Geriatric Trauma Outcome Score for Predicting Mortality among Older Korean Adults with Trauma: Is It Applicable in All Cases?
Article information
Abstract
Background
This study aimed to validate the Geriatric Trauma Outcome Score (GTOS) for predicting mortality associated with trauma in older Korean adults and compare the GTOS with the Trauma and Injury Severity Score (TRISS).
Methods
This study included patients aged ≥65 years who visited the Chungbuk National University Hospital Regional Trauma Center between January 2016 and December 2022. We used receiver operating characteristic curves and calibration plots to assess the discrimination and calibration of the scoring systems.
Results
Among 3,053 patients, the median age was 77 years, and the mortality rate was 5.2%. The overall GTOS-predicted mortality and 1–TRISS were 5.4% (interquartile range [IQR], 3.7–9.5) and 4.7% (IQR, 4.7–4.7), respectively. The areas under the curves (AUCs) of 1–TRISS and GTOS for the total population were 0.763 (95% confidence interval [CI], 0.719–0.806) and 0.794 (95% CI, 0.755–0.833), respectively. In the Glasgow Coma Scale (GCS) ≤12 group, the in-hospital mortality rate was 27.5% (79 deaths). The GTOS-predicted mortality and 1–TRISS in this group were 18.6% (IQR, 7.5–34.7) and 26.9% (IQR, 11.9–73.1), respectively. The AUCs of 1–TRISS and GTOS for the total population were 0.800 (95% CI, 0.776–0.854) and 0.744 (95% CI, 0.685–0.804), respectively.
Conclusion
The GTOS and TRISS demonstrated comparable accuracy in predicting mortality, while the GTOS offered the advantage of simpler calculations. However, the GTOS tended to underestimate mortality in patients with GCS ≤12; thus, its application requires care in such cases.
INTRODUCTION
As life expectancy increases worldwide, there is an increase in the older adult population. The proportion of older adults among all trauma patients is increasing, with subsequent increases in trauma-related deaths in this population. In Korea, as many as 200,000 severe trauma cases occur every year, approximately 40% of which occur in the population aged ≥60 years, more than half of whom are women. Moreover, the incidence of trauma in older adults has steadily increased.1)
Older adults have higher morbidity and mortality rates than younger patients owing to changes in physiological responses, frailty, and pre-existing comorbidities.2-5) Additionally, although older adult patients tend to experience less severe trauma and have a lower rate of hospitalization in the intensive care unit, owing to differences in the mechanism of trauma, their mortality rate is higher.6) Therefore, providing appropriate treatment through the early identification of risk factors and predicting prognosis is crucial in older patients with trauma.
Various scoring systems have been developed to predict the prognosis of patients with trauma,7,8) each of which has advantages and disadvantages. The Injury Severity Score (ISS), a trauma scoring system based on anatomical scores, is the most used to evaluate trauma severity; however, it does not reflect physiological changes in patients. The Trauma and Injury Severity Score (TRISS), a combination of anatomical and physiological scoring systems, is an effective scoring system for predicting survivability in patients with trauma.9,10) However, the TRISS has limitations,11) such as not adequately reflecting the effects of age in older adults and requiring an initial Glasgow Coma Scale (GCS) and vital sign values for calculation, which are often missing, especially in patients transferred from other hospitals.12) Several scoring systems have recently been developed to predict the prognosis and mortality in older adult patients with trauma, including the Geriatric Trauma Outcome Score (GTOS),13) Severity Characterization of Trauma,14) and Trauma-Specific Frailty Index.2) Among these, the GTOS, first introduced in 2015, is a simple scoring system, and several subsequent studies have reported good results in predicting the prognosis of patients with trauma.6,15) However, although they are easily calculated, these models may have limited metrics and may not provide accurate predictions for all patient groups. Moreover, little research has been conducted on older Korean adults.12,16)
Therefore, the present study aimed to verify the GTOS by comparing it with the TRISS through a subgroup analysis in older Korean patients with trauma.
