Intracerebral haemorrhage can have several causes. In younger patients (<40 years), intracranial vascular malformations are the most common single cause of intracerebral haemorrhage, usually lobar in location.¹ Cerebral amyloid angiopathy is thought to underlie about 30% of intracerebral haemorrhages in those aged >70 years and to cause mainly lobar haemorrhages.¹ Although hypertension is a major risk factor for intracerebral haemorrhage in general,^2–8 it is commonly considered to be associated more with patients having deep than with those having lobar haemorrhage. The current understanding of the arterial pathology underlying deep haemorrhage is largely based on studies conducted in the 1960s and 1970s by Fisher and others, with the meticulous examination and description of brains examined on autopsy of people with and without an intracerebral haemorrhage as the cause of their death.^9–13 Fisher’s findings led him to propose that hypertensive lipohyalinosis, affecting the small, deep, perforating, intracranial blood vessels, may lead to lacunar infarction in some circumstances and to deep intracerebral haemorrhage in others.^11,12 These conclusions, however, were based on small numbers (his own studies included <30 patients in total); the subjects in autopsy studies were sometimes deemed hypertensive or not on the basis of blood pressure soon after stroke, and uncertainty remains about whether the vascular pathology Fisher observed preceded or resulted from haemorrhage. Furthermore, lipohyalinotic changes have been found in the brains of normotensive patients with recent intracerebral haemorrhage.¹⁴ Despite these limitations, Fisher’s theory of the hypertensive arterial pathology of deep intracerebral haemorrhage has become so entrenched in the literature and in clinical teaching and practice^15–17 that deep haemorrhage has become virtually synonymous with hypertensive haemorrhage.

But is hypertension really more commonly a risk factor for patients with deep haemorrhage than for those with lobar haemorrhage? In this paper, we report the findings of a systematic review and meta-analysis of studies that compared the frequency of hypertension as a risk factor for patients with deep haemorrhage versus those with lobar supratentorial haemorrhage, and consider the effects of study methodology on the results.

METHODS

Study identification

We sought studies on patients with intracerebral haemorrhage, confirmed after brain imaging or after postmortem examination, published between January 1966 and December 2004 (inclusive), which provided data on the frequency of hypertension in patients with deep haemorrhage versus those with lobar supratentorial intracerebral haemorrhage. We did not exclude otherwise eligible studies that grouped deep supratentorial and posterior fossa haemorrhages together, as the proportion of such studies was small.

We identified studies with a comprehensive electronic search strategy (see appendix), supplemented by searching through the reference lists of all relevant articles identified; our own and colleagues’ collections of papers on intracerebral haemorrhage; and textbooks on stroke.

Data extraction

We extracted data from included studies on the total number and source of the patients studied; the numbers with deep or lobar haemorrhage, before and after any exclusions; the numbers with hypertension as a risk factor in each of the groups with deep and lobar haemorrhage; the mean (or median) age of the patient in each group; the definitions of deep and lobar haemorrhage and of hypertension; and whether the study population included first-ever strokes only. Where possible, to reduce noise in our analyses, we excluded patients with haemorrhage due to identified secondary causes unrelated to prior hypertension (such as intracranial vascular malformation, blood dyscrasia, anticoagulant treatment or drug misuse).

As the location of the bleeding source for some large haemorrhages is unclear from the brain scan, their classification as deep or lobar is potentially subject to various sources of bias. We therefore also extracted any available information on the type (computed tomography or magnetic resonance imaging) and timing of diagnostic brain scans; whether any haemorrhages were considered to be unclassifiable; whether brain scans were reported blinded to hypertension status; who reported or reviewed the scans (neuroradiologist, neurologist, etc); and the intrarater and inter-rater reliability of the classification of haemorrhage location.

