DVT indicates deep venous thrombosis; NICU, neurologic intensive care unit; and PE, pulmonary embolism.
Shaded area indicates presence of a variable; IVH, intraventricular hemorrhage; and SAH, subarachnoid hemorrhage.
Decision tree indicates optimal identification of febrile, critically ill neurologic and neurosurgical patients in whom antibiotics may be safely discontinued. Only patients with all variables available are included. IVH indicates intraventricular hemorrhage; SAH, subarachnoid hemorrhage.
Hocker SE, Tian L, Li G, Steckelberg JM, Mandrekar JN, Rabinstein AA. Indicators of Central Fever in the Neurologic Intensive Care Unit. JAMA Neurol. 2013;70(12):1499-1504. doi:10.1001/jamaneurol.2013.4354
Copyright 2013 American Medical Association. All Rights Reserved. Applicable FARS/DFARS Restrictions Apply to Government Use.
Fever is common in critically ill neurologic patients. Knowledge of the indicators of central fever may allow greater antibiotic stewardship in this era of rapidly developing super-resistant microorganisms.
To develop a model to differentiate central from infectious fever in critically ill neurologic patients with fever of an undetermined cause.
Design, Setting, and Participants
Retrospective data collection from January 1, 2006, through December 31, 2010, at a 20-bed neurologic intensive care unit of a large teaching hospital. Consecutive patients 18 years and older admitted for 48 hours or longer with a core body temperature higher than 38.3°C on at least 1 measurement for 2 consecutive days. Patients with alternative identified causes of noninfectious fever were excluded. In total, 526 patients were included in the final analysis.
Main Outcomes and Measures
Percentage incidence and odds ratios of variables associated with central fever. Fever was classified as infectious if there was culture growth of a pathogenic species or documented clinical diagnosis of infection treated with antibiotics. Remaining patients were considered to have central fever. Continuous fever lasting longer than 6 hours for 2 or more consecutive days was considered persistent.
Fever was central in 246 patients (46.8%). Patients with infectious fever were older (mean, 57.4 vs 53.5 years; P = .01) and had a longer length of stay in the neurologic intensive care unit (mean, 12.1 vs 8.8 days; P < .001). Central fever was more likely to occur within 72 hours of admission to the neurologic intensive care unit (76.4% vs 60.7%; P < .001) and tended to be persistent (26.4% vs 18.6%; P = .04). Blood transfusion (odds ratio [OR], 3.06; 95% CI, 1.63-5.76); absence of infiltrate on chest x-ray (3.02; 1.81-5.05); diagnosis of subarachnoid hemorrhage, intraventricular hemorrhage, or tumor (6.33; 3.72-10.77); and onset of fever within 72 hours of hospital admission (2.20; 1.23-3.94) were independent predictors of central fever on multivariable analysis. The combination of negative cultures; absence of infiltrate on chest radiographs; diagnosis of subarachnoid hemorrhage, intraventricular hemorrhage, or tumor; and onset of fever within 72 hours of admission predicted central fever with a probability of .90.
Conclusions and Relevance
We provide a reliable model to differentiate central fever from infectious fever in critically ill neurologic patients, allowing clinicians to select patients in whom antibiotics may be safely discontinued despite ongoing fever.
