Fever in Acute Stroke Worsens Prognosis
A Prospective Study
Background and Purpose No definitive data are yet available on the effects of body temperature on neurological damage after cerebral ischemia in humans. Experimental animal models have provided much evidence, but to our knowledge, only two studies on the relationship between fever and prognosis of stroke in humans have been published. The aim of our study was to investigate the prognostic role of fever in the first 7 days of hospitalization in a cohort of patients admitted to our hospital for acute stroke.
Methods We analyzed the data of 183 patients included in a prospective observational prognostic study. Vital status at 30 days was considered the main outcome and was obtained for all patients. Age, level of consciousness, and glycemia at the time of hospitalization were considered covariates for an exact logistic regression analysis. The maximum temperature recorded during the first 7 days dichotomized as “no or low fever” versus “high fever” was added to the model. Death within 10 days, taken as a secondary outcome suggestive of death from neurological causes, was analyzed with exact permutation tests.
Results Of the 183 patients analyzed in this study, 43% had fever during the first 7 days after hospitalization. The mean value of the maximum temperature recorded during the first 7 days in the 78 febrile patients was 38.3°C, and the median was 37.9°C. Onset of fever occurred in only 15% of febrile patients during the first day and in 49% on the second. The prognostic roles of age, level of consciousness, and glycemia were confirmed by exact logistic regression. Degree of consciousness impairment was the strongest prognostic variable, with an odds ratio (OR) of 11.4 (95% confidence interval [CI], 4.4 to 31.6). High fever (maximum temperature recorded during the first 7 days ≥37.9°C) was an independent factor for a worse prognosis, with an OR of 3.4 (95% CI, 1.2 to 9.5). The OR of dying within 10 days versus dying between 11 and 30 days was 4.9 (95% CI, 1.2 to 25.2) in patients with high fever with respect to all other patients.
Conclusions Fever in the first 7 days was an independent predictor of poor outcome during the first month after a stroke. No data were available on the underlying causes of fever, but the higher risk of death in the first 10 days, most frequently attributed to neurological mechanisms, suggested that high temperature was an independent component of poor prognosis and not only an epiphenomenon of other complications in the course after a stroke. In agreement with animal studies, we found that patients with higher temperature had a worse stroke outcome.
No definitive data are yet available in humans to explain and quantify the protective effects of hypothermia and the detrimental effects of hyperthermia on neurological damage after global or focal cerebral ischemia found in animals.1 2 Experimental mammalian models provide evidence that even mild hyperthermia up to 2°C above normal significantly increases ischemic neuronal injury.3 Timing is also relevant: the protective effect of hypothermia is enhanced when applied up to 1 hour after focal permanent ischemia.4 However, only two studies investigated the prognostic role of fever in stroke in humans: Hindfelt5 found that a rise in temperature of as little as 0.5°C was associated with a significantly worse outcome; Terent and Andersson6 found that fever greater than or equal to 38°C was significantly associated with poorer survival.
On the basis of these studies, fever is empirically considered prognostic of poor outcome in patients with acute stroke, and there is general agreement that fever of any cause should be treated with antipyretic agents to avoid its possible deleterious consequences.
In this report, data from a prospective observational prognostic study of patients admitted for acute stroke to the S. Orsola-Malpighi Community Teaching Hospital in Bologna, Italy, were analyzed to evaluate the role of fever on outcome.
Subjects and Methods
S. Orsola-Malpighi is one of the two major general hospitals in the city of Bologna, Italy. It contains 2543 beds distributed in 67 wards, including 14 medical departments where patients with acute stroke are usually treated. The data collected and analyzed in this report are part of a wider prospective observational study planned in three stages. First, the baseline prognosis of patients with acute stroke was evaluated (stage 1). Then guidelines were introduced to improve and standardize the whole approach (nursing, diagnostic, and therapeutic) to these patients in the hospital (stage 2). Stage 3, data collection after the introduction of the guidelines, is in progress. Sample size was calculated to detect a decrease of 1-month case-fatality rate from 33% in stage 1 to 20% in stage 3, with a 5% significance level and 80% test power. The case-fatality rate in stage 1 was estimated from a previous pilot study. A total of 395 patients were needed (195 for each stage). The patients considered in this report are those from stage 1.
