Injuries among children and adolescents represent a significant global health challenge, with more than 1,600 individuals under the age of nineteen dying daily due to trauma. This burden is particularly pronounced in low- and middle-income countries, where both the incidence and impact of injuries exceed those observed in high-income regions (1, 2). Among individuals aged 5–29, 3 of the 5 leading causes of death are injury-related, including traffic accidents, falls, homicides, and suicides (3). While unintentional injuries account for the majority of these deaths globally (4), Colombia presents a distinct trauma epidemiology. In this country, interpersonal violence emerges as the leading cause of death starting at age 10 and ranks as the fourth leading cause from age 5 onward (5).

Both the World Health Organization (WHO) (6) and the Centers for Disease Control and Prevention (CDC) (7) recognize injuries as the number one killer of young people, underscoring the urgent need for effective strategies to mitigate the toll of trauma on this vulnerable population. Beyond fatalities, tens of millions of individuals worldwide suffer non-fatal injuries each year, resulting in long-term healthcare needs, rehabilitation, and substantial economic impacts due to premature death and disability (8–10).

To address this burden, international trauma care authorities emphasize strategies to improve trauma care quality and outcomes. Survival prediction models play a central role in these efforts, providing tools to estimate survival probabilities and guide clinical decision-making (11). However, while numerous prediction tools have been developed and validated for adult trauma populations, similar tools specifically tailored to pediatric trauma remain scarce. This gap is particularly concerning, given the distinct physiological and epidemiological characteristics of pediatric patients, which differ significantly from those of adults. Accurate predictive models for pediatric trauma are essential to guide clinical decision-making, improve outcomes, and address the unique needs of children.

The Trauma and Injury Severity Score (TRISS) is one of the most widely used tools in trauma centers for assessing patient survival likelihood, enabling healthcare providers to identify high-risk patients and prioritize care (12–16). Additionally, TRISS supports trauma center quality improvement by comparing observed vs. expected outcomes, facilitating benchmarking, and identifying gaps in care (17, 18). This approach aligns with the WHO's Guidelines for Trauma Quality Improvement Programmes, which advocate for strategies to reduce the burden of trauma worldwide (19).

However, TRISS was originally developed for adult populations, and its predictive accuracy is limited in pediatric patients (20, 21). This limitation arises from the model's use of age as a binary variable, which overlooks the nuanced impact of age on survival in children (22, 23). Studies have shown that modifying TRISS to treat age as a continuous variable or recalculating its coefficients can significantly enhance its predictive power for pediatric populations (24–26).

Despite these advancements, pediatric-specific adaptations of TRISS remain limited, as most survival prediction models have been optimized for adults rather than children. This issue is particularly relevant in Colombia, where the unique trauma epidemiology—characterized by high rates of violence and other trauma-related causes—emphasizes the need to test and adapt predictive tools for pediatric populations in Latin America.

In this study, we aimed to adapt TRISS to account for pediatric age as a continuous variable (Peds-TRISS) and evaluate its performance among children treated at our trauma center in Colombia. Rather than creating a new model, we focused on refining the existing TRISS framework to improve its applicability to pediatric trauma, ultimately contributing to evidence to improve quality of care and outcomes in this vulnerable population.

A total of 1,013 children were enrolled in the study, with a complete data analysis performed after excluding 34 patients due to missing Glasgow Coma Scale information (Figure 1). The overall percentage of missing data was low (3.2%), and no imputation was conducted. Sociodemographic and clinical characteristics are summarized in Table 1 and were published in another article with a slightly larger sample that did not require the exclusion of the mentioned patients (46). The median age of participants was 12 years (IQR, 5–15 years), 73% being male. Age distribution was as follows: 21.2% were aged 0–4 years, 21.8% were aged 5–9 years, 23.9% were aged 10–14 years, and 33.1% were aged 15–17 years. Most patients (66.2%) sustained blunt trauma, while 33.8% experienced penetrating trauma. The subsidized health insurance system provided coverage for 32.9% of the patients, whereas 67.1% were insured under the contributory system, which includes private insurance. The leading causes of injury were traffic accidents (31.1%), falls (28.8%), and assaults (28.7%) (Figure 2), with firearms responsible for 97% of the violence-related deaths. Thirty-two percent of patients presented with severe trauma (ISS ≥16).

