The Risk Factors for Mortality among Septic Trauma Patients: A Retrospective Cohort Study Using the National Trauma Data Bank

Introduction In trauma patients, the development of sepsis as a hospital complication is significantly associated with morbidity and mortality. We aimed to assess the risk factors associated with in-hospital mortality among trauma patients who developed sepsis during their hospital stay. Material and methods. Using the 2017 National Trauma Data Bank, a retrospective cohort study was conducted to identify adult trauma patients who developed sepsis during their hospital stay. The primary outcome of interest was in-hospital mortality. Multivariate analysis was used to determine the risk factors associated with in-hospital mortality. Results 1782 trauma patients developed sepsis. 567 patients (31.8%) died during their hospital stay. The following patient factors were associated with higher odds of in-hospital mortality: age (OR = 1.045 95% CI = 1.036–1.054), chronic renal failure (OR = 2.564 95% CI = 1.528–4.301), and liver cirrhosis (OR = 3.699 95% CI = 2.267–6.033). Patients who developed cardiac arrest (OR = 4.994 95% CI = 3.381–7.378), acute kidney injury (OR = 3.808 95% CI = 2.837–5.110), acute respiratory distress syndrome (OR = 1.688 95% CI = 1.197–2.379), and stroke (OR = 1.998 95% CI = 1.075–3.714) during their hospital stay had higher odds of mortality. Higher Glasgow Coma Scale (13–15) at presentation was associated with lower odds of mortality (OR = 0.467 95% CI = 0.328–0.667). Conclusion Among trauma patients who developed sepsis, age, chronic renal failure, cirrhosis, the development of cardiac arrest, acute kidney injury, acute respiratory distress syndrome, and stroke in the hospital were associated with in-hospital mortality. These factors can be used to identify patients who are at higher risk of adverse outcomes and implement standardized or protocol-driven methods to improve patient care.


Introduction
Sepsis and septic shock remains a global public health burden. Sepsis is the most common condition among intensive care unit (ICU) patients [1]. In North America and Europe, it accounts for 10.4% of ICU admissions with an associated 28-day mortality of 36.7% [2]. Trauma, another worldwide public health burden, is a major cause of death for individuals below the age of 44 [3]. In 2014, it resulted in 136053 deaths and was responsible for 29 million Emergency Department (ED) visits in the United States [3]. Multiple factors are associated with worse outcomes in trauma patients who survive initial resuscitation such as sepsis [4,5]. Loss of tissue integrity and disrupted host defense mechanisms increase infection risk in trauma patients [6][7][8]. In a study by Eguia et al., six percent (N � 320450) of the total trauma population in the United States developed sepsis during their hospital stay [9]. A systematic review showed variability in mortality rates among trauma patients who developed sepsis during their hospital stay; four studies showed a mortality rate of less than seven percent while six others reported a mortality rate of 10% to 23% [10]. Sepsis in trauma patients was associated with an increased overall length of hospital and ICU stay as well as higher rates of single or multiple organ failure [11,12]. Several risk factors associated with the development of sepsis among trauma patients have been identifed. Tese include injury severity score (ISS), lower Glasgow Coma score (GCS), pre-existing medical conditions, revised trauma score (RTS), age, male gender, number of red blood cell units transfused, and number of operative procedures [11][12][13][14].
Studies on the predictors of mortality in trauma patients who develop sepsis during their hospital stay are scarce. Te primary aim of the study was to determine the risk factors associated with mortality among trauma patients who develop sepsis during their hospital stay using the National Trauma Data Bank (NTDB). Te NTDB collects data on average from one million trauma patients from more than 900 centers yearly and is considered the largest trauma registry in the United States.

Study Design.
Tis was a retrospective cohort study conducted by Emergency Medicine faculty members from the American University of Beirut in Lebanon. Te 2017 National Trauma Data Bank (NTDB) dataset was used. Te NTDB collects data on average from one million trauma patients from more than 900 centers yearly and is considered the largest trauma registry in the United States. Tis dataset provides insight into prehospital and hospital patient characteristics, hospital complications, outcomes, and trauma-related specifcs, such as trauma type and severity [15][16][17].
All trauma patients admitted to the hospital aged ≥16 who developed sepsis as a hospital complication were included. According to the NTDB, the diagnosis of sepsis must have been documented in the patient's medical record and must have occurred during the patient's initial stay at the hospital. Te exclusion criteria were missing age (N � 88), ED discharge disposition was not known/not recorded/not applicable (N � 97), or transferred to another hospital (N � two), and hospital discharge disposition was not known/not recorded (N � nine) or underwent interhospital facility transfer (N � 633).

Variables of Interest and
Outcome. Data regarding patient demographics, comorbidities, trauma characteristics, interventions during hospital stay, hospital complications, and other outcomes (such as ED and hospital disposition and length of hospital stay) were obtained from the database. Te primary outcome of interest was in-hospital mortality. Te secondary outcomes were ED discharge disposition, length of hospital stay, length of ICU stay, and total mechanical ventilation days.

