Prediction of mortality in critically-ill elderly trauma patients: a single centre retrospective observational study and comparison of the performance of trauma scores

The population demographics of trauma patients are shifting, with admissions of increasingly older (≥ 65 years) and frail patients with multiple comorbidities [1]. As the population continues to age, this trend is set to continue. Older patients with severe injuries are at risk of more adverse outcomes when compared with younger patients, even where injury patterns are similar [2‐5]. Critically-injured elderly trauma patients admitted to the intensive care unit (ICU) are particularly vulnerable with increased risk of hospital mortality or discharge to a supported care facility or nursing home [6].

Although generic intensive care-based scoring systems such as the Acute Physiology and Chronic Health Evaluation (APACHE) score are frequently used for risk stratification of critically ill patients, they are not specific for quantitative prognostication after severe traumatic injury. Several trauma-based scoring systems exist, such as the Injury Severity Score [7] (ISS), the Trauma and Injury Severity Score [8] (TRISS) and the more recent age-specific scoring system, the geriatric trauma outcome score (GTOS) [9]. However, none of these trauma scores have been tried and validated in the cohort of critically-ill, severely traumatised elderly patients.

We, therefore, performed a retrospective cohort study to investigate the outcome of all elderly trauma patients admitted to our intensive care unit and evaluated the performance of each scoring system in predicting mortality. Given the age-specific nature of the GTOS, we hypothesise that it would out-perform the TRISS and ISS scoring systems in our critically-ill, elderly patients. Furthermore, the Trauma Audit and Research Network (TARN) have developed a prediction model to calculate the probability of survival (Ps17) for all patients entered into the trauma database [10]. It has not been validated for prognostication in the individual patient, but it has been included to see how this prediction model performed, compared with the scoring systems. Appropriate early prognostication may help with patient and family centred goal-setting, enabling an objective clinical decision-making process to minimise suffering and reduce inappropriate interventions.

This is a retrospective study of all trauma patients aged 65 years and above admitted to the University Hospital Southampton Hospital NHS Foundation Trust from January 2012 through to December 2017. This information was obtained from the trauma databases and included pre-calculated ISS and Ps17 scores. The Ps17 was the version of TARN’s probability of survival score in use during the study period. We were unable to calculate TRISS scores for many of our patients due to lack of relevant clinical details, especially physiological data, and as a result, we have not pursued the use of this score. Additional clinical details were obtained from all available electronic hospital databases. This study was part of a larger retrospective cohort study (CRIT-CO) and had the approval of local research and development (R&D) and National Health Research Authority (HRA).

GTOS was calculated using the formula: age + (2.5 x ISS) + 22 (if packed red blood cells (PRBCs) were given during the first 24 h). This value can then be converted to percentage mortality using the nomogram as published by the original authors [9], or using an online calculator [11]. The Ps17 estimates a probability of survival instead of the probability of mortality. Table 1 summarises the components of each of the scoring systems.

Histograms for age, ISS, GTOS and Ps17 were all visually inspected and found not to be consistent with a normal distribution. Therefore, median and interquartile range (IQR) were used to describe the data, but mean and standard deviation have been included for completeness. Comparisons of age, ISS, GTOS and Ps17 of survivors vs non-survivors were carried out using the Mann-Whitney U test. Receiver operating characteristic (ROC) curves were then used to assess the performance of ISS, GTOS and the Ps17 for mortality prediction and age has been included for comparison. The optimal cut-off point for the prediction of mortality was obtained from the ROC curve using the Youden Index (Sensitivity + Specificity). Statistical analysis was carried out using MedCalc (version 19.4.0), SPSS (version 24) and Excel (version 16.36).

There were 714 elderly trauma patients hospitalised to our major trauma centre between January 2012 and December 2017. Of these, 255 required intensive care admission to either the general (GICU) or neuro-intensive care (NITU) units. Sixteen patients did not survive beyond the emergency department (Fig. 1).

Of those admitted to the intensive care units, 67% (n = 171) were male with a median age of 74 years (range 65–95 years). We divided these patients into two groups; 193 survivors (76%) and 62 non -survivors (24%), reflecting on the 30-day survival rate. The most common mechanism of injury was vehicular incident (54%), followed by fall from > 2 m (36%). The most commonly injured body region was head (40%), followed by multiple regions (19%). Most of these patients had medical comorbidities (86%), which included: hypertension (38%), ischaemic heart disease (29%), chronic obstructive pulmonary disease/asthma (18%) and diabetes mellitus (11%). A summary of patient demographics and injury mechanisms are presented in Table 2.

Outcome data was collected at hospital discharge, or 30-days if still an inpatient. ICU and 30-day survival rates were 88 and 76% respectively. Survivors tended to be younger with a lower ISS score and head injury rate. The median ICU and hospital stay for the survivors was 5 days (range 1–42) and 16 days (range 1–362) respectively. The median time to death from admission to ICU was 2 days and death following intensive care discharge was 1 day (range 1–12 days). Overall outcome data (mortality and neurological outcome defined by Glasgow Outcome Scale (GOS)) was known for 230 patients. Thirty-eight per cent had an unfavourable outcome, which included death or severe neurological disability and the rest had a favourable outcome, which is considered survival with a moderate neurological disability or good functional recovery according to the measured GOS at discharge. The discharge disposition in most cases was to another acute hospital (48%) followed by home (25%), rehabilitation facility (20%), nursing home (4%) and other institutes (3%).

