Development of a nomogram to predict in-hospital mortality of trauma patients in the ICU: an analysis of the MIMIC-IV database

Abstract

Treatment of patients with severe trauma remains challenging. This study aimed to identify risk factors for all-cause mortality in ICU trauma patients to construct a predictive model. 2205 trauma patients were selected from the MIMIC-IV database, and 49 ICU indicators were obtained. All trauma patients were divided into training and testing datasets in a ratio of 7:3. Standardized mean difference (SMD) were conducted to ensure no significant difference between the two datasets. Subsequently, the least absolute shrinkage and selection operator and multivariate logistic regression analyses were conducted to identify the core variables from all ICU indicators, followed by constructing and evaluating a nomogram model. The regression analyses selected hepatopathy, obesity, chloride, body temperature, white blood cell (WBC) count, and acute physiology score III (APS III) as core variables from the remaining indicators. Furthermore, the nomogram model showed that six core variables influenced the mortality of trauma patients. Additionally, the calibration curves, decision curve analysis, and area under the receiver operating characteristic curves (p > 0.05) all verified the good prediction performance of the model.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-026-38251-x.

Severe trauma refers to injuries that cause significant physiological disruptions and pose an immediate threat to life. These injuries can result from various mechanisms, including blunt force, penetrating trauma, or secondary injuries that arise from complex clinical processes¹. Epidemiological studies indicate that trauma is a leading cause of morbidity and mortality globally, with an estimated 4.4 million deaths related to trauma each year¹. With the establishment of advanced trauma systems, the overall mortality rate has improved; however, there has been no significant improvement in mortality from major trauma². The treatment and management of severely injured patients remain challenging, highlighting the need for systematic improvements in severe trauma care, covering aspects such as early triage, multidisciplinary collaboration, and technological innovation^3,4. Given the significant healthcare burden posed by severe trauma¹, research focused on improving management strategies and predictive modeling is essential for enhancing patient outcomes and refining clinical protocols.

In the intensive care unit (ICU), healthcare professionals utilize various indicators to assess the severity of a patient’s condition and predict mortality risk, particularly in cases of severe trauma. Common metrics include the Injury Severity Score (ISS), which quantifies overall injury severity, and the Glasgow Coma Scale (GCS), which evaluates the neurological status. Additionally, physiological parameters (e.g., blood pressure, heart rate) and laboratory values (e.g., serum lactate and hemoglobin levels) are crucial in this evaluation process. They can help guide clinical decisions by providing clear insights into a patient’s condition^5,6. However, these assessment tools often have limitations; they may overlook key factors, such as the patient’s age, existing comorbidities, and the dynamic responses of critically ill patients. These shortcomings underscore the pressing need for more comprehensive and adaptable assessment frameworks that can effectively capture the complexities of each patient’s unique situation^7,8.

The Medical Information Mart for Intensive Care (MIMIC-IV) database is a valuable resource for researchers, offering free access to de-identified health data from over 50,000 ICU admissions at the Beth Israel Deaconess Medical Center in Boston, Massachusetts, between 2008 and 2019⁹. This extensive dataset enables the exploration of outcomes related to various clinical variables in critically ill patients. Our study focused on identifying 2,205 trauma patients using ICD-9 and ICD-10 codes¹⁰. Utilizing LASSO regression and multivariable logistic regression analysis, we identified crucial clinical indicators associated with the risk of in-hospital mortality in this patient group. Furthermore, we developed a nomogram to estimate the risk of trauma based on these clinical variables, and employed calibration curves, decision, and ROC curves to evaluate the predictive accuracy of our model. These findings underscore the significance of specific ICU indicators in the management of trauma cases and aim to create a framework for improving prognosis in severely injured patients, ultimately guiding clinical practices to enhance patient outcomes.