Statistical Analysis.
Te statistical package for the social sciences (SPSS, version 24.0; Inc, IBM Corp, Chicago, IL) was adopted to perform all statistical analyses. Continuous variables were summarized by their means and standard deviations, whereas categorical variables were presented through their frequencies and percentages. To compare the patients' demographic and clinical characteristics by outcome (died: yes/no), Pearson's Chi-Square or Fisher's Exact Test and the Mann-Whitney test were used for categorical and continuous variables, respectively. Te Shapiro-Wilk test, Kurtosis and Skewness Z-score, and visualization of histograms revealed a violation of the normality distribution of all continuous variables (systolic blood pressure, pulse rate, respiratory rate, pulse oximetry, transfusion blood (4 hours), transfusion blood (24 hours), transfusion platelets (4 hours), transfusion platelets (24 hours), length of stay (days), total ICU length of stay, and total vent days). Terefore, the parametric test "Student's T test" was not carried out because none of the continuous variables had a normal distribution and this justifed the reason for using the Mann-Whitney test to compare all continuous variables by the outcome (died: yes/no).
A multivariate analysis was conducted to identify the associated factors with patients' mortality. More specifcally, a logistic regression model using a stepwise selection method was carried out by controlling for all clinically and statistically signifcant factors. Te variables that were entered into the model were: age (years); sex; race; ethnicity; smoking history; hypertension; diabetes mellitus; chronic renal failure; congestive heart failure; chronic obstructive pulmonary disease (COPD); myocardial infarction (MI); cerebrovascular accident (CVA); peripheral arterial disease (PAD); cirrhosis; disseminated cancer; currently receiving chemotherapy for cancer; bleeding disorder; anticoagulant therapy; functionally dependent health status; steroid use; substance abuse disorder; systolic blood pressure; pulse rate; respiratory rate; pulse oximetry; trauma type; mechanism of injury; nature of injury; body region; signs of life; injury severity score; Glasgow coma scale; blood transfusion (4 hours)-central line-associated bloodstream infection (CLABSI); deep surgical site infection; cardiac arrest with CPR; catheter-associated urinary tract infection (CAUTI); pulmonary embolism; extremity compartment syndrome; unplanned intubation; acute kidney injury; myocardial infarction; organ/ space surgical site infection; acute respiratory distress syndrome (ARDS); stroke/CVA; superfcial incisional surgical site infection; ventilator-associated pneumonia (VAP). Almost all study variables were included in the multivariate logistic regression except for those having a high percentage of unrecorded data (surgery for hemorrhage control type (72.0%), blood transfusion at 24 hours (72.0%), platelet transfusion at 4 hours (72.3%), and platelet transfusion at 24 hours (72.3%)) and those being as outcomes (such as length of hospital stay, ED discharge disposition, total ICU length of stay, and total ventilation days). Te variance infation factors of all independent variables in the regression model were found to be less than 10, indicating the absence of a multicollinearity problem. Te fnal model was a good ft to the data as indicated by the Hosmer and Lemeshow test (p � 0.870) and it had a good discrimination between the two categories of outcome (died/survived) as shown by the cstatistic (area under the curve � 0.774; p value <0.001; 95% CI: 0.750-0.797). A p value of ≤0.05 was used to denote statistical signifcance.

Missing Data and Outliers.
Te descriptive analysis showed that the percentage of missing data of all variables was inconsequential because it was less than 5% and therefore it was not handled through multiple imputation procedures. It is worth noting to mention that one of the presented variables "surgery for hemorrhage control type" was not included in the multivariate analysis because it had a very high percentage of unrecorded data (72.0%) that cannot be treated by multiple imputation methods. Te Cook's distance of all cases was found to be less than one, revealing that the estimated parameters were not afected by any infuential observations.

