Prediction of Functional Outcome and In-Hospital Mortality After Admission With Oral Anticoagulant–Related Intracerebral Hemorrhage
Background and Purpose—Early survival of patients with intracerebral hemorrhage in general is known to be most strongly dependent on the Glasgow Coma Scale score on admission. The aim of this study was to examine the factors determining functional outcome and in-hospital mortality of patients admitted with an intracerebral hemorrhage related to oral anticoagulant (OAC) use.
Methods—Correlation studies and multiple logistic regression analyses were performed on data from a retrospective series of 42 patients admitted with OAC-related intracerebral hemorrhages over a 6-year period to a tertiary care center in the north of Scotland.
Results—The functional outcome after an OAC-related intracerebral hemorrhage was dependent on maximum diameter of hematoma on CT scan (R=−0.72, P<0.001) and international normalized ratio (INR) (R=−0.35, P=0.024). Hematoma diameter and INR were not themselves strongly correlated (R=0.31, P=0.099). In-hospital mortality can be predicted by the Glasgow Coma Scale score alone (R2=0.36, overall predictive accuracy 68%) but more accurately by a logistic regression model including hematoma diameter and CT signs of cerebrovascular disease (R2=0.70, predictive accuracy 83%).
Conclusions—Neither functional outcome nor in-hospital mortality appears to be strongly dependent on INR measured on admission. CT scan, however, provides essential information and allows accurate predictions about the short-term outcome of OAC-related intracerebral hemorrhages.
For intracerebral hemorrhage in general, short-term mortality is strongly related to decreased levels of consciousness on presentation (odds ratio [OR] 4.8 to 11.8).1 2 3 Thirty-day (or in-hospital) mortality is also strongly dependent on the volume of hematomas measured on CT scan (eg, OR of up to 28.5 for volumes >60 cm3).1 3 Location of the hemorrhage in the posterior fossa2 and extension of the hemorrhage into the ventricular system1 2 4 are bad prognostic factors. Displacement of the midline structures is less clearly related to short-term mortality (OR 2.7).1 4 One reason that the above factors are not equally important in predicting early survival is the strong colinearity of the factors considered in multivariate analysis. Hence, the introduction of a combined variable “mass effect” in one of the studies mentioned.4 Mortality at 6 months, but not short-term mortality after an intracerebral hemorrhage, has been shown to be strongly dependent on age (OR 11.8 for each increase in age by 10 years).4
The present study of oral anticoagulant (OAC)-related intracerebral hemorrhages explores the possible correlations between the diameter of hematomas on CT scan and the intensity of OAC therapy measured on admission (international normalized ratio [INR]) and the relationships between outcome and hematoma diameter on one hand and outcome and INR on the other. Another objective of the present study was to identify clinical and/or radiological features that may accurately predict early mortality. This has never before been attempted in a systematic way for intracerebral hemorrhages related to OAC use.
Subjects and Methods
From January 1993 until March 1999, a total of 1512 patients had been admitted to Aberdeen Royal Infirmary (ARI) with intracranial hemorrhages. For most patients, we have been able to determine whether an intracerebral hemorrhage may have been associated with OAC use or not. For such cases, the following patient-related information was collected by a retrospective review of patient notes: sex, age, presence of hypertension, diabetes mellitus, hypercholesterolemia, cardiovascular disease (ie, atrial fibrillation, ischemic heart disease, or congestive heart failure), cerebrovascular disease, venous thromboembolism, alcohol abuse, liver and renal disease, cancer, and previous OAC-related hemorrhages. Furthermore, we recorded the following treatment characteristics: duration of OAC therapy, number of concomitant drugs, aspirin use, INR as measured on admission, most recently checked INR value, time interval between the latter 2 INR measurements, and concordance with the Scottish Intercollegiate Guidelines Network guidelines.5 When reviewing the notes of patients known to have been discharged from ARI with a diagnosis of OAC-related intracerebral hemorrhage between January 1993 and March 1999, we also recorded the following data for every patient: means of diagnosis, preceding head trauma, duration of symptoms before admission, Glasgow Coma Scale (GCS) score on admission, maximum diameter of the hematoma on CT scan (if performed), displacement of the midline (possibly associated intraventricular blood), location in the posterior fossa, signs of cerebral ischemia on CT scan, need for surgical intervention, short-term outcome (death or no, partial, or full recovery on discharge from ARI), and length of in-hospital stay.