Materials and methods

Patient selection

The MIMIC-IV (Version 2.0, https://physionet.org/content/mimiciv/2.0/) is a widely used public database containing exhaustive clinical information on ICU patients from 2008 to 2019, with all participants providing informed consent before admission. It is important to note that the use of this database was approved by the Massachusetts Institute of Technology (MIT) and BethIsrael Deaconess Medical Center (BIDMC). One author (Yiqian Zeng) completed the Collaborative Institutional Training Initiative (CITI) examination, received authorization to enter the MIMIC IV database, and obtained certification (ID: 54997518). The entire study was performed in accordance with the relevant guidelines and regulations.

In this study, trauma was defined as sudden tissue damage resulting from violence or an accident¹¹. Subsequently, as shown in Fig. 1, clinical data (ICU indicators) of 2205 ICU patients, diagnosed with trauma—based on ICD-9 (800 to 959) or ICD-10 (S00 to S99, T00 to T14, and T20 to T32) codes¹⁰—were extracted from the MIMIC-IV database by using to the following filtered criteria: (i) patients were between 18 and 89 years of age; (ii) only the first hospital admission was included for patients with multiple hospital or ICU admissions; (iii) patients with ICU stays of less than 1 day or more than 180 days were excluded; (iv) patients missing key information, such as age or sex were also excluded. Trauma types primarily included: traumatic brain injury (such as intracranial hemorrhage and cerebral contusion), spinal cord injury, thoracoabdominal visceral injury (such as cardiac, pulmonary, hepatic, and splenic ruptures), severe pelvic fractures, and other musculoskeletal trauma. The specific coding ranges were provided in Table S1. Moreover, in-hospital mortality was recorded for each patient, and all patients were divided into either the survival or the dead group.

Fig. 1.

The diagram illustrates the selection process of patients included.

Data acquisition

The indicators of trauma patients within 24 h of ICU admission, including baseline information, comorbidities, scoring systems, laboratory parameters, and vital signs, are shown in Table S2. Among these, laboratory parameters and vital signs were extracted from the first recorded values within 24 h of ICU admission (where multiple readings occurred, the initial value was used). If a specific indicator was not recorded within the first 24 h after ICU admission, it was considered a missing value. In the subsequent statistical analysis for that indicator, the corresponding patient’s record for that variable was excluded. Additionally, comorbidities (e.g., liver disease, obesity) were defined based on ICD-9/ICD-10 diagnostic codes, with specific coding ranges provided in Table S3. Subsequently, using the createDataPartition function from the caret package (v 6.0.93)¹², all 2205 patients were stratified and randomly sampled in a 7:3 ratio to form the training and test datasets. To evaluate and ensure the comparability of baseline characteristics between the training and test sets, the standardized mean difference (SMD) was calculated. An SMD < 0.1 indicates good balance for that variable between the groups.

Screening for core variables

In the training dataset, LASSO regression analysis was performed using the glmnet package (v 4.1.2), with the minimum lambda identified through ten-fold cross-validation, to select characteristic variables from all ICU indicators. The reported lambda value was log-transformed. Furthermore, based on the training dataset, a multivariate logistic model was constructed using all LASSO characteristic variables to further screen the core variables (p < 0.05) using the glmnet package (v 4.1.2)¹³. Meanwhile, the adjusted odds ratio (OR) and 95% confidence interval (CI) for all LASSO-selected variables were computed to assess their relationship with trauma patients’ mortality, where OR > 1 indicated a risk factor and OR < 1 indicated a protective factor.

Construction and evaluation of a nomogram based on core variables

To further explore the influence of core variables on trauma patient mortality, a nomogram model was constructed based on all core variables using the rms package (v 6.5.0)¹⁴. In this nomogram model, a higher total score—calculated as the sum of points for each core variable—corresponded to a higher probability of death in trauma patients. Subsequently, calibration curves, decision curve analysis (DCA), and receiver operating characteristic (ROC) curves were plotted for both training and testing datasets to evaluate the predictive performance of the nomogram model. Among them, calibration curves were generated with 1,000 bootstrap resamples using the calibration function from the rms package (v 6.5.0). The Hosmer-Lemeshow goodness-of-fit test yielded a p-value > 0.05, indicating that the predicted probabilities did not significantly deviate from the observed outcomes and that the model was well-calibrated; DCA curves were plotted using the ggDCA package (v 1.2)¹⁵, and ROC curves were drawn using the pROC package (v 1.18.0)¹⁶, where an area under the curve (AUC) greater than 0.7 suggested the good prediction accuracy of the nomogram model. In addition, the Hosmer-Lemeshow test was employed to assess the prediction performance of the nomogram model with a p-value greater than 0.05.