Results
Te total number of trauma patients included in the database was 997970. A total of 2507 trauma patients who developed sepsis as a hospital complication were identifed from the NTDB. After removing the patients who met one of the exclusion criteria, a total of 1782 were included in the data analysis ( Figure 1 Te percentage of nonsurvivors who presented with a systolic blood pressure less than 90 mmHg was higher compared to survivors (17.2% vs. 12.4%, p � 0.007). Tere was no signifcant diference between both groups with regard to other vital signs at presentation. Trauma type at presentation was signifcantly diferent between both groups (p < 0.001), with blunt trauma being the most common type of trauma in both groups. Te nonsurvivor group had a slightly higher number of blunt trauma patients (85.8% versus 82.5%). Te Glasgow Coma Scale at presentation was signifcantly diferent between both groups (p � 0.047), with the nonsurvivors having more patients presenting with aGCS ≤ 8 (22.5% versus 18%). Tere was no signifcant diference between both groups with regard to the nature of the injury, body region of injury, injury severity score, and signs of life at presentation ( Table 2).
In addition, there was no signifcant diference in the amount of blood transfusion and platelet transfusion between both groups at 4 and 24 hours after presentation. Similarly, no diference was observed between survivors and nonsurvivors with regard to the type of surgery for hemorrhage control (Table 3).
With regard to hospital complications, the total rates of central line associated blood stream infection, catheterassociated urinary tract infection, and ventilator-  More specifcally, some patients who had interhospital facility transfer had an ED disposition as one of the excluded categories. Also, some patients whose age was not recorded or were 15 years or younger were transferred or had an ED disposition as one of the excluded categories. Tese overlaps explain why the fnal number on which the data analysis was conducted cannot be calculated just by subtracting the number of excluded patients from the selected sample. * Note: Tere are overlaps among the categories of the excluded variables. More specifcally, some patients who had interhospital facility transfer had an ED disposition as one of the excluded categories. Also, some patients whose age was not recorded or were 15 years or younger were transferred or had an ED disposition as one of the excluded categories. Tese overlaps explain why the fnal number on which the data analysis was conducted cannot be calculated just by subtracting the number of excluded patients from the selected sample.

Discussion
A total of 1782 trauma patients developed sepsis during their hospital stay and 31.8% (N � 567) died throughout. In these patients, age, chronic renal failure, and cirrhosis were found to be associated with in-hospital mortality. Patients who developed cardiac arrest, acute kidney injury, ARDS, and Various studies reported that age was an independent risk factor associated with in-hospital mortality of sepsis and septic shock, even with adequate early intervention [18][19][20]. Brakendrige et al. examined the association between age and the innate immune response in a cohort study involving septic trauma and surgery patients. Te cohort of aged patients (>55 years old) had worse clinical outcomes and a higher rate of 28-day mortality compared to the young cohort [21]. Increased levels of catabolic biomarkers were reported in the older cohort. Te persistence of these biomarkers for up to 28 days indicated a state of immune Emergency Medicine International suppression in older patients with a history of sepsis [21]. In our study, age was also associated with in-hospital mortality in septic trauma patients. Medical comorbidities have been associated with mortality in sepsis and trauma. Te 28-day and 1-year mortality rates of septic shock were higher in patients with CKD compared with non-CKD patients (70% vs. 50%, p � 0.03, and 82% vs. 64 p � 0.03, respectively). CKD remained signifcantly associated with 28-day and 1-year mortality after adjusting for potential confounders [22]. Compared to    [23]. Liver cirrhosis, despite advancements in management, remains a risk factor for both trauma and septic shock-related mortality. Te mortality rate of cirrhotic patients who develop septic shock was found to range between 60 and 80%, potentially due to the disruption of the innate immune response and excessive stimulation of proinfammatory cytokines [24][25][26][27]. Lastly, a study using the National Trauma Data Bank (NTDB) showed a signifcantly higher mortality rate in patients with cirrhosis (15.8% vs. 8.1%, p < 0.0001) and a meta-analysis showed higher odds of mortality for cirrhotic trauma patients compared to noncirrhotic patients (4.52, 95% CI [3.13-6.54]) [28,29]. In our study, CKD and cirrhosis were associated with in-hospital mortality in septic trauma patients. Te results of this study have shown that septic trauma patients with higher GCS scores had lower odds of inhospital mortality (OR � 0.467, 95% CI � 0.328-0.667, p < 0.001 for GCS � 13-15 compared to ≤8). Historically, Te Glasgow Coma Score has been found to be associated with increased mortality in trauma patients. Sewalt et al. identifed 154476 patients between 2013 and 2016 from the Trauma Audit and Research Network. Tey found that the GCS score was among the three most important predictors of in-hospital mortality [30]. It is possible that lower GCS patients are perceived as higher-risk patients who would require more prompt and aggressive care, as well as better monitoring and followup in the hospital.
Septic trauma patients who had a history of cardiac arrest or stroke during their hospital stay were found to have higher mortality. Postsepsis, MI, or stroke was independently associated with increased 180-day (HR � 2.16, 95% CI 1.69-2.76) and 365-day (HR � 1.90, 95% CI 1.54-2.32) mortality [31]. In addition, traumatic cardiac arrest is associated with poor outcomes; studies have reported overall survival rates of 2.6% and 5.6% [32,33]. In another study, patients who developed stroke after a traumatic brain injury had higher mortality rates than those who only sustained a TBI (10.2% vs. 3.2%, p < 0.0001) with an adjusted relative risk of death (RR) of 1.46 (95% CI � 1.15-1.84) [34]. Similarly in our study, the development of cardiac arrest or stroke among septic trauma patients during hospital stay was associated with increased mortality.
In critically ill patients, acute kidney injury (AKI) is a common complication and is associated with increased mortality [35]. Sepsis is one of the major contributors to the development of AKI in critically ill patients [36]. In a study of 5680 septic and septic shock patients, 57.7% developed AKI. Te AKI group had a higher mortality rate (16.4% vs. 2.9%) and increased AKI severity was associated with increased mortality [37]. Trauma is also a major contributor to the development of AKI [38,39]. Out of 413 trauma patients admitted to the ICU of a level 1 trauma center, 24.9% developed AKI during the frst week. AKI was associated with an increase in 30-day (17.5% vs. 5.8%) and 1 year mortality (26.2% vs. 7.1%) [38]. In a meta-analysis of 25,182 trauma patients admitted to the ICU, the incidence of AKI was 24% [39]. Trauma patients who developed AKI had a signifcantly higher length of ICU stay and risk of death (RR � 3.4) [39]. Similarly, in our study, septic trauma patients who developed AKI during their hospital stay had higher odds of inhospital mortality.
Finally, Sepsis is the most common cause of ARDS [40]. In a study by Mikkelsen et al., the incidence of ARDS among septic patients was 6.2%. Patients who developed ARDS had higher rates of in-hospital mortality (60% vs. 14%) [41]. Traumatic injury represents the third most common cause of ARDS [42]. In a study using the 2010-2014 national Trauma quality improvement dataset, the incidence of ARDS went from 3% to 1.1%, but associated mortality increased (18% to 21%, p < 0.001) [43]. Severe traumatic brain injury patients were found to have an incidence of 20-25% for ARDS. ARDS in those patients was associated with increased mortality [44]. Similarly, in our study, septic trauma patients who developed ARDS during their hospital stay had higher odds of in-hospital mortality.
Finally, older age, CKD, and cirrhosis put trauma patients who develop sepsis in a higher risk category. Tese patients may beneft from more frequent and careful (years)-Sex-Race-Ethnicity-Comorbidities [current smoker; hypertension; diabetes mellitus; chronic renal failure; congestive heart failure; chronic obstructive pulmonary disease (COPD); myocardial infarction (MI); cerebrovascular accident (CVA); peripheral arterial disease (PAD); cirrhosis; disseminated cancer; currently receiving chemotherapy for cancer; bleeding disorder; anticoagulant therapy; functionally dependent health status; steroid use; substance abuse disorder]-sbp-pulse rate-respiratory rate-pulse oximetry-trauma type-mechanism of injury-nature of injury-body region-signs of life-iss-gcs-transfusion blood (4 hours)-hospital complications [Central line-associated bloodstream infection (CLABSI); deep surgical site infection; cardiac arrest with cpr; catheter-associated urinary tract infection (CAUTI); pulmonary embolism; extremity compartment syndrome; unplanned intubation; acute kidney injury; myocardial infarction; organ/space surgical site infection; acute respiratory distress syndrome (ARDS); stroke/CVA; superfcial incisional surgical site infection; ventilator-associated pneumonia (VAP)]. monitoring and management. Similarly, septic trauma patients who develop AKI, cardiac arrest, stroke, or ARDS have a poor prognosis, and aggressive management to prevent these conditions should occur in the ICU setting.

