Mortality differences among patients with trauma based on systolic blood pressure and pulse pressure: insights from a generalized additive model diagram

1. Introduction

Trauma is a leading cause of morbidity and mortality worldwide, demanding rapid and accurate clinical assessment for effective intervention. In the pre-hospital and early hospital phases, clinicians rely heavily on vital signs, with blood pressure as a key indicator of circulatory status and tissue perfusion. However, reliance on a single parameter, such as systolic blood pressure (SBP), is limited in its ability to reflect the complexities of hemodynamic changes, particularly when vascular resistance is altered.

To overcome these limitations, additional indices—such as diastolic blood pressure, mean arterial pressure, pulse pressure (PP) and shock index (SI)—have been explored to enhance risk stratification. Typically, SBP below 90 mmHg is indicative of hypotension, and a PP constituting less than 25% of SBP is considered abnormal [1]. Despite these established thresholds, the isolated use of these variables has not consistently provided optimal predictive accuracy for mortality, thereby contributing to variability in clinical decision-making.

Recognizing that the interplay between these hemodynamic parameters may offer superior prognostic insight, our study employs a generalized additive model (GAM). GAM provides a significant advantage over traditional linear models by capturing complex, nonlinear relationships and interactions between variables. This approach facilitates a more nuanced understanding of how SBP and PP jointly influence mortality risk, which may enhance early risk stratification and support tailored clinical interventions during the critical initial phase of trauma care.

The primary objective of this study is to compare the predictive performance of SBP, PP and SI for mortality in trauma patients. Additionally, we seek to elucidate the interaction between SBP and PP using a GAM, aiming to refine early risk assessment and guide management decisions in emergency settings.

2. Methods

3. Results

3.1 Baseline characteristics by survival

The study enrolled 62,680 patients with traumatic injuries during the study period (Fig. 1).

Fig. 1.Flowchart diagram of the study population. Final enrollment is assigned except for missing or incomplete data.

Table 1 presents the general characteristics and categorical variables according to survival status. The study cohort included 62,620 survivors and 60 deceased individuals. The median ages of the two groups were relatively similar, with no significant differences. This cohort had a notable male predominance, particularly among those who died. A significant portion of the cohort was transported to tertiary emergency medical centers (EMCs). Mortality rates were significantly associated with the Korean Triage and Acuity Scale (KTAS). Most deaths were due to non-intentional injuries, and there was no significant variation in mortality rates based on the specific mechanism of injury.

Table 1.Demographic and general characteristics of mortality among patients with trauma.

