Comparison of the revised Shock Index multiplied by the Glasgow Coma Scale and the Trauma and Injury Severity Score for predicting survival in paediatric patients with trauma: a nationwide retrospective cohort study in Japan

Abstract

Objectives

The Trauma and Injury Severity Score (TRISS) is widely used to predict survival in patients with trauma; however, its predictive accuracy in paediatric populations remains suboptimal. In contrast, the revised Shock Index multiplied by the Glasgow Coma Scale (rSIG), a simple physiological marker, has shown potential utility in adults and children. Therefore, we aimed to compare the predictive performance of rSIG and TRISS in paediatric patients with trauma.

Design

Retrospective cohort study.

Setting

Japan Trauma Data Bank.

Participants

Paediatric patients with trauma aged ≤17 years who were registered in the Japan Trauma Data Bank between 2009 and 2021.

Primary and secondary outcome measures

The optimal cut-off value for rSIG was determined using the derivation cohort (2009–2018). In the validation cohort (2019–2021), the predictive accuracy of rSIG and TRISS for in-hospital survival was compared using the area under the receiver operating characteristic curve (AUC).

Results

In the derivation cohort, the optimal cut-off value for rSIG was 10.13. In the validation cohort, the overall AUC for rSIG was 0.857 (95% CI 0.796 to 0.918) compared with 0.740 (95% CI 0.661 to 0.820) for TRISS, demonstrating the significantly superior predictive accuracy of rSIG (p=0.006). Age-stratified analyses revealed that rSIG significantly outperformed TRISS in the 7–12 and 13–17 age groups.

Conclusions

The rSIG demonstrated higher predictive accuracy for in-hospital survival among paediatric patients with trauma than TRISS. These findings suggest that rSIG is a more effective prognostic tool for paediatric trauma care. Validation and the establishment of age-specific cut-off values are warranted to enhance its clinical applicability.

STRENGTHS AND LIMITATIONS OF THIS STUDY.

• This study used a large, nationwide trauma registry to compare prognostic model in paediatric trauma patients.

• Optimal revised Shock Index multiplied by the Glasgow Coma Scale cutoff values was derived using a predefined derivation cohort and validated in a separate cohort.

• As an observational study, the results may be affected by missing data and potential verification bias.

• The database includes only selected emergency and trauma centres, which may limit generalisability.

• Physiological measurements are age-dependent, and assessment accuracy may be limited in younger children.

Introduction

Trauma scoring systems used for survival prediction help to evaluate the quality of trauma care.¹ These scoring tools enable survival rates comparisons between patients with varying anatomical and physiological injury severity or among those with equivalent injury severity, as well as across different clinical settings. In Japan, the Trauma and Injury Severity Score (TRISS) method is the most widely used scoring system in trauma care.

TRISS was developed in 1987 and has since been extensively adopted as a standard method for predicting survival outcomes.² It combines the Injury Severity Score (ISS), which assesses anatomical injury severity, with the Revised Trauma Score (RTS), a physiological index and incorporates variables such as the Glasgow Coma Scale (GCS), systolic blood pressure (SBP), respiratory rate (RR) and patient age.

Several limitations have been identified in the application of TRISS, particularly in cases of severe trauma and in younger paediatric populations. Previous studies have reported decreased predictive accuracy of the TRISS model in these subgroups.^{3 4} In a previous investigation,⁵ TRISS had a particularly poor performance in neonates and infants compared with older children. Moreover, in a subgroup analysis based on predicted survival probability (Ps), TRISS demonstrated notably reduced performance in patients with Ps ≤0.950.⁵ Recalibration efforts incorporating updated coefficients were undertaken to address the decreasing in-hospital mortality associated with recent advances in trauma care.⁶ However, they failed to improve model accuracy.