Statistical analysis

All analyses were performed utilizing the R programming language (v 4.2.2)¹⁷. Statistical analysis was conducted using the SMD. If SMD < 0.1, intergroup balance was considered satisfactory. Categorical variables are expressed as weighted percentages, while continuous variables are presented as weighted means or medians and interquartile ranges (IQR).

Results

Six ICU indicators were selected as core variables for trauma patients

Initially, a total of 2205 trauma patients were divided into the training (1544 patients) and testing (661 patients) datasets at a ratio of 7:3. Subsequently, based on the results of the SMD, most ICU indicators showed no significant differences between the training and testing datasets (SMD < 0.1). Although there was a slight imbalance in ICU length of stay (LOS) between the two datasets (SMD = 0.124), this imbalance had minimal impact on model performance as LOS was not included as a predictor variable. Detailed results were presented in Table 1. These findings suggested that all remaining indicators were randomly distributed in both the training and testing datasets and were therefore selected for further analyses. A total of 31 ICU indicators were identified as LASSO characteristic variables based on log(lambda.min) = − 5.324272 (Fig. 2A,B). These variables were then subjected to multivariate logistic regression analysis, and those significantly associated with the outcome of interest were selected. The model equation was as follows: logit(P) = − 3.0941 + 0.3057 × Sepsis + … + 0.1123 × Charlson_index. After meticulous screening, only six of the 31 variables met the stringent criteria. The results of the multivariable logistic regression model (including regression coefficients, standard errors, etc.) were detailed in Table S4. Specifically, hepatopathy showed a strong association [OR(95% CI) = 3.122 (2.093 to 4.651), p < 0.001], identifying it as a prominent risk factor for the death of trauma patients. In contrast, obesity showed an inverse relationship [OR(95% CI) = 0.379 (0.195 to 0.705), p = 0.013], making it a protective factor. Additionally, chloride [OR(95% CI) = 0.980 (0.960 to 1), p = 0.005], body temperature [OR(95% CI) = 0.887 (0.764 to 1.028), p < 0.001], WBC [OR(95% CI) = 1.002 (0.981 to 1.022), p = 0.005] and APSIII [OR(95% CI) = 1.010 (0.999 to 1.021), p = 0.008] also passed the multivariate logistic regression analysis (Table 2). These six variables were then designated and renamed as core variables for trauma patients, laying a solid foundation for subsequent research and clinical decision-making in the ICU setting.

Table 1.

Baseline characteristics of trauma patients in the training and validation datasets.

