Predictors of mortality for blunt trauma patients in intensive care: A retrospective cohort study

Study design

This is a retrospective study of all critically ill trauma patients aged ≥18 years admitted to the General Intensive Care Unit (GICU) at Southampton General Hospital, between January 2018 and December 2019.¹⁵ Major trauma patients were defined as those who had sustained significant injuries due to trauma, resulting in requirement for organ support. Only these patients met the admission requirements for GICU and only GICU patients were included in this study. Patients admitted to other clinical areas including the high dependency unit (HDU) and the neurosciences intensive care unit (NICU) were excluded. These patients did not meet major trauma admission criteria for GICU because they had suffered either minor trauma or isolated neurological trauma, thus they were deemed outside the scope of this study. Penetrating trauma patients were not deliberately excluded, however, our dataset contained only blunt trauma patients as a consequence of the local epidemiology. The sample size was determined by the number patients admitted during the defined time period and was comparable to studies of similar design. Authors did not have access to information that could identify individual participants during or after data collection. Ethical approval was obtained through Ethics and Research Governance Online (ERGO) by the Faculty of Medicine at Southampton University on 4 August 2020, Reference 56519. This study was part of the large CRIT-CO study (Outcomes of Patients Admitted with Critical-Illness to the General Intensive Care Unit – a Retrospective Observational Study) IRAS Reference 232922. This study used retrospective analysis of non-identifiable patient data, thus the need for individual informed consent was waived.

Baseline data and outcomes

The following variables were collected from all available Southampton General Hospital databases: age, sex, comorbidities, mechanism of injury, and injury distribution.¹⁵ The following scores were recorded based on each patient’s condition on admission:

Glasgow Coma Score (GCS) – assessment of impairment of conscious level in response to defined stimuli, initially developed for assessment of traumatic brain injury. It has a minimum score of 3 and a maximum score of 15 and is the most widely-used score we evaluated.¹⁶

Injury Severity Score (ISS) – used to describe severity of injury in trauma patients. Injury severity of each of 6 body systems are scored according the Abbreviated Injury Scale (AIS) 0-6. The three body systems with the highest AIS scores are used to calculate ISS. Each is squared and the sum of these three scores gives the ISS which ranges in value from 3-75. If any system has an 'unsurvivable' injury (AIS = 6), the total score automatically becomes 75.¹⁷

TARN Probability of Survival Score (Ps19) – an updated composite score based on Trauma Score and Injury Severity Score (TRISS). It incorporates age, sex, trauma type (blunt/penetrating), Revised Trauma Score (RTS), Injury Severity Score (ISS), Glasgow Coma Scale (GCS), number of comorbidities and outcomes at 30 days to give a probability of survival.¹⁸

Acute Physiology and Health Evaluation II (APACHE II) score – a physiological scoring system that incorporates age, serum laboratory values (pH, sodium, potassium, creatinine, haematocrit, white blood cell count), patient signs (temperature, mean arterial pressure, heart rate, respiratory rate, FiO2), and both acute and chronic diseases (acute renal failure, history of immunocompromise and organ failure). APACHE II is normally used for predicting mortality in ICU patients. The score ranges from 0-71 with increasing score associated with higher mortality.¹⁹

Charlson’s comorbidity index (CCI) – most widely used comorbidity index. Determines survival rate (1 year and 10 year) in patients with multiple comorbidities. CCI has been adapted multiple times since its inception in 1987 and for the purposes of our study, CCI was determined using an online calculator.²⁰ 0-4 points were assigned for advancing age and between 1 and 6 points are assigned for comorbidities based on their severity including myocardial infarction, congestive heart failure, peripheral vascular disease, cerebrovascular disease, dementia, chronic pulmonary disease, rheumatological disease, peptic ulcer disease, liver disease, diabetes, hemiplegia, paraplegia, renal disease, malignancy, leukaemia, lymphoma and acquired immunodeficiency syndrome (AIDS). Scores were summed to provide a total score to predict mortality. For the multivariate analysis, a modified CCI score was calculated by subtracting the age component of CCI from the total CCI score to avoid confounding between age and CCI.

The outcomes evaluated in our study were duration of mechanical ventilation, ICU and hospital length of stay, and 28-day all cause hospital mortality.