The revised Shock Index multiplied by the Glasgow Coma Scale (rSIG) has recently gained attention as a simple and rapidly obtainable indicator based solely on vital signs. The rSIG has potential utility in predicting outcomes and the need for emergency interventions in adult and paediatric patients with trauma,^{7 8} with studies showing its strong discriminatory ability in predicting in-hospital mortality among children.^{8 9} Nonetheless, the clinical application of rSIG requires the establishment of reliable cut-off values. Age-specific cut-off values have been proposed⁹,¹¹; however, the cut-off values tend to differ among studies depending on the specific outcome of interest. Lower cut-off values (typically ≤10) are frequently reported in studies aimed at mortality prediction, whereas higher values are often used for predicting the need for emergency interventions.¹⁰,¹² However, a standardised threshold has not yet been established.

Thus far, none of the studies in Japan have directly compared the predictive performance of TRISS and rSIG, and the clinical utility of rSIG relative to TRISS remains unclear. Moreover, no previous research has established an optimal rSIG cut-off value for predicting mortality in Japanese paediatric patients with trauma. Therefore, we aimed to evaluate whether the simplified rSIG score offers a useful alternative to the existing TRISS model for predicting survival in paediatric patients with trauma and identify an optimal cut-off value for clinical use.

Methods

Study design and setting

In this retrospective cohort study, we used data from the Japan Trauma Data Bank (JTDB), which is maintained by the Japanese Association for the Surgery of Trauma (JAST). The JTDB contains comprehensive records, including demographic information, comorbidities, injury types, mechanisms of injury, transportation methods, vital signs, Abbreviated Injury Scale (AIS) scores, ISS, prehospital and in-hospital interventions, trauma diagnoses based on AIS and clinical outcomes.¹³ In most cases, data are entered online by physicians trained in AIS coding using the 1990 revision of the AIS. Data collection for the JTDB commenced in 2003 involving 55 hospitals. The number of participating institutions has increased steadily, reaching 280 by March 2019. This includes 92% of the government-designated tertiary emergency medical centres in Japan. The JAST authorises open access and updates to medical data in the registry, and the Japanese Association for Acute Medicine evaluates the submitted information.

Figure 1 illustrates the patient selection flowchart. We used the JTDB dataset covering the period from 1 January 2009 to 31 December 2021, which included data on 407 545 patients with trauma. The inclusion criterion was patients with trauma aged ≤17 years. Patients were excluded if they were aged ≥18 years, had burn injuries or an unknown mechanism of injury, were not directly transported to the hospital from the scene, presented with cardiac arrest on arrival or had missing data regarding in-hospital outcomes or TRISS-predicted survival.

Figure 1. Flow diagram of patient selection process. GCS, Glasgow Coma Scale; HR, heart rate; JTDB, Japan Trauma Data Bank; SBP, systolic blood pressure; TRISS, Trauma and Injury Severity Score.

Data collection

The following variables were extracted from the JTDB: age (years), sex, injury site, AIS, RTS, ISS, Ps, SBP, heart rate (HR), RR, GCS on arrival and in-hospital mortality.

Ps was estimated using the TRISS score, which ranges from 0 (certain death) to 1 (certain survival). TRISS was calculated using the following formula:

$${\displaystyle {\displaystyle \mathrm{T}\mathrm{R}\mathrm{I}\mathrm{S}\mathrm{S}=\mathrm{P}\mathrm{s}=1/(1+{\mathrm{e}}^{-\mathrm{b}})(\mathrm{b}={\mathrm{b}}_0+{\mathrm{b}}_1(\mathrm{R}\mathrm{T}\mathrm{S})+{\mathrm{b}}_2(\mathrm{I}\mathrm{S}\mathrm{S})+{\mathrm{b}}_3(\mathrm{a}\mathrm{g}\mathrm{e}))}}$$

$${\displaystyle {\displaystyle \mathrm{R}\mathrm{T}\mathrm{S}=0.9368\times \mathrm{G}\mathrm{C}\mathrm{S}+0.7326\times \mathrm{s}\mathrm{B}\mathrm{P}+0.2908\times \mathrm{R}\mathrm{R}}}$$

rSIG was calculated using SBP, HR and GCS measured at the time of hospital arrival, as follows:

$${\displaystyle {\displaystyle \mathrm{r}\mathrm{S}\mathrm{I}\mathrm{G}=\mathrm{s}\mathrm{y}\mathrm{s}\mathrm{t}\mathrm{o}\mathrm{l}\mathrm{i}\mathrm{c}\mathrm{b}\mathrm{l}\mathrm{o}\mathrm{o}\mathrm{d}\mathrm{p}\mathrm{r}\mathrm{e}\mathrm{s}\mathrm{s}\mathrm{u}\mathrm{r}\mathrm{e}/\mathrm{h}\mathrm{e}\mathrm{a}\mathrm{r}\mathrm{t}\mathrm{r}\mathrm{a}\mathrm{t}\mathrm{e}\times \mathrm{G}\mathrm{C}\mathrm{S}}}$$