Name	Description	Alive (N = 62,620)	Death (N = 60)	Total (N = 62,680)	p
Age (yr)	Mean ± SD	53.5 ± 22.9	68.7 ± 17.7	53.5 ± 22.9	<0.001
Sex
	Female	26,989 (43.1%)	14 (23.7%)		0.004
	Male	35,631 (56.9%)	45 (76.3%)
Hospital level
	Regional EMC	19,287 (30.8%)	26 (43.3%)		0.049
	Tertiary EMC	43,333 (69.2%)	34 (56.7%)
KTAS
	Level 1	125 (0.2%)	10 (16.7%)		NA*
	Level 2	3006 (4.8%)	39 (65.0%)
	Level 3	20,602 (32.9%)	10 (16.7%)
	Level 4	37,008 (59.1%)	1 (1.7%)
	Level 5	1879 (3.0%)	0 (0.0%)
Emergency
	Yes	58,237 (93.0%)	60 (100.0%)		0.021
	No	4383 (7.0%)	0 (0.0%)
Intention
	Nonintentional	57,548 (91.9%)	57 (94.8%)		0.144
	Self-harm/Suicide	1753 (2.8%)	3 (5.2%)
	Assault	3256 (5.2%)	0 (0.0%)
	Others	63 (0.1%)	0 (0.0%)
Injury mechanism
	Traffic accident	17,659 (28.2%)	27 (45.0%)		NA*
	Falling	7327 (11.7%)	23 (38.3%)
	Slip down	22,292 (35.6%)	7 (11.7%)
	Collision	10,019 (16.0%)	2 (3.3%)
	Cut/Stab	4947 (7.9%)	1 (1.7%)
	Machine	376 (0.6%)	0 (0.0%)
Transport Interval	Mean ± SD	33.4 ± 17.0	39.6 ± 22.2	33.4 ± 17.0	0.047
Total GCS	Mean ± SD	14.6 ± 1.7	9.8 ± 5.0	14.5 ± 1.8	<0.001
ISS	Mean ± SD	9.4 ± 8.1	19.4 ± 22.2	9.4 ± 8.3	<0.001
SBP	Mean ± SD	139.4 ± 26.2	113.8 ± 40.3	139.4 ± 26.2	<0.001
DBP	Mean ± SD	80.8 ± 15.0	71.9 ± 24.2	80.8 ± 15.0	0.006
MAP	Mean ± SD	100.4 ± 16.7	85.9 ± 28.5	100.4 ± 16.8	<0.001
PP	Mean ± SD	58.7 ± 20.3	41.9 ± 23.2	58.7 ± 20.3	<0.001
PR	Mean ± SD	85.2 ± 16.0	101.9 ± 29.9	85.2 ± 16.1	<0.001
RR	Mean ± SD	18.9 ± 2.1	21.0 ± 4.8	18.9 ± 2.1	0.002
BT	Mean ± SD	36.7 ± 0.5	36.1 ± 1.0	36.7 ± 0.5	<0.001
SpO2	Mean ± SD	97.7 ± 2.2	94.2 ± 5.9	97.7 ± 2.2	<0.001
ROUTE
	Direct	62,019 (99.0%)	60 (100.0%)		1.000
	Transfer	573 (0.9%)	0 (0.0%)
	Outpatient referral	22 (0.0%)	0 (0.0%)
	Others	6 (0.0%)	0 (0.0%)
Transport interval, time interval from scene to emergency department (ED); GCS: Glasgow Coma Scale; ISS: Injury Severity Score; SBP: systolic blood pressure; DBP: Diastolic blood pressure; MAP: Mean arterial pressure; PP: Pulse pressure; PR: Pulse rate; RR: Respiratory rate; BT: Body temperature; SpO2: oxygen saturation; ROUTE: Visit route; SD: Standard deviation; EMC: Emergency medical center; KTAS: Korean Triage and Acuity Scale. *p-value could not be calculated using chi-square approximation due to small expected counts; exact test was applied.

In the deceased group, GCS, SBP, DBP (Diastolic blood pressure), MAP (Mean arterial pressure), PP, body temperature (BT) and oxygen saturation (SpO₂) values were lower, while the transport interval from the scene to the emergency room (aggregated as Transport Interval), the Injury Severity Score (ISS), PR (Pulse rate), and respiratory rate (RR) were higher.

3.4 Interaction analysis of SBP and PP

ROC analysis (Fig. 2) demonstrated that SI, PP, and SBP each exhibited moderate and comparable discriminative ability for mortality prediction, with area under the curve (AUC) values of 0.724 (95% CI, 0.653–0.796), 0.688 (95% CI, 0.630–0.746), and 0.691 (95% CI, 0.614–0.768), respectively (all p < 0.001). The optimal cutoff values for SI, PP, and SBP were 0.86, 34 mmHg, and 112 mmHg, respectively.

Fig. 2.ROC curve analysis of SBP, PP and SI. Comparison among the three models revealed no significant differences under DeLong’s test, even though each model individually demonstrates statistical significance. ROC: Receiver operating characteristic; SI: Shock index; PP: Pulse pressure; SBP: Systolic blood pressure; AUC: area under the curve. A diagonal dotted line presents the performance of a random classifier.

No significant differences were observed in DeLong’s test for SBP vs. PP (p = 0.8347), SBP vs. SI (p = 0.3077) or PP vs. SI (p = 0.4433).

Although SBP, PP and SI each showed statistical significance, no significant differences were observed between these variables. Because of the correlation between SBP and SI, a GAM analysis was performed with SBP and PP. GAM revealed a significant interaction between SBP and PP with adjusted confounding factors (p < 0.001, deviance explained = 21%).

An interaction diagram was constructed to identify the relationship between SBP and PP; the interaction diagram between SBP and PP demonstrated a bimodal association between SBP and mortality, and a concurrent increase in mortality risk with low PP (Fig. 3).

Fig. 3.Interaction on mortality diagram of SBP and PP performed by GAM. (A) Mortality over SBP shows a bimodal pattern. Mortality over PP shows a decrescendo pattern. The interaction between extremes of SBP and low PP dramatizes mortality. (B) The figure on the right shows added confidence intervals with ±1 standard error (SE). ED_SBP: Systolic blood pressure measured at the emergency department; ED_PP: Pulse pressure measured at the emergency department; GAM: Generalized additive model.