The diameter of intracerebral hemorrhages and the corresponding INR values were nonnormally distributed; therefore, the nonparametric Spearman rank test was applied to confirm a possible correlation between both variables. In the same way, possible correlations were checked for outcome (death or no, partial, or full recovery) of intracerebral hemorrhages and hematoma diameter and for outcome and INR. The 5% significance level was adopted for hypothesis testing, and scatterplots were produced as illustrations. Multiple logistic regression analysis was used to investigate which of the patient and treatment characteristics and which of the clinical and radiological features of intracerebral hemorrhages are important for the prediction of in-hospital mortality (forward method) and to determine how much weight each of these factors carries (enter method). Where applicable, 95% CIs were calculated for the estimated ORs. Sensitivity, specificity, and positive and negative predictive values have been determined to assess the constructed predictive models. All statistical analyses were performed by use of SPSS 8.0.
Between January 1993 and March 1999, 1512 patients had been admitted to ARI because of a confirmed intracranial hemorrhage. From this total, 68 patients were found to have been treated with OACs at the time of sustaining their intracranial hemorrhage. Out of this total of 68 hemorrhages, 42 (62%) had been diagnosed as intracerebral. Data were missing with regard to the duration of OAC therapy for 1 patient, the INR value as checked on admission for 1 patient, and the time interval between the preceding and the most recent check for 11 patients. These 3 variables were, together with age and number of concomitant drugs, the only clinical variables under consideration that were continuous. The mean age of all 42 patients admitted with an OAC-related intracerebral hemorrhage was 71±10 years. The mean duration of OAC therapy was 29±43 months. The mean number of concomitant drugs was 3.0±1.9. The mean INR value on admission to hospital was 3.6±2.1 (Figure 1⇓). The INR values had, on average, been checked 26±25 days before hospital admission.
The above continuous variables have been further broken down into categorical variables: age categories (<65, 65 to 79, and >79 years), duration of OAC therapy (≤12, 13 to 95, and ≥96 months), use of ≤3 or >3 concomitant drugs other than warfarin, INR on admission (<2, 2 to 4.5, and >4.5), and most recently checked INR value (≤42 or >42 days ago). The frequencies for these respective variables, as well as for the previously mentioned dichotomous variables (eg, sex, presence of hypertension), are shown in Table 1⇓.
Among the 42 patients with OAC-related intracerebral hemorrhages, the diagnosis was made by means of CT scan in 36 patients, MRI in 1 patient, and autopsy in 2 patients. Size estimates based on CT scan were available for 30 patients (6 patients had been scanned elsewhere before referral to ARI). There were no missing data for any outcome variable other than those checked on the basis of CT scan.
Minor head trauma had preceded OAC-related intracerebral hemorrhages in 14% of cases. The mean duration of symptoms before admission to the hospital because of OAC-related intracerebral hemorrhages was 39±59 hours. The mean GCS score on admission was 11.9±4.4 (Figure 2⇓). On CT scan, 40% of all OAC-related intracerebral hemorrhages were associated with signs of cerebral ischemia. The mean maximum diameter of the hematomas was 40±21 mm. The distribution of diameters was clearly nonnormal (Figure 3⇓). Intraventricular blood was present on 45% of CT scans, the midline was displaced in 21% of cases, and 21% of the hematomas were located in the posterior fossa. One patient demonstrated the characteristics of a hemorrhagic infarct on CT scan. Surgical management was undertaken in 12% of the patients. At the time of discharge from the hospital, the distribution of clinical outcomes was as follows: 43% death, 9% no recovery, 41% partial recovery, and 7% full recovery. The mean duration of in-hospital stay was 14±13 days.
The relationship between hematoma diameter and the corresponding INR value found on admission is shown in Figure 4⇓. The diameter of OAC-related intracerebral hemorrhages was clearly correlated with clinical outcome (Figure 5⇓). The relationship between INR value and clinical outcome is shown in Figure 6⇓.
Although clinical outcome, as defined by the 4 above-mentioned categories, was not strongly correlated with INR at the time of admission, this would not preclude a strong relationship between in-hospital mortality and INR. To further investigate this, logistic regression was used to identify prognostic factors for in-hospital mortality. When all intracranial hemorrhages were considered together (n=68), short-term mortality appeared to be largely predictable on the basis of 3 variables: type of hemorrhage, admission within 12 hours of onset of symptoms, and level of consciousness on admission (GCS score <14) (R2=0.66, 74% sensitivity, and 100% specificity). Although included in the analysis, there was no place for any of the above-mentioned patient- or treatment-related characteristics in the model (eg, hypertension or INR). On the contrary, short-term mortality for all OAC-related intracranial hemorrhages could be predicted from the clinical characteristics of the hemorrhage alone.