Statistical analysis

To determine the optimal cut-off value for survival prediction using rSIG, 13 883 patients from 2009 to 2018 were designated as the derivation cohort. To compare the predictive performance between rSIG and TRISS, 2814 patients from 2019 to 2021 were assigned to the validation cohort.

In the derivation cohort, the optimal cut-off value for predicting emergency intervention was determined by calculating the Youden index from the receiver operating characteristic curve analysis. Continuous variables are presented as medians with IQRs, and categorical variables are summarised as frequencies and percentages. The survival and non-survival groups were compared using the Mann-Whitney U test for continuous variables and Fisher’s exact test for categorical variables. A two-sided p value of <0.05 was considered statistically significant.

In the validation cohort, the sensitivity, specificity, positive predictive value, negative predictive value and the area under the receiver operating characteristic curve (AUROC) were calculated for rSIG and TRISS, with corresponding 95% CIs. Differences between AUROCs were assessed using the DeLong test.

To assess whether the prognostic performance of the rSIG differed according to injury type, we performed a prespecified sensitivity analysis in patients with isolated traumatic brain injury (TBI). Isolated TBI was defined as head AIS score ≥3 in the absence of severe extracranial injury (AIS <3 in all other body regions).

Within this subgroup, the optimal rSIG cut-off value for in-hospital mortality was derived in the derivation cohort using the same receiver operating characteristic curve-based approach as in the primary analysis and subsequently evaluated in the validation cohort.

Patient and public involvement

Patients and/or the public were not involved in the design, conduct, reporting or dissemination plans of this review.

Results

Participant selection

Of those who met the inclusion criterion, 1776 patients (8.5%) had missing survival outcome data, and 3562 (17%) lacked TRISS-Ps, precluding the calculation of Ps using the TRISS method.

Participant characteristics: the derivation and validation cohorts

The derivation cohort comprised 13 562 patients in the survival group and 321 in the non-survival group. Patients in the non-survival group tended to be older than those in the survival group (median age: 15 vs 12 years, p<0.001). Mortality rates varied across paediatric age groups: 2.2% in neonates and infants, 1.3% in school-aged children and 3.1% in adolescents. Severe injuries (AIS ≥3) to the head and chest were significantly more common in the non-survival group than in the survival group (head injuries: 88% vs 32%, p<0.001; chest injuries: 57% vs 18%, p<0.001). rSIG and TRISS scores were significantly lower in the non-survival group than in the survival group (rSIG: median 3.7 vs 18.1, p<0.001; TRISS: median 45.9 vs 99.3, p<0.001) (table 1). The optimal rSIG cut-off value for predicting survival, derived from the derivation cohort, was 10.13. When stratified by age group, the cut-off values differed: 7.94 for children aged ≤6 years, 10.98 for those aged 7–12 years and 11.93 for those aged 13–15 years (table 2).

Table 1. Comparison of demographic and clinical characteristics between survival and non-survival groups in the derivation cohort.