Statistical analyses

We calculated study-specific and summary odds ratios (ORs) of hypertension in patients with deep versus lobar haemorrhage using Cochrane Revman V.4.2.¹⁸ We carried out several sensitivity subgroup analyses, comparing the results from studies that clearly fulfilled each of several predefined methodological quality criteria with studies that did not. These were as follows:

1. A definition of hypertension based solely on raised blood pressure before stroke (as blood pressure is often raised after intracerebral haemorrhage, so that blood pressure after stroke does not necessarily reflect that before stroke)

2. A population-based study, which we defined as being either community based or hospital based, where the hospital was the only centre serving a defined population and all admissions with intracerebral haemorrhage were included

3. Inclusion of patients with first-ever stroke only (as risk factors and the distribution of deep and lobar haemorrhage may differ between the first and recurrent strokes).

We then compared results from studies that met all three methodological quality criteria with those that did not.

In a retrospective subgroup analysis, we compared results from studies recruiting people unselected for age against those specifically recruiting younger patients. We used the χ² test to assess statistical heterogeneity between studies and groups of studies.

RESULTS

Study characteristics

We found 32 potentially relevant studies from a total of 1611 publications identified by our search. From these, we excluded two studies that reported frequency of hypertension among patients with recurrent intracerebral haemorrhage,^19,20 and two from which it was impossible to extract relevant data.^21,22 This left us with 28 studies on a total of 3930 patients (2196 with deep and 1734 with lobar haemorrhage) for inclusion in our analyses.^23–50 Table 1 and fig 1 summarise the characteristics of these studies.

Lobar haemorrhage was consistently defined as that occurring in the temporal, parietal, occipital or frontal lobes, whereas deep haemorrhage was generally defined as that arising in the basal ganglia region. Two studies reported the frequency of hypertension for deep supratentorial and posterior fossa haemorrhages together, and were included because the proportion of cases with posterior fossa haemorrhages was small (eg, 9 posterior fossa v 90 supratentorial deep haemorrhages in the larger of the two studies).^32,49

In all, 7 studies (n = 1352 patients) were population based,^{23,25,26,35,40,45,46} 13 (n = 1856) used a pre-stroke definition of hypertension,^{23,25–27,29,30,36,38–40,45,46,49} 4 (n = 614) included only patients with first-ever stroke^25,45,46,50 and only 3 (n = 601) fulfilled all three of these criteria (table 1).^25,45,46

Twenty studies reported the mean (or in three studies median) age, with an overall study size-weighted average (mean or median) of 59 (range 27–73) years. Fifteen studies recruited people unselected for age (weighted average 64 years), whereas five specifically recruited a younger population (generally <45 years, weighted average 32 years; fig 1A and table 1). In the eight studies that reported mean or median age separately for patients with deep and lobar haemorrhage, patients with deep haemorrhage were slightly younger than those with lobar haemorrhage (weighted average 66 v 71 years; table 1).

Six studies^{23–25,27,41,50} did not exclude people from their study population with haemorrhages from secondary causes unrelated to hypertension. The other studies had already excluded people with haemorrhages from secondary causes or provided data that allowed us to do so. The proportion of patients excluded because of a secondary cause was available from half of the studies.^{26,28,29,31,33,34,36–39,42–44,47,48} It was highly variable, ranging from 6% to 73%, and was generally higher in studies that specifically recruited a younger population (fig 1B, table 1). Furthermore, in studies that reported on exclusions from the deep and lobar haemorrhage groups separately, the proportion excluded was consistently higher—sometimes substantially so—in the group with lobar haemorrhage (table 1). No study reported the proportion of patients with deep and those with lobar haemorrhage undergoing further investigation for a secondary cause. But two studies on younger patients, in which about a third of all patients underwent catheter angiography, reported that most patients with hypertension did not have a catheter angiogram, especially if their haemorrhage was deep.^28,37

Most studies used computed tomography brain imaging, but only nine reported the time from onset of symptoms to that of the scan (table 1).^{23,24,29,33,38–40,45,48} Only a few studies mentioned any difficulty in classifying the location of the haemorrhages, and only two actually reported any unclassifiable haemorrhages.^23,25 No study reported inter-rater or intrarater reliability of classification of haemorrhage location. Few studies commented on who reviewed the scans, and some classified the location based solely on information in computed tomography scan reports. Only one study reported that haemorrhage location was classified blind to the hypertension status of the patient.³⁶

