Introduction
In the USA, the incidence of traumatic brain injury (TBI) of all severity is 579 per 100,000 persons, equating to roughly 1.7 million cases of TBI per year.Reference Faul and Coronado1 Estimates of TBI in Canada range between 22 and 52 per 100,000 persons.Reference Colantonio, Saverino and Zagorski2 TBI is one of the leading causes of death and disability in the USA, with roughly 30% of all injury-related deaths involving TBI.Reference Faul and Coronado1 Approximately 25% of patients who acquire TBI die – with 17% dying at the site of injury and 6% in an acute care setting.Reference Faul and Coronado1
In-hospital death rates following TBI vary widely, with estimates ranging between 4% and 28%.Reference Colantonio, Croxford, Farooq, Laporte and Coyte3,Reference Utomo, Gabbe, Simpson and Cameron4 Epidemiologic studies have identified a number of risk factors for in-hospital mortality following TBI, including older age, race, number of medical co-morbidities, total number of other traumatic injuries, and TBI severity.Reference Colantonio, Escobar and Chipman5–Reference Selassie, Fakhry and Ford7 Imaging classification systems such as the Marshall CT classification,Reference Steyerberg, Mushkudiani and Perel8 the Rotterdam CT score,Reference Maas, Hukkelhoven, Marshall and Steyerberg9 the Stockholm CT score,Reference Nelson, Nystrom and MacCallum10 and the Helsinki CT scoreReference Raj, Siironen, Skrifvars, Hernesniemi and Kivisaari11 are all validated in predicting outcomes in patients with TBI. Clinical biomarkers including glial fibrillary acidic protein and S100B have also shown promise in the prognostication of patients with TBI.Reference Vos, Jacobs and Andriessen12
Nosocomial complications in patients with TBI are myriad and include venous thromboembolism, postoperative complications, healthcare-acquired infections, and sepsis. Patients with acute neurologic injury are at higher risk of developing respiratory failure and ICU-acquired sepsis, when compared with other critically injured patients.Reference Mascia, Sakr and Pasero13 To date, few studies have investigated the role of sepsis in patients with TBI.Reference Selassie, Fakhry and Ford7,Reference Cardozo Junior and Silva14,Reference Corral, Javierre, Ventura, Marcos, Herrero and Manez15
In this study, our objectives were to: (a) describe the epidemiology of sepsis in our TBI cohort and (b) to determine whether sepsis was associated with increased mortality or healthcare utilization in patients with TBI. We hypothesized that patients with TBI and sepsis would have increased mortality and longer hospital lengths of stay (LOS), when compared with TBI patients without sepsis.
Materials and Methods
We conducted a retrospective, observational, single-center study. The Research Ethics Board at the University of Alberta approved the study and obviated the need for informed consent (study number Pro00071672). STROBE guidelines for reporting of observational studies were followed.Reference von Elm, Altman and Egger16
Inclusion Criteria
All adult patients (aged 18 or greater) with TBI admitted to the University of Alberta Hospital general and neurosurgical ICUs from January 1, 2012, to December 31, 2016, were included in the study.
Patients with TBI were identified using International Statistical Classification of Diseases and Related Health Problems, Tenth Revision, Canada (ICD-10-CA) codes using the Data and Health Information Resources-Alberta Health Services Data Reporting Repository (DIMR-AHSDRR).17 ICD-10-CA codes used to identify patients with TBI included S06 (S06.0–S06.1), S06.2 (S06.25–S06.26), S06.3 (S06.35–S06.36, S06.4–S06.6, and S06.9).
Patient characteristics including age, sex, Acute Physiology and Chronic Health Evaluation (APACHE II) score, Glasgow Coma Scale (GCS) score, Sequential Organ Failure Assessment (SOFA) score, ICU admission date, ICU discharge date, hospital discharge date, and ICU and hospital mortality were identified using an ICU medical information system (TRACER). Severe TBI was defined as GCS ≤ 8.
