Trauma is a major cause of morbidity and mortality in the United States. In 2013 it was the leading cause of death for people ages 35–44, and the fourth leading cause of death for the whole population (Murphy et al. 2016). Trauma includes intentional and unintentional injury from motor vehicle crashes, penetrating or blunt violence, falls, firearms, poisoning, and burns. Pre-hospital systems have been studied extensively to optimize the initial care of trauma patients (Williams et al. 2013; McQueen et al. 2015; McNeill and Bryden 2013; Ringburg et al. 2009; Wilson et al. 2015; National Institute for Health and Care Excellence 2016), but there is a shortage of data describing the clinical course of patients admitted to the intensive care unit (ICU) after traumatic injury. The aim of this study was to describe the clinical characteristics of adult patients admitted to the ICU after traumatic injury, including an assessment of the risk factors for hospital complications and hospital mortality in this population.

This was a population-based, multicenter retrospective study of adult patients admitted to the ICU at Level 1 and Level 2 trauma centers after traumatic injury in 2013, using the National Trauma Data Bank (NTDB) National Sample Program. The NTDB is the largest registry of trauma data in the United States, and the National Sample Program includes data from a nationally representative sample of 100 hospitals with Level 1 or Level 2 American College of Surgery (ACS) Trauma designation. The NTDB National Sample Program data are collected prospectively by trained data abstractors at contributing hospitals, and validated centrally before pooling. Steps taken to ensure the validity of the data include the use of a data dictionary, data collection tutorials, and electronic validation of incomplete or inconsistent data. Inclusion and exclusion criteria vary between contributing hospitals. All patients were followed until hospital discharge. The National Sample Program uses a stratified sampling design, with 16 strata based on United States Census regions (Northeast, Midwest, South, and West), level of trauma care designation (ACS Levels 1 and 2), and NTDB reporting status (NTDB-contributors, NTDB-non-contributors). The final weights for hospitals are also adjusted by Emergency Room monthly volume.

Data collected as part of the National Sample Program include general patient demographics, pre-hospital emergency medical services, trauma severity scores (Injury Severity Score (ISS) (Baker et al. 1974)), injury class (e.g., blunt versus penetrating), intention (assault, self-harm, unintentional, other), Emergency Room vital signs (e.g., blood pressure, Glasgow Coma Scale Score (GCS) (Teasdale and Jennett 1974)), preexisting clinical diagnoses, in-hospital diagnosis codes, hospital length of stay, intensive care unit length of stay, and discharge status (mortality, discharge destination). Patients were included for analysis if they were 18 years or older and were admitted to the hospital. Patients were excluded if they were discharged from or died in the Emergency Room. Patients with primary burn injuries were also excluded, because Injury Severity Scores are not validated in this population and major burn injuries are frequently cared for at designated burn centers rather than trauma centers (Pruitt et al. 2012).

We first assessed transitions of care by summarizing the locations of patients after hospital admission from the Emergency Room. We were unable to assess the temporal flow of patient transitions during the hospitalization. We were also unable to account for or exclude readmissions, and patients may have been admitted more than once during the sampling period. We then summarized the national incidence rates of ICU admission with 95 % confidence intervals (CI) based on United States 2013 population census data by age, gender, race, injury type, and census region (Appendix, Table 5) (United States Census Bureau 2013).

We divided all admissions into two groups: patients admitted to the ICU during the hospital course and patients not admitted to the ICU during the hospital course. We defined those who were admitted to the ICU as patients who were admitted to ICU at any time during the index hospitalization, not necessarily those admitted directly to the ICU from the Emergency Room. We summarized the general demographics (e.g., age, gender), clinical scores (e.g., Glasgow Coma Scale, Injury Severity Score) injury types (e.g., blunt, penetrating, other) and the presence of hypotension on admission (systolic blood pressure < 90 mmHg), which has been demonstrated as a prognostic factor in trauma patients (Parks et al. 2006). We summarized the proportion of patients who received mechanical ventilation. We assessed the prevalence of preexisting comorbidities, which were identified using codes collected and reported within the dataset for each patient (Appendix, Table 6). Preexisting comorbidities included coronary artery disease, congestive heart failure, diabetes mellitus, cerebrovascular accident, peripheral vascular disease, pulmonary disease, chronic kidney disease (including stages 1–5 based on the National Kidney Foundation practice guidelines (Levey et al. 2003)), alcoholism, and a current smoking history. Hospital length of stay and ICU length of stay were summarized using calendar days.