Statistical analysis

Continuous variables are expressed as median and interquartile range (IQR). Mann Whitney U was used as the statistical analysis for continuous variables and chi-square for categorical variables. The distribution of variables was assessed and if they had a non-normal distribution they were dichotomised into categorical variables with equal sized groups. Univariate analysis using logistic regression to investigate if variables that varied significantly between the survival and non-survival group, were also significant predictors of mortality. Prior to a multivariate analysis a correlation matrix was done to assess the collinearity between each of the significant predictors using Spearman’s test. This informed the subsequent multivariate analysis using a logistic regression to identify independent significant predictors of survival. Predictors were deemed significant if p < 0.05. Additionally, receiver operating characteristics (ROC) area under the curve (AUC) graphs were constructed to assess each variables performance in predicting mortality. Data analysis was done in SPSS Version 25 (RRID:SCR_016479) and RStudio Version 1.4.1103 (RRID:SCR_000432) using packages: dplyr, ggplot2, lme4 and pROC.¹⁵

Demographics

A total of 414 critically injured trauma patients were admitted to the Intensive Care Unit between January 2018 and December 2019. Of these, 69.3% (n = 287) were male and 30.7% (n = 127) were female. The median age was 54 years (IQR 34–72). Of those admitted, 66.2% (n = 274) had at least one co-morbidity and the median CCI was 1 (IQR 0–3). The most common mechanism of injury was vehicle incident (46.1%), followed by fall <2 metres (23.9%). The most common body part injured was chest (29.2%), followed by other (20%), multiple injuries (19.1%), head (12.1%), abdominal (8.7%), spinal (6.3%), limbs (3.6%) and facial injuries (1%) ( Figure 1). The median GCS, ISS and APACHE II scores were 15, 22 and 11 respectively.

Figure 1. The mechanism of injury and proportion of injuries in different body regions.

Overall, the 28-day all cause hospital survival was 81.4% (n = 337), with survivors being on average younger than non-survivors (51 (32–68) vs 74-years-old (55–85). There were no survival differences between male and female patients. Survivors had fewer comorbidities than non survivors (CCI medians of 1 (0–4) and 4,^1–4 respectively). Among non-survivors, fall of <2 metres was the most common mechanism of injury (39%) reflecting the older age of this group, followed by an RTC (32.5%). Among survivors, the most common mechanism of injury was RTC (49.3%), followed by fall from <2 metres (20.5%). Non-survivors had lower GCS at presentation than survivors (9 vs 15). They also had higher ISS (25 vs 20) and higher APACHE-II scores (13 vs 11) than survivors at presentation.

The Ps19 predictive model was significantly lower for non-survivors (59 vs 98). The type of body region injured also varied between survivors and non-survivors. The non-survivors had increased frequency of head injury and the survivors had more chest injuries. Abdominal injuries were more common in survivors than non-survivors (10.1% vs 2.6%), whereas limb injuries were more common in non-survivors than survivors (10.4% and 2.1% respectively) ( Table 1).

We performed univariate logistic regression analysis to assess the association between common variables that demonstrated significant difference between survivor and non-survivor groups with 28-day hospital survival ( Table 2). In the univariate analysis, the following factors were found to be significant predictors of mortality: age (OR 1.04, CI 1.03-1.06, p < 0.001), CCI (OR 1.51, CI 1.32-1.74, p < 0.001, fall <2 metres (OR 3,17, CI 1.74–5.84, p < 0.001), GCS <15 (OR 3.79, CI 2.21–6.63, p < 0.001), ISS 41–60 (OR 3.10, CI 1.46–6.46, p = 0.00269), Ps19 < 81 (p = 0.001), number of surgeries (OR 0.627, CI 0.467–0.806, p < 0.001) and the most severely injured body region of the head (OR 11.1, CI 4.87–27.1, p < 0.001), multiple injuries (OR 2.60, CI 1.12–6.31, p = 0.0288) and other injuries (OR 12.7, 95% CI (3.84–44.0, p = 0.001) ( Table 2). A multivariate analysis was conducted using these variables which demonstrated that age 70-80 (OR 3.267, CI 1.102-9.404, p = 0.029), age >80 (OR 27.043, CI 10.228-78.264, p < 0.001) and GCS < 15 (OR 8.728, CI 4.273-19.504, p < 0.001), were independent predictors of mortality (Table 3). Modified CCI, fall <2 metres, and number of surgeries were not found to independently predict mortality in the multivariate analysis.