	Level	Training set	Validation set	SMD
n		1544	661
OS (%)	0	1334 (86.4)	577 (87.3)	0.026
	1	210 (13.6)	84 (12.7)
Sepsis (%)	No	1256 (81.3)	538 (81.4)	0.001
	Yes	288 (18.7)	123 (18.6)
Acute_kidney_failure (%)	No	1010 (65.4)	434 (65.7)	0.005
	Yes	534 (34.6)	227 (34.3)
LOS (median [IQR])		4.31 [2.12, 9.17]	5.10 [2.45, 11.18]	0.124
Age (median [IQR])		62.50 [46.00, 75.00]	62.00 [46.00, 76.00]	0.008
Gender (%)	F	528 (34.2)	220 (33.3)	0.019
	M	1016 (65.8)	441 (66.7)
Weight (median [IQR])		79.80 [67.47, 95.53]	80.40 [66.75, 96.40]	0.039
Height (median [IQR])		173.00 [164.75, 178.00]	173.00 [163.00, 178.00]	0.01
BMI (median [IQR])		26.93 [23.40, 31.53]	26.89 [23.45, 31.62]	0.041
Heart_failure (%)	No	1244 (80.6)	547 (82.8)	0.056
	Yes	300 (19.4)	114 (17.2)
Hypertension (%)	No	752 (48.7)	308 (46.6)	0.042
	Yes	792 (51.3)	353 (53.4)
Chronic_lung_disease (%)	No	1271 (82.3)	532 (80.5)	0.047
	Yes	273 (17.7)	129 (19.5)
Hepatopathy (%)	No	1356 (87.8)	580 (87.7)	0.002
	Yes	188 (12.2)	81 (12.3)
Rheumatoid_arthritis (%)	No	1520 (98.4)	654 (98.9)	0.044
	Yes	24 (1.6)	7 (1.1)
Obesity (%)	No	1411 (91.4)	604 (91.4)	< 0.001
	Yes	133 ( 8.6)	57 (8.6)
Diabetes (%)	No	1198 (77.6)	519 (78.5)	0.022
	Yes	346 (22.4)	142 (21.5)
Anemia (%)	No	1198 (77.6)	519 (78.5)	0.022
	Yes	346 (22.4)	142 (21.5)
Aniongap (median [IQR])		12.00 [11.00, 15.00]	13.00 [11.00, 15.00]	0.016
Creatinine (median [IQR])		0.80 [0.70, 1.10]	0.80 [0.60, 1.10]	0.08
Chloride (median [IQR])		103.00 [99.00, 106.00]	103.00 [99.00, 106.00]	0.014
Glucose (median [IQR])		114.00 [98.00, 136.00]	115.00 [97.00, 136.00]	0.002
Bicarbonate (median [IQR])		22.00 [19.00, 24.00]	21.00 [19.00, 24.00]	0.057
Hematocrit (median [IQR])		30.10 [25.20, 35.00]	29.50 [25.30, 35.10]	0.038
Hemoglobin (median [IQR])		10.20 [8.50, 11.80]	10.00 [8.50, 11.70]	0.05
Platelets (median [IQR])		166.00 [117.00, 219.00]	165.00 [115.00, 226.00]	0.031
Potassium (median [IQR])		3.90 [3.50, 4.20]	3.80 [3.50, 4.20]	0.078
PT (median [IQR])		12.70 [11.60, 14.20]	12.60 [11.60, 14.00]	0.041
PTT (median [IQR])		27.00 [24.48, 30.90]	27.20 [24.90, 30.70]	0.049
INR (median [IQR])		1.10 [1.00, 1.30]	1.10 [1.00, 1.30]	0.031
Sodium (median [IQR])		138.00 [135.00, 140.00]	138.00 [135.00, 140.00]	0.023
Calcium (median [IQR])		8.00 [7.50, 8.50]	8.00 [7.50, 8.50]	0.013
BUN (median [IQR])		15.00 [11.00, 23.00]	15.00 [10.00, 23.00]	0.053
WBC (median [IQR])		9.40 [7.00, 12.30]	9.40 [6.80, 12.30]	0.049
RBC (median [IQR])		3.57 [3.04, 4.16]	3.59 [3.05, 4.10]	0.028
RDW (median [IQR])		14.10 [13.30, 15.50]	14.10 [13.20, 15.70]	0.024
MCV (median [IQR])		91.00 [88.00, 95.00]	91.00 [87.00, 95.00]	0.031
Heart_rate (median [IQR])		69.00 [59.00, 81.00]	69.00 [60.00, 82.00]	0.029
MBP (median [IQR])		59.00 [51.00, 66.00]	60.00 [52.00, 66.00]	0.043
Resp_rate (median [IQR])		12.00 [10.00, 14.00]	12.00 [10.00, 15.00]	0.085
SBP (median [IQR])		90.00 [79.00, 101.00]	91.00 [81.00, 100.00]	0.03
DBP (median [IQR])		46.00 [39.00, 53.00]	47.00 [40.00, 53.00]	0.031
SpO2 (median [IQR])		93.00 [91.00, 96.00]	93.00 [91.00, 96.00]	0.024
Temperature (median [IQR])		36.50 [36.06, 36.78]	36.50 [36.06, 36.78]	0.011
SAPSII (median [IQR])		34.00 [26.00, 44.00]	35.00 [26.00, 43.00]	0.043
SOFA (median [IQR])		4.00 [2.00, 7.00]	4.00 [2.00, 7.00]	0.022
APSIII (median [IQR])		43.00 [30.00, 57.00]	41.00 [32.00, 56.00]	0.04
OASIS (median [IQR])		33.00 [28.00, 39.00]	33.00 [28.00, 39.00]	0.004
GCS (median [IQR])		15.00 [13.00, 15.00]	15.00 [14.00, 15.00]	0.095
Charlson_index (median [IQR])		3.00 [1.00, 6.00]	3.00 [1.00, 6.00]	0.023