When OAC-related intracerebral hemorrhages were considered separately, the GCS score on its own was found to predict in-hospital mortality most accurately (R2=0.36, sensitivity 46%, and specificity 83%). Other patient-related (eg, hypertension) or treatment-related (eg, INR) characteristics did not add to the predictive value of GCS alone. When the radiological (ie, CT scan) features were considered, a best predictive model was found on the basis of hematoma diameter and the presence of associated signs of cerebral ischemia. Details of this model are listed in Table 2⇓. It demonstrates a sensitivity of 73%, a specificity of 89%, a positive predictive value of 80%, and a negative predictive value of 85%. The most striking finding about this model is the negative sign of the parameter estimate for cerebral ischemia; ie, evidence of ischemia on CT scan appears protective against short-term mortality. This conclusion is reinforced when considering the multivariate ORs of in-hospital mortality for the different possible radiological features of OAC-related intracerebral hemorrhages (Table 2⇓).
The present study is the first to apply logistic regression to the short-term prognosis of patients admitted with OAC-related intracerebral hemorrhages. Because of the relative lack of patients with complete data, we have performed regression analysis twice, once for all clinical (ie, patient-, treatment-, and bleeding-related) characteristics and once for the radiological features. We have thereby been able to identify GCS score and hematoma diameter as the 2 most important independent prognostic factors for short-term mortality. This is in keeping with the previous findings for intracerebral hemorrhages in general and for those related to OAC use.
Diminished consciousness on admission was an important prognostic factor of mortality according to 2 studies,6 7 in accordance with the conclusions reached for the short-term prognosis of intracerebral hemorrhages in general.1 2 3
Five reports have commented on the effect of hematoma size on mortality. Three clearly showed the average size of hematomas to be larger in deceased patients than in survivors.8 9 10 Another report merely claimed that mortality increased with hematoma size,11 whereas the last report contradicted this finding.7 Hematoma size has, of course, also been found to be an important prognostic factor for the short-term survival of patients admitted with an intracerebral hemorrhage in general.1 3
For intracerebral hemorrhages in general (ie, without associated OAC use), there is some evidence to suggest that functional outcome, and not just mortality, is related to hematoma size.2 Two previous reports claimed that there was no such (strong) correlation between size and outcome for OAC-related hemorrhages,6 7 although neither provided any details to substantiate this claim. Our data suggest a strong correlation between functional outcome and the diameter of intracerebral hematomas. However, this correlation may be dominated by the effect of hematoma diameter on (short-term) mortality. Volumetric measurement of hematoma size is widely accepted as a reliable measurement but would have been available to us in only a proportion of patients who were examined in the most recent version of our CT scanner. However, to maximize the number of patients on whom some measure of hematoma size was available, we chose to use maximum diameter instead, in the knowledge that this measurement is readily available to clinicians in routine clinical practice even in the absence of sophisticated software.
Unexpectedly, radiological signs of cerebral ischemia appeared to have a significant protective influence. This finding cannot easily be explained, especially because we have previously found a past medical history of cerebrovascular disease to be a significant risk factor for intracerebral bleeding on warfarin (J.B. and J.W., unpublished data, 1999). One possibility is that gliosis resulting from previous major stroke may have permitted the accommodation of acute intracranial bleeding better than in patients without previous cerebral damage. Our classification of cerebral ischemia included gliosis, lacunae, and periventricular ischemia in single, multiple, or diffuse distributions. Further explanation of such appearances would be of interest. The few previous studies reporting the occurrence of OAC-related intracerebral hemorrhages in patients with old cerebral infarcts on CT scan have not provided any separate information on the case-fatality rate for those patients.1 6 10 12 In the above studies of prognostic factors for the short-term survival of patients with intracerebral hemorrhages in general, radiological or clinical evidence of existing cerebrovascular disease has not emerged as a risk factor.1 2 3 4 In previous reports, midline displacement,9 posterior fossa location,2 11 and intraventricular hemorrhage1 2 4 were associated with a high mortality. We found similar trends in univariate analysis, although they did not contribute significantly in the logistic regression model. It should be emphasized that our prognostic model deals with the chance of dying in the short term of an OAC-related intracerebral hemorrhage and not with the chance of sustaining such a hemorrhage in the first place.