Each day one half of the patients admitted to the emergency department of the hospital with a suspected diagnosis of acute stroke, excluding those with subarachnoid hemorrhage, were randomized into the study by extracting the patients’ names from a box. This randomization was necessary because we did not have the resources to apply the intensive follow-up schedule in real time to all consecutive patients. The study period went from February 1, 1992, to February 28, 1993, the day the sample size planned for the overall study was reached. Each randomized patient was examined within 48 hours by one of three neurologists (G.A., L.F., F.N.), who diagnosed the patients according to the criteria for stroke proposed by the World Health Organization.7 Two hundred nineteen patients not meeting these criteria were excluded. All the 204 patients diagnosed as having an acute stroke and included in the first stage of the study represented the initial sample for this analysis.
Variables recorded in the general database included date of birth, sex, time of onset of symptoms and of arrival in the hospital, risk factors for cerebrovascular diseases, laboratory tests, electrocardiography, CT scan, medical therapy, rehabilitation, and nursing. Level of consciousness was classified according to Plum and Posner.8 The Canadian Neurological Scale for fully alert or drowsy patients and the Glasgow Coma Scale for patients with a consciousness impairment greater than drowsiness were applied at days 1, 3, 10, 17, and 30 after hospitalization. The Barthel Index score was determined before admission and after 1, 3, and 6 months. Axillary temperature on the unaffected side of the body was recorded at least twice daily during the first 7 days of hospitalization according to the schedules of the various wards. MTEMP was defined as the maximum temperature observed during the first 7 days. Day of onset of fever (up to day 7) was also recorded.
Three variables associated with outcome were considered candidates for covariates.9 10 11 Patients were classified according to age (≤75 versus >75 years), level of consciousness at first examination (fully alert versus drowsy or comatose), and glycemia (≤6.7 versus >6.7 mmol/L) at the time of hospitalization. Vital status at 30 days was considered the main outcome. Absence of fever was defined as MTEMP of 37.2°C or lower. The median value of MTEMP in patients with fever was taken as the threshold between low and high fever; the latter included the threshold value.
After descriptive documentation of the relationship among the variables, exact logistic regression was performed and the exact odds ratios (ORs) estimated by means of the program LogXact.12 Student’s unpaired t test was used to compare means, and the Mann-Whitney nonparametric test was used to compare medians.
To estimate the importance of covariates, hierarchical backward stepwise regression was performed, with all interactions in the first model (saturated model). Nonsignificant interactions and factors were removed to arrive at a simple model, to which fever could then be added. Logistic regression with trichotomized fever described by two dichotomized variables was used to decide how many dichotomized variables, if any, to keep in the final model.
Death within 10 days was taken as a secondary outcome, suggestive of death from neurological causes. MTEMP was dichotomized on the basis of the results obtained from LogXact. Odds of dying within 10 days compared with dying between 11 and 30 days were analyzed in the group of patients who died within 30 days. Only the most important covariate from the first analysis was put into the model, to keep cell size adequate.
The aforementioned odds of dying within 10 days compared with dying between 11 and 30 days obtained from this fit were estimated with the use of permutation tests performed “exactly” by StatXact.13 The effect of the stratification factor was evaluated. Exact confidence intervals (CIs) were provided.
In the initial sample of 204 patients, values of glycemia were not available for 5 patients, 12 patients had no data on body temperature, and in 6 febrile patients the case notes did not allow exact quantification. Vital status at 30 days was known for all patients. Complete data were thus available for 183 patients (mean±SD age, 77.2±10.1 years; range, 45 to 95 years; 78 men). One hundred thirty-seven patients (75%) were fully alert, and 78 patients (43%) had fever during the first 7 days after hospitalization. For the whole group the mean value of glycemia was 7.5 mmol/L (SD, 3.4; range, 3.3 to 25.2 mmol/L). Twenty-four (18%) of the 137 fully alert patients had complete paresis, as did 43 (93%) of the 46 patients with impaired consciousness. Twenty-eight (42%) of the 67 patients with complete paresis developed fever of 37.9°C or greater compared with 14 (12%) of the 116 patients without complete paresis. The mean MTEMP in the 78 febrile patients was 38.3°C, and the median was 37.9°C. Onset of fever occurred in 15% of febrile patients (n=12) during the first day and in 49% (n=38) on the second. The overall 30-day mortality rate was 29% (53 patients, 36 of whom died within 10 days). Univariate OR suggested a prognostic role for age, glycemia, level of consciousness, fever, and early onset, whereas the OR for sex was not significant (Table 1⇓).