Variables	Survivor n=13 562	Dead n=321	P value
Age, years (median IQR)	12 (7–16)	15 (8–16)	<0.001
Patients age 0–6, n (%)	2691 (2)	60 (17)	<0.001
Patients age 7–12, n (%)	4396 (32)	59 (18)	<0.001
Patients age 13–17, n (%)	6475 (47)	202 (63)	<0.001
Male, n (%)	9736 (72)	228 (71)	0.765
Blunt injury, n (%)	13 337 (98)	320 (99)	0.059
Injury region with AIS ≥3, n (%)
Head injury	4395 (32)	266 (88)	<0.001
Facial injury	116 (1)	8 (2)	0.002
Neck injury	21 (0.2)	6 (2)	<0.001
Chest injury	2382 (18)	183 (57)	<0.001
Abdominal and pelvic injury	683 (5)	59 (18)	<0.001
Spinal injury	570 (4)	21 (7)	0.009
Upper extremity injury	1050 (8)	12 (4)	0.008
Lower extremity injury	2110 (16)	71 (22)	0.001
rSIG, (median IQR)	18.1 (14.0–22.1)	3.7 (2.2–7.5)	<0.001
Systolic blood pressure at hospital arrival, (median IQR)	123 (111–136)	105 (70–137)	<0.001
Heart rate at hospital arrival, (median IQR)	95 (81–111)	119 (80–142)	<0.001
Glasgow Coma Scale at hospital arrival, (median IQR)	15 (14–15)	3 (3–6)	<0.001
Survival probability by TRISS methods, % (mean, IQR)	99.3 (98.4–99.5)	45.9 (17.5–78.5)	<0.001

AIS, Abbreviated Injury Scale; rSIG, revised Shock Index multiplied by the Glasgow Coma Scale; TRISS, Trauma and Injury Severity Score.

Table 2. Age-specific cutoff values of rSIG for predicting survival derived from the derivation cohort.

Variables	rSIG optimal cut-off point
All paediatric patients	10.13
Patients age 0–6	7.94
Patients age 7–12	10.98
Patients age 13–17	11.93

rSIG, revised Shock Index multiplied by the Glasgow Coma Scale.

The predictive accuracy of rSIG and TRISS for survival was compared in the validation cohort (table 3). In the entire cohort of paediatric patients with trauma (n=2814), the area under the receiver operating characteristic curve (AUC) for rSIG was 0.857 (95% CI 0.796 to 0.918), which was significantly higher than that for TRISS (AUC 0.740, 95% CI 0.661 to 0.820; p=0.006). Age-stratified analysis showed that in the 7–12 years age group, rSIG had the highest AUC of 0.914 (95% CI 0.831 to 0.996), which was significantly greater than that of TRISS (AUC 0.706, 95% CI 0.560 to 0.852; p=0.033). In the 13–17 years group, rSIG also demonstrated superior performance, with an AUC of 0.873 (95% CI 0.799 to 0.947), compared with TRISS (AUC 0.745, 95% CI 0.638 to 0.852; p=0.016). In contrast, in the 0–6 years group, no significant difference was observed between rSIG (AUC 0.836, 95% CI 0.640 to 1.000) and TRISS (AUC 0.797, 95% CI 0.557 to 1.000; p=0.699).

Table 3. Comparison of predictive accuracy of rSIG and TRISS for in-hospital survival in the validation cohort.

	No. of patients	Optimal cut-off point	Sensitivity, %	Specificity, %	PPV, %	NPV, %	AUC (95% CI)	P value
All paediatric patients
rSIG	2814	10.1	89.3	82.1	99.7	9.8	0.857 (0.796 to 0.918)	0.006
Survival probability by TRISS methods	2814	50.0	99.3	48.7	99.3	50.0	0.740 (0.661 to 0.820)
Patients age 0–6
rSIG	576	7.9	89.5	80.0	99.8	2.9	0.836 (0.640 to 1.000)	0.699
Survival probability by TRISS methods	576	50.0	99.4	60.0	99.7	50.0	0.797 (0.557 to 1.000)
Patients age 7–12
rSIG	908	11.0	91.1	91.7	99.9	15.7	0.914 (0.831 to 0.996)	0.033
Survival probability by TRISS methods	908	50.0	99.6	41.7	99.2	55.6	0.706 (0.560 to 0.852)
Patients age 13–17
rSIG	1330	11.9	88.2	86.4	99.6	14.5	0.873 (0.799 to 0.947)	0.016
Survival probability by TRISS methods	1330	50.0	99.1	50.0	99.2	47.8	0.745 (0.638 to 0.852)

AUC, area under the curve; NPV, negative predictive value; PPV, positive predictive value; rSIG, revised Shock Index multiplied by the Glasgow Coma Scale; TRISS, Trauma and Injury Severity Score.

