Comparison of Glasgow coma scale, motor component, eye component, and simplified motor scale for predicting trauma outcomes: a 13-year multicenter retrospective cohort study

Abstract

Background

To address the complexity and time-consuming nature of the full Glasgow Coma Scale (GCS), various simplified tools for assessing brain function have been proposed, such as the eye (GCSE) and motor component of the GCS (GCSM), and the Simplified Motor scale (sMS). However, few studies have evaluated the predictive ability of these scoring systems.

Methods

A 13-year cohort study was conducted using the trauma database of Tzu Chi Hospital to compare the accuracy of the full GCS with those of the GCSE, GCSM, and sMS for predicting short- and long-term mortality (3-day mortality, 7-day mortality, and in-hospital mortality), intensive care unit (ICU) stay of ≥ 14 days, and hospital stay of ≥ 30 days in patients with trauma.

Results

This study included 41,297 patients with trauma. The full GCS achieved slightly higher area-under-the-receiver-operating-characteristic-curve (AUROC) values for predicting 3-day mortality (full GCS vs. GCSM vs. sMS: 0.899 vs. 0.894 vs. 0.890), 7-day mortality (0.871 vs. 0.864 vs. 0.861), in-hospital mortality (0.833 vs. 0.817 vs. 0.815), ICU length of stay (LOS)of ≥ 14 days (0.645 vs. 0.628 vs. 0.628), and hospital LOS of ≥ 30 days (0.607 vs. 0.587 vs. 0.587). The GCSE exhibited inferior discriminative ability for all clinical outcomes. The AUROC values for the ability of the sMS to predict 3-day mortality, 7-day mortality, and in-hospital mortality were comparable to those of the GCSM but lower than those of the full GCS for patients aged ≥ 65 years, aged < 65 years, with or without cardiovascular diseases, and with or without traumatic brain injury. For predicting ICU LOS of ≥ 14 days and hospital LOS of ≥ 30 days, the discriminative accuracy of the full GCS was marginally higher than those of the GCSE, GCSM, and sMS across the aforementioned subgroups. However, GCSE, GCSM, and sMS had similar discriminative accuracy.

Conclusions

Although the full GCS assessment exhibited higher accuracy in predicting 3-day mortality, 7-day mortality, in-hospital mortality, ICU LOS of ≥ 14 days, and hospital LOS of ≥ 30 days compared with the GCSE, GCSM, and sMS, the marginally higher accuracy of the full GCS may be negligible given its time-consuming nature. Furthermore, use of the GCSM provides no substantial advantage over use of the simpler sMS, which has comparable predictive accuracy.

Clinical trial number

Not applicable.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12873-025-01246-4.

Introduction

Trauma, a leading cause of death among individuals aged < 45 years, results in considerable physical disability, psychological disorders, and substantial social costs [1, 2].The timely identification of high-risk patients and the rapid transport of patients to trauma centers for definitive care can prevent catastrophic complications and optimize the allocation of limited emergency department resources [3, 4].

The Glasgow Coma Scale (GCS) is one of the crucial predictors of severe injury in patients with trauma and is commonly used for field triage. The GCS has three independent components: eye opening (E), verbal response (V), and motor response (M), each with distinct scoring criteria; the total GCS score ranges from 3 to 15 [5].Increasing evidence suggests that the three independent components of the GCS represent distinct neurological dimensions, with each component score having distinct meanings and weights [6]. Therefore, summing these scores can dilute their original meaning and lead to an overemphasis on the total score. Previous research has also demonstrated that the reliability of a single GCS component is higher than that of the total score, possibly because normal scores for the other components can inflate the overall total. Moreover, Feldman et al. reported that field providers correctly calculated the total GCS score in only 40% of cases [7]. Although the full GCS score represents the results of a detailed and multifaceted assessment, the complexity and time-consuming calculation involved in the assessment process can hinder accurate evaluation, prolong the on-scene time, and potentially delay patient transport. To address these challenges, several simplified tools for assessing brain function have been proposed, including the eye (GCSE) and motor component of the GCS (GCSM), and the Simplified Motor Scale (sMS).

The GCS-E component assesses a patient’s eye-opening response on a 4-point scale: spontaneous eye opening, eye opening to verbal command, eye opening to pain, and no eye opening. The GCS Motor (GCS-M) component is a 6-point scale that evaluates motor responses, ranging from no motor response to localization of pain and the ability to obey commands [8–11]. Both the GCSE and GCSM have been demonstrated to have strong correlations with the full GCS score, and they are useful for rapid assessment of brain function. The 3-point sMS, a trichotomized version of the 6-point GCSM, provides a simpler and more user-friendly alternative to the full GCS and GCSM [12]. Studies have reported that the sMS exhibits discriminative ability similar to that of the GCS for predicting the need for intubation in the emergency department, neurosurgical intervention, brain injury, and mortality [13, 14]. In prehospital care, the sMS serves as a quick and practical tool for assessing brain function in real time. Although various modified GCS assessments have been developed and demonstrated to provide prediction accuracy and discriminative ability comparable to those of the full GCS score for patients with trauma, few studies have directly compared the GCSE, GCSM, and sMS with the full GCS in terms of predicting mortality and other clinical outcomes [12, 14–16].

In the present study, we evaluated and compared the discriminative ability of the full GCS with those of the GCSE, GCSM, and sMS for predicting short- and long-term mortality and other clinical outcomes in patients with trauma.