We compared the ISS and GTOS scoring systems, and the Ps17 prediction model for their performance in predicting hospital mortality, in severely injured elderly patients admitted to the intensive care unit. Compared with survivors, the non-survivor group was older and had significantly higher ISS and GTOS scores, and significantly lower Ps17 scores (Table 3). The AUROC (Fig. 2) was statistically significant for all 3 of the scoring systems (Table 4). Age was a poor predictor of mortality, with an AUROC of 0.60 (95% confidence interval (CI) 0.536–0.662) (Fig. 2). Using the Ps17 we also compared predicted survival with our observed survival (Fig. 3). The observed survival rate closely matched the expected survival rate in all age ranges except in those aged ≥90 where observed survival was less than expected.

Severe trauma in elderly population imposes a significant health care burden and is associated with substantial morbidity and mortality [2, 3, 5]. Specific studies investigating the clinical outcomes of such patients in the intensive care setting are lacking. There were 255 trauma patients aged ≥ 65 years old admitted to our intensive care units, representing 36% of all elderly trauma patients admitted to our hospital during the 6 years. This proportion is slightly higher than the recently published United Kingdom (UK) National Trauma data, where just over a quarter of all trauma patients, regardless of their age, were admitted to either an ICU or high dependency facility [12]. However, our ICU admission rate was very similar to a large retrospective observational study of elderly trauma patients [13].

Intensive care unit outcomes of traumatic elderly patients are rarely reported. This study confirms that even within the defined elderly cohort, older age is associated with increased mortality. Each additional decade showed an increment in 30-day mortality, where ages 65–74, 75–84, ≥ 85 were associated with a mortality of 19, 29 and 40% respectively. Our overall ICU and 30-day mortality were 12 and 24% respectively. A larger multi-centre Spanish study of elderly trauma patients admitted to ICU with a median ISS score of 21, demonstrated an ICU mortality of 22 and 30% for patients aged 66–75 years and over 75 years, respectively [14]. Similarly, an American observational study of slightly younger patients with trauma (aged 55 and older) admitted to a surgical intensive care unit had a 28-day in-hospital mortality of 18% [15]. In general, our findings are comparable to those internationally published outcome data on elderly trauma ICU patients.

In our study, both the ISS and GTOS trauma scoring systems were predictive of mortality, with the AUROC being very similar at 0.66 and 0.68, respectively. Our AUROC for GTOS was much lower than the original publication and the follow-up validation study, where AUROC was 0.82 (95% CI 0.807–0.831) and 0.86 (95% CI 0.857–0.867), respectively [9, 16]. Several factors could account for this variation including differences in the patient population (general vs critically ill), validation sample size, and the availability of regional trauma services. By comparison, our patients were a subset of patients who were critically injured, with a higher ISS and had already survived from admission to intensive care. To the best of our knowledge, there are no other studies specifically looking at the use of scoring systems in this subset of elderly trauma patients.

The ISS is purely an anatomical scoring system [8], lacking the inclusion of physiological variables, age or underlying comorbidities, and although the AUROC was statistically significant, it lacked robust association. This finding is in line with a previous work, which has also demonstrated that the ability of the ISS to predict mortality in patients aged 65 and over is poor [17]. Therefore, neither the ISS or GTOS were robust enough to be helpful with ongoing individualised decision-making process in our intensive care setting. The Ps17, despite its better performance, with an AUROC of 0.79, is not designed as a prognosticative scoring system to guide decision making in individual patients. In our cohort, the Ps17 overestimated survival in the 90+ age range. This group included only 8 patients and is therefore too small to draw further conclusion as to why this was the case.

The limitations of this study include that this was an observational study relying on retrospective data collection from existing databases. Moreover, this study was conducted in a single major trauma centre and may not reflect the practices from other centres or be transferable to other trauma units worldwide. As a major trauma centre most of our patients were transferred from other areas. Consequently, we were not able to assess their longer-term mortality outcomes and discharge destination beyond 30 days. Given the retrospective nature of this study, our data set is not entirely exhaustive or comprehensive as some patients, despite their severity of trauma, may not have been admitted to the intensive care unit due to various reasons including prior established limitations of care. We had hoped to calculate the TRISS score to determine its utility in this population, however, the lack of complete physiological data set precluded us from doing so, with physiological variables most likely to be missing from the most critically ill. This problem is not unique to this dataset [18].

Larger, prospective, multicentre studies are needed to evaluate the utility of trauma scoring systems for the prognostication of elderly traumatised patients in the intensive care setting. Moreover, future studies should specifically focus on combination scoring systems to include organ specific measures and perhaps with the inclusion of frailty markers to assess outcome measures in this population. Besides, elderly trauma scores developed and validated in the USA may not be transferable universally, and an ICU specific UK based scoring system needs to be established. Furthermore, in addition to the short-term mortality outcomes, exploration of patient-centred outcomes such as quality of life, degree of functional and cognitive dependency and discharge destination may be of value.

About a third of ≥ 65-year olds of all hospitalised trauma patients were admitted to an intensive care unit setting with a 30-day survival of 76%. Early prognostic prediction may help to support the clinical decision-making process. Both ISS and GTOS scores performed similarly but lacked robust association. Ps17 survival prediction model had a stronger association. Larger prospective studies are needed to establish robust scoring systems in this patient cohort and to evaluate, not only the mortality, but also ongoing quality of life associated with critically ill trauma in the elderly.