ICD-10-CA codes were furthered used to identify patients with TBI and sepsis. ICD-10-CA codes used to identify sepsis included A40 (A40.0–A40.3 and A40.8–A40.9), A41 (A41.0–A41.4), A41.5 (A41.50–A41.52 and A41.58), A41.8 (A41.80 and A41.88), A41.9, B37.7, R57.2, and R65.1. Charts of patients with TBI and sepsis were then manually reviewed for medical co-morbidities, sepsis source, and microbial etiology. For timing, the development of sepsis was defined as the time of in-hospital antibiotic prescription.
Statistical Analysis
Baseline characteristics of the cohort were expressed as means with standard deviations (SD) or medians with interquartile ranges (IQR), as appropriate, for continuous variables. Categorical variables were expressed as a number and percentage.
Non-parametric tests (Mann–Whitney U) were used to determine differences between TBI patients with and without sepsis for continuous variables (as all were non-normal in distribution). Chi-square and Fisher’s exact tests were used to compare categorical variables. Tests were two-tailed with an alpha of 0.05. Univariate logistic regression was performed to test for associations with mortality and reported as unadjusted odds ratios (OR). Variables of clinical significance and those with p-values ≤ 0.1 in univariate analyses were included in multivariable logistic regression modeling to evaluate the association with mortality. Adjusted odds ratios (aOR), 95% confidence intervals (95% CI), and p-values were generated to quantify the magnitude and precision of the estimate. Statistical analyses were performed using IBM SPSS Statistics, Version 23.0 (IBM Corp., Armonk, NY).
Results
A total of 486 patients with TBI were identified. Baseline characteristics can be found in Table 1. Of note, most patients were young males and had severe TBI (49%). In the entire cohort, 363 patients (74.7%) required intubation and mechanical ventilation. Median LOS in the ICU was 3.9 days (IQR 1.7–8.6) and in-hospital was 15.7 days (IQR 5.1–31.8). Eighty-six (17.7%) patients with TBI died in the ICU, while 119 (24.5%) with TBI died in the hospital.
Baseline characteristics and univariate analyses examining associations with sepsis in TBI patients
APACHE = Acute Physiology and Chronic Health Evaluation; GCS = Glasgow Coma Score; ICU = intensive care unit; IQR = interquartile range; LOS = length of stay; SOFA = Sequential Organ Failure Score; TBI = traumatic brain injury.
Severe TBI defined as GCS ≤ 8.
Bold values indicate statistical significance.
Of the 486 patients with TBI, 16 (3.3%) developed sepsis (Table 1). All 16 patients were male, with 13 patients (81.3%) requiring intubation and mechanical ventilation. One patient of the 16 septic patients (6.3%) developed septic shock while in the ICU. Pneumonia was the predominant source of sepsis (94%, Table 2). Additional sources included skin and soft-tissue infection, colitis, and urinary tract infection. Staphylococcus aureus was the most commonly identified pathogen (Table 2, 12; 75%) – of which nine were methicillin-sensitive (MSSA) and three were methicillin-resistant (MRSA). Co-morbidities were rare, with alcohol use disorder being the most common (31%, Table 2). Eleven of 16 (68.8%) patients with TBI and sepsis had isolated head injuries, while the remaining 5 (31.2%) had additional traumatic injuries including long-bone fractures (three patients), pneumothorax (two patients), intra-abdominal injury (two patients), and hemothorax (one patient). Median time to the development of sepsis was 4.9 days (IQR 1–6).
Co-morbidities, sepsis source, and microbial etiology in TBI patients with sepsis
AD = Alzheimer’s disease; BC = Bacillus cereus; BPH = benign prostatic hypertrophy; CA = cancer; C diff = Clostridium difficile; DMI = diabetes mellitus type I; DMII = diabetes mellitus type II; EC = Escherichia coli; ETOH = ethanol; HI = Haemophilus influenzae; HTN = hypertension; IVDU = intravenous drug use; MP = Mycoplasma pneumoniae; MRSA = methicillin-resistant Staphylococcus aureus; MSSA = methicillin-sensitive Staphylococcus aureus; PA = Pseudomonas aeruginosa; PUD = peptic ulcer disease; SCZ = schizophrenia; SSTI = skin and soft-tissue infection; SM = Serratia marcescens; UTI = urinary tract infection.