Methods

Study design and setting

The present retrospective cohort study used data from Tzu Chi Hospital’s trauma database and was approved by the hospital’s institutional review board (approval number: 12-XD-077). For the Trauma Registry, registered case managers must undergo certification and continuing training courses provided by the Formosa Association for the Surgery of Trauma [17, 18]. The trauma database contains the data of all hospitalized patients with traumatic injuries, including data pertaining to 152 variables, such as demographics, injury mechanisms, injury types, Injury Severity Scores (ISSs), vital signs, and in-hospital mortality [19–22]. The present study was conducted in accordance with the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines (Supplementary Table 1) [23].

Patient selection

We included all patients with trauma recorded in the Tzu Chi Hospital trauma database between January 2009 and December 2021. Patients were excluded if they (1) were younger than 20 years (because of variations in normal vital sign ranges and brain function management in the pediatric population), (2) were transferred to another hospital, or (3) had insufficient data on in-hospital mortality or GCS scores. The GCS, GCSE, and GCSM were employed to assess brain function for each patient, and the discriminative abilities of these tools were compared for five major clinical outcomes. These comparisons were also conducted within subgroups based on age (geriatric: ≥65 years vs. nongeriatric: <65 years), cardiovascular disease status (CVD vs. non-CVD), traumatic brain injury (TBI vs. non-TBI), and injury severity (major trauma: ISS ≥ 16 vs. minor trauma: ISS < 16). Cardiovascular disease was defined as the presence of any of the following conditions: congestive heart failure, valvular heart disease, coronary artery disease, hypertension, or others. These comorbidities were identified using ICD-10 codes I00–I52 documented in the electronic medical record.

Variable measurements

Brain function measures

The GCS is widely used to assess a person’s level of brain function, and it is regarded as a more robust discriminator of mortality compared with traditional hemodynamic variables, such as systolic blood pressure (SBP), diastolic blood pressure (DBP), and heart rate [24, 25].The full GCS score is calculated by summing the scores from its E, V, and M components. Although the full GCS score is based on a detailed and comprehensive evaluation of brain function, the assessment process is time-consuming and may delay the on-scene response.

The GCSE score assesses eye-opening response on a 4-point scale: 1 (spontaneous eye opening), 2 (eye opening to verbal command), 3 (eye opening to pain), and 4 (no eye opening). The GCSM score represents the motor component of the GCS and uses a 6-point scale, with endpoints ranging from 1 (no motor response) to 6 (obeys commands), to predict mortality in patients with injuries. The GCSM was further trichotomized to develop the 3-point sMS, which assigns scores as follows: obeys commands = 3; localizes pain = 2; withdraws from pain or worse = 1. The sMS is a more straightforward alternative to the full GCS and GCSM, facilitating patient assessment. Studies have reported that the sMS demonstrated discriminative ability similar to that of the full GCS in predicting the need for intubation in emergency departments, neurosurgical intervention, brain injury, and mortality [13, 14].

Prehospital brain function evaluations were typically performed by certified emergency medical technicians (EMTs), while in-hospital assessments were conducted by triage nurses in the emergency department. In Taiwan, individuals must complete formal education and clinical training and pass a national certification exam to become licensed EMTs. EMTs are classified into three levels: EMT-1 (basic), EMT-2 (intermediate), and EMT-P (paramedic), requiring a minimum of 40, 280, and 1,280 h of training, respectively. In-hospital triage is carried out by certified triage nurses. For instance, at Taipei Tzu Chi Hospital, nurses must complete at least three years of foundational clinical training in general medicine, observation units, emergency care, intensive care, pediatrics, trauma, and toxic/environmental injury care. They must also be certified through the Clinical Ladder System of the Taiwan Nurses Association and complete formal triage classification training. Triage nurse candidates are supervised by senior triage nurses and must perform triage evaluations on at least 100 patients over five consecutive days. Only after passing a formal assessment can they be certified as triage nurses. Currently, 25 certified emergency triage nurses are practicing at Taipei Tzu Chi Hospital. Both EMTs and triage nurses are required to undergo regular continuing education and training to maintain consistency in assessment. Although inter-rater variability cannot be completely eliminated due to the retrospective nature of the study, the Tzu Chi Trauma Database follows a standardized data collection protocol. All data are entered by trained personnel at participating institutions. In addition, monthly data audits are conducted to ensure the accuracy and reliability of the recorded information.

Covariates

The present study retrieved and analyzed the included patients’ basic clinicodemographic characteristics, including age, sex, underlying diseases, triage level, vital signs, injury mechanisms, and injury severity. Triage levels, determined using the Taiwan Triage and Acuity Scales, range from level I (most urgent) to level V (least urgent). The vital signs recorded upon hospital arrival were heart rate, SBP, DBP, and GCS score. The Abbreviated Injury Scale (AIS) categorizes injuries into nine predefined anatomical regions and assigns a severity score from 1 (minor) to 6 (maximal, currently untreatable). The AIS forms the basis for calculating the ISS, which reflects the overall trauma burden. The ISS is calculated by identifying the highest AIS score in each of the three most severely injured body regions, squaring these values, and summing the results. The ISS ranges from 1 to 75, with major trauma defined as an ISS ≥ 16 [26, 27]. TBI status was categorized as TBI (head Abbreviated Injury Scale [AIS] score ≥ 3) or non-TBI (head AIS score < 3). The identified injury mechanisms were motor vehicle collisions, low falls (fall height < 1 m), high falls (fall height ≥ 1 m), and other mechanisms (e.g., drowning, burns, and cold injuries).