Sensitivity analysis in patients with isolated TBI

In the derivation cohort, 3332 patients (24.0%) met the criteria for isolated TBI, defined as head AIS ≥3 without severe extracranial injury, among whom 94 patients (2.8%) died. Using this subgroup, the optimal rSIG cut-off value for predicting in-hospital mortality was identified as 9.67.

When this cut-off value was applied to the validation cohort, 118 patients (20.4%) were classified as having an abnormal rSIG group. The predictive performance of rSIG for mortality remained high in this subgroup, with an AUROC of 0.879 (95% CI 0.859 to 0.899).

Discussion

Our findings demonstrated that the optimal cut-off value of rSIG for survival prediction was 10.13, and rSIG exhibited significantly higher predictive accuracy than TRISS. This superiority was particularly pronounced in patients aged ≥7 years.

The cut-off value identified in this study was generally consistent with those reported in previous studies focusing on mortality prediction in paediatric populations.⁸,¹⁰ Notably, several studies have also proposed age-specific cut-off values and a consistent trend across these studies is that the cut-off value increases with age.^{8 10} This is likely attributable to age-related physiological characteristics unique to paediatric patients, specifically higher HRs and lower SBPs in younger children.¹⁴ It should be emphasised that the optimal rSIG cut-off value identified in this study was derived to maximise discriminative performance for mortality prediction. Cut-off values optimised for prognostic accuracy, particularly those based on maximum AUROC, may not be directly applicable to clinical triage settings, where minimising undertriage is a primary priority. In such settings, higher cut-off values that prioritise sensitivity and negative predictive value may be more appropriate, even at the expense of overall discrimination. Therefore, while rSIG represents a rapid and simple physiological indicator, the cut-off value proposed in this study should be interpreted according to its intended use and alternative thresholds may be required for real-time triage decision-making.

Previous large-scale multicentre studies, mainly in adult trauma populations, have demonstrated that TRISS is a robust predictor of trauma-related mortality and is widely used as an internationally standardised prognostic tool with a long track record of clinical use.¹⁵,¹⁷ In many settings, TRISS has been incorporated into emergency department workflows for severity assessment and benchmarking of trauma care. However, despite these established strengths, several factors may explain why rSIG showed superior predictive performance in the present paediatric cohort. First, the TRISS model dichotomises age at 55 years and does not account for the non-linear relationship between age and mortality in paediatric populations. In this study, mortality rates varied across paediatric age groups: 2.2% in neonates and infants, 1.3% in school-aged children and 3.1% in adolescents. These variations suggest that a simple age adjustment is insufficient for accurately estimating prognosis in children. Therefore, scoring systems such as rSIG, which allow for age-specific cut-off values, may better reflect the physiological differences associated with paediatric age groups.

Second, the discrepancy between the context in which the TRISS model was developed and the current state of trauma care in Japan may also contribute to its limitations. The TRISS methodology was constructed using data from the USA prior to 1990, which differs significantly from Japan’s current prehospital care system, trauma care infrastructure and data registration standards. Furthermore, advancements in trauma care have dramatically reduced in-hospital mortality rates for severely injured patients (defined as AIS ≥3 or ISS ≥16–18) from >20% to approximately 9–12.3% in recent years,^{18 19} leading to concerns that TRISS may overestimate mortality risk.^{3 20} The limitations of outcome prediction based solely on anatomical severity scores like the ISS have become evident, and the use of models incorporating physiological indicators adjusted for age has gained recognition. Scoring systems such as the Pediatric Trauma Score and the Shock Index Pediatric Adjusted have demonstrated high predictive performance by evaluating physiological parameters in an age-appropriate manner.²¹,²³ Moreover, in paediatric populations, physiological indicators vary significantly by age, which limits the applicability of models like TRISS, originally designed for adults. The RTS, which includes coded values (0–4) for SBP, RR and level of consciousness, may not be appropriate for paediatric use. Previous studies have also reported that TRISS tends to underestimate survival probabilities in patients with low predicted survival and younger children, highlighting the need for model recalibration and age-specific assessment.⁵