OS, Overall Survival; BMI, Body mass index; SBP, Systolic blood pressure; DBP, Diastolic blood pressure; SpO2, Peripheral oxygen saturation; GCS, Glasgow Coma Scale; APSIII, Acute Physiology Score III; SOFA, Sequential Organ Failure Assessment; WBC, White blood cell; PT, Prothrombin time; PTT, Partial Thromboplastin Time; INR, International normalized ratio; BUN, Blood urea nitrogen; LOS, Length of stay; RBC, Red Blood Cell (count); RDW, Red cell Distribution Width; MCV, Mean Corpuscular Volume; MBP, Mean Blood Pressure; SAPSII, Simplified Acute Physiology Score II; OASIS, Oxford Acute Severity of Illness Score.

Data are presented as mean ± standard deviation or median (interquartile range) for continuous variables, and as number (%) for categorical variables. Standardized mean difference (SMD) values < 0.1 indicate good balance between the training and validation datasets.

Fig. 2.

LASSO’s clinical feature selection. (A) The variation features of the LASSO regression coefficient of 31 variables, and each curve in the figure indicates the variation trace of the coefficient of each independent variable. (B) Partial likelihood deviance was plotted versus log(λ), and the dotted vertical lines were drawn at the optimal values using the minimum criteria and the 1-SE criteria.

Table 2.

Logistics regression model.

	OR	95% CI		p
Hepatopathy	3.1220	2.0926	4.6509	< 0.001
Obesity	0.3791	0.1948	0.7049	0.0128
Chloride	0.9801	0.9604	1.0003	0.0053
WBC	1.0017	0.9807	1.0222	0.0055
Body temperature	0.8869	0.7640	1.0279	< 0.001
APSIII	1.0098	0.9987	1.0209	0.0079

Six ICU indicators influencing mortality in patients with trauma

Based on the previous analyses, six ICU—hepatopathy, obesity, chloride, body temperature, WBC, and APS III—were selected and designated as the core variables. Identified as the most influential and relevant factors, these variables served as the foundation for constructing a nomogram model, illustrated in Fig. 3. Upon construction and exploration of the nomogram model, APSIII emerged as the dominant factor, exerting the most profound and substantial influence on the probability of death among trauma patients. To assess the predictive performance of the nomogram, a series of validation procedures was implemented. Calibration curves, representing the agreement between predicted and actual outcomes, showed that the training dataset closely approached the reference line with a p-value of 0.0638. These results suggested that the model had a relatively good predictive performance regarding trauma patients’ mortality, as depicted in Fig. 4A. This finding was further corroborated by the testing dataset, where the p-value was 0.186 (Fig. 4B). Next, in the training set, the calibration curve had an intercept close to 0 and a slope close to 1, with a Brier score of 0.103; in the test set, the calibration curve also showed an intercept close to 0 and a slope close to 1, with a Brier score of 0.1. These results indicated that the model’s predicted probabilities were well calibrated to the observed outcomes in both datasets. Remarkably, AUC values in both the training and testing datasets were greater than 0.7 (training dataset: AUC = 0.753 (95% CI 0.717–0.789); testing dataset: AUC = 0.715 (95% CI 0.653–0.778)). These results indicated that the nomogram model achieved a satisfactory level of accuracy in predicting trauma patients mortality, as illustrated in Fig. 5. In addition, the DCA curves for the training and testing datasets are shown in Fig. 6. These curves provided insight into the net benefit of using the model across different threshold probabilities. When the threshold probability was between 0.25 and 0.50, the model exhibited a higher net benefit compared to alternative scenarios involving the six ICU indicators, including those where all or none of the indicators were considered. Notably, among the six ICU indicators, APS III consistently demonstrated the greatest net benefit. This result reaffirmed the significant impact of APS III on the trauma patient mortality and further validated the nomogram model as a valuable tool for clinical decision-making and risk stratification in ICU trauma management.

