Deep learning-based survival outcomes of REBOA vs resuscitative thoracotomy in trauma: a nationwide cohort study in South Korea

Resuscitative endovascular balloon occlusion of the aorta (REBOA) has been used for trauma resuscitation and hemorrhage control, traditionally managed with resuscitative thoracotomy (RT). Despite growing evidence supporting REBOA, most studies have employed Kaplan–Meier (KM) analysis due to limited covariate data1. Therefore, using a nationwide, multi-center trauma cohort in South Korea, we aimed to compare survival between REBOA and RT, and evaluate whether a deep learning model could improve risk adjustment over KM by incorporating detailed clinical covariates. We utilized data from the Korean Trauma Data Bank, a nationwide registry capturing all consecutive trauma cases by integrating data from all 19 designated major trauma centers in South Korea from 1 January 2017 to 31 December 2022 (dataset No. K2024-01-2-03-29). This study was approved by the Institutional Review Board of Cheju Halla General Hospital (CHH-2024-L07), with a waiver of informed consent due to the secondary use of administrative data. Among 227 567 patients initially enrolled, 224 patients who received RT or REBOA were included (Supplemental Digital Content Table S1, available at: https://links.lww.com/JS9/G811). To compare survival outcomes, we constructed KM survival curves and developed a deep learning-based survival model following the TRIPOD + AI and TITAN 2025 guidelines on the use of AI in clinical research2. We employed neural multi-task logistic regression (N-MTLR), incorporating demographics and prehospital or in-hospital vital signs as covariates (Supplemental Digital Content Table S2, available at: https://links.lww.com/JS9/G811). Additionally, as a sensitivity analysis, we applied least absolute shrinkage and selection operator (LASSO) regularization to preselect informative covariates and retrained the N-MTLR model to assess the robustness of our survival estimates (Supplemental Digital Content Table S3, available at: https://links.lww.com/JS9/G811). This approach enabled multivariable-adjusted, nonlinear survival prediction and more accurate modeling of individual risk trajectories3. To enhance robustness, 95% confidence intervals (CIs) were estimated using a bootstrap resampling procedure. The predictive accuracy of KM and N-MTLR was evaluated using the integrated Brier score (IBS; Supplemental Digital Content Figure S1, available at: https://links.lww.com/JS9/G811), and interpretability was assessed with Shapley Additive exPlanations analysis (Supplemental Digital Content Figure S2, available at: https://links.lww.com/JS9/G811). Based on the better-performing model, subgroup analyses were conducted by (1) the presence of prehospital emergency care and (2) injury location (urban vs rural). Among the 224 patients, 162 underwent RT and 62 received REBOA mean age: 48.9 (SD, 3.6) years for RT; 51.0 (4.0) years for REBOA; Table 1. Mean survival probabilities were higher in the REBOA group than in the RT group by both KM 0.20 (95% CI, 0.10–0.31) versus 0.09 (0.05–0.14); log-rank P-value <0.05 and N-MTLR 0.20 (0.06–0.41) versus 0.17 (0.08–0.28); P-value <0.05. A similar pattern was observed in the N-MTLR model with LASSO regularization 0.29 (0.10–0.47) versus 0.13 (0.06–0.21); P-value <0.05; Table 1. Table 1 - Overall comparison Survival REBOA Resuscitative thoracotomy Log-rank P-value No. of deaths/no. of patients (%) 49/62 (79.03) 147/162 (90.74) Survival probability, mean (95% CI) Kaplan–Meier 0.20 (0.10–0.31) 0.09 (0.05–0.14) <0.05 N-MTLR* 0.20 (0.06–0.41) 0.17 (0.08–0.28) <0.05 N-MTLR** 0.29 (0.10–0.47) 0.13 (0.06–0.21) <0.05 Integrated Brier score Kaplan–Meier 0.048 0.085 N-MTLR* 0.035 0.039 *Covariates used in the N-MTLR model are as follows: sex (male or female), age (0–19, 20–29, 30–49, 40–49, 50–59, 60–69, 70–79, and ≥80 years), consciousness on arrival (yes or no), blood pressure on arrival (normal or abnormal), pulse on arrival (normal or abnormal), temperature on arrival (normal or abnormal), oxygen saturation on arrival (normal or abnormal), GCS score on arrival (mild, moderate, or severe), CCI (normal or abnormal), injury severity score (mild, moderate, severe, or critical), AIS severity (ordinal score from 1 to 6), AIS injury category (head/neck, thorax, abdomen/pelvis, extremities, or other), and transfusion (performed or not performed).