We identified those who were diagnosed with hospital complications using codes collected and reported within the dataset for each patient (Appendix, Table 7). Hospital complications included acute kidney injury, acute respiratory distress syndrome (ARDS), cardiac arrests, cerebrovascular accidents, decubitus ulcer, deep vein thrombosis, alcohol or drug withdrawal, myocardial infarction, pneumonia, pulmonary embolism, unplanned intubation, urinary tract infection, and sepsis. The frequency and types of complications were reported for all patients, and for patients admitted to the ICU, complications were also reported by injury mechanism. Patients may have had more than one complication during the hospitalization. Bivariate analyses were conducted to assess the relationship between patient factors and the development of hospital complications using the χ² and t test, as appropriate.

Multiple logistic regression analyses were performed to assess factors associated with the development of hospital complications and hospital mortality. Based on the results of Thompson et al. in creating the Mortality Risk for Trauma Comorbidity Index (Thompson et al. 2010) and the recommendation for rigorous risk-adjusted analysis of trauma mortality by Haider et al.(2012), the multivariable logistic models for hospital complications and hospital mortality included variables that assessed the mechanism of injury (e.g., blunt, penetrating), the physiologic severity (e.g., presence of hypotension on admission, head injury with severity ≥4 on Abbreviated Injury Scale (AIS)), anatomic severity (e.g., Injury Severity Score (ISS)), age, and gender, as well as preexisting comorbidities.

Statistical anlysis was performed using Stata 12.1. We used the Stata survey procedures to account for the sampling design and sampling weights to account for differential probability of selection between strata. Weighted frequencies and proportions for each group were calculated on the basis of the relative weights for patients in each facility within the sample. Continuous variables are presented as medians with interquartile range (IQR). Missing data were rare and less than 5 % for all variables included in the analysis. The study was reviewed and approved by the New York Presbyterian-Columbia University Medical Center Institutional Review Board.

The NTDB 2013 National Sample Program included 2,104,210 weighted records of hospital admission for traumatic injury at Level 1 and Level 2 trauma centers in 2013, and 1,028,817 (48.9 %) included ICU admission during the index hospitalization. The majority (61.3 %) of these patients were treated in ACS Level 1 trauma hospitals. Patients were admitted from the Emergency Room to various units in the hospital, including 711,731 (33.8 %) admitted directly to the ICU (Fig. 1). The national incidence of adult ICU admission after trauma was 3.3 per 1000 (95 % CI 3.2–3.3). The incidence was highest amongst those over 80 years of age (7.8 per 1000 (95 % CI, 7.7–8.0)) (Table 1).Fig. 1

Patient flows from Emergency Room. *Numbers do not add up exactly due to rounding of weighted estimates. **Does not include all patients admitted to ICU, as temporal flow of admissions not included in dataset

This study describes the characteristics and outcomes for adult patients admitted to the ICU at Level 1 and Level 2 trauma hospitals after traumatic injury in the United States. We found that almost half of patients hospitalized after trauma were admitted to the ICU. This population was primarily composed of young males with blunt unintentional traumatic injuries. Hospital complications developed in almost a quarter (22.6 %) of patients who required ICU admission and were associated with a higher severity of injury (median ISS 26 (IQR 17–34)). Although hospital mortality for patients admitted to the ICU (10.7 %) was not high compared to other ICU cohorts in the US, hospital mortality amongst those with hospital course complications (16.9 %) was significantly higher and similar to that of ICU populations nationwide (Lilly et al. 2011).

There is scant epidemiological literature describing patients who require critical care services after trauma in the United States. Epidemiological investigations of trauma patients admitted to the ICU have been conducted in other high-income countries (Curtis et al. 2012), developing countries (Chalya et al. 2011; Adenekan 2009) and in military settings (Brown et al. 2011), but these studies lack generalizability to the United States because of variations in the availability and structure of pre-hospital systems and different ICU bed availability. Although descriptive studies have been conducted in the United States, these studies are often single-center (Ong et al. 2009) or focused on very specific subsets of patients (Brown et al. 2011; Majidi et al. 2014; Nishijima et al. 2013; Lustenberger et al. 2011; Recinos et al. 2009; Sangthong et al. 2006), which also limits generalizability. One recent multicenter study described the case mix, complications, and outcomes of 11,064 patients admitted to ICUs after trauma, and found that hospital complications were associated with age, gender, and traumatic CNS injury (Mondello et al. 2014). The rigor of our study is attributable to the large number of patients from hospitals across the country, the prospective validated data collection process, and the representative nature of the data.

