Risk Factors for Intracranial Hemorrhage in Outpatients Taking Warfarin

Intracranial hemorrhage is the most feared complication of anticoagulant therapy. Outcomes are frequently catastrophic, often resulting in death or severe neurologic disability. The effect of intracranial hemorrhage is as great as that of the thromboembolic events warfarin is used to prevent. As a result, the risk for intracranial hemorrhage is a critical feature of the decision to use anticoagulation [1]. The indications for use of anticoagulants are expanding, particularly among older patients. For example, anticoagulant therapy is now recommended to prevent stroke in patients with atrial fibrillation [2-6], and it has been found to significantly improve outcome in patients after myocardial infarction [7]. Clinical trials of warfarin as primary preventive therapy for ischemic heart disease are ongoing [8]. With this increase in the use of long-term warfarin therapy, there is a pressing need to identify clinical features that raise the risk of its most severe bleeding complications. No previous study has identified independent risk factors for intracranial hemorrhage among patients taking warfarin. Previous prospective analytic studies that focused on the entire spectrum of major bleeding complications have contained few cases of intracranial hemorrhage [9-11]. We designed a casecontrol study focused exclusively on intracranial hemorrhage occurring among outpatients taking warfarin. We drew on an 11-year experience of one general hospital to provide case-patients and used the same hospital's large anticoagulant therapy unit to provide contemporaneous controls who were also taking warfarin. This design was chosen to increase statistical power for detecting risk factors for intracranial hemorrhage while reducing possible bias. Methods Case-Patient Identification and Eligibility Using a discharge log of consecutive admissions to the Massachusetts General Hospital during the period from 1 January 1981 through 31 December 1991, we identified 1881 patients with a principal diagnosis of intracranial hemorrhage (ICD-9 codes 430, 431, 432.0, 432.1, and 432.9). Forty-one (2%) medical records for these patients could not be located. In the remaining 1840 case-patients, warfarin use was determined from review of the neurologist, neurosurgery resident, and attending staff physician admission notes. Intracranial hemorrhage was verified by computed tomographic (CT) scanning, lumbar puncture, or postmortem examination in all but one case-patient. In this latter case-patient, the diagnosis was based on clinical grounds because the patient died before diagnostic studies were done and no autopsy was performed. To be eligible for the study, patients had to be at least 18 years old and taking warfarin as an outpatient. Patients with an anatomic abnormality or underlying bleeding diathesis predisposing to intracranial hemorrhage, regardless of anticoagulant therapy, were not included. Hemorrhages sustained as a result of major head trauma (with skull fracture and loss of consciousness) also were not eligible. Of 131 patients with intracranial hemorrhage identified, 10 were excluded as ineligible: Four patients had subarachnoid hemorrhage resulting from angiographically identified intracranial aneurysms; 2 had hemorrhage into primary or metastatic tumors; 2 had acute subdural hemorrhage after major head trauma; 1 bled after multiple craniotomies and radiation therapy for a recurrent craniopharyngioma; and another patient had aplastic anemia. Controls: Source and Matching Controls were selected from the registry of the Massachusetts General Hospital's anticoagulant therapy unit. During the study period, this unit managed warfarin dosing for approximately 8000 patients referred from all hospital clinical services. The most common indications for anticoagulation were atrial fibrillation, previous stroke, presence of prosthetic heart valves, and venous thromboembolism. Approximately 40% of the patients managed by the anticoagulant therapy unit have prothrombin time tests done at the Massachusetts General Hospital. An additional 20% have their prothrombin times measured at one large commercial laboratory (via home phlebotomy services). The remaining 40% have their tests done at various local laboratories. Each case-patient was matched to three randomly selected controls taking warfarin at the time of the case-patient's intracranial hemorrhage. Matching was accomplished by first identifying all patients managed by the anticoagulant therapy unit at the date of hospital admission for the given case-patient. Each potential control was assigned a random number. The three controls with the lowest random numbers were selected. No control was used more than once. Three controls subsequently became case-patients. Matching was done to control for any change in background risk over the 11-year study period (for example, from changes in prothrombin time targets or technique of measuring the prothrombin