Fig. 3.

A nomogram model for predicting in-hospital all-cause mortality in ICU trauma patients.

Fig. 4.

Analysis of calibration curves in the training (A) and validation (B) cohorts. The horizontal axis depicts the nomogram-predicted likelihood of in-hospital survival, whereas the vertical axis depicts actual in-hospital mortality.

Fig. 5.

Analysis of the receiver operating characteristic curves in the training (A) and validation (B) cohorts to predict in-hospital all-cause mortality.

Fig. 6.

Decision curve analysis for the training (A) and validation (B) cohorts, revealing the net advantage of using the nomogram.

Discussion

Severe trauma remains a major public health concern in the modern era¹¹. Its treatment necessitates the coordination of various medical disciplines, and the management of complex injuries under challenging circumstances, often hindering rescue efforts and contributing to elevated rates of disability and mortality^18–21. Prompt and effective management of multiple injuries is imperative to reduce mortality²². The ICU plays an instrumental role in facilitating collaborative care across various medical specialties, thus functioning as a pivotal platform for the management of trauma patients²¹. This management encompasses not only the initial stabilization phase but also the subsequent management of complications such as circulatory failure, respiratory failure, acute kidney failure, and severe infections^23–25. Consequently, the development of a rapid and accurate predictive model for trauma patients in the ICU is essential.

This study retrieved the clinical and survival data of 2,205 trauma patients from the MIMIC-IV database. To identify the indicators of in-hospital mortality among these patients, a combination of LASSO regression and multivariate logistic analysis was employed. Six significant indicators were identified: hepatopathy, obesity, chloride, body temperature, WBC count, and APSIII. Subsequently, a nomogram was developed to predict outcomes. The findings suggest that this approach is discriminatory and precise in terms of calibration. AUC calibration curve measurements demonstrated that this innovative nomogram was effective in both the training and validation cohorts.

Patients with severe trauma are more likely to develop organ dysfunction, which indicates a poorer prognosis²⁶. ASP III is an indicator for assessing the extent of physiologic disorders in critically ill patients, with higher scores indicating more severe conditions and a higher risk of mortality, and is often used to predict prognosis²⁷. Previous reports have indicated that the APS III score significantly distinguishes between ICU survivors and those who did not survive in the ICU²⁸. APS III was significantly associated with increased mortality in trauma patients, as demonstrated by our study’s outcome. Consequently, the observed findings suggest the potential of the APS III score as a prognostic indicator for all-cause in-hospital mortality in trauma patients.

Notably, hepatopathy was the most significant independent risk factor for all-cause mortality in hospitalized trauma patients. This observation is consistent with findings reported in previous studies. Especially, cirrhosis has been shown to have a significant impact on trauma outcomes, leading to an increased mortality rate and an elevated risk of sepsis, acute respiratory distress syndrome, and trauma-associated coagulopathy^29–32. These outcomes are associated with prolonged ICU and hospital stays³³. The underlying pathophysiological mechanisms include coagulation-anticoagulation imbalances, platelet dysfunction, immune suppression, and systemic inflammatory dysregulation^33,34. The implementation of multidisciplinary critical care protocols is crucial. Dynamic coagulation monitoring, prudent surgical decision-making, and prolonged monitoring to mitigate delayed hepatic failure risks also play pivotal roles in management strategies³⁵.