Fever is common in the neurologic intensive care unit (NICU), occurring in up to 70% of patients.1- 5 At least half of these febrile episodes are caused by infection.2,4- 6 Central fever results from complex disturbances of central mechanisms of thermoregulation.7,8 In the NICU, central fever is frequently considered after infection and other noninfectious causes of fever have been excluded, such as drug administration or withdrawal, venous thromboembolism, or blood transfusion reactions. Identifying and treating the cause of fever is imperative because of the ramifications of failing to identify a treatable condition; the negative consequences of antibiotic overuse, including adverse events and selection of multidrug-resistant organisms in the NICU; and the detrimental effect of hyperthermia on brain-injured patients.9- 14
Because of the concern for infection in critically ill patients, broad-spectrum antimicrobials are often administered at the onset of fever after obtaining appropriate culture specimens and may be continued unnecessarily in cases of central fever. With the proliferation of multidrug-resistant pathogens, antibiotic stewardship is becoming increasingly important. An improved understanding of predictors of central fever may allow for earlier antibiotic discontinuation in a population in whom fever is common and often persistent.2 The few studies that have specifically evaluated clinical characteristics or predictors of central fever had a limited scope.4- 6
The objective was to develop a model to differentiate central from infectious fever in critically ill neurologic patients to identify a population in whom antibiotics could be safely discontinued despite the persistence of fever.
This study was approved by the Mayo Clinic Institutional Review Board and did not require patient permission because it is a retrospective review.
Consecutive patients admitted for more than 48 hours to the NICU of Saint Marys Hospital (Rochester, Minnesota) between January 1, 2006, and December 31, 2010, were identified using our electronic data search system and screened for the occurrence of fever. Fever was defined as a core body temperature higher than 38.3°C (101°F) documented on at least 1 measurement for 2 consecutive days. Causes of noninfectious, noncentral fever, including venous thromboembolism, drug fever, and blood transfusion reactions, were identified using predefined diagnostic criteria15,16 and were excluded from the analysis.
Collected data included (1) patient demographics; (2) neurologic diagnosis; (3) fever onset and duration; (4) maximum temperature; (5) presence of persistent fever (continuous fever for >6 hours for ≥2 consecutive days); (6) cultures of urine, blood, respiratory secretions, cerebrospinal fluid, ventricular fluid, shunt fluid, peritoneal fluid, stool, sinus aspirates, and Clostridium difficile polymerase chain reaction; (7) findings on chest radiographs within the first 48 hours from fever onset; (8) presence of an endotracheal tube, ventriculostomy, indwelling bladder catheter, or arterial or central venous catheter at fever onset; (9) leukocyte count and percentage of neutrophils on the first day of fever; (10) presence of criteria for systemic inflammatory response syndrome at the onset of fever17; (11) duration of mechanical ventilation; (12) NICU and hospital length of stay; and (13) mortality. For each patient, we divided the number of days without fever by the number of days in the NICU to form a “fever-free ratio.” A positive culture was defined as the growth of any microbial species in any normally sterile bodily fluid. In patients with subarachnoid hemorrhage (SAH), we documented the presence of vasospasm, defined as transcranial Doppler velocity of more than 120 cm/s in the anterior circulation or vessel narrowing on angiography.
Body temperature was measured continuously by bladder probes and documented hourly throughout patients’ NICU stay. Fever was treated uniformly with a practice protocol consisting of acetaminophen and mechanical cooling measures, including cooling blankets and ice packs.
Classification of fever was based on review of all available data for the episode of care. Fever was classified as infectious (1) if any culture from a normally sterile source grew a pathogenic species or (2) there was a documented clinical diagnosis of infection or sepsis syndrome in the medical records and the patient received a full therapeutic course of antibiotics. For this study, all species were considered pathogenic except (1) Candida species growing in a sputum culture, (2) coagulase-negative Staphylococcus growing in one-third or one-fourth bottles of blood culture, and (3) Candida species growing in a urine culture obtained from an indwelling bladder catheter. The remaining patients were classified as having central fever. Fever classification was performed by a research fellow (L.T.). Expert reviewers (S.E. H., J.M.S., and A.A.R.) performed a masked review of 60 cases to classify the fever type without using the study definitions, and results were compared with the classification of the research fellow. In addition, when fever classification was deemed questionable, cases were assessed by 2 reviewers who did not initially review the case, and classification was reached by consensus. When consensus could not be reached, the fever was classified as infectious.
