Three Important Subgroups of Hypertensive Persons at Greater Risk of Intracerebral Hemorrhage
Abstract—Hypertension as a risk factor for intracerebral hemorrhage (ICH) is poorly quantified, particularly in the setting of the use of modern antihypertensive agents. To investigate this, we studied 331 consecutive hospital cases of primary ICH verified by computed tomography or autopsy, occurring during the period 1990 through 1992, and 331 age- and sex-matched community-based control subjects in a city-wide study involving 13 hospitals. Hypertension approximately doubled the risk of ICH (adjusted odds ratio [OR], 2.45; 95% confidence interval [CI], 1.61 to 3.73). The OR associated with hypertension was significantly greater among those who had ceased taking medications, supervised and unsupervised (OR, 4.98; 95% CI, 2.25 to 11.02), compared with those who had not (OR, 1.95; 95% CI, 1.20 to 3.16), were under the age of 55 years (OR, 7.68; 95% CI, 2.65 to 22.5), or were current smokers (OR, 6.12; 95% CI, 2.29 to 16.35). The presence of hypertension did not influence size or location of the hemorrhage. However, those dying from ICH displayed a greater risk of ICH due to hypertension than survivors, with the ratio of the two ORs being 5.47 (95% CI, 1.23 to 24.44). These findings provide evidence for a greater increase in risk of ICH due to hypertension among younger persons, current smokers, and those discontinuing antihypertensive therapy. This is the first direct evidence for a link between stopping antihypertensive medication use and stroke risk; targeting these individuals for more intensive monitoring and education on the importance of risk factor modification may help to reduce the impact of this form of stroke.
Hypertension has been regarded as the most prevalent and powerful of the risk factors for ICH. The relationship has been established in several epidemiological studies in which a graded relationship between systolic and diastolic blood pressure levels and the risk of ICH has been reported.1 2 3 4 5 6 7 The mechanism is thought to involve a hypertension-induced degeneration of the walls of small arteries (lipohyalinosis) that leaves them prone to rupture.8
The majority of epidemiological studies linking elevated blood pressure to an increased risk of ICH were conducted before the widespread effective treatment of hypertension. Many also preceded the introduction of CT scanning and diagnosis was based on less reliable clinical criteria.1 2 3 4 5 Several of the larger studies were undertaken among Japanese or Japanese-American populations as part of the investigation of the high incidence of ICH in this racial group.4 6 7
The present study provided the opportunity to reassess the relationship between hypertension and ICH in a predominantly white population with a high level of detection and treatment of elevated blood pressure. It also allowed the assessment of risks in specific subgroups of hypertensive subjects.
The methods used have been described in detail elsewhere9 and are similar to those in previous studies of ischemic stroke involving the Melbourne Stroke Risk Factor Study (MERFS) Group.10 We identified consecutive cases of ICH occurring during the period from 1990 through 1992 by surveillance of discharge records from the 13 major city hospitals and by regular inspection of coroner’s reports. These hospitals manage most cases occurring in the Melbourne and metropolitan areas (population, 3.5 million), apart from those occurring among already disabled individuals in nursing homes.
Inclusion and Exclusion Criteria
Case subjects aged between 18 and 80 years were included when experiencing a first-episode primary ICH that was verified by CT (94.9%), autopsy (4.8%), or MRI (0.3%). Patients with any prior stroke or in whom the index stroke was a subarachnoid hemorrhage or secondary hemorrhagic infarction (transformation of nonhemorrhagic infarction to hemorrhagic infarction) were not included. Also excluded were those in whom a secondary cause of the stroke was identified (eg, arteriovenous malformation, tumor, bleeding diathesis) or in whom the stroke followed the ingestion of sympathomimetic drugs. Case subjects residing in nursing homes at the time of their stroke were likewise excluded from the series.
Control subjects of the same sex and similar age (±5 years) were identified from among neighbors of the case subject. Recruitment and interview of the control were carried out by the nurse who interviewed the matching case subject. A house-to-house approach was used to obtain the first eligible control living in the same (or nearest) street as the case. To ensure that unemployed or disabled persons were not overrepresented in the control series, repeated attempts at contact were made during evenings and weekends when a targeted control subject was not at home at the time of the initial visit.