Frequency of hypertension in patients with deep versus lobar haemorrhage

The pooled OR for all studies suggested that frequency of hypertension was about twice as common in patients with deep as in those with lobar haemorrhage (OR 2.10, 95% confidence interval (CI) 1.82 to 2.42; fig 2). However, we found statistically significant heterogeneity between individual studies (χ²₂₇ = 75.4; p<0.001), some of which seemed to be explained by differences in study methods. For each of the three predefined methodological quality criteria, the summary OR was substantially lower for studies that fulfilled the criterion than for those that did not, with significant heterogeneity between the two groups of studies in each case (fig 3). The pooled OR for the three studies (601 patients) meeting all three criteria (OR 1.50, 95% CI 1.09 to 2.07) suggested a smaller but still statistically significant excess of hypertension in deep versus lobar haemorrhage. The OR for these methodologically more rigorous studies was substantially lower than that for the remaining studies (OR 2.27, 95% CI 1.94 to 2.66), with significant heterogeneity between the two groups (fig 3).

The pooled OR for studies that included only younger patients (OR 12.32, 95% CI 6.13 to 24.77) was far larger than for studies that recruited a population unselected for age (OR 1.91, 95% CI 1.65 to 2.21).

DISCUSSION

The pooled result from all the studies we analysed suggested that hypertension was twice as common a risk factor for patients with deep haemorrhage than with lobar haemorrhage without an identified secondary cause. We found an excess of hypertension among patients with deep haemorrhage in the methodologically more rigorous studies, albeit smaller and statistically less certain. As only three studies actually met all of our rather modest criteria for methodological quality, this summary estimate is based on quite small numbers of patients (601, less than one sixth of the total in all studies included). The differences between subgroups in the various sensitivity analyses are striking and emphasise the effect of study methods on the results.

We could not quantify the effect of the additional potential sources of bias that we considered, but these are also likely to have affected the results. Firstly, some misclassification of haemorrhage location must surely have occurred, yet only two studies acknowledged this, none reported reliability of the classification of the haemorrhage location and only one mentioned that scans were reported blind to hypertension status. As the concept of deep hypertensive haemorrhage is entrenched in clinical teaching and practice, the knowledge of a patient’s hypertension status may have influenced the classification of haemorrhage location and may have generated a spurious association between hypertension and deep versus lobar haemorrhage. This may be particularly true in the classification of large haemorrhages, for which the location of the original bleeding source may be uncertain. Many such corticosubcortical haemorrhages are thought to originate from the basal ganglia, but evidence for this is limited.¹⁵

Secondly, few studies reported on the timing of the brain scan relative to the onset of symptoms. Many may have included some patients with haemorrhagic transformation of a cerebral infarct, although it is difficult to know just how this may have affected our results.

Thirdly, investigation bias is likely to have affected the results of these studies to a variable degree. The study populations included in our analyses depended on the extent of patient investigation, which may well have differed between groups with deep and lobar haemorrhage and according to age. The available data showed a larger proportion of exclusions from younger study populations than from those unselected for age, and from groups of patients with lobar rather than deep haemorrhage. This could be because of real differences in the proportions with an underlying secondary cause, higher rates of investigation on younger patients and on those with lobar haemorrhage, or both. Some support for differential investigation bias being part of the explanation comes from two studies on younger patients that reported lower rates of catheter angiography among patients with hypertension and deep haemorrhage,^28,37 as well as from the (admittedly anecdotal) observation that, in clinical practice, a non-hypertensive structural cause is less often considered and sought if a history of hypertension and brain imaging shows a deep rather than a lobar haemorrhage. Some of the excess of hypertension found among patients with deep haemorrhage may therefore be accounted for by differential investigation, resulting in secondary causes remaining undetected among those having deep haemorrhage with coincidental hypertension.