Agreement among the reviewers was assessed using the κ statistic, with a κ value of greater than 0.75 considered excellent.18Descriptive summaries were reported as mean (SD) for continuous variables and counts and frequencies for categorical variables. Univariate comparisons between subgroups were performed using the χ2 test or Fisher exact test for categorical variables. Continuous variables were compared using the Wilcoxon rank sum test. All tests were 2-sided, and P < .05 was considered statistically significant. We used univariate and multivariate logistic regression analyses with fever as a binary outcome variable to assess the associations with predictor variables. Magnitude of association with outcome was expressed as odds ratio and 95% CI. Prediction probabilities were estimated based on the final multivariate model. We chose variables to include in the probability table that would be applicable early in the hospital course because our goal was to develop a model clinicians could use to predict central fever to discontinue empirical antibiotics. Recursive partitioning was then used to develop a decision tree. We used JMP, version 9.0 (SAS, Inc) and SAS, version 9.2 (SAS, Inc) to analyze the data.
Among 8761 patients with research authorization admitted to the NICU during the study period, 1302 (14.9%) experienced a febrile episode. In total, 526 patients were included in the final analysis after excluding those who (1) were younger than 18 years; (2) were admitted to the NICU for shorter than 48 hours; (3) had an identified noninfectious, noncentral cause of fever (37/952; 3.9%); or (4) had fever that occurred on only 1 day (Figure 1). Mean (SD) age was 55.6 (19.1) years, and 322 (61.2%) were men. Fever was central in 246 patients (46.8%). There was excellent agreement on fever classification between the expert reviewers and the research fellow (κ, 0.8; 0.64-0.95).
There were no differences between the central and infectious fever groups in sex; maximum temperature; rate of use of central venous catheters, indwelling bladder catheters, ventriculostomy catheters, or endotracheal tubes; leukocyte count; or the presence of systemic inflammatory response syndrome at the onset of fever. The patients with infectious fever were significantly older (mean, 57.4 vs 53.5 years; P = .01) and were mechanically ventilated longer (mean, 6.8 vs 3.5 days; P < .001) than those with central fever. Patients with status epilepticus tended to have infectious fever more commonly (3.9% vs 0.8%; P = .02), while central fever predominated in patients with brain tumors (9.8% vs 3.2%; P = .002), intraventricular hemorrhage (IVH) (42.7% vs 25.0%; P < .001), and SAH (25.2% vs 14.3%; P = .002). Vasospasm was present in 43 patients with SAH (42.2%) and was predictive of central fever (50.8% vs 27.5%; P = .02). Blood transfusions were more common in the central fever group (25.6% vs 11.1%; P < .001). Central fever was more likely to occur within 72 hours of NICU admission (76.4% vs 60.7%; P < .001) and tended to be persistent (26.4% vs 18.6%; P = .04). Fever-free ratio was lower in the central fever group (0.5 vs 0.6; P < .001). Patients with infectious fever had significantly longer NICU admissions (mean, 12.1 vs 8.8 days; P < .001). There was no difference in NICU or hospital mortality between groups. Table 1 summarizes the comparison between the patients with infectious and central fever on univariate analysis.
Multivariate analysis confirmed that having received a blood transfusion; absence of infiltrate on chest radiograph; diagnosis of SAH, IVH, or tumor; and onset of fever within 72 hours of admission were strongly predictive of central fever (Table 2). The combination of negative cultures; absence of infiltrate on chest radiograph; diagnosis of SAH, IVH, or tumor; and onset of fever within 72 hours of admission predicted central fever with a probability of .90 (Figure 2). A decision tree using these variables is shown in Figure 3.