Prognostic Roles of Age, Consciousness Impairment, and Glycemia
Lethality of stroke ranged from 3% in younger fully conscious patients with normal glycemia to 91% in older patients with impaired consciousness and high glycemia. The effects of the three factors were confirmed by exact multivariate logistic regression. None of the interactions among the factors in the hierarchical models were significant. Their effect was to add “noise” and enlarge the CIs, and they were all removed from the final model (Table 2⇓). Degree of consciousness impairment was the strongest prognostic variable, with an OR of 11.4 (95% CI, 4.4 to 31.6).
Prognostic Role of Body Temperature
Lethality increased overall and in almost all strata as a function of temperature. Patients were divided into three subgroups, with no fever (≤37.2°C), lower fever (37.3°C to 37.8°C), and higher fever (≥37.9°C) than the median value. Both trichotomized MTEMP as a factor and “high fever” versus “low or no fever” were statistically significant. Ockham’s razor suggested the use of the simplest model, which is shown in Table 3⇓. Observed and predicted mortality are compared in the Figure⇓. With respect to the model without MTEMP (Table 2⇑), the prognostic values of age, degree of consciousness impairment, and glycemia were very similar in order of magnitude, and the OR of the dichotomized factor high fever versus low or no fever was 3.4 (95% CI, 1.2 to 9.5) (Table 3⇓).
Body Temperature and Early Death
A concentration of deaths within 10 days compared with those occurring between 11 and 30 days is evident among the patients with high fever, both in alert patients and in those with consciousness impairment (Table 4⇓). On the basis of the logistic regression model shown in Table 3⇑, MTEMP was dichotomized as no or low fever versus high fever.
The exact permutation test with stratification on level of consciousness gave the common OR as 4.6 (95% CI, 1.1 to 23.1). Without stratification, the OR was 4.9 (95% CI, 1.2 to 25.2), very similar to that found with the stratified analysis.
In this study we found that fever during the first 7 days of hospitalization occurred in 43% of our stroke patients and that onset of fever occurred in the first 2 days in 64% of febrile patients. These rates are similar to those described by Hindfelt,5 while Przelomski et al14 and Terent and Andersson6 described a lower incidence of fever. However, comparisons among various studies are difficult because of differences in the definition and measurement of fever. Considering death within 30 days after hospital admission as the main measure of outcome, we found that maximum temperature of at least 37.9°C was confirmed in a multiple logistic regression analysis to be significantly related to a poorer outcome, while fever per se (temperature ≤37.2°C versus >37.2°C) did not enter the model. It was not possible to estimate a threshold above which fever seemed detrimental. Probably a range of cutoffs would have given significant results. A dose effect, if present, should be detectable in a larger sample.
In our analysis we did not separate ischemic from hemorrhagic stroke because CT scan was performed in only 60% of our patients and therefore we could not estimate the different consequences of fever in the two types of stroke. This fact depended on the main design of the study, for which we needed to avoid whatever bias could have modified the diagnostic and therapeutic approaches to the patients before introducing the guidelines.
Our data did not provide information on underlying causes of fever in our patients. Thus, we could not exclude a priori that temperature of at least 37.9°C in the first 7 days after a stroke as a marker of poor prognosis might to some extent or entirely be an epiphenomenon of some common causes of fever (eg, pulmonary or urinary tract infections, sepsis, or pulmonary embolism from deep venous thrombosis). We could also not distinguish cases in which the stroke was elicited by infectious disease. However, if fever were the etiologic mechanism of the stroke, this could not have occurred except in some of the 12 patients who had fever during the first day of hospitalization.
We observed a concentration of early as opposed to later deaths among patients with fever of at least 37.9°C. Patients still alive at 30 days were irrelevant to the comparison. The good agreement between the stratified and unstratified analyses suggested an underlying relationship between high fever and early death.
