5 years (n=10) and various other time points (n=24). We chose to summarise outcome at 1 year because it was the most frequently reported time point and 5 years because this was the latest time point most frequently reported.
One-year survival: Of 37 studies reporting survival at 1 year, six cohorts were selective (patients in intensive care units,s184, s202 patients on chronic dialysis22 or survivor cohorts27 ,47 ,53) and three cohorts selected patients according to ICH location (supratentorial15 ,16 or pontines196), leaving 28 studies10 ,12–14 ,17 ,18 ,20–24 ,26 ,28 ,29 ,31–36 ,38 ,41 ,45 ,48–52 for analysis (figure 1). The pooled 1-year survival estimate in nine population-based studies13 ,17 ,21 ,23 ,31 ,32 ,35 ,36 ,48 was 46.0% (95% CI 43.4 to 48.6; n/N=1102/2408; I2=27.2%; figure 1), but we did not meta-analyse the results of hospital-based studies because of their inconsistency (I2=97.5%; figure 1). There was no temporal trend in 1-year survival from years 1983 to 2004 (annual increase of 0.2% per year, 95% CI −0.1 to 0.5, p=0.5).
One-year survival, stratified by ICH location: Of the 37 studies reporting survival at 1 year, six stratified their results by ICH location.15 ,16 ,29 ,48 s184, s196 One of these studies was restricted to patients in intensive care unitss184 and another was restricted to patients with pontine ICH,s196 leaving four studies, one of which was a population-based study,48 in which the influence of ICH location on 1 year survival in univariate analyses varied: 1-year survival was 45.4–59.1% after lobar ICH and 45.4–59.5% after deep ICH (see online supplementary table 2).
Five-year survival: Of the 10 studies reporting survival at 5 years, two studies recruited survivors alone30 ,34 and another study was restricted to patients with supratentorial ICH,16 leaving seven studies for analysis (figure 2).13 ,17 ,20 ,21 ,25 ,33 ,44 The pooled 5-year survival estimate in three population-based studies13 ,17 ,21 was 29.2% (95% CI 25.7 to 32.8; n/N=203/699; I2=6.4%; figure 2), but we did not meta-analyse the results of hospital-based studies because of their inconsistency (I2=91.9%; figure 2). There was no temporal trend in 5-year survival from years 1983 to 1997 (annual increase of 0.3% per year, 95% CI −0.02 to 0.6, p=0.5).
Recurrent ICH: summary
Of 21 studies reporting data on recurrent ICH, the risk was reported at 1 year (n=13), 2 years (n=8), 3 years (n=7), 5 years (n=6) and various other time points (n=13). We chose to summarise the risk of recurrent ICH at 1 year because it was the most frequently reported time point.
Recurrent ICH after ≥1 year: Of the 13 studies reporting the risk of recurrent ICH over at least 1 year, one was restricted to patients with lobar ICH (see next section)54 and three studies were not survivor cohorts,42 s227, s240 leaving nine studies of survivor cohorts (table 1).11 ,20 ,30 ,34 ,37 ,39 ,40 ,45 ,47 Excluding the study that reported only the risk of fatal recurrent ICH,47 the annualised rate of ICH recurrence was 2.0–2.4% in five studies20 ,34 ,37 ,39 ,40 and the proportion of patients having a recurrent ICH by 1 year ranged from 1.8%11 to 7.2–7.4%.30 ,45
Recurrent ICH after ≥1 year, stratified by ICH location: Four studies reported the risk of recurrent ICH stratified by the location of the ICH at study entry, but none of these studies was population based and only one reported a mean follow-up period.54 Lobar ICH location was associated with a higher risk of recurrent ICH in two out of three studies that compared it with a concurrent group of patients with non-lobar ICH (table 2).30 ,42
Other events: summary
Twelve studies reported the frequency of ICH and vaso-occlusive events with follow-up at different time points using different analytical methods.15 ,17 ,19–21 ,27 ,34 ,37 ,43 ,53 s213, s221 Two studies only recruited patients with supratentorial ICH15 ,20 and three studies did not separately quantify vaso-occlusive events.19 ,21 ,27 The remaining seven cohorts reported the rates of ICH and ischaemic stroke (but not other vaso-occlusive events). Three of these studies reported these rates over a mean follow-up period and their results are not summarised here.17 ,53 s213 Comparing the rates of the two events in each of the remaining four studies,34 ,37 ,43 s221 ischaemic stroke appears to be at least as common as recurrent ICH over 3 years (table 3).