We identified a set of variables associated with central fever in neurocritical patients. Patients with negative cultures, chest radiographs without infiltrates, and onset of fever within 3 days of hospitalization are likely to have central fever, especially if their primary diagnosis is SAH or tumor or they have IVH. In these patients, antibiotics might be safe to discontinue. The importance of identifying the central nature of the fever is highlighted by our findings that fever burden is greater and onset is earlier when fever is caused by acute neurologic injury rather than by an infection. Central fever tends to appear within 72 hours of admission, persists longer, and is present on more days after accounting for NICU length of stay. Because central fever starts earlier and lasts longer than infectious fever, the possibility of antibiotic overuse, with the associated risk of the emergence of resistant microorganisms, is particularly high in these cases.
The frequency of fever in our NICU (14.9%) is lower than in some previously reported neurocritical care and neurosurgical populations.2,4 This is likely because the frequency was calculated before excluding patients admitted for less than 48 hours, a substantial proportion of whom were elective admissions following surgical procedures. A NICU admission less than 24 hours has been associated with a fever frequency of 15%.2 The lower fever rate may also reflect differences in patient acuity or the effect of our treatment practices to maintain normothermia. Noninfectious, noncentral causes of fever were identified in 3.9% of patients. This is consistent with rates reported in other NICUs.4,6 We cannot report the true incidence of central or infectious fever in this study because we did not classify the fever until patients with noninfectious, noncentral fever and those with fever occurring on only 1 day had been excluded.
Having a brain tumor, IVH, or SAH was strongly associated with the development of central fever, and in patients with SAH, vasospasm was associated with central fever. Fever has been associated with SAH and, more specifically, with the development of delayed cerebral vasospasm in several retrospective studies4,12,19 and 1 prospective study.6 The relationship between IVH and fever has been described in animal studies20,21 and in patients with intracerebral hemorrhage accompanied by IVH.11 Yet, to our knowledge, ours is the first study to directly demonstrate a strong association between the presence of IVH and the development of noninfectious fever. Intraventricular hemorrhage is thought to alter hypothalamic function, resulting in temperature set-point elevation, which is hypothesized to be related to direct damage to thermoregulatory centers in the preoptic region, stimulation of prostaglandin generation, and impaired inhibitory feedback from the lower midbrain that typically serves to suppress thermogenesis.7
Fever has been associated with neoplasms; however, the relationship between brain tumors and fever has not been systematically examined. Fever has been observed in tumors located in the sella, diencephalon, and intraventricular region.22 We did not collect data on the location of the tumors in the brain and thus cannot postulate whether particular locations may increase the likelihood of central fever. These tumors are often treated surgically, and thus these patients may be prone to develop postoperative fever. Others have postulated that fever may be provoked by irritation of the leptomeninges either by the tumor or its breakdown products.23
In our study, blood transfusion was associated with central fever. Blood transfusion is more common in patients with greater severity of brain injury, as we have previously shown in our patients with aneurysmal SAH.24 Therefore, we cannot exclude that selection bias may have influenced this association. Patients in whom fever occurred as a direct reaction to the blood transfusion were excluded from the analysis.
Central fever was more likely to occur within 72 hours of NICU admission and tended to be persistent. Fever-free ratio was lower in the central fever group, indicating more days of fever after accounting for the NICU length of stay. These findings suggest a higher fever burden among patients with central fever. Early onset has been associated with central fever.6 Sustained fever is also thought to be suggestive of central fever,25 and this is supported by the association of persistent fever and a central origin in our study.
We found that patients with infectious fever were older and had a longer duration of mechanical ventilation and NICU length of stay. Older patients are likely predisposed to infection because of decreased immunity and comorbidities, which may delay liberation from mechanical ventilation26 and prolong NICU length of stay. The increased rate of infection is probably reflective of the additional risk of ventilator-associated pneumonia in these patients.27 Status epilepticus was also associated with infectious fever, which likely relates to prolonged immobility and impaired immunity because of anesthesia, hypothermia, and immunosuppressants, which are often used for the treatment of refractory seizures.28,29
It is remarkable that criteria for systemic inflammatory response syndrome and leukocytosis were similarly prevalent in patients with central and infectious fever. This underscores the difficulty in differentiating central and infectious fever prospectively in the critically ill population. The extreme physiologic stress provoked by acute neurologic injury can cause a rise in inflammatory markers and an increased sympathetic response.30,31 The percentage of neutrophils was higher in patients with infectious fever, suggesting that, while leukocytosis may not be a reliable clinical variable to decide whether to use empirical antibiotics or discontinue antibiotics early, the presence of a left shift remains useful.