Univariate comparisons between septic and non-septic TBI cohorts (Table 1) did not reveal significant differences in age, APACHE II score, or TBI severity. The percentage of males in the septic cohort, however, was higher compared to the non-septic cohort. SOFA scores were also higher in the septic cohort (8 [IQR 7–11]) versus the non-septic cohort (6 [IQR 4–8]). While ICU and hospital mortality did not differ between groups, ICU LOS was longer in the septic group (12.2 days [IQR 4.4–23.5]) versus 3.7 days [IQR 1.7–8.2] in those without sepsis. Furthermore, hospital LOS was also longer in the septic group (28.0 days [IQR 11.8–41.4]), as compared with the group without sepsis (15.3 days [IQR 5.0–30.9]).
Univariate analyses revealed that severe TBI, APACHE II, and SOFA scores were associated with increased ICU mortality (Table 3). Similarly, severe TBI, APACHE II, and SOFA scores, in addition to age and sex, were all associated with increased hospital mortality (Table 4). Sepsis was not associated with mortality.
Univariate logistic regression analyses for ICU mortality in patients with TBI
Bold values indicate statistical significance.
Univariate logistic regression analyses for hospital mortality in patients with TBI
Bold values indicate statistical significance.
Multivariable logistic regression confirmed that sepsis was not a predictor of ICU (aOR 0.51, 95% CI 0.12–2.27, p = 0.38, Table 5) or hospital (aOR 0.78, 95% CI 0.21–2.96, p = 0.78, Table 6) mortality. Older age (aOR 1.02, 95% CI 1.00–1.04, p = 0.04, for hospital mortality), severe TBI (aOR 3.71, 95% CI 1.52–9.08, p = 0.004, for ICU mortality and 4.19, 95% CI 1.95–8.65, p < 0.001, for hospital mortality), and higher illness acuity (aOR 1.19 per point in APACHE II score, 95% CI 1.11–1.28, p < 0.001, for ICU mortality and 1.22, 95% CI 1.14–1.31, p < 0.001, for hospital mortality) were independently associated with mortality (Tables 5 and 6).
Multivariable logistic regression analyses for ICU mortality in patients with TBI
Bold values indicate statistical significance.
Multivariable logistic regression analyses for hospital mortality in patients with TBI
Bold values indicate statistical significance.
Discussion
This retrospective study of patients with TBI admitted to intensive care over a 5-year period did not demonstrate a significant association between the development of sepsis and ICU or hospital mortality. This is in contrast to the findings of Selassie et al. who demonstrated, in the largest study investigating the role of sepsis in TBI to date, that sepsis was associated with an increased risk of in-hospital death.Reference Selassie, Fakhry and Ford7
In another study investigating the impact of non-neurological complications in patients with severe TBI, Corral et al. demonstrated that septic shock (but not sepsis) was associated with an increased risk of hospital mortality.Reference Corral, Javierre, Ventura, Marcos, Herrero and Manez15 Similarly, in a retrospective analysis of 175 patients with TBI and sepsis, severe sepsis, or septic shock, Cardozo Junior et al. showed that only septic shock (and again, not sepsis) was associated with an increased risk of patient mortality.Reference Cardozo Junior and Silva14
In our analysis, we did not discriminate between sepsis and septic shock. Though our study is similar in size to that of Corral et al. and Cardozo Junior et al., it is challenging to make comparisons between the studies, given a number of key differences. In the study by Corral et al., only patients with severe TBI (defined as GCS < 9 in their analysis) were included in their analysis, whereas our cohort included mostly (49%), but not solely, patients with severe TBI. Furthermore, the percentage of patients who developed sepsis in our study (3.3%) were markedly lower compared with their study (75%; which seems very high). While Cardozo Junior et al. did include patients with varying degrees of TBI, they subcategorized patients with a diagnosis of TBI into those with sepsis, severe sepsis, and septic shock in their analysis.