Clinical outcomes

The present study examined five primary clinical outcomes: two short-term outcomes (3-day mortality and 7-day mortality) and three long-term outcomes (in-hospital mortality, ICU length of stay [LOS] of ≥ 14 days, and hospital LOS of ≥ 30 days).

Statistical analysis

Categorical and nominal variables were analyzed using Pearson’s chi-squared test or Fisher’s exact test. For comparisons, ordinal variables (e.g., GCS, GCSE, GCSM, and sMS) were evaluated using nonparametric analysis of variance or Mann–Whitney U tests. The discriminative ability of the four scoring systems (GCS, GCSE, GCSM, and sMS) for each clinical outcome was evaluated using the area under the receiver operating characteristic curve (AUROC). We use the DeLong test to compare two AUROC values, which is a widely used non-parametric method in medical research that evaluates the difference between two correlated ROC curves without assuming any specific data distribution [28]. By calculating the difference in AUCs and their standard error, a p-value is obtained. If the p-value is below the significance threshold, it indicates a statistically significant difference in diagnostic or predictive performance. Patients with missing data pertaining to GCS, GCSE, GCSM, or sMS scores were excluded from the analysis. The proportion of missing data was 4.5% for GCS scores, and 3.0% each for GCSE, GCSM, and sMS scores. Because these scores were essential for predicting clinical outcomes inpatients with trauma, imputation was not performed to avoid potential biases. The demographic characteristics of the patients who were included and excluded (due to missing data) are provided in Supplementary Table 2.

Subgroup analyses were conducted to assess the discriminative performance of the scoring systems in various patient populations, including those stratified by injury severity (minor vs. major), CVD status (non-CVD vs. CVD), age (geriatric: ≥65 years vs. nongeriatric: <65 years), and TBI status (non-TBI vs. TBI). All statistical tests were two-sided, with the significance level set at P < 0.05. Data analyses were conducted using SPSS version 25.0 for Windows (IBM, Armonk, NY, USA) under a valid copyright license.

Results

Characteristics of study participants

Of the 48,524 eligible patients who were identified, 7,227 were excluded for the following reasons: age < 20 years (n = 3,780), transfer to another hospital (n = 1,081), missing in-hospital mortality data (n = 1,171), or missing data for GCS, GCSE, GCSM, or sMS score (n = 1,195). Ultimately, 41,297 patients were included in the analysis. Figure 1 presents a detailed patient selection flowchart.

Fig. 1.

Flow diagram for selection of study participants

The prevalence of 3-day mortality, 7-day mortality, and in-hospital mortality was 1.4%, 1.9%, and 3.1%, respectively. Geriatric patients accounted for 56.4% of in-hospital deaths, and the prevalence of CVD in the cohort was 25.4%. The overall prevalence of major injuries (ISS ≥ 16) was 12.8%, and major injuries occurred in 77.8% of the patients who died in hospital. Falls were the most common injury mechanism (40.4%), comprising low falls (32.7%) and high falls (7.7%), followed by traffic accidents (35.9%). Head injuries were reported for 18.0% of the patients. Regarding LOS, 4.5% of patients had a total hospital LOS of ≥ 30 days, and 9.4% had an ICU LOS of ≥ 14 days. Table 1 provides percentages for each demographic variable as they relate to mortality. Patients with the following characteristics demonstrated a higher in-hospital mortality rate: age ≥ 65 years, male sex, triage level 1, and a GCS score of 3–8. In terms of specific subcomponents, those with an eye-opening score of 1, a motor response score of 1, and a sMS score of 1 also had increased mortality. Additionally, patients with an Injury Severity Score (ISS) ≥ 16, those sustaining non-penetrating trauma, and those injured by high-level falls were associated with significantly higher mortality. Subgroup comparisons of the four scoring systems are summarized in Table 2. Patients who died within 3 days, 7 days, or during hospitalization had significantly lower GCS, GCSE, GCSM, and sMS scores than did the survivors. Similarly, patients with triage level I had significantly lower scores in all four systems relative to those with triage levels II to V.

Table 1.