Risk Factor Ascertainment
Nurse interviewers conducted in-person interviews using a structured questionnaire. Information was obtained about current and past cigarette smoking, alcohol consumption, physical activity, and past medical history (including a past history of hypertension). A detailed history of medication use was also recorded. This was validated by reference to prescription records in 70% of cases and 74% of controls. Although patient medical records were available for case subjects, care was taken to avoid accessing these information sources to reduce the likelihood of ascertainment bias.
The questionnaire has previously been validated with information from patients’ medical records.11 Agreement between the two sources for the major risk factors under study was excellent (κ=0.98).
When an eligible patient had died or was mentally impaired, dysphasic, or otherwise incapable, information was obtained from the closest available informant. To minimize information bias, participants acting as controls for these case subjects were asked to nominate a relative (of the same relationship as that of the case’s proxy to the case). Proxy interviews were obtained for 43% of cases and 31% of controls.
Because of the nature of ICH, interviewers could not be blinded to the case-control status of the interviewee. However, participants were informed that this was a study of lifestyle factors and stroke and were not told of the specific hypotheses under investigation.
Discussion of procedures for recently obtained interviews and scrutiny of response rates occurred at weekly quality-control meetings to ensure adherence to study protocols.
Site and Size of Hemorrhage
Each hemorrhage seen on CT, MRI, or autopsy was sited as follows: lobar, deep, subcortical, cerebellar, and pontine. Volumes of hemorrhages on CT were calculated manually by using an overlay grid scale, with each square representing 1 cm2 of the brain. The numbers of squares that were more than half filled were summed and then multiplied by the thickness of the tomographic slice. All tomographic slices showing hemorrhage were summed in this manner.12
Definitions were as follows: intracerebral hemorrhage: a sudden onset of an acute focal neurological event with confirmation of intraparenchymal ICH provided by CT, MRI, or autopsy; hypertension, diabetes, previous cardiovascular disease, and high cholesterol: a history of the condition as reported to the patient by a medical practitioner; claudication: a history of walking-induced calf pain that was relieved by rest; antihypertensive use: “ever” use of antihypertensive medications in the subject’s lifetime (this definition excluded patients taking antihypertensive medications for other conditions, eg, diuretics for chronic heart failure); currently using antihypertensives: use of antihypertensive medications at the time of the ICH (cases) or interview (controls); ceasing use: subjects who were not taking antihypertensive medication at that time but had reported using such medications in the past (up to 24 hours before the ICH [cases] or interview [controls]); hypertensive but never treated: individuals reporting a history of hypertension but who had never taken medication for the condition; never smoker: a person who had never smoked at least 1 cigarette, cigar, or pipe per day for at least 3 months at some period in his or her lifetime; current smoker: a person smoking at least 1 cigar, cigarette, or pipe per day for the preceding 3 months (this category was further divided into those who smoked <20 cigarettes per day on average and those who smoked ≥20 cigarettes per day); ex-smoker: a person who did not meet the criteria for never or current smoking; lifetime sedentary disposition: never exercising a minimum of once a week in the subject’s lifetime; alcohol consumption: ever drinking alcohol regularly in his or her lifetime, subcategorized into never drinker, previous drinker, and three levels of current alcohol consumption (light, medium, and heavy drinkers); body mass index: self-reported weight (kilograms) divided by the square of height (meters); survivors: patients who survived their initial hospitalization; and deceased: patients who died during their initial hospitalization or whose death was reported through the state coroner.
The OR of ICH was estimated for subjects in various categories of hypertensive status. Individuals who had reported never being advised of having hypertension and who had also never taken antihypertensive medications were the reference category.
Conditional logistic regression (using EGRET statistical software13) was used to compute ORs approximating the relative risks of ICH for various exposures. Initially, univariate ORs were calculated for hypertension and potentially confounding variables. We initially included all plausible potential confounding factors in the multivariate analyses. However, this resulted in substantially reduced numbers of matched sets with complete data. We then excluded those variables for which data were especially incomplete (aspirin and claudication) and assessed the impact on the hypertension ORs. Exclusion of these two factors altered the hypertension OR by <10%, and we therefore did not consider them further for confounding control. Interactions between hypertension and potential confounding variables (age, smoking, alcohol use, cholesterol, exercise, diabetes, gender, and previous cardiovascular disease) were assessed by the likelihood ratio statistic.14 CIs for ORs were based on large sample theory for conditional maximum likelihood estimators.14 Two-sided significance levels were used throughout.