Moreover, our findings indicate that elevated body temperature is associated with a poor clinical prognosis in trauma patients. The incidence of hypothermia in severely traumatized patients has been reported at 12.6%, and it is significantly associated with increased mortality and hospital complications^36–38. Traumatic hypothermia, defined as a body temperature below 35 °C, raises the risk of mortality via two primary mechanisms. First, hypothermia inhibits the coagulation cascade; a 1 °C decrease in temperature results in a 10% reduction in thrombin generation, significantly impairing platelet function^39,40. Second, hypothermia impairs citrate metabolism, leading to hypocalcemia and reduced myocardial contractility, which can further exacerbate shock severity³⁹. Consequently, preventing hypothermia is essential in the management of severely traumatized patients.

Notably, our study found that obesity was independently associated with a negative correlation with mortality in trauma patients. Several studies have demonstrated that obesity exerts multifaceted effects on patients with traumatic injuries. Overweight and mild obesity may reduce the risk of abdominal organ injuries in blunt force trauma—through mechanisms such as fat buffering, as evidenced by a lower incidence of acute fascial compartment syndrome and liver and kidney injuries after lower extremity fracture—the “obesity paradox,” wherein the mortality rate of such patients is lower than that of individuals with normal weight, has been observed in certain studies^41–44. However, it should be noted that morbid obesity significantly increases the risk of complications, prolonged hospitalization, and ICU admission following blunt force trauma^45,46. The prognosis varies significantly by trauma type and BMI stratification and must be assessed in conjunction with clinical features. In the present study, the implementation of a stratified assessment of patients with varying degrees of obesity proved to be infeasible because of the paucity of available data.

Laboratory tests have identified the serum chloride level as a potential risk factor. Hypochloremia may be an important marker of poor prognosis in trauma patients, but related research is extremely limited. Rodríguez-Triviño found that hypochloremia in patients with severe traumatic brain injury was significantly associated with mortality⁴⁷. Kimura et al. found that hypochloremia in postoperative ICU patients independently increased mortality risk⁴⁸. Although both studies support the prognostic value of hypochloremia in different trauma scenarios, existing evidence is scarce, and further research is urgently needed to explore its mechanisms. Interestingly, our study found that WBC count was a statistically significant predictor of all-cause mortality in hospitalized trauma patients, which is consistent with previous research. Akköse’s study showed that leukocyte counts in patients with blunt force trauma are positively correlated with the severity of trauma⁴⁹. In cases of significant elevation of leukocytes after multiple injuries, antibiotic intervention is necessary to prevent infection⁵⁰. This underscores the critical importance of meticulous white blood cell monitoring and timely adjustment of treatment regimens when necessary.

This study has several limitations. This was a single-center retrospective study, which may have introduced selection bias and affected the accuracy of the result. Additionally, the MIMIC-IV database lacks key variables, such as ISS scores. Finally, the results were only internally validated, and external validation through large multicenter studies is required to improve their applicability.

Conclusion

This study identified hepatopathy, obesity, chloride, body temperature, WBC count and APSIII as independent risk factors for severely trauma patients. A unique nomogram model was constructed to facilitate clinical decision-making and predict all-cause mortality in hospitalized patients. Nevertheless, prospective external validation is warranted to ensure long-term reliability of the model.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Y.Z. and E.Q. conceived and designed the study. N.T., X.H. and S.P. collected the primary data. Y.Z., E.Q., and S.P. performed data analysis. Y.Z. and N.T. developed analytical tools. Y.Z., S.P., and N.T. contributed to data interpretation. E.Q., N.T., and X.H. created visualizations and figures. N.T. and S.P. validated the methodology and findings. Y.Z. drafted the manuscript with input from all authors. E.Q. and S.P. supervised the study and critically revised the manuscript. All authors read and approved the final version of the manuscript.

Funding

This work was supported by the Hunan Province Natural Science Foundation of China (NO.2024JJ9554 to YQ Zeng). The study did not involve any potential impact caused by the sponsor.

Data availability

The original data of this study are from the MIMIC-IV database, which is open to the public. Detailed data from this study may be obtained by contacting the corresponding author.

Declarations

Competing interests

The authors declare no competing interests.