Demographic characteristics of included patients stratified by mortality

	Total patients	3-day mortality		7-day mortality		In-hospital mortality
Characteristics	Number (%)	Number (%)	p-value	Number (%)	p-value	Number (%)	p-value
Patient number	41,297(100%)	583(1.4%)	----	797(1.9%)	----	1293(3.1%)	----
Age (years)
Age < 65ys	23,598(57.1%)	305(1.3%)	0.018	374(1.6%)	< 0.001	564(2.4%)	< 0.001
Age ≥ 65ys	17,699(42.9%)	278(1.6%)		423(2.4%)		729(7.1%)
Sex, n (%)
Female	19,341(46.8%)	186(1.0%)	< 0.001	269(1.4%)	< 0.001	432(2.2%)	< 0.001
Male	21,956(53.2%)	397(1.8%)		528(2.4%)		861(3.9%)
Triage
1	2752(6.7%)	477(17.3%)	< 0.001	600(21.8%)	< 0.001	587(31.1%)	< 0.001
2	14,785(35.9%)	63(0.4%)		119(0.8%)		261(1.8%)
3	23,088(56.1%)	42(0.2%)		76(0.3%)		172(0.7%)
4 and 5	505(1.3%)	1(0.2%)		1(0.2%)		2(0.4%)
GCS score
3–8	1876(4.5%)	453(24.1%)	< 0.001	565(30.1%)	< 0.001	782(41.7%)	< 0.001
9–12	1135(2.7%)	35(3.1%)		61(5.4%)		126(11.1%)
13–15	38,286(92.7%)	95(0.2%)		171(0.4%)		385(1.0%)
Eye subscale
1	1399(3.4%)	418(29.9%)	< 0.001	515(36.8%)	< 0.001	696(49.7%)	< 0.001
2	391(0.9%)	28(7.2%)		40(10.2%)		70(17.9%)
3	1047(2.5%)	21(2.0%)		36(3.4%)		78(7.4%)
4	38,460(93.1%)	116(0.3%)		206(0.5%)		449(1.2%)
Motor subscale
1	1019(2.5%)	363(35.6%)	< 0.001	434(42.6%)	< 0.001	589(57.8%)	< 0.001
2	101(0.2%)	19(18.8%)		28(27.7%)		39(38.6%)
3	144(0.3%)	25(17.4%)		34(23.6%)		43(29.9%)
4	557(1.3%)	41(7.4%)		64(11.5%)		109(19.6%)
5	1370(3.3%)	43(3.1%)		73(5.3%)		138(10.1%)
6	38,106(92.3%)	92(0.2%)		164(0.4%)		375(1.0%)
sMS score
1	1821(4.4%)	448(24.6%)	< 0.001	560(30.8%)	< 0.001	780(42.8%)	< 0.001
2	1370(3.3%)	43(3.1%)		73(5.3%)		138(10.1%)
3	38,106(92.3%)	92(0.2%)		164(0.4%)		375(1.0%)
Injury severity
ISS < 16	35,480(87.2%)	68(0.2%)	< 0.001	126(0.4%)	< 0.001	272(0.8%)	< 0.001
ISS ≥ 16	5215(12.8%)	506(9.7%)		658(12.6%)		955(18.3%)
Traumatic brain injury
Non-TBI	33,880(82.0%)	145(0.4%)	< 0.001	192(0.6%)	< 0.001	398(1.2%)	< 0.001
TBI	7417(18.0%)	438(5.9%)		605(8.2%)		895(12.1%)
Injury type
Penetrative	2124(5.1%)	12(0.6%)	0.001	13(0.6%)	< 0.001	22(1.0%)	< 0.001
Non-penetrative	39,173(94.9%)	571(1.5%)		784(2.0%)		1271(3.2%)
Mechanism of injury
Traffic road injury	14,838(35.9%)	223(1.5%)	< 0.001	306(2.1%)	< 0.001	487(3.3%)	< 0.001
Low fall	13,502(32.7%)	124(0.9%)		204(1.5%)		398(2.9%)
High fall	3171(7.7%)	73(2.3%)		98(3.1%)		141(4.4%)
Others	9786(23.7%)	163(1.7%)		189(1.9%)		267(2.7%)
Comorbidity
CVD	10,475(25.4%)	122(1.2%)	0.013	184(1.8%)	0.135	322(3.1%)	0.698
Non-CVD	30,822(74.6%)	461(1.5%)		613(2.0%)		971(3.2%)
Hospitalization
Total LOS < 30 days	39,287(95.5%)	583(1.5%)	< 0.001	797(2.0%)	< 0.001	1116(2.8%)	0.011
Total LOS ≥ 30 days	1843(4.5%)	0(0.0%)		0(0.0%)		71(3.9%)
ICU admission
ICU LOS < 14 days	7154(90.6%)	167(2.3%)	< 0.001	307(4.3%)	< 0.001	457(6.4%)	< 0.001
ICU LOS ≥ 14 days	744(9.4%)	1(0.1%)		2(0.3%)		97(13.0%)

Abbreviations: CVD, cardiovascular disease; TBI, traumatic brain injury; ISS, injury severity score; LOS, length of stay; ICU, intensive care unit; sMS, simplified motor score

Table 2.

Glasgow coma scale (GCS), E subscale, M subscale, and simplified motor scale (sMS) scores in subgroup analysis

Characteristics	GCS		E subscale		M subscale		sMS score
	Median (IQR)	P-value	Median (IQR)	P-value	Median (IQR)	P-value	Median (IQR)	P-value
Level I	10(3–15)	< 0.001	2(1–4)	< 0.001	4(1–6)	< 0.001	1(1–3)	< 0.001
Level II	15(15–15)		4(4–4)		6(6–6)		3(3–3)
Level III	15(15–15)		4(4–4)		6(6–6)		3(3–3)
Level IV/V	15(15–15)		4(4–4)		6(6–6)		3(3–3)
Injury severity
ISS < 16	15(15–15)	< 0.001	4(4–4)	< 0.001	6(6–6)	< 0.001	3(3–3)	< 0.001
ISS ≥ 16	15(11–15)		4(3–4)		6(5–6)		3(2–3)
TBI status
Non-TBI	15(15–15)	< 0.001	4(4–4)	< 0.001	6(6–6)	< 0.001	3(3–3)	< 0.001
TBI	15(13–15)		4(3–4)		6(5–6)		3(2–3)
ICU
ICU LOS < 14 days	15(15–15)	< 0.001	4(4–4)	< 0.001	6(6–6)	< 0.001	3(3–3)	< 0.001
ICU LOS ≥ 14 days	15(9–15)		4(2–4)		6(4–6)		3(1–3)
Hospital Stay
Days < 30	15(15–15)	< 0.001	4(4–4)	< 0.001	6(6–6)	< 0.001	3(3–3)	< 0.001
Days ≥ 30	15(14–15)		4(4–4)		6(5–6)		3(2–3)
3-day mortality
Survival	15(15–15)	< 0.001	4(4–4)	< 0.001	6(6–6)	< 0.001	3(3–3)	< 0.001
Death	3(3–9)		1(1–2)		1(1–4)		1(1–1)
7-day mortality
Survival	15(15–15)	< 0.001	4(4–4)	< 0.001	6(6–6)	< 0.001	3(3–3)	< 0.001
Death	4(3–13)		1(1–4)		1(1–5)		1(1–2)
In hospital mortality
Survival	15(15–15)	< 0.001	4(4–4)	< 0.001	6(6–6)	< 0.001	3(3–3)	< 0.001
Death	7(3–15)		1(1–4)		3(1–6)		1(1–3)