4. Discussion

Although all patients experiencing shock require rapid intervention, traumatic injuries tend to worsen rapidly. Studies have indicated that a significant proportion of deaths resulting from trauma with shock occur within the first 24 hours owing to hypovolemic shock [2, 3]. In cases of major trauma, the majority of fatalities occur within the initial hours post-injury, with hemorrhage responsible for 30%–40% of these deaths [4]. Early resuscitation and transfusion are vital in preventing mortality; however, hypotension is often not identified until the patient reaches a class III shock status, characterized by blood loss exceeding 30% [5, 6].

Current methods for predicting mortality in trauma vary widely, as no single hemodynamic index has been definitively identified as superior. Although SBP is generally considered a reliable indicator of cardiac performance, its correlation with cardiac output may be unreliable in conditions such as cardiogenic shock [7]. DBP, which reflects vascular tone, can provide additional insights into clinical outcomes. For instance, high DBP in pediatric patients with refractory septic shock has been associated with a better prognosis, potentially because it indicates a cardiogenic origin [8]. Similarly, a DBP >70 mmHg measured after resuscitation has been shown to be an independent predictor of survival [9]. However, DBP does not always correlate with overall diastolic perfusion pressure owing to variations in systemic vascular resistance and arteriolar critical closing pressure.

PP and SI further illustrate the challenges of hemodynamic assessment. PP, indicative of regional blood flow and arterial stiffness, is variably associated with mortality, with a narrow PP and high DBP often serving as early markers of blood loss and decreased venous capacitance [10]. In addition, while high PP values in the first 24 hours after extracorporeal cardiopulmonary resuscitation have been reported as independent predictors of survival [11], PP >48 mmHg in septic shock populations has been independently associated with 28-day mortality [12]. Moreover, a J-shaped relationship between PP and all-cause mortality was observed in patients with ischemic heart failure, with statistical significance reached only when SBP exceeded 110 mmHg [13]. Similarly, both a narrow PP and elevated SI have been linked to increased mortality in trauma patients requiring massive transfusion [14], and a novel PP/Heart rate ratio has demonstrated moderate predictive ability for transfusion needs [15]. Despite these findings, neither mean arterial pressure (MAP) nor PP fully accounts for age-related changes in blood pressure [16].

Although each of these hemodynamic variables shows promise for predicting mortality, their associations vary significantly with the underlying pathology. Understanding the complex interactions among these indices may provide a more robust framework for mortality prediction than relying on any single parameter alone. Our findings emphasize the importance of examining these interactions to capture the nuances of the shock state. By visualizing the interconnected nature of these variables, we gain insights that could lead to the development of a more accurate and comprehensive risk stratification model. Future research should integrate these interactions into predictive models to facilitate prompt recognition of severely ill patients and guide timely, effective interventions [17, 18, 19].

In Korea, where this study was based, the KTAS level and emergency status are determined at the triage level upon hospital arrival. In contrast, the selection of the transport hospital and transportation time are decided during the pre-hospital phase, led by the EMS. The KTAS scale begins by evaluating life-threatening conditions. Then, it assesses the patient’s main complaint while considering additional factors such as vital signs, consciousness level, pain severity, mechanism of injury, and blood sugar levels. The assigned levels are resuscitative, urgent, emergent, semi-emergency and non-emergent from Level 1 to Level 5 [20]. According to this study, patients who ultimately died were classified under emergency cases with low KTAS levels, indicating that the triage evaluation was, to some extent, appropriate. However, it was found that not only the transportation time was longer, but a higher proportion were transferred to regional hospitals rather than tertiary centers in the deceased group, suggesting that the pre-hospital evaluation of the transport phase was not appropriate. We believe that clarification of the relationship between hemodynamic indices and mortality would significantly enhance the accuracy of severity evaluation and hospital selection in the pre-hospital phase, as assessments are currently inadequate.