The κ coefficient15 was used to test agreement between reports of medication use for hypertension obtained from medical practitioners and self-reported data.
This study was approved by the responsible ethics committees at Monash University and each of the participating hospitals. Informed consent was obtained from each participant after explanation of the purpose and methods of the study by a nurse interviewer. All procedures undertaken were in accordance with institutional guidelines.
A total of 370 eligible consecutive cases were identified. Of these, 29 refused to participate and 10 were not contactable, leaving an eventual case series of 331. To obtain the same number of control subjects, 342 age-, gender-, and geographically matched controls were identified, 11 of whom refused to participate. Matching by geographical area enabled some degree of matching for socioeconomic status. There were more men than women, and women were slightly older than men on average (Table 1⇓). Over 90% of subjects were white. More case than control subjects reported being hypertensive, using α-adrenergic antagonists and calcium channel blocking agents, never exercising regularly, and currently drinking heavily. More controls reported having a history of cardiovascular disease, having high cholesterol, and using angiotensin-converting enzyme inhibitors, β-adrenergic antagonists, or diuretics. The κ value for the agreement between patient recollections and medical practitioners’ records (n=487) was 0.67 (95% CI, 0.60 to 0.74), indicating a good measure of agreement (data not shown).
Table 2⇓ provides crude and adjusted ORs for ICH according to hypertensive status. The adjusted OR for ICH for those ever being told that they had elevated blood pressure was 2.45 (95% CI, 1.61 to 3.73). When a self-report of ever having used antihypertensive medication was used as a surrogate for hypertension, the adjusted OR was 2.44 (95% CI, 1.56 to 3.82). This was nearly identical to the OR obtained using a past history of hypertension.
Individuals who were ever told that their blood pressure was elevated were further classified within four age ranges. The impact of hypertension was greatest among those <55 years of age, and a decreasing trend was observed as age increased. The linear component of this decreasing trend in the (logarithm of) ORs resulted in an estimated decrease of 42% (95% CI, 13% to 62%) in OR for each 10-year increase in age (P=0.006), computed using age as a linear term in the interaction with hypertension.
The stroke risk among hypertensive subjects varied according to their current or past use of antihypertensive medication. The OR was 1.95 (95% CI, 1.20 to 3.16) for current users of antihypertensive medication compared with never users, whereas among those who had received antihypertensive agents in the past but had subsequently ceased this medication, the OR was 4.98 (95% CI, 2.25 to 11.02), with the ratio of the two ORs being significantly different from unity (ratio of two ORs, 2.56; P=0.002). The majority ceasing their antihypertensive medication had done so for more than 1 month (33 cases and 11 controls), while a minority had ceased taking medication within the relevant week (4 cases and 1 control). None of these antihypertensive agents was long-acting.
The OR associated with hypertension also varied according to subgroups defined by smoking status, with the interaction being statistically significant (P=0.018, Table 2⇑). When a comparison was made between hypertensive current smokers and normotensive current smokers, the OR was 6.12 (95% CI, 2.29 to 16.35). When the same comparison was made for never smokers, the OR was 2.92 (95% CI, 1.62 to 5.27). Because of the post hoc nature of this analysis, the finding must be interpreted with caution.
Further analyses were undertaken to determine whether the hypertension OR differed according to gender, cardiovascular disease history, alcohol use, serum cholesterol, diabetes history, or exercise. However, none of these differences reached statistical significance (all interaction values were P>0.5).
ORs are presented separately according to whether the response was from an index or proxy subject (Table 3⇓). For those individuals who were ever told that their blood pressure was elevated, the OR for ICH was 2.04 (95% CI, 1.32 to 3.15) among index subjects and 2.97 (95% CI, 1.50 to 5.89) among proxies. The ratio of these two ORs was not significantly different from unity (OR=0.69, P=0.364). Similar results were obtained for those reporting ever using antihypertensive medications, with an OR for ICH among index subjects of 2.07 (95% CI, 1.30 to 3.28) and among proxies of 2.81 (95% CI, 1.35 to 5.86). The ratio of these two ORs was similarly not significant (OR=0.74, P=0.490).