Limitations
One of the limitations of this study is its retrospective nature. To obtain our sample, we used ICD-10 CM for trauma and sepsis. Tere can be variations in coding from one hospital to another. Tis can afect the selection of the sample by not including all eligible septic trauma patients. Patients with a missing disposition from the ED were excluded; this can lead to selection bias. Our study included patients with a diagnosis of sepsis on their medical record. Sepsis-3 criteria were not used thus some patients with sepsis may have not been included. Tere were some potential confounders that we would have liked to adjust for but were missing from the database, such as laboratory parameters, imaging results, etiology of sepsis, and sepsis severity scores. In addition, the database only includes hospitals located in the United States, which can limit the generalizability of this study to other populations. However, this database consists of the largest trauma registry in the United States, making the results of this study generalizable to the United States health system. Tere might be diferences in the data quality collected from the diferent hospitals in the NTDB. Despite these limitations, the data is under continuous monitoring and reviewing in order to maintain its high quality.

Conclusion
Tis retrospective study was conducted using data obtained from the NTDB. We aimed to identify predictors of inhospital mortality among trauma patients who develop sepsis as a hospital complication. Older Age, CKD, liver cirrhosis, and the development of cardiac arrest, stroke, AKI, and ARDS were associated with increased in-hospital mortality. A higher GCS score was associated with decreased odds of in-hospital mortality.

Data Availability
Te datasets used and/or analyzed during the current study are available from the corresponding author upon reasonable request.

Ethical Approval
Tis is an observational study. Te American University of Beirut Research Ethics Committee has confrmed that no ethical approval is required.