A total of 124 patients with median age of 45.00 were enrolled. Among them, 29 (23.40%) was female, and 95 (76.60%) were male. The injury mechanisms included car accidents (76/61.30%), fall from heights (31/25.00%), and other causes (17/13.70%), respectively. The hospital mortality was 26.7% (33/124). Compared with survivors, non-survivors demonstrated higher levels of APTT, PT, INR, Scr, and BUN, and lower levels of admission temperature, MAP, Fib, and PLT, which indicated a more severe coagulopathy combined with unstable vital signs. A greater incidence of TIC (63.60% vs. 19.80%, P < 0.001) and platelet transfusion (33.30% vs. 8.80%, P = 0.001) were found in non-survivors than survivors, with higher chance of receiving vasoactive drug and CRRT treatment. Furthermore, non-survivors exhibited significantly higher ISS, APACHE II, and SOFA scores with lower GCS score (all P-value < 0.001) (Table 1).

Univariate logistic regression analyses revealed correlations between hospital mortality and risk factors (temperature ≤ 35℃, Lac, APTT, PT, INR, Scr, BUN, TIC, mechanical ventilation, vasoactive drugs, CRRT, platelet transfusion, ISS, APACHE II score, SOFA scores, severe coma, and severe injury) and protective factors (MAP, PLT, Fib, and GCS) (Table S1). After multivariate logistic regression (Forward LR), TIC (OR 4.238, 95% CI 1.46–12.28, P = 0.008) and BUN (OR 1.397, 95% CI 1.09–1.78, P = 0.008) were identified as independent risk factors, while GCS score (OR 0.720, 95% CI 0.61–0.85, P < 0.001) was identified as independent protective factor (Table S2). Accordingly, we constructed a prediction model based on TIC, BUN and GCS score (Table S2). The equation is as follows:

$$\left[logit (\text)= -1.444+1.444\times \text-0.329\times \text+0.334\times \text\right]$$

$$\left[\text=\frac}}\right]$$

A nomogram and calibration curve were constructed based on the prediction model (Fig. 2A, B). The calibration curve demonstrated satisfactory agreement between the predicted and actual probabilities. The Hosmer–Lemeshow test confirmed that the prediction model was well-fitted (χ2 = 9.8576, df = 8, P = 0.2752 > 0.05). As shown in DCA (Fig. 2C), the prediction model provided a greater clinical net benefit compared to both the “treat-all” and “treat-none” strategies across a wide range of threshold probabilities. Furthermore, the CIC (Fig. 2D) revealed a robust consistency between the high-risk patients identified by the model and those who experienced adverse outcomes. Overall, the prediction model demonstrated superior clinical net benefit and satisfactory clinical impact.

Comprehensive evaluation of the prediction model. A nomogram of the prediction model; B calibration curve of prediction model; C decision curve analysis (DCA) comparison of the prediction model with APACHE II and SOFA scores; D clinical impact curve (CIC) of the prediction model

ROC curve analysis was conducted for the prediction model and each predictive factor to assess their performance in predicting hospital mortality (Fig. 3A). The AUC for TIC in predicting hospital mortality was 0.719 (95% CI 0.626–0.812), with a sensitivity of 63.64%, specificity of 80.22%, and a Youden index of 0.4386. The AUC for GCS score at admission was 0.854 (95% CI 0.782–0.926), with a sensitivity of 78.79%, specificity of 83.52%, and a Youden index of 0.6230. The AUC for BUN at admission was 0.694 (95% CI 0.590–0.798), with a sensitivity of 84.85%, specificity of 48.35%, and a Youden index of 0.3320. The combined regression model incorporating TIC, GCS, and BUN for predicting hospital mortality yielded an AUC of 0.898 (95% CI 0.834–0.962), with a sensitivity of 90.91%, specificity of 80.22%, and a Youden index of 0.7113. The prediction model demonstrated a significantly higher AUC compared to TIC and BUN (P < 0.001), but no statistical difference in AUC was observed between the prediction model and GCS score (P = 0.152) (Table 2).

ROC curves of the prediction model and other factors. A ROC curve comparison of the prediction model with TIC, GCS, and BUN. B ROC curve comparison of the prediction model with APACHE II and SOFA scores

Additionally, ROC curve analysis was performed for APACHE II and SOFA scores to predict hospital mortality (Fig. 3B). The AUC for the SOFA score was 0.822 (95% CI 0.734–0.910), with a sensitivity of 75.76%, specificity of 81.32%, and a Youden index of 0.5708. The AUC for APACHE II score was 0.894 (95% CI 0.827–0.960) with a sensitivity of 81.82%, specificity of 86.81%, and a Youden index of 0.6863. When compared to traditional scoring systems, the AUC of the prediction model was not significantly different from that of the APACHE II score (P = 0.870), but it was higher than that of SOFA score (P = 0.037) (Table 3).