We performed a probability of survival analysis based on variables including age and the scoring systems Ps19, ISS, GCS, and APACHE-II ( Figure 2). For Ps19 ( Figure 2A), patients with a low Ps19 0–20 had an almost linear decrease in survival probability up until 14 days, had a lower survival probability and were more likely to die sooner. Patients with Ps19 scores between 21–60 had similar survival probabilities until day seven, at which point they diverge with the 41–60 group having the lowest survival probability at 28 days (28%), and the 21–40 group having a survival probability of 44%. Patients with Ps19 scores of 61 or higher had significantly higher probability of survival than the other groups, with the 61–80 group demonstrating more than 70% chance of survival at 28 days. The Ps19 score >80 group demonstrated a survival probability of over 90%.

Figure 2. Survival probability and trauma scoring systems.

Probability of survival against total number of days in hospital. Patients are categorised into groups based on their scores in scoring systems. A: Ps19 Survival probability. B: APACHE II Survival probability. APACHE II scores could not be calculated for 139 patients, so this is taken into consideration with a N/A line. One patient was excluded from the APACHE II graph due to being the only one who had a score >30. C: GCS Survival probability. D: ISS Survival probability.

APACHE II: Acute physiology and chronic health evaluation; GCS: Glasgow coma scale; ISS: Injury severity score; Ps19: Probability of survival.

For APACHE II score, likelihood of survival at 28 days decreased with increasing APACHE II scores ( Figure 2B). Until the seventh day there was a similar survival curve for all APACHE II scores groups, after which patients with a score of 0–10 clearly show a higher probability of survival compared to patients with an APACHE II score of 11–20 or 21–30, (92%, 89% and 87% at 28 days, respectively).

Patients with reduced GCS ( Figure 2C), had lower likelihood of survival compared to those admitted GCS 15 (73% and 94% at 28 days respectively).

For ISS scores ( Figure 2D), patients in the highest score range (61–80) had only a 50% chance of survival at 28 days. Patients with a score of 41–60 had a 70% survival probability, those scoring 21–40 had an 84% chance of survival whereas the group scoring 1–20 had a 90% probability of survival. Thus, a higher ISS score was associated with a lower probability of survival.

The difference between the predicted Ps19 and observed mortality for the cohort is shown in Figure 3. The Ps19 predicted score was similar to the expected mortality for most ages, except for the groups >80 years of age ( Figure 3).

Figure 3. Observed vs Ps19 predicted survival in different age groups.

The area under the receiver operator curve (AUROC) was statistically significant for all variables ( Figure 4). Ps19 was the best predictor of mortality with an AUROC of 0.90 (95% CI 0.85–0.96) followed by GCS AUROC of 0.75 (95% CI 0.64–0.86) and age 0.73 (95% CI 0.62–0.85). ISS, APACHE II and number of surgeries were less predictive of mortality in comparison to these variables with ISS being the worst predictor with an AUROC of 0.66 (95% CI 0.50–0.76).

Figure 4. ROC curves comparing predictors of mortality.

Calculated for APACHE II Score, ISS, GCS, Ps19 and Number of surgeries in mortality correlation prediction.

ROC: receiver operator characteristic, AUROC: area under the receiver operator curve, APACHE II: Acute physiological assessment and chronic health evaluation, ISS: Injury Severity Score, GCS: Glasgow Coma Scale, Ps19: Probability of Survival Score.

Overall, 190 patients (45.9%) required invasive mechanical ventilation and the proportion was higher in non-survivors, compared with survivors (68.8% vs 40.7%) and for both groups the duration of mechanical ventilation was three days. Median ICU and hospital length stay were 3 (IQR 1, 7) and 13 (IQR 7, 26) respectively. There was no statistically significant difference between the ICU length of stay (LOS) for survivors and non-survivors; however, survivors had a longer hospital LOS (15 vs 7 days p < 0.01) ( Table 4).