A bivariate analysis of characteristics by survival status at discharge is presented in Table 1. The median age of deceased patients was 15 years (IQR, 13–17 years), compared to 11 years (IQR, 5–15 years) in survivors (p < 0.001). Most deaths (83%) occurred among patients aged 10–17 years, with males representing 90% of this group. Among the deceased, 60.3% were covered by subsidized health insurance. Violence was the leading cause of death, followed by traffic accidents (61.3% and 33.8%, respectively). The median ISS in deceased patients was 31 (IQR, 25–37), compared to a median ISS of 9 (IQR, 4–16) in survivors. The most frequent fatal injury was traumatic brain injury (91.3%). The overall mortality rate was 5.7%, with deaths occurring at a median of 15 h post-injury (IQR, 3.8–49 h), and no deaths occurred beyond 20 days of hospitalization.

The coefficients of the Peds-TRISS are presented in Table 2. In terms of performance, a comparison between the Peds-TRISS and the original TRISS was conducted. Both demonstrated strong discrimination (Table 3). The area under the receiver-operating characteristic curve (AUROC) was 0.971 (95% CI, 0.945–0.996) for the original TRISS and 0.984 (95% CI, 0.961–0.993) for the Peds-TRISS. The DeLong test revealed no statistically significant difference in AUROCs (p > 0.05) (Figure 3). However, calibration was satisfactory only in the Peds-TRISS model, as indicated by the Hosmer–Lemeshow statistic (HL = 9.7, p = 0.3 for the Peds-TRISS and HL = 16.6, p = 0.03 for the original TRISS). Additionally, the Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC) were lower for the Peds-TRISS (AIC = 187, BIC = 197) compared to the original TRISS (AIC = 228, BIC = 238), supporting its selection based on model fit. An additional analysis of Peds-TRISS performance by age subgroup was conducted, demonstrating robust performance of the score across all age groups, as measured by discrimination and calibration (Table 4).

The internal validation of the Peds-TRISS demonstrated strong performance across multiple metrics, including discrimination, calibration, and model fit, as shown in Figure 4.

For the benchmarking assessment, the M statistic was 0.93, and the Z statistic was 0, indicating no statistically significant difference in survival rates between our patient cohort and the expected outcomes as defined by the comparison standard.

This study assessed the performance of Peds-TRISS, an age-adjusted TRISS-based survival prediction model in a pediatric population in Colombia, using trauma center data from patients under 18 years old collected between 2011 and 2019. A significant proportion of deaths within this cohort were due to violent trauma mechanisms, and the Peds-TRISS performance was compared with that of the original TRISS. Our findings differ from the global literature, which predominantly reports trauma-related deaths due to unintentional injuries (4, 8), but align with studies conducted in Latin America (47, 48). For example, research from Brazil highlights that intentional trauma, particularly homicides, plays a significant role in adolescent mortality (49). Similarly, data from Mexico identify assaults as a major cause of trauma among adolescents, emphasizing the influence of violence on mortality patterns (50). Consistent with these findings, the SALURBAL study reports that homicides among adolescents and young adults represent a serious public health problem in Latin America, reinforcing the region's designation as one of the most violent in the world (44).

In Colombia, injuries constitute the first three causes of death from age 10, with homicide leading the list (5). This reflects a public health problem of significant magnitude, requiring substantial investment of human, financial, technical, and technological resources across the continuum of care—from pre-hospital services to long-term rehabilitation. Additionally, the emotional, social, and economic implications are profound, particularly due to the loss of productive years through premature death or disability (51–53). This shared regional burden underscores the need for effective interventions and predictive tools that are tailored to the specific epidemiological patterns observed in low- and middle-income countries (LMICs).

The implementation of survival prediction tools in pediatric trauma is strongly recommended by international authorities, such as the American College of Surgeons Committee on Trauma (ACS COT) in Advanced Trauma Life Support and the National Institute for Health and Care Excellence (NICE) in the United Kingdom (54, 55). These organizations recognize the utility of applying these scores for effective trauma system functioning. One of the most commonly used survival prediction tools in trauma research is TRISS, which has critical implications for patient management and trauma center quality assessment (12–16, 20) but has been less evaluated and documented in pediatric patients. TRISS incorporates anatomical and physiological variables, with the latter playing a vital role in children's adaptive response to trauma (30, 56–58). However, the parameters used (respiratory rate, systolic blood pressure) are based on findings from adult populations, limiting their reliability in pediatric settings, as these values vary significantly with age.