Abbreviations: CVD, cardiovascular disease; TBI, traumatic brain injury; ISS, injury severity score; LOS, length of stay; ICU, intensive care unit; sMS, simplified motor scale

Discriminative accuracy for clinical outcomes

Table 3 summarizes the AUROC values comparing the discriminative ability of the four scoring systems for predicting 3-day mortality, 7-day mortality, in-hospital mortality, ICU LOS of ≥ 14 days, and hospital of LOS ≥ 30 days. The discriminative ability of the sMS for all clinical outcomes was similar to that of the GCSM but poorer than that of the full GCS. The full GCS achieved a significantly higher AUROC value than did the GCSM and sMS in predicting 3-day mortality (GCS vs. GCSM vs. sMS, 0.899vs 0.894 vs. 0.890), 7-day mortality (0.871vs 0.864 vs. 0.861), in-hospital mortality (0.833vs 0.817 vs. 0.815), ICU LOS of ≥ 14 days (0.645vs 0.628 vs. 0.628), and hospital LOS of ≥ 30 days (0.607vs 0.587 vs. 0.587). Of the four scales, the GCSE exhibited the lowest AUROC values for predicting all clinical outcomes.

Table 3.

GCS, E subscale, M subscale, and sMS scores in terms of the area under the receiver operating characteristic curve (AUROC) for prediction of five trauma outcomes

Scoring systems	AUROC (95% CI)	Difference of AUROC
		Total GCS score	E subscale	M subscale	sMS score
Predicting 3-day mortality
Total GCS score	0.899(0.879–0.919)***	----	----	----	----
E subscale	0.875(0.853–0.897)***	0.025(0.015–0.034)***	----	----	----
M subscale	0.894(0.873–0.914)***	0.005(0.001–0.011)*	0.019(0.008–0.030)**	----	----
sMS score	0.890(0.870–0.910)***	0.009(0.004–0.015)**	0.015(0.004–0.026)*	0.004(0.003–0.004)***	----
Predicting 7-day mortality
Total GCS score	0.871(0.852–0.890)***	----	----	----	----
E subscale	0.839(0.819–0.860)***	0.032(0.022–0.041)***	----	----	----
M subscale	0.864(0.844–0.883)***	0.008(0.003–0.013)*	0.024(0.014–0.035)***	----	----
sMS score	0.861(0.841–0.880)***	0.010(0.005–0.016)**	0.021(0.011–0.032)**	0.003(0.002–0.003)***	----
Predicting inhospital mortality
Total GCS score	0.833(0.816–0.849)***	----	----	----	----
E subscale	0.790(0.772–0.807)***	0.043(0.034–0.052)***	----	----	----
M subscale	0.817(0.800-0.834)***	0.016(0.010–0.021)***	0.027(0.018–0.037)***	----	----
sMS score	0.815(0.798–0.832)***	0.018(0.012–0.024)***	0.025(0.016–0.035)***	0.002(0.002–0.003)***	----
Predicting ICU length of stay ≥ 14 days
Total GCS score	0.645(0.621–0.669)***	----	----	----	----
E subscale	0.625(0.600–0.650)***	0.020(0.009–0.032)**	----	----	----
M subscale	0.628(0.604–0.653)***	0.017(0.007–0.026)**	0.004(-0.009-0.016)	----	----
sMS score	0.628(0.604–0.652)***	0.017(0.008–0.026)**	0.003(-0.009-0.015)	0.001(-0.001-0.001)	----
Predicting prolonged hospital length of stay ≥ 30 days
Total GCS score	0.607(0.591–0.622)***	----	----	----	----
E subscale	0.578(0.563–0.594)***	0.028(0.022–0.035)***	----	----	----
M subscale	0.587(0.571–0.602)***	0.020(0.015–0.026)***	0.008(0.002–0.015)*	----	----
sMS score	0.587(0.572–0.602)***	0.020(0.014–0.025)***	0.009(0.002–0.015)*	0.001(0.001–0.001)***	----

*P < 0.05; **P < 0.01; ***P < 0.001

Discriminative accuracy for clinical outcomes in subgroups

Table 4 presents the results of the subgroup analysis. The AUROC values for the ability of the sMS to predict 3-day mortality, 7-day mortality, and in-hospital mortality were similar to those for the GCSM but lower than those for the GCS across all the subgroups, namely patients aged < 65 years, aged ≥ 65 years, with TBI, without TBI, with CVD, and without CVD. For all subgroups, the GCSE had the lowest discriminative accuracy for all clinical outcomes. For predicting ICU LOS of ≥ 14 days and hospital LOS of ≥ 30 days, the full GCS had the highest discriminative accuracy across all subgroups. However, the accuracies in predicting ICU LOS and hospital LOS were similar among the GCSE, GCSM, and sMS. The AUROC values for all four brain function assessment tools (GCS, GCSE, GCSM, and sMS) were consistently higher in the TBI group compared to the non-TBI cohort across all outcome measures. Overall, the GCS demonstrated the highest AUROC among the four tools for each of the five clinical outcomes and across all six subgroups, with the sole exception of ICU length of stay ≥ 14 days in the non-TBI population.