*Covariates used in the N-MTLR model are as follows: sex (male or female), age (0–19, 20–29, 30–49, 40–49, 50–59, 60–69, 70–79, and ≥80 years), consciousness on arrival (yes or no), blood pressure on arrival (normal or abnormal), temperature on arrival (normal or abnormal), oxygen saturation on arrival (normal or abnormal), injury severity score (mild, moderate, severe, or critical), AIS severity (ordinal score from 1 to 6), and AIS injury category (head/neck, thorax, abdomen/pelvis, extremities, or other). Survival probabilities remained consistently higher in the REBOA group than the RT group throughout the follow-up period, as shown by both models (Fig. 1A). The IBS for N-MTLR was lower than that of KM (0.035 vs 0.048 for REBOA; 0.039 vs 0.085 for RT; Table 1), indicating greater accuracy for N-MTLR. Subgroup analyses showed higher survival with REBOA in patients with prehospital care 0.16 (0.07–0.23) vs 0.07 ([0.03–0.11) and without prehospital care 0.32 (0.16–0.47) vs 0.13 (0.06–0.20). Survival probabilities were similar across urban and rural settings (Table 2 and Figure 1B-1C). Figure 1.: Comparison of survival curves between REBOA and resuscitative thoracotomy using Kaplan–Meier and N-MTLR models*, including subgroup analysis.* (A) Overall survival curves estimated using Kaplan–Meier and N-MTLR models. (B) Subgroup analysis based on prehospital emergency care. (C) Subgroup analysis based on the location of injury (urban or rural). Shaded areas represent 95% confidence intervals. Table 2 - Subgroup comparison by the N-MTLR model* Subgroup Group 1 Group 2 Log-rank P-value Prehospital emergency care Yes NO No. of deaths/no. of patients (%) REBOA 36/42 (85.71) 13/20 (65.00) Resuscitative thoracotomy 98/106 (92.45) 49/56 (87.50) Survival probability, mean (95% CI) REBOA 0.16 (0.07–0.23) 0.32 (0.16–0.47) <0.05 Resuscitative thoracotomy 0.07 (0.03–0.11) 0.13 (0.06–0.20) 0.397 Location of injury Urban Rural No. of deaths/no. of patients (%) REBOA 18/22 (81.82) 25/33 (75.76) Resuscitative thoracotomy 85/90 (94.44) 35/42 (83.33) Survival probability, mean (95% CI) REBOA 0.23 (0.08–0.37) 0.23 (0.15–0.32) 0.564 Resuscitative thoracotomy 0.06 (0.03–0.11) 0.17 (0.08–0.28) 0.145 AIS, Abbreviated Injury Scale; CCI, Charlson Comorbidity Index; GCS, Glasgow Coma Scale; N-MTLR, neural multi-task logistic regression; REBOA, resuscitative endovascular balloon occlusion of the aorta.*Covariates used in the N-MTLR model are as follows: sex (male or female), age (0–19, 20–29, 30–49, 40–49, 50–59, 60–69, 70–79, and ≥80 years), consciousness on arrival (yes or no), blood pressure on arrival (normal or abnormal), pulse on arrival (normal or abnormal), temperature on arrival (normal or abnormal), oxygen saturation on arrival (normal or abnormal), GCS score on arrival (mild, moderate, or severe), CCI (normal or abnormal), injury severity score (mild, moderate, severe, or critical), AIS severity (ordinal score from 1 to 6), AIS injury category (head/neck, thorax, abdomen/pelvis, extremities, or other), and transfusion (performed or not performed). The higher survival rates observed following REBOA may reflect its less invasive nature compared to RT. REBOA enables rapid proximal aortic occlusion without requiring thoracic entry4, reducing the physiological burden and risks of infection or iatrogenic injury5. Interestingly, survival did not significantly differ across injury locations, suggesting consistent management of severe hemorrhage across regions, possibly due to standardized triage and transport protocols despite broader system disparities. Nonetheless, this study has limitations. As a retrospective observational study, unmeasured confounding may persist. The lack of standardized indications for REBOA and RT, coupled with reliance on clinician judgment, may have introduced selection bias and unobserved differences in baseline severity despite statistical adjustment. Furthermore, the recent introduction of a REBOA procedure code in South Korea may have led to underreporting of earlier cases6, resulting in a small sample size. Although the small sample size may limit the generalizability, our goal was not to develop a clinical predictive model but to use deep learning as a comparative tool to assess whether nonlinear adjustment improves survival estimation. Finally, because this study was conducted with the Korean trauma care system, which has distinct organizational characteristics, caution is warranted when applying these findings to other settings. Because no comparable external database exists, external validation was not feasible. To mitigate overfitting, we performed internal validation using repeated bootstrap resampling. Using a nationwide cohort from all-registered trauma centers in South Korea, we found that REBOA may be associated with higher survival, and a deep learning model could improve survival estimation over KM analysis. With proper training and clinical guidelines, REBOA implementation could further enhance survival in patients with traumatic hemorrhage. This novel framework, deep learning-based N-MTLR survival analysis, may contribute to improving survival prediction in emergency and surgical medicine, providing a potential direction for enhancing clinical decision-making and outcome forecasting in these fields.