The percentage of patients who developed sepsis in our study, as mentioned earlier, were surprisingly low (3.3%), but similar to the incidence (1.0%) reported by Selassie et al. Reference Selassie, Fakhry and Ford7 Sellasie et al., however, had a much larger sample size, with a total of 41,395 patients, and higher statistical power. As such, we believe our study was underpowered to detect an association between sepsis and mortality even if one does exist.
In keeping with other epidemiologic studies of TBI,Reference Colantonio, Escobar and Chipman5–Reference Selassie, Fakhry and Ford7 we found that TBI severity and illness acuity (as measured by the APACHE II score) were independent predictors of both ICU and hospital mortality. Age was also an independent predictor for hospital mortality. In addition, ICU and hospital LOS were significantly longer in patients who developed sepsis, when compared with patients who did not develop sepsis.
In patients who developed sepsis the putative source was almost exclusively pulmonary (94%) and the microbial etiology was mostly Staphylococcus aureus (75%). The median time to the development of sepsis was 4.9 days. In the 15 patients who developed pneumonia, 6 (40%) did so within 2 days of admission, suggesting a community-acquired pneumonia. The remaining nine developed pneumonia after 2 days of admission (range 3–20 days), implying hospital acquisition. Mechanistically, it remains unclear as to why head-injured patients appear to preferentially develop Staphylococcus aureus pneumonia, though the addition of blunt traumaReference Rodriguez, Gibbons, Bitzer, Dechert, Steinberg and Flint18,Reference Rello, Ausina, Castella, Net and Prats19 and/or pre-colonization of the nares by Staphylococcus aureus Reference Campbell, Hendrix, Schwalbe, Fattom and Edelman20 are established risk factors.
According to the Canadian Institute for Health Information, the daily cost of an ICU admission in Canada is substantial, with an average cost per day of $3592.21 Notably, this value increases to $4186 in teaching ICUs across Canada. Moreover, patients with TBI are particularly costly to the healthcare system. This is evidenced by a Canadian study that demonstrated that the mean acute care cost of patients with TBI was $19,083.Reference Chen, Bushmeneva, Zagorski, Colantonio, Parsons and Wodchis22 Similarly, a European study by Raj et al. showed that in patients with TBI, the mean university hospital cost was €19,568.Reference Raj, Bendel and Reinikainen23 Taken together, and in the context of the present study, sepsis has the potential to dramatically increase ICU-associated healthcare costs for a given patient with TBI.
Despite its strengths, our study has a number of limitations. Although we included patients from multiple ICUs, the study was performed at a single center and, thus, may not be generalizable. Also, the use of an administrative database and the use of ICD-10-CA codes may have resulted in misclassification (for either sepsis and/or TBI). Moreover, our study was comparatively small, and likely underpowered to show an association between sepsis and mortality. In addition, APACHE II and SOFA scores were collected at the time of admission and may not have been truly reflective of illness and/or organ dysfunction at the time of sepsis diagnosis. And as with all observational studies, we were not able to infer causality, only associations. Lastly, there exists the possibility of residual confounding despite risk adjustment.
Conclusions
In this study, sepsis was rare in patients with TBI admitted to ICU, with an incidence of 3.3%. In those who did develop sepsis, pneumonia was the most common source and Staphylococcus aureus was the predominant pathogen. In keeping with previous studies, older age, TBI severity, and higher illness acuity were independently associated with mortality. Sepsis in patients with TBI was not associated with increased mortality; however, it was associated with increased healthcare utilization (ICU and hospital LOS).
Acknowledgements
We thank David McKinley for electronic data retrieval, as well as the staff at DIMR, TRACER, and AHS medical records.
Conflict of Interest
The authors have no conflicts of interest to declare.
Statement of Authorship
DA designed the study, performed the chart review, and drafted the manuscript. DJK revised the manuscript. WIS designed the study, performed the statistical analysis, and revised the manuscript.