We may expect studies on younger patients to be particularly useful in that these patients should be more extensively investigated and thus allow a more accurate assessment of the contribution of hypertension to deep and lobar haemorrhages that have no identifiable secondary cause. We found that the frequency of hypertension in patients with deep versus lobar haemorrhage was much higher in studies on only young patients than in those studies that did not select on age. This result, however, is difficult to interpret, as the number of patients included in the studies on younger patients (n = 238) was very small, making the results imprecise. Furthermore, none of these studies met any of our predefined methodological quality criteria, and the result could reflect greater investigation bias in younger patients. None the less, the apparently more extreme excess of hypertension in younger patients with deep versus lobar haemorrhages is an interesting finding that deserves further study.

Given that cerebral amyloid angiopathy is thought to be particularly important in elderly people, it would have been interesting to assess the effect of age on the results of studies in which patients were unselected for age. Only two studies, however, reported a frequency of hypertension by age group; they used different age cut-offs and the numbers of patients were too small to allow any meaningful analyses. This is a limitation of our systematic review and further large studies that allow age-specific analyses are required to explore this issue.

Our analyses have considered the contribution of chronically raised blood pressure to lobar and deep haemorrhage. Intracerebral haemorrhage may also arise as a result of acutely raised blood pressure, particularly in previously normotensive people. Such acute sudden rises in blood pressure may be of more importance for deep than for lobar haemorrhage, but this would be particularly difficult to study in humans, because it is impossible to distinguish raised blood pressure arising as a result of the intracerebral haemorrhage from sudden raised blood pressure precipitating the haemorrhage. Although there are reports in the literature of instances where intracerebral haemorrhage was assumed to be the result of acutely raised blood pressure, no clear evidence indicates that this occurs more often in patients with deep than with lobar haemorrhage.^1,51

In summary, pooled results from observational studies suggest that hypertension is more frequently a risk factor for deep than for lobar haemorrhage, perhaps particularly in younger age groups. These findings are, however, heavily influenced by studies with less robust methods. In the methodologically more rigorous studies, we found a smaller, but still statistically significant, excess of hypertension among patients with deep haemorrhage. This may, however, be accounted for by additional, unquantified sources of bias. Further large, methodologically robust studies are needed to determine accurately the relative contribution of hypertension to deep and lobar haemorrhage in different age groups. After the exclusion of secondary causes, raised blood pressure may make an important contribution to the arterial pathologies underlying lobar haemorrhages—for example, cerebral amyloid angiopathy—as well as to those leading to rupture of small, deep, penetrating arteries. Thus, the terms hypertensive haemorrhage and deep haemorrhage should not be considered synonymous, as this implies that all deep haemorrhages are attributable solely to hypertension. This may cause some patients with deep haemorrhages routinely to be excluded from further investigation on potentially treatable non-hypertensive causes (such as arteriovenous malformations), especially if they are elderly.

APPENDIX

Medline search

1. exp intracranial hemorrhages/

2. ((brain$ or cerebral or intracranial or intracerebral) adj5 (haemorrhage$ or hemorrhage$ or bleed$)).tw.

3. ich.tw.

4. exp intracranial hemorrhages/ or ((brain$ or cerebral or intracranial or intracerebral) adj5 (haemorrhage$ or hemorrhage$ or bleed$)).tw. or ich.tw

5. (lobar or deep or subcortical or sub-cortical).tw.

6. exp cerebral amyloid angiopathy/

7. cerebral amyloid angiopath$.tw.

8. exp cerebral amyloid angiopathy/ or cerebral amyloid angiopath$.tw.

9. (lobar or deep or subcortical or sub-cortical).tw. or exp cerebral amyloid angiopathy/ or cerebral amyloid angiopath$.tw.

10. exp intracranial hemorrhages/ or ((brain$ or cerebral or intracranial or intracerebral) adj5 (haemorrhage$ or hemorrhage$ or bleed$)).tw. or ich.tw and (lobar or deep or subcortical or sub-cortical).tw. or exp cerebral amyloid angiopathy/ or cerebral amyloid angiopath$.tw.

A similar search was used for Embase.