We created a probability table to predict central fever on the basis of odds ratios for every variable. The combination of negative cultures; absence of infiltrate on chest radiographs; diagnosis of SAH, IVH, or tumor; and onset of fever within 72 hours of admission predicted central fever with a probability of .90. We used this combination of readily available clinical variables to develop a decision tree that clinicians can follow to identify patients with a high probability of central fever. Misclassification of fever type is possible given the retrospective nature of the study. However, misclassification is far more likely to have occurred in the direction of infectious fever. Although the interrater agreement of the classification system was substantial, cases in which there was no agreement were classified as infectious. This is because study definitions were chosen to favor the classification of fever as infectious over central because misclassification of fever as central could result in inappropriate withholding of antimicrobial therapy. As a consequence, it is possible that the prevalence of central fever in our analysis is underestimated, and our decision tree is probably more specific than sensitive in identifying patients with a high probability of central fever.
There are limitations to our study. The study was a retrospective analysis of a derivation cohort and will require prospective validation. Our hospital is a referral center located in a mostly rural area, which may affect the generalizability of results, again necessitating a prospective multicenter validation. We do have a large referral base and a similar disease distribution to large general community hospitals, with the exception of a lower volume of trauma patients. Patients may not have had complete diagnostic evaluations in some cases. This is likely not a large factor because patients with negative cultures in whom the physician documented suspicion of infection or treated with a therapeutic course of antibiotics were classified as having infectious fever. Our study design did not account for potential overlap of fever origin. Some patients with documented infection may also have had impaired regulation of temperature due to acute neurologic injury. Because we collected data only on the first episode of fever in the NICU, our results do not fully reflect the situation of patients with prolonged admissions who may have had multiple episodes of fever from different infectious and noninfectious causes. In addition, we did not collect information on antibiotic administration before admission to our NICU, which could affect culture positivity results. Procalcitonin levels were not collected in our patients. We acknowledge that central fever is a diagnosis of exclusion that cannot be positively proven; however, it is a common phenomenon in neurocritical patients, and we used an inclusive definition for the diagnosis of infectious fever to minimize the chances of missing any infectious cases.
In conclusion, our results provide a model to differentiate central from infectious fever among neurocritical patients. The use of the decision tree developed in this study may allow clinicians to confidently select patients with acute neurologic illness in whom antibiotics may be safely discontinued despite ongoing fever. Validation of this model on an independent population and prospective testing of its performance with longitudinal data are necessary to confirm the usefulness of this tool to identify central fever and optimize antibiotic stewardship.
Accepted for Publication: July 15, 2013.
Corresponding Author: Sara E. Hocker, MD, Mayo W8-B, Mayo Clinic, 200 First St SW, Rochester, MN 55905 (firstname.lastname@example.org).
Published Online: October 7, 2013. doi:10.1001/jamaneurol.2013.4354.
Author Contributions: Drs Hocker and Tian contributed equally to the manuscript. Drs Hocker and Mandrekar had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.
Study concept and design: Hocker, Tian, Li, Steckelberg, Rabinstein.
Acquisition of data: Hocker, Tian, Li.
Analysis and interpretation of data: All authors.
Drafting of the manuscript: Hocker, Tian, Li.
Critical revision of the manuscript for important intellectual content: Hocker, Tian, Steckelberg, Mandrekar, Rabinstein.
Statistical analysis: Hocker, Tian, Li, Mandrekar.
Study supervision: Steckelberg, Rabinstein.
Conflict of Interest Disclosures: None reported.