MATERIALS AND METHODS
Patients and Data Collection
This retrospective study included patients with trauma who visited at the Chungbuk National University Hospital Regional Trauma Center and the Regional Emergency Medical Center between January 2016 and December 2022. We analyzed data from the medical records of patients admitted to this hospital and registered in the Korean Trauma Database. The study protocol was approved by the Institutional Review Board of the Chungbuk National University Hospital (Approval No. 2023-04-001), which waived the requirement for informed consent owing to the retrospective nature of this study.
We excluded from this study patients <65 years of age at the time of hospital visit, those who were transferred to another hospital, those who were discharged with no hope of recovery, those who were pronounced dead upon arrival in the emergency room or who did not survive after cardiopulmonary resuscitation, and those who died ≤ 24 hours or >30 days after admission.
Patients with a systolic blood pressure (SBP) of ≤90 mmHg at the time of the emergency room visit were considered hypotensive. We analyzed the distribution of patients according to the Abbreviated Injury Scale score. Based on the Injury Severity Score (ISS), an indicator of the severity of trauma, we classified the patients into mild and severe groups using a cutoff score of 15. Based on the GCS score, we also classified the patients into groups with severe (3–8 points), moderate (9–13 points), or mild (14–15 points) conditions. We applied the TRISS17) and GTOS13) to predict mortality. The Revised Trauma Score (RTS) was calculated using the GCS, SBP, and respiratory rate (RR) as follows:
TRISS (probability of survival) = 1/(1+e-b)
bBlunt = -0.4499 + 0.8085 × RTS - 0.0835 × ISS -1.7340 × Age Index
bPenetrating = -2.5355 + 0.9934 × RTS - 0.0651 × ISS -1.1360 × Age Index
(RTS = 0.9368 × GCS + 0.7326 × SBP + 0.2908 × RR)
GTOS = Age + (2.5 × ISS) + 22 (if packed red blood cells were transfused in the first 24 hours after injury)
GTOS-predicted mortality = e[-6.9115+0.03912×GTOS] / (1+ e[-6.9115+0.03912×GTOS]).
The probability of survival was calculated using the TRISS equation, while the secondary predicted mortality was calculated as 1–TRISS. The TRISS equation used the constant revised in 1995, and the Age Index was calculated as 1 for patients aged ≥55 years and 0 for those <55 years. The GTOS-predicted modality was calculated using the GTOS value.13)
Statistical Analysis
We performed the statistical analyses using R software (version 4.2.2; https://www.r-project.org/). Continuous variables that did not satisfy normality are expressed as medians with interquartile ranges (IQR; 25th–75th percentiles). Categorical variables are expressed as percentages. Chi-square or Fisher exact tests were used to analyze nominal variables, while the t-test or Mann–Whitney U test was used for continuous variables, depending on the normality of their distributions.
We used the receiver operating characteristic (ROC) curve and calibration plot to evaluate discrimination and calibration, respectively.18-20) To compare the predictability of mortality between GTOS and TRISS, we plotted ROC curves from 1–TRISS and GTOS. Comparison of the areas under the curves (AUCs) were performed as described by DeLong et al.21) We performed similar analyses for subgroups of patients with ISS ≥15 and with GCS ≤12. We prepared a calibration plot using the standard rms statistical software package in R.22)
RESULTS
Basic Demographics and Characteristics
We included and analyzed data from a total of 3,053 patients who met the inclusion criteria during the study period. Fig. 1 is a flowchart depicting the patient selection for this study. Table 1 presents the basic characteristics of the study population. This study included 1,406 men (46.1%). The median patient age was 77 years (range, 71–82 years). A total of 868 patients (28.3%) were admitted to traumatic intensive care, 115 patients (3.8%) experienced hypotension, and 465 patients (15.2%) required emergency surgery. The median ISS, GTOS, and TRISS in the overall study cohort (n=3,053) were 9 (IQR, 9–13), 104 (IQR, 93.5–119), and 0.953 (IQR, 0.953–0.953), respectively. The in-hospital mortality rate was 5.2% (n=159).