Table 4.

AUROCs of GCS, E subscale, M subscale, and sMS scores for prediction of five trauma outcomes in subgroup analysis

Subgroup	Age < 65 years	Age ≥ 65 years	Non- CVD	CVD	Non-TBI	TBI
Prediction of 3-day mortality
Total GCS score	0.931 (0.909–0.952)***	0.851 (0.817–0.885)***	0.927 (0.908–0.946)***	0.799 (0.742–0.856)***	0.800 (0.749–0.851)***	0.908 (0.888–0.928)***
E subscale	0.914 (0.891–0.938)***	0.819 (0.783–0.856)***	0.907 (0.886–0.929)***	0.756 (0.695–0.816)***	0.775 (0.723–0.828)***	0.881 (0.858–0.903)***
M subscale	0.927 (0.905–0.949)***	0.843 (0.808–0.877)***	0.922 (0.902–0.942)***	0.791 (0.733–0.849)***	0.793 (0.741–0.844)***	0.905 (0.885–0.925)***
sMS score	0.923 (0.901–0.945)***	0.840 (0.806–0.874)***	0.917 (0.898–0.937)***	0.789 (0.731–0.847)***	0.792 (0.741–0.843)***	0.886 (0.866–0.906)***
Prediction of 7-day mortality
Total GCS score	0.945 (0.924–0.966)***	0.822 (0.794–0.851)***	0.907 (0.888–0.926)***	0.760 (0.712–0.809)***	0.750 (0.703–0.798)***	0.882 (0.862–0.901)***
E subscale	0.929 (0.905–0.952)***	0.780 (0.749–0.812)***	0.879 (0.858-0.900)***	0.717 (0.666–0.767)***	0.726 (0.678–0.774)***	0.849 (0.827–0.871)***
M subscale	0.942 (0.920–0.963)***	0.812 (0.782–0.841)***	0.901 (0.882–0.920)***	0.747 (0.698–0.796)***	0.747 (0.700-0.794)***	0.877 (0.857–0.897)***
sMS score	0.937 (0.916–0.958)***	0.810 (0.780–0.839)***	0.898 (0.878–0.917)***	0.746 (0.697–0.795)***	0.746 (0.699–0.793)***	0.861 (0.842–0.881)***
Prediction of inhospital mortality
Total GCS score	0.906 (0.886–0.926)***	0.780 (0.757–0.803)***	0.872 (0.855–0.889)***	0.724 (0.688–0.761)***	0.733 (0.700-0.766)***	0.845 (0.826–0.864)***
E subscale	0.880 (0.857–0.903)***	0.728 (0.703–0.753)***	0.836 (0.816–0.855)***	0.663 (0.625–0.701)***	0.703 (0.669–0.737)***	0.802 (0.781–0.823)***
M subscale	0.896 (0.874–0.917)***	0.761 (0.737–0.785)***	0.859 (0.840–0.877)***	0.702 (0.665–0.740)***	0.725 (0.691–0.758)***	0.833 (0.814–0.852)***
sMS score	0.892 (0.871–0.914)***	0.759 (0.735–0.783)***	0.856 (0.838–0.874)***	0.701 (0.664–0.739)***	0.724 (0.691–0.757)***	0.821 (0.802–0.841)***
Predicting ICU length of stay ≥ 14 days
Total GCS score	0.662 (0.626–0.698)***	0.632 (0.599–0.665)***	0.653 (0.624–0.681)***	0.630 (0.584–0.676)***	0.518 (0.478–0.558)	0.707 (0.679–0.734)***
E subscale	0.652 (0.616–0.688)***	0.607 (0.573–0.640)***	0.637 (0.608–0.666)***	0.600 (0.553–0.646)***	0.517 (0.477–0.557)	0.676 (0.647–0.706)***
M subscale	0.656 (0.620–0.692)***	0.606 (0.573–0.639)***	0.637 (0.609–0.666)***	0.609 (0.563–0.655)***	0.526 (0.486–0.566)	0.677 (0.648–0.706)***
sMS score	0.656 (0.620–0.692)***	0.605 (0.572–0.638)***	0.637 (0.608–0.666)***	0.609 (0.563–0.655)***	0.526 (0.486–0.566)	0.676 (0.647–0.705)***
Prediction of prolonged hospital length of stay ≥ 30 days
Total GCS score	0.596 (0.576–0.617)***	0.619 (0.596–0.642)***	0.612 (0.594–0.630)***	0.589 (0.558–0.621)***	0.534 (0.515–0.553)***	0.637 (0.614–0.659)***
E subscale	0.577 (0.557–0.598)***	0.580 (0.558–0.603)***	0.584 (0.566–0.601)***	0.561 (0.530–0.592)***	0.522 (0.503–0.541)*	0.606 (0.582–0.629)***
M subscale	0.584 (0.564–0.605)***	0.589 (0.567–0.612)***	0.593 (0.576–0.611)***	0.564 (0.533–0.595)***	0.527 (0.508–0.546)**	0.612 (0.589–0.635)***
sMS score	0.585 (0.564–0.605)***	0.590 (0.567–0.612)***	0.594 (0.577–0.611)***	0.564 (0.533–0.595)***	0.527 (0.508–0.546)**	0.615 (0.592–0.638)***

*P < 0.05; **P < 0.01; ***P < 0.001 (P-values are based on the null hypothesis that AUROC = 0.5 (no predictive discrimination), using the Mann–Whitney U test. 95% confidence intervals were calculated using the method of Hanley & McNeil.)