Table 4⇓ provides crude and adjusted ORs partitioned according to the size and location of the hemorrhage and according to the clinical outcome (fatal or otherwise). The impact of hypertension was similar regardless of the site of the hemorrhage or the size of the resulting hematoma, with neither of the interactions being statistically significant (P>0.9 for both interactions). However, hypertension was a greater risk factor for fatal ICH (OR, 10.84; 95% CI, 2.63 to 44.66) than for nonfatal hemorrhage (OR, 1.98; 95% CI, 1.23 to 3.21), with the ratio of the two ORs being significantly different from unity (ratio of ORs, 5.47; P=0.026).
The present study confirms the significance of hypertension as a risk factor for ICH. Among those ever told that they had hypertension, the risk of ICH was more than doubled. Among those who had been treated with antihypertensive drugs, the OR was significantly higher in those who had ceased therapy than in those continuing treatment.
Patients rarely cease antihypertensive therapy because their hypertension has resolved. More commonly, cessation of therapy represents a failure of compliance with medical advice. The higher risk of ICH in such patients emphasizes the importance of continuing therapy and adequate blood pressure monitoring, as well as encouraging compliance in patients with established hypertension. Although this concept might appear to be obvious based on trials of antihypertensive agents, there is currently no direct evidence that ceasing antihypertensive therapy is associated with increased stroke risk, particularly ICH. A possible reason for the increase in risk with cessation of medication use is that the thinning of the arterial wall that occurs with the use of many antihypertensive agents might render the wall more subject to injury when exposed to the high pressure that may occur when medications are ceased. It is not possible, however, to attribute this increased risk of ICH among persons ceasing antihypertensive therapy to the degree of management of hypertension because blood pressure was not measured in these subjects before their ICH.
Another striking finding of this study was the steep decline in relative risk with age, with the OR falling from 7.7 among those aged 15 to 54 years to 1.3 in those aged 65 to 74 years. A similar gradient was observed in the Honolulu Heart Program.16 In that study, the risk associated with hypertension decreased from 6.1 to 2.6 to 1.6 in the groups of ages 45 to 54, 55 to 64, and 65 to 81 years, respectively.
Several possible explanations exist for the lower relative risk of ICH among elderly patients. Risk factor prevalence surveys have shown that in this community a higher proportion of elderly hypertensives are identified and treated. For example, in the 1989 Australian Risk Factor Prevalence study, more than 85% of individuals aged 60 years and over reported having their blood pressure measured during the preceding 12 months.17 Among those aged 30 to 60 years, the corresponding figure was 72%.
The relationship between increasing age and a declining impact of hypertension might be explained by the vascular thickening that accompanies aging.18 19 Ultrastructural studies have shown that the small arteries seen in elderly patients are accompanied by medial hypertrophy, even in the absence of elevated blood pressure. There may be other conditions, eg, amyloid angiopathy, that increase risk among those without hypertension in the older age groups,20 accounting for a lower relative risk of ICH associated with hypertension. Conversely, the greater risk of ICH among hypertensive smokers might be due to a weakening of vessel walls resulting from adverse effects of nicotine (and possibly other components of cigarette smoke) on the endothelium and media of small vessels.21 22 23 This may potentiate any vessel wall weakening induced by hypertension. Although smoking has not previously been shown to increase the risk of ICH among hypertensives, a similar effect of smoking has been noted on the risk of all stroke.24
The relative risk imparted by hypertension was similar regardless of the position or size of the hemorrhage, although CIs were wide in some instances. This finding is in contrast with those of some previous studies, in which hypertension was associated principally with deep lesions.25 26 However, only cases with an outcome of fatality were included in these investigations, and these may not be representative of ICH in general. A more recent study that included both fatal and nonfatal outcome cases also found similar ORs among lobar and deep hemorrhages, in keeping with the results of this study.27
The relative risk of ICH due to hypertension was substantially higher in cases in which the outcome was fatal. This finding seems unlikely to be due to bias introduced by proxy respondents because the OR for hypertension among individuals for these two groups was markedly different (OR, 10.84 for fatal ICH and 2.97 among proxies). Furthermore, the ratio of the two ORs for index and proxy responses was not significantly different from unity (P=0.364), whereas the ratio of the hypertension ORs for fatal and nonfatal ICHs was significantly different (P=0.026). This finding may help to explain the observed decline in mortality from ICH28 (but possibly not in incidence) over recent years as blood pressure control has improved.