Observed and Predicted Risks of Mortality
The median GTOS-predicted mortality of the whole cohort was 5.4% (IQR, 3.7–9.5), and a total of 159 patients (5.2%) died after admission for trauma. The GTOS demonstrated a fair ability to predict in-hospital mortality in older adult patients overall (AUC=0.794, 95% confidence interval [CI], 0.755–0.833). The 1–TRISS of the entire cohort was 4.7% (IQR, 4.7–4.7), with an AUC of 0.763 (95% CI, 0.719–0.806) (Tables 2, 3).
The same analysis was conducted in patients with ISS ≥15 and GCS ≤12. Among the 674 patients with ISS ≥15, the in-hospital mortality rate was 14.5% (98 deaths). The GTOS-predicted mortality and 1–TRISS in this group were 18.9% (IQR, 10.7–30.8) and 11.9% (IQR, 4.7–26.9), respectively (Table 3). The AUC values for the GTOS and 1–TRISS were 0.773 (95% CI, 0.679–0.781) and 0.822 (95% CI, 0.774–0.870), respectively (Table 2). Among the 287 patients with GCS ≤12, the in-hospital mortality rate was 27.5% (79 deaths). The GTOS-predicted mortality and 1–TRISS in this group were 18.6% (IQR, 7.5–34.7) and 26.9% (IQR, 11.9–73.1), respectively (Table 3). The AUC values for GTOS and 1–TRISS were 0.744 (95% CI, 0.685–0.804) and 0.800 (95% CI, 0.776–0.854), respectively (Table 2). Figs. 2 and 3 depict the ROC curves of the GTOS and 1–TRISS for the entire patient group as well as the subgroups.

ROC curves of GTOS-predicted-mortality. Values are presented as AUC (95% CI). GTOS, Geriatric Trauma Outcome Score; ISS, Injury Severity Score; GCS, Glasgow Coma Scale; ROC, receiver operating characteristic; AUC, area under the curve; CI, confidence interval.
Overall Calibration
The calibration of the GTOS-predicted mortality was reasonable when the predicted mortality was <50%. The accuracy of the prediction was reduced for >50% owing to the small number of cases. The model tended to overestimate mortality compared with that observed in the entire cohort. (Fig. 4A). However, the GTOS tended to underestimate mortality in patients with GCS ≤12 (Fig. 4B).

Calibration curves of GTOS-predicted-mortality: (A) in entire cohort and (B) in GCS ≤12. This figure was generated using standard statistical software: the rms package for R (http://cran.r-project.org/package=rms).
DISCUSSION
Increased life expectancy and active lifestyles in older adults expose them to trauma.1,23) Trauma in this population is associated with substantial morbidity and mortality and imposes a significant healthcare burden. Accurate prognostic prediction plays a vital role in the treatment of older adult patients with trauma. In addition, accurate predictions can optimize resource management in hospitals and offer support in decision making to ensure sufficient care.
The trauma-based scoring systems included the ISS and TRISS,17) as well as the more recently proposed GTOS age-specific scoring system.13) Favorable predictability of the GTOS and TRISS have been reported6,15,24); however, few studies have compared the predictability of the GTOS in various patient groups.
We also compared subgroups of patients with ISS ≥15 and GCS ≤12. The results demonstrated that GTOS-predicted mortality had a reasonable ability to discriminate (AUC=0.794) 30-day mortality associated with trauma in older adult patients and that the GTOS was not inferior to the TRISS in the overall population.
Accurate prediction reflecting various physiological indicators and patient factors is crucial for predicting the prognosis of patients with trauma; however, simple bedside assessments are also important. Given its simplicity of calculation, the GTOS could be useful for predicting mortality in geriatric patients with trauma. The GTOS revealed good discrimination in the older adult group (≥65 years); however, the discrimination was lower in cases with ISS ≥15 or GCS ≤12.