Third, assessment of consciousness level using the GCS may serve as a critical prognostic factor in paediatric trauma. Head injuries constitute a substantial proportion of severe trauma cases,^{24 25} and in the present study, 88% of the non-survivors sustained head trauma. The rSIG comprises three physiological parameters—SBP, HR and GCS—with the GCS potentially exerting the greatest effect on the score. Prognostic models that incorporate GCS have demonstrated superior predictive accuracy to those that do not. In particular, rSIG has been reported in several studies to outperform other physiology-based scores, such as the Shock Index Pediatric Adjusted, in terms of predictive performance.^{7 9 10} These findings underscore the importance of including GCS in trauma scoring systems. However, the limitations of GCS in paediatric populations should be noted, especially in younger children with possible underdeveloped verbal and motor responses. In the present study, rSIG did not demonstrate superior predictive performance compared with TRISS, potentially reflecting limitations in GCS assessment and marked age-related physiological variability.

In response to concerns that the optimal rSIG cut-off value may differ according to injury type, we performed a sensitivity analysis focusing on patients with isolated TBI. In this subgroup, rSIG maintained strong discriminatory performance for in-hospital mortality; however, the optimal cut-off value (9.76) was lower than that derived from the overall paediatric cohort. This finding suggests that the prognostic threshold of rSIG may vary depending on the presence of brain injury. These results are consistent with prior adult trauma studies reporting that optimal rSIG cut-off values differ between patients with and without brain injury, reflecting the substantial influence of neurological impairment on mortality prediction.²⁶ In paediatric trauma, where brain injury constitutes a major determinant of outcome, reliance on a single universal cut-off value may therefore lead to misclassification in certain subgroups. In addition, penetrating injuries were rare in the present cohort, precluding mechanism-specific analyses; therefore, caution is warranted when extrapolating these findings to trauma systems with a higher prevalence of penetrating trauma, particularly outside Japan.

This study has some limitations. First, as with all epidemiological observational studies, the completeness and validity of the data, as well as potential verification bias, are inherent limitations. Second, the JTDB does not include all emergency and trauma centres nationwide, introducing the possibility of selection bias. Furthermore, this study was conducted using data from designated emergency centres in Japan, which may limit the generalisability of the findings. In addition, the cut-off value in this study is based on the current state of Japanese healthcare, and future validation through international multicentre studies is necessary. Third, physiological assessments such as the GCS, SBP and RR are highly age-dependent, and accurate evaluation in younger paediatric patients may be challenging, potentially introducing age-specific bias. Fourth, this study did not include comparisons with other prognostic scoring tools capable of predicting survival, such as the Shock Index, Shock Index Paediatric Adjusted, Base deficit, International Normalised Ratio, GCS or the Paediatric Trauma Score.

Conclusion

This study demonstrated that the rSIG score provides superior predictive accuracy for survival compared with the TRISS method in paediatric patients with trauma. The rSIG may serve as a valuable tool for the initial assessment of paediatric patients with trauma as it is simple and can be applied rapidly. Notably, the ability of rSIG to reflect age-related physiological variability through age-specific cut-off values represents a potential advantage over traditional models that incorporate age adjustment in a limited manner.

As this study was limited to a comparison between rSIG and TRISS, future research should include comparative evaluations with other paediatric trauma scores. To facilitate the clinical implementation of rSIG, establishing reliable, age-stratified cut-off values is essential. Future multicentre prospective studies are warranted to validate these findings and develop more universally applicable standards. Moreover, in younger paediatric populations, where the limitations of GCS assessment are well recognised, alternative methods for evaluating consciousness or complementary scoring systems should be explored. Incorporating clinical laboratory parameters, such as blood lactate levels and arterial blood gas analysis (pH and base excess), may further enhance the accuracy of risk stratification.^{27 28} In the future, the development of a comprehensive prognostic model integrating these variables could support rapid triage decisions in prehospital settings and during initial clinical assessment.

Data availability statement

Data are available upon reasonable request.