High fever thus seemed to be associated with a higher probability of early death. This was evidence that fever of at least 37.9°C indicated poor early prognosis even without a consideration of its causes. Since the majority of the experimental studies documented the direct effects of temperature on neurological damage,1 2 our data suggest that fever could worsen prognosis through direct neurological damage.
Only two studies have investigated the prognostic significance of fever in stroke. Hindfelt5 found that a mean body temperature above 37.5°C from any cause in the first 7 days was associated with poor prognosis at 2 months after the stroke. However, his study was retrospective, the sample excluded patients who died within 2 months, and the measurement of outcome was not validated, so that his conclusions are not easily generalized to the whole population of stroke patients.
In a prospective study of 281 patients with stroke, Terent and Andersson6 found that in patients with mean body temperature of 38°C or more during the first week, chest x-ray films revealed bronchopneumonia in half. Fever as defined above indicated a significantly worse prognosis, but their data did not distinguish between the consequences of complete paresis and body temperature. We could not consider complete paresis because of its strong association with level of consciousness impairment, reported as having a superior prognostic value for early mortality.9
Przelomski et al14 prospectively investigated the frequency and causes of fever in a sample of 104 consecutive stroke patients. In particular, these authors studied the possible association between fever, almost always secondary to infections, and the size of the lesion. Comparison with our study is difficult because the authors excluded brain stem infarcts and intraventricular hemorrhage, conditions in which “neurogenic fever” is most likely, and they did not evaluate the prognostic value of fever.
In conclusion, this is the first prospective study specifically investigating the prognostic role of fever in patients with acute stroke. Although we did not distinguish between hemorrhagic and ischemic strokes and no data were available on the underlying causes of fever, our results indicated that patients with higher temperature have a worse prognosis. Further studies distinguishing between the two types of stroke would be useful to investigate the relationship between intracranial hemorrhage and rise in temperature and to evaluate the role of infections. Our results support a lowering of fever, as previously recommended in almost all guidelines for medical therapy of stroke. Controlled trials to evaluate the effects of lowering body temperature are also necessary.
This study was supported in part by Ministero dell’Università e della Ricerca Scientifica e Tecnologica (60%). This study would not have been possible without the collaboration of the registered nurses Vera Serra, Rovena Rubini, and Eleonora Conti, who were responsible for daily review of the emergency department reports and collection of laboratory data. We wish to thank Ubaldo Montaguti, MD, Daniela Celin, MD, and Giambattista Spagnoli, MD, for their invaluable help in setting up this study. We also thank Anne Collins for revising the English.
- Received March 6, 1995.
- Revision received August 8, 1995.
- Accepted August 16, 1995.
- Copyright © 1995 by American Heart Association
WHO Task Force on Stroke and Other Cerebrovascular Disorders. Stroke—1989: recommendations on stroke prevention, diagnosis and therapy. Stroke. 1989;20:1407-1431.
Plum F, Posner JB. The pathologic physiology of signs and symptoms of coma. In: Plum F, Posner JB, eds. The Diagnosis of Stupor and Coma. Philadelphia, Pa: FA Davis Co; 1980:1-5.
Rasmussen D, Kohler O, Worm-Petersen S, Blegvad N, Jacbsen HL, Bergmann I, Egeblad M, Friis M, Nielsen NT. Computed tomography in prognostic stroke evaluation. Stroke. 1992;23:506-510.
Kiers L, Davis SM, Larkins R, Hopper J, Tress B, Rossiter SC, Carlin J, Ratnaike S. Stroke topography and outcome in relation to hyperglycemia and diabetes. J Neurol Neurosurg Psychiatry.. 1992;55:263-270.
Mehta C, Patel N, eds. LogXact-Turbo Logistic Regression Software Featuring Exact Methods: User Manual. Cambridge, Mass: CYTEL Software Corp; 1993.
Mehta C, Patel N, eds. StatXact-Turbo, Statistical Software for Exact Nonparametric Inference: User Manual. Cambridge, Mass: CYTEL Software Corp; 1992.
Przelomski MM, Roth RM, Gleckman RA, Marcus EM. Fever in the wake of a stroke. Neurology. 1986;36:427-429.