With the exception of smoking status, the risk imparted by hypertension was generally similar among subgroups, including those defined by gender, alcohol consumption, exercise habits, diabetes, and the presence of preexisting cardiac disease. An increased risk of ICH has been described previously among those consuming substantial quantities of alcohol.29 30 31 32 However, there was no evidence in this study that alcohol potentiates the risk of ICH associated with hypertension.
In general, the magnitude of the relative risk associated with hypertension was less than that reported in most previous studies of the condition (Table 5⇓).29 30 32 33 34 35 However, it is similar to that found in the only other major Australian study31 and may partly be explained by the high level of detection and management of hypertension in the Australian community, as well as other population differences, including the prevalence of smoking and other risk factors for ICH.36 It is unlikely to be explained by the broad definition of hypertension (ie, “ever told by a doctor” that hypertension is present) because the OR remained virtually unchanged when only those who had received drug therapy were included. It is also unlikely to be due to reporting bias influenced by restricted access and utilization of medical care, since medical care is either free or requires a small copayment and level of utilization is high.17 Furthermore, comparison of those subjects over 55 years of age in the 1989 Risk Factor Prevalence Study with controls in the present study produced similar proportions of persons who were not hypertensive (60% versus 63%, respectively), were hypertensive and taking medications for hypertension (26% versus 27%), and were not taking medication for their hypertension (14% versus 9%). The results provide support for the value of community-wide hypertension programs in reducing this form of stroke.
As with all case-control studies, the results are potentially influenced by various sources of bias and confounding. The most important of these is differential ascertainment of information from case and control subjects resulting from intellectual impairment among ICH survivors. In practice, this would be expected to reduce the observed ORs. However, the close agreement observed between drug histories obtained at interview and those obtained from the treating doctor suggests that bias resulting from this source was relatively small. Similar results were also obtained among subgroups with and without proxy interviews, suggesting that this methodological approach was also free of major bias. In support of this, previous authors37 have validated the use of proxies for the presence of hypertension and use of antihypertensive medications.
There may have been other potential confounding factors that were neither considered nor included in the analyses. Even though the participants were matched by age and neighborhood of residence, the matching may not have been adequate to account for these variables. Inclusion of age in the multivariate model to allow for case-control differences in age within each matched pair did not alter the results.
In summary, this study has revealed an increased risk of ICH among hypertensives, with a substantially higher risk among those who have ceased their antihypertensive therapy. The increased risk among hypertensives was also substantially greater in young individuals and among current smokers. The increased risk imparted by hypertension is considerably less than that found in most previous studies and is in keeping with the success of community-wide programs for the detection and management of hypertension in this community.
Selected Abbreviations and Acronyms
|MRI||=||magnetic resonance imaging|
This study was made possible by financial support from the Victorian Health Promotion Foundation, the National Health and Medical Research Council, Alfred Hospital Research Trust, and the National Stroke Foundation. We would also like to acknowledge the assistance of research nurses Fiona Ellery, Annie Crowe, Judy Snaddon, and Belinda Muir and the computer assistance of Lichun Quang.
Reprint requests to Dr A. Thrift, Department of Neurology, Austin and Repatriation Medical Centre, Studley Rd, Heidelberg, Victoria 3084, Australia.
- Received May 9, 1997.
- Revision received June 26, 1997.
- Accepted January 16, 1998.
Paffenbarger RS, Williams JL. Chronic disease in former college students, V: early precursors of fatal stroke. Am J Public Health. 1967;57:1290–1299.
Okada H, Horibe H, Ohno Y, Hayakawa N, Aoki N. A prospective study of cerebrovascular disease in Japanese rural communities, Akabane and Asahi, I: evaluation of risk factors in the occurrence of cerebral hemorrhage and thrombosis. Stroke. 1976;7:599–607.
Abu-Zeid HAH, Choi NW, Maini KK, Hsu PH, Nelson NA. Relative role of risk factors associated with cerebral infarction and cerebral hemorrhage: a matched pair case-control study. Stroke. 1977;8:106–112.
Lin CH, Shimizu Y, Kato H, Robertson TL, Furonaka H, Kodama K, Fukunaga Y. Cerebrovascular diseases in a fixed population of Hiroshima and Nagasaki, with special reference to relationship between type and risk factors. Stroke. 1984;15:653–660.