Discussion

The primary finding of our study was that the full GCS achieved slightly higher AUROC values than did the GCSE, GCSM, and sMS in predicting 3-day mortality, 7-day mortality, in-hospital mortality, ICU LOS ≥ 14 days, and hospital LOS ≥ 30 days in adult patients with trauma, including specific subgroups such as older adults, patients with TBI, and those with CVD. The GCSE consistently demonstrated the poorest performance in predicting all clinical outcomes. Although the sMS exhibited discriminative ability similar to that of the GCSM, it stands out for its speed and ease of use.

Consistent with the results of other studies, the GCS had slightly higher accuracy for predicting all clinical outcomes, particularly short-term mortality, compared with the GCSE, GCSM, and sMS [29–33]. However, its advantage was found to be smaller when predicting long-term mortality and prolonged LOS. Notably, the GCS also has key limitations, including moderate interrater reliability, challenges in calculating scores for specific patients, and the potential for disparate eye, motor, and verbal profiles to yield identical total scores [34].For instance, a GCS score of 4 corresponded to mortality rates of 48%, 27%, and 19% when calculated as E1V1M2, E1V2M1,and E2V1M1, respectively [35].

The 2021 National Guideline for the Field Triage of Injured Patients recommends replacing the criterion of a full GCS score of ≤ 13 with “Unable to follow commands (motor GCS < 6)” [36]. In a meta-analysis of 18 head-to-head studies, the GCS demonstrated slightly greater discriminative ability relative to the GCSM and sMS, but the difference between the AUROC values for the GCS and GCSM was minimal. This suggests that the clinical benefit of using the GCS over the GCSM is limited, particularly given the simplicity and practicality of the dichotomized cutoff point used in field triage [31]. The findings of that meta-analysis align with ours, indicating only a marginal difference in discriminative ability between the GCS, GCSM, and sMS. Therefore, the full GCS should be replaced with the GCSM. Although this would simplify the assessment process, the GCSM is nonetheless the most challenging of the three GCS components to assess [37].

Because geriatric patients often have low physiological reserves and multiple neurological injuries, predictions based on a single component of the GCS are less reliable for these populations. Geriatric patients, in particular, are more likely to present with non-motor-only deficits compared with younger adults. This raises concerns about using the GCSM as the new triage criterion because it may increase the risk of undertriage, particularly in the aforementioned populations [38]. Studies on triage systems—such as the prehospital shock index multiplied by the AVPU score, the reverse shock index multiplied by the sMS score, and the reverse shock index multiplied by the GCS score—have highlighted their lower predictive ability for geriatric patients and patients with preexisting chronic conditions or medications, particularly those with cardiovascular dysfunction [21, 39–42]. Our study supports these findings, revealing lower predictive accuracy for all scoring systems. Notably, the GCSE exhibited the greatest decline in accuracy, whereas the declines in the accuracies of the GCS, GCSM, and sMS scores were comparable.

Interestingly, contrary to previous studies reporting lower predictive ability of brain function–based triage scores (such as AVPU, sMS, and GCS combined with shock indices) in TBI populations, our findings revealed that all four brain function assessment tools—full GCS, GCSE, GCSM, and sMS—demonstrated consistently higher AUROC values for predicting clinical outcomes in the TBI group compared to the non-TBI group. This discrepancy may stem from the fact that these tools were originally developed to evaluate brain function and are therefore more directly applicable to TBI patients, in whom consciousness changes are primarily due to neurological injury. In contrast, altered mental status in non-TBI patients often results from systemic causes such as hypovolemic or obstructive shock, where impaired consciousness is not a direct reflection of neurological injury. Therefore, the predictive performance of brain function assessment tools may be inherently limited in non-TBI populations. Our findings highlight the importance of matching the physiological basis of scoring systems to the patient population in which they are applied and further support the rationale for combining neurological assessments with physiological indicators such as the shock index, particularly in non-TBI populations.

In the dynamic prehospital setting, distinguishing between abnormal flexion (M3) and extension (M2) can be challenging. This difficulty is one of the key factors driving our comparison of the GCSM and sMS. Studies have reported higher mortality in patients with lower GCSM scores (M1–4), which indicate more severe conditions, compared with those with GCSM scores of > 4, regardless of whether the assessment was conducted in the field or upon hospital admission [43]). Mortality rates exceed 50% when the GCSM score is < 3 [44]). Therefore, categorizing the higher mortality group (M1–4) into a single severity group in the sMS enables faster triage compared with performing the full GCS or GCSM assessment. In the full GCS assessment, additional time is needed to evaluate the less strongly predictive E component. Although the GCSM has similar accuracy to the sMS, it is more challenging to assess, which could lead to a longer on-scene duration and increase the error rate, particularly when it is used by an inexperienced emergency medical technician. In settings where shorter assessment times, improved accuracy, and reproducibility are prioritized, use of the most efficient tool, the sMS, appears to be the optimal choice. Although its AUROC is slightly lower compared to GCS, it is faster and easier to use in the field. However, there is no doubt that despite requiring more time for assessment, the GCS offers slightly higher accuracy in measuring brain function, which is particularly important for patients with head trauma. Its multi-dimensional evaluation of brain function also provides neurosurgeons with more comprehensive information. Therefore, assessment tools should be selected based on the appropriateness of the setting—with rapid, simplified tools favored in the prehospital phase, and more detailed assessments like GCS reserved for the in-hospital phase—to facilitate optimal patient management.