Eggleston et al.25) have reported that ISS and GTOS revealed low predictive power, with AUC values of 0.66 and 0.68, respectively, in critically ill older adult patients with trauma. Ryu et al.16) have reported the good predictive ability of the TRISS for in-hospital mortality in geriatric patients with severe trauma. In the present study, we classified patients with an ISS ≥15 as having severe trauma. This subgroup is particularly crucial, as it represents patients with significant trauma and a higher risk of mortality. The observed AUC for GTOS in this subgroup was 0.773, whereas that for TRISS was 0.822. Therefore, the TRISS has marginally better discriminative power in predicting mortality than GTOS within this group. Despite its lower AUC than TRISS, the GTOS remains valuable owing to its simplicity and ease of use. This tool may be useful in settings requiring rapid and straightforward assessments. However, in cases with severe trauma where precision is critical, supplementary tools or more comprehensive systems, such as the TRISS, might be preferred.
Patients with a GCS score ≤12 are at a higher risk due to significant neurological impairment. Our results indicated an AUC for the GTOS in this patient group of 0.744, whereas that for the TRISS was 0.800. Thus, the TRISS marginally outperformed the GTOS in predicting mortality in patients with impaired consciousness.
In our study, patients with ISS ≥15 and GCS ≤12 revealed higher AUCs for TRISS compared with GTOS, although the difference was not statistically significant. Compared with the overall trauma group, TRISS, a scoring system that reflects physiological indicators, may provide more accurate predictions in patients with severe trauma. The TRISS model incorporates the RTS, which includes GCS as a component. This inclusion directly addresses neurological impairments and improves the predictive accuracy in patients with significant brain injuries. The better performance of the TRISS in this subgroup highlights the importance of accounting for neurological status in trauma prognosis.17,26)
To date, many studies have presented only the AUC of prognostic prediction score systems. Although the AUC is a useful statistical indicator of the prognostic prediction model, it only reflects the discrimination of dichotomized results. Thus, the AUC alone does not indicate how accurately these results can be predicted in real-world settings. Therefore, a well-calibrated prognostic model must be developed using a calibration curve.20) Thus, the current study generated calibration curves for TRISS and GTOS to provide a more accurate validation.
Barea-Mendoza et al.27) reported that the GTOS and TRISS revealed excessive mortality predictions in critically ill patients. In our study, the GTOS-predicted mortality in patients with GCS ≤12 was underestimated compared with the observed value. The reason for this difference in this patient group is likely due to the significant impact of brain injuries on death, which the GTOS does not consider. Therefore, the GTOS may not adequately reflect the high mortality risk associated with severe brain injury.
The results of this study highlight the need for trauma-scoring systems tailored to specific patient subgroups. Models that account for the interplay between anatomical severity and physiological indicators are crucial for accurate mortality prediction.
This study has several limitations. First, this was a retrospective study and may have been affected by selection bias and unreliable data. Second, this study was based on data from a single center; thus, the results may not be generalizable to all patients with trauma. Third, in the subgroup analysis, the predicted mortality rate of >50% group was imprecise owing to the small number of patients.
In conclusion, the study findings demonstrated similar accuracies between the GTOS and TRISS in predicting mortality and that the GTOS can be easily applied to mortality prediction because of its simpler calculations. However, predicted mortality was underestimated in patients with GCS ≤12. Thus, this underestimation should be considered when applying the GTOS to predict mortality in patients with traumatic brain injury. Prospective multicenter studies are needed to refine and validate trauma-scoring systems for various situations and injury types. Developing scoring systems that integrate more detailed physiological, anatomical, and demographic factors will enhance the predictive accuracy and clinical utility of these tools.
Notes
CONFLICT OF INTEREST
The researchers claim no conflicts of interest.
FUNDING
None.
AUTHOR CONTRIBUTIONS
Conceptualization, HJH; Data curation, HJH, YSY, SJP, LJY, KSH, YJB, KHR, LJS, SYH; Formal analysis, HJH, LJY; Investigation, HJH, KHR; Methodology, HJH, LJY; Software, HJH; Validation, HJH, YSY; Visualization, HJH; Writing – original drafting, HJH; Writing – review & editing, HJH, YSY, SJP, LJY, KSH, YJB, KHR, LJS, SYH.