Our evaluation of the original TRISS model in our pediatric cohort of patients revealed limitations in its performance. Previous studies have identified similar constraints, noting increased predictive power with modifications to variables like age, such as expanding age group categorizations, recalculating coefficients, or treating age as a continuous quantitative variable rather than a binary category (20, 24–26). Despite these limitations, these studies acknowledge TRISS's utility as a tool for evaluating trauma care quality improvement systems, making it the most widely used scale in injury survival studies. Additionally, a systematic review of trauma mortality prediction models underscores the importance of including demographic predictors, ideally quantitative, to enhance research quality and model performance (11).

Consistent with previous studies, we aimed to enhance the TRISS model by incorporating age as a continuous variable and recalculating all model coefficients. Our analysis revealed that the ROC curve for the Peds-TRISS model was nearly identical to that of the original TRISS, indicating that the age adjustment contributes minimally and without statistical significance to its discriminatory power. However, the adjusted model showed superior performance in terms of calibration, suggesting that the age-adjusted version aligns predicted and observed outcomes more accurately across all risk levels. This underscores the potential of tailored modifications to refine predictive models for specific populations. Furthermore, earlier studies have reported improvements in model performance following adjustments to variables such as age (20, 24, 25), highlighting the value of these modifications in enhancing predictive tools for specific clinical and epidemiological contexts.

The development and refinement of models like Peds-TRISS emphasize the importance of leveraging innovative tools to enhance trauma care, particularly in regions facing significant public health challenges. Recent advancements in artificial intelligence (AI) provide a promising avenue for further improving trauma prediction models, enabling real-time, personalized outcome predictions. By integrating large-scale data—such as imaging, vital signs, and patient history—AI-driven systems could optimize triage, refine risk stratification, and support clinical decision-making. However, for AI to achieve its full potential in pediatric trauma, it is crucial to adapt these technologies to the unique physiological characteristics of children and ensure access to high-quality data. Future research integrating AI into models like Peds-TRISS could facilitate more precise and actionable insights for trauma care in diverse clinical settings (59).

In addition to refining predictive models, comparative evaluation of patient outcomes provides a valuable measure of trauma center care quality (17–19). Comparative statistics enable the assessment of observed survival rates for injured patients at a trauma center against expected survival rates from national or international reference centers. The literature reports varied results (41–43); however, our findings showed no statistically significant differences between our patients and the MTOS group (Z = 0), with statistically valid comparability between both groups (M = 0.93). Integrating advancements such as AI into these frameworks could further enhance their utility, enabling real-time, personalized predictions while facilitating inter-hospital comparability.

We believe that both tools, survival prediction and outcome comparison, are valuable in clinical practice. The Trauma Audit and Research Network (TARN), the largest trauma registry in Europe, employs survival probability for prediction and comparative outcome analysis through DEF statistics (M, Z) for inter-hospital comparability (60). These tools are invaluable for objectively addressing unfavorable outcomes, facilitating the identification of contributing factors to unexpected results, and improving the quality of trauma center care. Internal validation of Peds-TRISS confirmed strong performance across all metrics (Figure 3).

Given the nature of our patients' injuries and Peds-TRISS's performance metrics, it could serve as a local reference or be applied in similar contexts, underscoring the value of developing new models based on regional data (61, 62). Moreover, its applicability extends beyond Colombia, offering potential value for regions with similar epidemiological challenges, including countries across Latin America where violence-related injuries remain a substantial burden.

The critical burden of violence-related injuries in our region, disproportionately affecting children and adolescents, underscores the urgent need for tailored solutions in trauma care. Models like Peds-TRISS offer a starting point for addressing these challenges, providing a foundation for both clinical decision-making and broader public health strategies. Healthcare professionals and researchers must actively engage in developing and disseminating evidence-based tools that identify gaps, inform interventions, and ultimately reduce the societal impact of pediatric trauma. Collaborative efforts, supported by advancements in predictive technologies and an understanding of regional epidemiological patterns, are essential to creating sustainable solutions for improving outcomes in vulnerable populations.

In conclusion, this study revealed a high rate of violence-related deaths among adolescents, highlighting a critical public health challenge. While the discriminatory power of the Peds-TRISS model was nearly identical to that of the original TRISS, its superior calibration underscores its potential to more accurately align predicted and observed outcomes across varying levels of risk. Given the characteristics of our cohort and the results obtained, the model appears promising for similar contexts in Latin America. Further validation through prospective, multicenter studies is recommended to strengthen the evidence supporting its use and to inform the development of tailored and effective pediatric trauma care interventions