Few reports have commented on a possible relation between clinical outcome and the intensity of OAC therapy measured on admission. Graphical data (without statistical analysis) provided in one report suggest that there is no such correlation for intracerebral hemorrhages.8 In the present study, we have found this correlation to be statistically significant for intracerebral hemorrhage, although not for intracranial hemorrhages in general (n=68).
Previously, studies reported on a possible correlation between the size of intracerebral hemorrhages and the corresponding intensity of OAC therapy measured on admission. One report claimed such a correlation, which was based on rather simple data (mean sizes were 91 cm3 for INR >3.6, 65 cm3 for INR 2.0 to 3.6, and 59 cm3 for INR <2.0) and no statistical analysis.1 Other reports found no strong correlation between hematoma size and intensity values, but there was no formal analysis.6 8 Only one report has appropriately provided the results of a statistical analysis (n=27, R=0.14, P=0.5).9 The results of the present study confirm the absence of a significant correlation between hematoma hemorrhage and anticoagulant intensity.
Two logistic regression models had previously been reported to predict the short-term mortality of patients with intracerebral hemorrhages in general.3 11 The present study was limited by incomplete data in a number of patients, but we still believe that our model(s) may be clinically relevant and merit validation in an independent patient sample, eg, for clinical decision making or advising patients and family about prognosis. Another important application is that future experimental studies of OAC-induced intracerebral hemorrhage (eg, into reversal of OAC therapy) may now be controlled for the factors identified in our analyses (particularly GCS score and hematoma diameter).
Finally, our data also confirm the importance of early clinical assessment and radiological confirmation of a suspected intracranial hemorrhage on warfarin.12 13 14 15 First of all, urgent CT scanning confirms the diagnosis and allows a rational decision to be made about reversal of anticoagulation. Second, it helps in the prediction of eventual outcome.
This study was conducted as part of a project leading to a master of science degree in clinical pharmacology (J. Berwaerts) at the University of Aberdeen.
- Received February 7, 2000.
- Revision received May 24, 2000.
- Accepted August 9, 2000.
- Copyright © 2000 by American Heart Association
Rådberg JA, Olsson JE, Rådberg CT. Prognostic parameters in spontaneous intracerebral hematomas with special reference to anticoagulant treatment. Stroke. 1991;22:571–576.
Steiner I, Gomori JM, Melamed E. The prognostic value of the CT scan in conservatively treated patients with intracerebral hematoma. Stroke. 1984;15:279–282.
Broderick JP, Brott TG, Duldner JE, Tomsick T, Huster G. Volume of intracerebral hemorrhage: a powerful and easy-to-use predictor of 30-day mortality. Stroke. 1993;24:987–993.
Daverat P, Castel JP, Dartigues JF, Orgogozo JM. Death and functional outcome after spontaneous intracerebral hemorrhage: a prospective study of 166 cases using multivariate analysis. Stroke. 1991;22:1–6.
Scottish Intercollegiate Guidelines Network (SIGN). Antithrombotic Therapy: A National Clinical Guideline. Edinburgh, UK: Royal College of Physicians; 1999. SIGN publication No. 36.
Forsting M, Mattle HP, Huber P. Anticoagulant-related intracerebral hemorrhage. Cerebrovasc Dis. 1991;1:97–102.
Franke CL, de Jonge J, van Swieten JC, Op de Coul AAW, van Gijn J. Intracerebral hematomas during anticoagulant treatment. Stroke. 1990;21:726–730.
Fogelholm R, Eskola K, Kiminkinen T, Kunnamo I. Anticoagulant treatment as a risk factor for primary intracerebral hemorrhage. J Neurol Neurosurg Psychiatry. 1992;55:1121–1124.
Kase CS, Robinson RK, Stein RW, DeWitt LD, Hier DB, Harp DL, Williams JP, Caplan LR, Mohr JP. Anticoagulant-related intracerebral hemorrhage. Neurology. 1985;35:943–948.
Portenoy RK, Lipton RB, Berger AR, Lesser ML, Lantos G. Intracerebral hemorrhage: a model for the prediction of outcome. J Neurol Neurosurg Psychiatry. 1987;50:976–979.
Lacroix P, Portefaix O, Boucher M, Ramiandrisoa H, Dumas M, Ravon R, Christides C, Laskar M. Conditions de survenue des accidents hémorragiques intracrâniens des antivitamines K. Arch Mal Coeur. 1994;87:1715–1719.
Volans AP. The risks of minor head injury in the warfarinised patient. J Accid Emerg Med. 1998;15:159–161.