In two studies, the risk of epileptic seizure(s) at 1 year and 5 years were 14%10 and 25.2%,16 respectively and one study reported a higher 5-year seizure rate for lobar ICH than deep ICH (54% vs 10.3%).16
Functional outcome: summary
Fifty-seven studies reported functional outcome, most frequently using the mRS, at varying time points: 90 days (n=29), 6 months (n=14), 1 year (n=12) and other time-points (n=7). Because of the statistical inconsistency in estimates of case fatality and the varying eligibility criteria in hospital-based studies (figure 1), we restricted our summary to the four population-based studies, which found independence (mRS 0–2) among 32.8–42.4% of all ICH (53.7–83.7% of survivors) at 6 months46 ,55 and 16.7–24.6% of all ICH (53.8–57.1% of survivors) at 1 year (see online supplementary figure 2).3 ,36
Predictors of long-term survival and ICH recurrence: summary
Survival: 32 cohort studies tested or adjusted for the association of covariates with long-term survival (median time point 1 year) using multivariable analyses.15 ,17 ,18 ,22–24 ,26 ,27 ,29 ,37 ,40 ,47 ,51 s185, s186, s194, s200, s204, s206, s208–210, s214, s215, s220, s223, s234, s235, s238, s239, s246, s251 We have summarised the 12 covariates explored or adjusted for in five or more studies, but five studies22 ,23 s208, s220, s228 did not examine any of these 12 covariates so are not included in our summary table (Table 4). Seven studies examined all the variables in the most frequently used prognostic model.56 The predictors that were most frequently studied, and significantly associated with death more often than not across the studies, were: increasing age, decreasing Glasgow Coma Scale score, increasing ICH volume, presence of intraventricular haemorrhage and deep/infratentorial ICH location (table 4).
ICH recurrence: Six studies,20 ,30 ,40 ,54 s183, s249 tested or adjusted for the association of covariates with ICH recurrence (median time point 3 years) using multivariable analyses: two studies reported a significant association with lobar ICH location,20 ,30 two studies reported a significant association with prior ICH,54 s183 and one study reported a significant association with APOE genotype.54
This systematic review and meta-analysis of 122 longitudinal cohort studies reporting long-term (>30 days) outcome after spontaneous ‘primary’ ICH has shown that 1-year survival was 46.0% (figure 1) and 5-year survival was 29.2% (figure 2) in population-based studies. These long-term survival rates do not appear to have changed over time. The most frequently studied predictors of death in the long term (table 4), which were more often than not significantly associated with death in multivariable analyses in mostly hospital-based studies, were increasing age, decreasing Glasgow Coma Scale score, increasing ICH volume, presence of intraventricular haemorrhage and deep/infratentorial ICH location, which are the principal components of the ICH score.56
The long-term rate of recurrent ICH for early survivors in mostly hospital-based studies varied from 1.3–7.4% per year, over average durations of follow-up of 1–7 years. Some studies have found that the rate of ICH recurrence was higher after lobar ICH than after non-lobar ICH (possibly due to the frequency of underlying cerebral amyloid angiopathy or variation in the use of antihypertensive therapy),20 ,30 ,42 whereas others have not.37 ,40 Our findings about ICH recurrence update a systematic review published in 20013 with the inclusion of recent studies and the exclusion of studies that either did not meet our required sample sizes91, s171 or did not meet our requirement to exclude pure IVH.s40 Subsequently published studies have found no difference in the risks of ischaemic stroke and recurrent ICH in the same patients over 3 years after ICH, unlike the findings of the 2001 systematic review;3 therefore the risk of all vaso-occlusive events (ie, deep vein thrombosis, pulmonary embolism, acute coronary syndrome, peripheral arterial occlusion and mesenteric ischaemia, as well as ischaemic stroke) may be higher than the risk of recurrent ICH.