The present study has several strengths. First, in accordance with the new field triage guidelines for predicting trauma outcomes [36], we validated the clinical utility of the sMS, a rapid field triage scoring system, and compared it with the GCS, GCSE, and GCSM. This is clinically significant because the sMS can be easily implemented in prehospital or emergency department settings, facilitating the timely assessment of neurological status in patients with trauma. It enables emergency crews, even those with limited experience, to accurately evaluate patients. Second, we conducted extensive subgroup analyses to assess the accuracy of the sMS in predicting five major clinical outcomes. These analyses confirmed that the sMS functions as a rapid prognostic discriminator with predictive accuracy comparable to that of the GCS and GCSM while outperforming the GCSE, particularly in geriatric patients, patients with TBI, and patients with CVD.

The present study has several limitations. First, our results may not be fully generalizable to all patients with trauma in different emergency departments. This is because our cohort included a high proportion of geriatric patients, those with TBI, and those with a high prevalence of comorbidities. These factors could influence the effectiveness of triage tools in predicting mortality and other outcomes. However, our data reflect the real-world scenario in which an aging population and increasing comorbidities affect the predictive accuracy of triage scoring systems. These systems must account not only for diseases but also for the complexities of aging. Second, data collected at a single time point may not fully capture a patient’s overall condition or the severity of their injury. Among patients with trauma, sMS scores can vary, influenced by disease progression and factors such as alcohol consumption and drug use. These factors could result in overestimation or underestimation of clinical outcomes. Additionally, because of varying injury times, the sMS may be used in practice at different time points. Instead of relying on time-consuming assessments conducted at a single time point, emergency medical technicians could make multiple quick assessments at key points of clinical change to obtain a more accurate reflection of a patient’s condition and improve the accuracy of outcome predictions.

There is an inherent risk of bias in retrospective cohort studies, particularly due to missing data. In our study, most of the 2,366 patients who were excluded had missing values specifically for one or more of the brain function related variables (i.e., GCS, GCSE, GCSM, or sMS), rather than for demographic or other physiological parameters such as age or sex. Given that these neurological scores were the primary variables under investigation, we considered it inappropriate to impute these missing values, as doing so could introduce substantial bias and compromise the validity of our findings. Therefore, we further compared the basic characteristics of the included sample and the sample of individuals excluded due to missing data and found differences in patients’ demography and outcome in Supplementary Table 2. Although the excluded cases accounted for only 5.7% of the total sample, we acknowledge that our analysis may be susceptible to potential bias arising from differences in baseline characteristics between included and excluded patients, as well as from the missing data inherent to the retrospective study design.

Another key limitation is that our GCS, GCSM, GCSE, and sMS scores were derived from injury assessment records made upon the patient’s arrival at the emergency department instead of from data collected at the injury scene by emergency medical technicians. The most commonly used on-scene brain function assessment tool in our database is the AVPU scale and not the GCS, making it impossible to calculate GCSM and sMS scores for comparison of the prehospital versus in-hospital setting. Furthermore, prehospital on-scene assessments of brain function tend to be more susceptible to the influences of environmental factors and time pressure, which may affect accuracy. Therefore, we used the brain function data collected in emergency departments for comparison with existing predictive scores. In Taiwan, the prehospital transport time is typically < 20 min. Specifically, in Taipei, the median transport interval is 7 min, and the median prehospital interval is 23 min; these intervals are considerably shorter than those reported in other countries [45]. We believe that the brain function data collected in Taiwanese emergency departments are similar to those collected through prehospital on-scene assessments.

Finally, the low prevalence of the outcomes in our study could have contributed to low AUROC values and other validation metrics, particularly in the subgroup analyses. Given the relatively low event rates within the cohort, with only 3.1% experiencing in-hospital mortality as the primary outcome, the accuracy of the triage tools was likely influenced by their high specificity and negative predictive value. Therefore, a prospective clinical trial will be essential to further validate their clinical utility.

Conclusion

Although the full GCS assessment demonstrated slightly higher accuracy in predicting 3-day mortality, 7-day mortality, in-hospital mortality, ICU length of stay ≥ 14 days, and hospital length of stay ≥ 30 days compared to the GCSE, GCSM, and sMS, rapid and simplified tools such as the sMS are more appropriate for use in the prehospital setting to reduce assessment time. In contrast, during the in-hospital phase or for patients with traumatic brain injury, the GCS allows for a more comprehensive evaluation of the patient’s neurological status, supporting more informed clinical decision-making.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Author contributions

Shu-Jui Lee, Yu-Long Chen, Giou-Teng Yiang, and Meng-Yu Wu wrote the main manuscript text and Tsung-Hsien Wu, Chi-Yuan Liu, Chien-Hsing Wang, Chia-Hung Tsai, and Jui-Yuan Chung prepared figure and tables. All authors reviewed the manuscript.”

Funding

This study was supported by the grant of Taipei Tzu Chi Hospital (TCRD-TPE-113-RT-4, TCRD-TPE-114-38, TCRD-TPE-113-45, TCRD-TPE-112-39).

Data availability

The data used or analyzed in this study are available from the corresponding author upon reasonable request.

Declarations

Ethics approval and consent to participate

This study was approved by the Institutional Review Board of Taipei Tzu Chi Hospital and was conducted in accordance with the Declaration of Helsinki (approval number: 12-XD-077). This was an observational study that did not involve additional interventions in the usual care of patients. Therefore, the requirement for informed consent was waived by the Institutional Review Board of Taipei Tzu Chi Hospital.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.