In contrast to previous investigations that focused on the predictive value of individual hemodynamic indices, our study reveals that relying on isolated parameters may be insufficient for accurate mortality prediction in trauma patients. While conventional ROC analysis showed that SBP, PP and SI each had moderate predictive value, no significant differences were observed when these indices were considered separately. Notably, our results demonstrated a significant correlation between SBP and SI, which led us to explore the interaction between SBP and PP using a GAM. This analysis uncovered a novel, nonlinear and bimodal relationship between SBP and mortality—particularly highlighting that low PP markedly amplifies mortality risk at both extremes of SBP. These findings suggest that an integrated approach combining SBP and PP provides a more refined understanding of hemodynamic alterations in trauma patients, thereby enhancing early risk stratification and informing more targeted clinical interventions.

This study has several limitations. First, it is a retrospective analysis based on a pre-existing database, which did not include detailed information regarding interventions or first-aid measures that could have influenced mortality outcomes. Additionally, specific mechanisms of trauma, such as penetrating, blunt or head injuries, which may have differential impacts on patient outcomes, were not considered. The analysis also did not account for the effects of comorbidities or pre-existing health conditions. Second, the scope of this study was restricted to traumatic shock. Additional research is needed to explore the applicability of these findings to other types of shock, including septic and cardiogenic shock. Since the data were exclusively derived from Seoul, Korea, further research is required to determine whether these results are generalizable to different populations and regions. Lastly, while the study emphasizes the importance of promptly recognizing shock, future investigations should evaluate whether the insights gained from these hemodynamic indices can inform the development of optimal resuscitation targets in broader clinical contexts.

5. Conclusions

The interaction effect of PP and SBP on mortality was confirmed using GAM. This consideration is crucial for optimizing patient care and ensuring effective management of emergency trauma situations. This finding would not only assist triage evaluation, but also significantly enhance pre-hospital evaluation at the transportation level.

Availability of data and materials

The data that support the findings of this study are available from the Seoul Golden Time Board, but restrictions apply to the availability of these data, which were used under license for the current study and are not publicly available. However, the data are available from the authors upon reasonable request and with the permission of the Seoul Golden Time Board.

Author contributions

KL and YS—drafted the work and were major contributors to the manuscript. YC and GC—collected the data from the Seoul Golden Time Registry. YS and TP—analyzed and interpreted the data regarding hemodynamic indices and mortality in patients with trauma. All authors contributed equally to all the other sections. All the authors have read and approved the final version of the manuscript.

Ethics approval and consent to participate

This study utilized data from the Seoul Golden Time Registry, a registry designed for quality control and enhancement of the emergency medical system in Seoul, South Korea. The registry collects and analyzes data to improve patient outcomes in emergency care. Given the nature of the study, a waiver for obtaining participants’ informed consent was granted by the Institutional Review Board of Kangdong Sacred Heart Hospital (IRB No. KANGDONG 2024-09-002).

Acknowledgment

This study was based on the Seoul Golden Time registry, thanks to participating investigators. Investigators (Hospital) of Seoul Golden Time emergency medical system: Han Joon Kim (The Catholic University of Korea Seoul St. Mary’s Hospital), Jae Hoon Oh (The Catholic University of Korea, Eunpyeong St. Mary’s Hospital), Jae Guk Kim (Hallym University Kangnam Sacred Heart Hospital), Eun ah Han (Yonsei University Gangnam Severance Hospital), Gyu Chong Cho (Hallym University Kangdong Sacred Heart Hospital), Dong Hyuk Shin (Kangbuk Samsung Medical Center), Sin Young Kim (Konkuk University Medical Center), Jong Seok Lee (Kyung Hee University Medical Center), Sung Joon Park (Korea University Guro hospital), Su Jin Kim (Korea University Anam hospital), Jung In Ko (National Medical center), Kwang Hyun Cho (Nowon Eulji Medical Center), Kyung Jun Song (SMG-SNU Boramae Medical Center), Won chul Cha (Samsung Medical Center), Seok Yong Ryu (Inje University Sanggye Paik Hospital), Ki Jeong Hong (Seoul National University Hospital), Seung Mok Ryoo (Asan Medical Center), Keun Hong Park (Seoul Medical Center), Sang Il Kim (Soonchunhyang University Hospital), Hyun Soo Chung (Yonsei University Severance Hospital), Yoon Hee Choi (Ewha Womans University Medical Center), Chul Han (Ewha Womans University Medical Center), Jun Young Hong (Chung Ang University Hospital), Hyunggoo Kang (Hanyang University Seoul Hospital), June Seob Byun (Hanil General Hospital).

Funding

This research received no external funding.

Conflict of interest

The authors declare no conflict of interest.