Although ICU admission for traumatic injury, at least for some period of observation, is common practice in many centers (Kaufman et al. 2016), regional triage criteria and the actual utilization of ICU-level care (e.g., mechanical ventilation) are unclear. The severity of illness for patients admitted to the ICU in this study was only moderately high (median ISS 17, IQR 10–27). ICU admission for observation-only may theoretically result in unnecessarily high healthcare costs (Wunsch et al. 2008), exposure of patients to ICU-related complications (e.g., nosocomial infections (Grundmann et al. 2005) and medical errors during transfers of care (Bell et al. 2011)), and denial of ICU beds to other patients. Similarly, ICU admission for patients with extremely high expected mortality may be considered futile, and may also result in high healthcare costs and delayed ICU admission for other patients (Huynh et al. 2014). Admission decisions are often subjective, and for these reasons it is important to clarify the optimal use of ICU resources. This study may serve as a first step to informing ICU triage decisions for trauma patients. For example, these data demonstrate a higher severity of injury amongst patients with hospital course complications who were admitted to the ICU (median ISS 26, IQR 17–34). Although we cannot establish whether hospital course complications occurred before or after ICU admission, hospital course complications may be one clinical factor utilized to determine whether some proportion of patients may be safely treated in other hospital areas, such as intermediate care units, without adversely affecting outcomes.

Additionally, these data demonstrate that 50.3 % of trauma patients admitted to the ICU receive mechanical ventilation, while up to 9.3 % of trauma patients admitted to other hospital areas also require mechanical ventilation. With the exception of patients intubated for airway protection (i.e., central nervous system trauma, airway hemorrhage, penetrating chest trauma), alternatives to mechanical ventilation may be explored for some patients to reduce the prevalence of associated complications and mortality. For example, non-invasive pressure support ventilation has demonstrated a mortality benefit in adult trauma patients (Roberts et al. 2014; Chiumello et al. 2013). The reasons behind the relatively high provision of mechanical ventilation in non-ICU settings deserves further attention.

The hospital complication rate amongst patients admitted to the ICU was high (22.6 %). With increasing scrutiny on quality of care and recent linking of complication rates to reimbursement (Sipkoff 2008), healthcare systems nationwide are searching for ways to reduce in-hospital complications. An important target of future research should include clarifying the time course of hospital complications in hospitalized trauma patients, so as to better identify modifiable risk factors. Preexisting comorbidities were also common in the ICU cohort, especially alcoholism, smoking, and diabetes, despite a generally young population. Notably, these three common comorbidities were not strong predictors of hospital complications and were not associated with an increased risk of hospital mortality. Although the mechanism underlying this association is beyond the scope of this study, these findings are consistent with the results of a single-center study evaluating smoking and trauma outcomes (Ferro et al. 2010). The comorbidities most strongly associated with hospital complications and mortality (pulmonary disease, peripheral vascular disease, and chronic kidney disease) were rare in the cohort. This type of information may help guide triage decisions and future study design.

An important limitation of these data is the lack of temporal association between ICU admission and the development of hospital complications. We do not suggest that hospital complications lead to ICU admission, or vice versa. This important data would greatly enhance the ability to draw clinical conclusions from this study. This study also includes only those patients admitted to American College of Surgery Level 1 and Level 2 designated trauma centers. Despite well-developed trauma triage systems in the United States, some proportion of patients will be treated at Level III or IV centers, or non-trauma centers, and this study does not include these patients. Another limitation is the lack of detailed clinical data about intra-hospital patient transfers, staffing, rapid response teams, and ICU organization. Finally, although the sampling population of hospitals in the National Sample Program is intended to be nationally representative, this likely increases the heterogeneity of ICU and hospital organization within the cohort. Organizational hospital-level factors have been shown to affect outcomes (Sakr et al. 2015) and these limitations limit the generalizability and interpretation of these results.

ICU admission after traumatic injury in adults is common, and almost a quarter of these patients will develop hospital course complications. Hospital complications are associated with significantly higher hospital mortality for ICU patients, and more detailed clinical data is necessary to identify modifiable risk factors. Describing the characteristics and hospital course of patients admitted to the ICU after traumatic injury is an important first step to clarifying the needs of this population.

ACS, American College of Surgery; AIS, Abbreviated Injury Scale; ARDS, acute respiratory distress syndrome; CI, confidence intervals; GCS, Glasgow Coma Scale; ICU, intensive care unit; IQR, interquartile range; ISS, Injury Severity Score; LOS, length of stay; NTDB, National Trauma Data Bank; SBP, systolic blood pressure