This review benefited from sensitive overlapping search criteria, no restriction on the language of publication and explicit selection criteria intended to maximise the internal validity of the review. We were able to focus our meta-analyses on population-based studies, which demonstrated negligible statistical inconsistency unlike hospital-based studies (figures 1 and 2). Updating our literature search from February 2012 to September 2013 did not identify new studies that qualitatively or quantitatively changed our conclusions. This review has some limitations. We restricted our search to studies that had been published in journals indexed in Medline, but we combined medical subject headings and text words (which is known to improve the sensitivity and specificity of searches for prognostic studies in Medline8), screened included studies’ bibliographies and crosschecked the studies we identified with a recent, related systematic review.1 The design and methods of the included studies varied (see online supplementary table 2), especially in their methods and completeness of follow-up, their restriction to spontaneous ‘primary’ ICH, and their stratification according to important treatments like oral anticoagulant therapy before ICH57 and surgery after ICH. However, having demonstrated the consistency of information in population-based studies (figures 1 and 2), we were able to restrict our meta-analyses and description of functional outcome to this more generalisable design. The reporting of the included studies determined the time points at which we could summarise long-term outcome. The included studies’ statistical methods and reporting of outcomes other than survival (eg, annualised rates over average periods of follow-up and proportions of patients with recurrent ICH at particular time points; table 2) precluded meta-analyses. The quality of studies that reported predictors is limited by the incomplete inclusion of common covariates, such as those in the ICH score (table 4). The methodological inconsistencies and potential selective reporting rendered interpretation of covariates associated with these outcomes (eg, HRs, ORs, relative risks and sometimes just p values) difficult. Recent reporting guidelines for prognosis studies may improve this in the future.58
Future studies of long-term outcome after ICH should address remaining uncertainties about prognosis such as the influence of ICH location on ICH recurrence (as well as other biomarkers of the underlying cause of ICH59), whether the risk of recurrent ICH changes over time, the risks of other haemorrhagic and vaso-occlusive events after ICH (including the relative risk of these events over time), predictors of functional outcome and the influence of antithrombotic drugs on outcome.14 ,38 ,60 s235 The findings of this review indicate that these studies should be population based or community based, and they should adhere to recent recommendations for study design and reporting.58 Given the reasonably consistent findings concerning the major predictors of death (table 4), future studies should be sufficiently well powered to determine whether novel predictors add to these known predictors. An individual patient data prognosis meta-analysis of similar well-designed studies could achieve some of these objectives. In light of the risk of all vaso-occlusive events being at least as high as the risk of recurrent ICH, whose respective severities may vary, randomised controlled trials of whether or not to resume antithrombotic drugs after ICH appear justified (eg, http://www.RESTARTtrial.org, ISRCTN71907627, evaluating whether or not to start antiplatelet drugs after ICH).
Less than a half of patients with ICH survive 1 year and less than a third survive 5 years. Risks of recurrent ICH and ischaemic stroke after ICH appear similar after ICH, provoking uncertainties about the use of antithrombotic drugs.
The authors would like to thank Dora Miovic, Dmitry Tsigelnitskiy, Yuhki Takebayashi, Yoshi Bartlett-Imadegawa, Mireia Moragas, Paolo Candelaresi, Alice Tod and Carla Ferreira for their help with translation.
This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.
Files in this Data Supplement:
- Data supplement 1 - Online supplement
Contributors MTCP, AFF and RA-SS contributed to the conception and design of the article. MTCP, AFF and RA-SS all contributed to data collection and data analysis. MTCP and RASS contributed to data interpretation and drafting of the article. All authors contributed to revising the article for important intellectual content and gave final approval of the version to be published.
Provenance and peer review Not commissioned; externally peer reviewed.