time). Data Collected Clinical features of case-patients and controls were extracted primarily from hospital records, with supplementation in a few instances from physician office records. Variables were entered on a predesigned data form and included indication for anticoagulant therapy; race; sex; age; prothrombin time ratio (PTR); duration of warfarin therapy; history of diagnosed hypertension, stroke, transient ischemic attack, diabetes mellitus, myocardial infarction, atrial fibrillation, and congestive heart failure; and medications. Clinical features of the intracranial hemorrhages were also recorded. Clinical data for the controls were current at the admission date of the matched case-patient with intracranial hemorrhage. The prothrombin time ratio was expressed as the ratio of the patient's value divided by the simultaneously reported control value. For case-patients, we used the PTR on admission, or, if available, the PTR closest to the reported onset of symptoms. For controls, the PTR closest to that of the case-patient admission date was recorded from the anticoagulant therapy unit database. Because values for the international sensitivity index (ISI) for thromboplastins were not universally reported before 1988, we analyzed our data using the PTR rather than the international normalized ratio (INR). Since 1988, Massachusetts General Hospital's hematology laboratory has used Simplastin Automated (Organon Teknika Corporation, Durham, North Carolina), with values for the ISI ranging from 1.9 to 2.0. During the study period, this company (previously General Diagnostics) supplied the thromboplastin and assisted our laboratory in selecting lots of comparable sensitivity. Studies were routinely done to minimize year-to-year variation in thromboplastin sensitivity. Since 1988, the most frequently used commercial laboratory has used thromboplastins with ISI values ranging from 1.9 to 2.1. The PTR was missing for four case-patients; the data for case-patients and controls were otherwise complete. Selected Relevant Definitions Hypertension was defined as probable if the patient had such a diagnosis listed in the medical record and as definite if the patient was receiving antihypertensive medication. When hypertension was diagnosed, it was classified as definite 97% of the time for case-patients and 93% of the time for controls. All diagnoses of hypertension are used in the analyses. Patients with documented carotid or vertebrobasilar disease and those with a history of previous stroke were defined as having cerebrovascular disease. Diagnoses of carotid or vertebrobasilar disease or both were confirmed by angiography in 71% of case-patients and 93% of controls, by Doppler studies in 17% of case-patients and 2% of controls, and solely by assessment by a neurologist in 12% of case-patients and 5% of controls. Statistical Analysis Case-patients and controls were compared using chi-square tests and the Fisher exact test, where appropriate, for categorical variables and using the Student t-test for continuous variables. Univariate odds ratios were calculated using unmatched and matched techniques. The matched odds ratios were provided by the Mantel-Haenszel [12] summary statistic across matched sets. The unmatched and matched techniques provided very similar estimates of odds ratios. We report the unmatched results. Confidence intervals for odds ratios were calculated using the Taylor series method [13]. The test of trend was done using the Cochran-Mantel-Haenszel test [14]. Logistic regression models assessed the independent effect of multiple clinical features and the significance of interaction terms. Conditional logistic models [15] accounting for matching provided estimates similar to those of the unmatched logistic regression analyses. The estimates from the unmatched analyses are reported. Data were recorded in R:BASE (Microrim, Bellevue, Washington) from the paper data forms. Statistical analyses were done using SAS (SAS Institute Inc., Cary, North Carolina) and GLIM (Numerical Algorithms Group Limited, Oxford, United Kingdom). Results Clinical Course of Case-Patients During the 11-year study period, 121 patients with intracranial hemorrhage were eligible; 77 hemorrhages were intracerebral and 44 were subdural (Table 1). Three of the patients with subdural hemorrhage had a history of trivial head trauma; the others had no known antecedent head trauma. For the patients with intracerebral hemorrhages, headache was the most common presenting feature (53%), followed by nausea and vomiting (40%) and unresponsiveness (36%). Seventy-eight percent of the case-patients presented to the emergency department within 24 hours of the onset of symptoms and 87% within 48 hours. In contrast, only 36% of the patients with subdural bleeding came to medical attention within this same period. Forty-six percent of patients with intracerebral bleeding died, and 17% survived with major neurologic deficits that prevented subsequent independent living. Table 1. Clinical