Kagan A, Popper JS, Rhoads GG, Yano K. Dietary and other risk factors for stroke in Hawaiian Japanese men. Stroke. 1985;16:390–396.
Thrift AG, McNeil JJ, Forbes A, Donnan GA. Risk factors for cerebral hemorrhage in the era of well-controlled hypertension. Stroke. 1996;27:2020–2025.
You R, McNeil JJ, O’Malley HM, Davis SM, Donnan GA. Risk factors for lacunar infarction syndromes. Neurology. 1995;45:1483–1487.
Rowe CC, Donnan GA, Bladin PF. Intracerebral haemorrhage: incidence and use of computed tomography. BMJ. [Clin Res]. 1988;297:1177–1178.
Statistical and Epidemiology Research Corp. Epidemiological Graphics, Estimation, and Testing Package. Seattle, Wash: Statistical and Epidemiological Research Corp; 1991.
Breslow NE, Day NE. Conditional logistic regression for matched sets. In: Davis W, ed. Statistical Methods in Cancer Research: The Analysis of Case-Control Studies. Lyon, France: IARC Scientific Publications; 1980:247–279.
Altman DG. Practical Statistics for Medical Research. London, UK: Chapman & Hall; 1991.
Curb JD, Abbott RD, MacLean CJ, Rodriguez BL, Burchfiel CM, Sharp DS, Ross GW, Yano K. Age-related changes in stroke risk in men with hypertension and normal blood pressure. Stroke. 1996;27:819–824.
Risk Factor Prevalence Study Management Committee. Risk Factor Prevalence Study: Survey Number 3, 1989. Canberra, Australia: National Heart Foundation of Australia and Australian Institute of Health; 1990.
Vinters HV, Gilbert JJ. Cerebral amyloid angiopathy: incidence and complications in the aging brain, II: the distribution of amyloid vascular changes. Stroke. 1983;14:924–928.
Bonita R, Scragg R, Stewart A, Jackson R, Beaglehole R. Cigarette smoking and risk of premature stroke in men and women. BMJ. 1986;293:6–8.
Freytag E. Fatal hypertensive intracerebral haematomas: a survey of the pathological anatomy of 393 cases. J Neurol Neurosurg Psychiatry. 1968;31:616–620.
McCormick WF, Rosenfield DB. Massive brain hemorrhage: a review of 144 cases and an examination of their causes. Stroke. 1973;4:946–954.
Broderick J, Brott T, Tomsick T, Leach A. Lobar hemorrhage in the elderly: the undiminishing importance of hypertension. Stroke. 1993;24:49–51.
Calandre L, Arnal C, Ortega JF, Bermejo F, Felgeroso B, del Ser T, Vallejo A. Risk factors for spontaneous cerebral hematomas: case-control study. Stroke. 1986;17:1126–1128.
Monforte R, Estruch R, Graus F, Nicolas JM, Urbano-Marquez A. High ethanol consumption as risk factor for intracerebral hemorrhage in young and middle-aged people. Stroke. 1990;21:1529–1532.
Jamrozik K, Broadhurst RJ, Anderson CS, Stewart-Wynne EG. The role of lifestyle factors in the etiology of stroke: a population-based case-control study in Perth, Western Australia. Stroke. 1994;25:51–59.
Juvela S, Hillbom M, Palomäki H. Risk factors for spontaneous intracerebral hemorrhage. Stroke. 1995;26:1558–1564.
Brott T, Thalinger K, Hertzberg V. Hypertension as a risk factor for spontaneous intracerebral hemorrhage. Stroke. 1986;17:1078–1083.
Schütz H, Bödecker RH, Damian M, Krack P, Dorndorf W. Age-related spontaneous intracerebral hematoma in a German community. Stroke. 1990;21:1412–1418.
Inzitari D, Giordano GP, Ancona AL, Pracucci G, Masalchi M, Amaducci L. Leukoaraiosis, intracerebral hemorrhage, and arterial hypertension. Stroke. 1990;21:1419–1423.
Australian Institute of Health, and Welfare. Australia’s Health 1992: The Third Biennial Report of the Australian Institute of Health and Welfare. Canberra, Australia: Australian Government Publishing Service; 1992.