JMIR Publications

ABSTRACT

Background:

Preoperative and postoperative traumatic coagulopathy (PPTIC) in trauma patients is closely related to various adverse outcomes. However, there is still a lack of fast and accurate PPTIC prediction tools.

Objective:

1 To investigate the risk factors for preoperative and postoperative traumatic coagulopathy (PPTIC) in trauma patients. 2 To explore complications during hospitalization in patients with PPTIC. 3 To employ 10 machine learning (ML) models to predict the risk of PPTIC in trauma patients.

Methods:

We analyzed data from 13,235 trauma patients from four medical centers, including medical histories, lab results, and hospitalization complications. We developed 10 ML models in Python to predict PPTIC based on preoperative indicators. Data from 10,023 MIMIC patients were divided into training (70%) and test (30%) sets, with 3,212 patients from three other centers used for external validation. Model performance was assessed with five-fold cross-validation, bootstrapping, Brier score (BS), and SHapley Additive exPlanations (SHAP) values.

Results:

Univariate logistic regression identified PPTIC risk factors as prolonged activated partial thromboplastin time (APTT), prothrombin time (PT), and international normalized ratio (INR); decreased levels of hemoglobin, hematocrit (HCT), red blood cells (RBC), calcium, and sodium; lower admission diastolic blood pressure (DBP); elevated alanine aminotransferase (ALT), aspartate aminotransferase (AST) levels, and admission heart rate (HR); and emergency surgery and perioperative transfusion. Multivariate logistic regression revealed that PPTIC patients faced significantly higher risks of sepsis (1.75-fold), heart failure (1.5-fold), delirium (3.08-fold), abnormal coagulation (3.57-fold), tracheostomy (2.76-fold), mortality (2.19-fold), and urinary tract infection (1.95-fold), along with longer hospital and ICU stays. The random forest (RF) was the most effective ML model for predicting PPTIC, achieving an AUROC of 0.91, AUPRC of 0.89, accuracy of 0.84, sensitivity of 0.80, specificity of 0.88, precision of 0.88, F1 score of 0.84, and BS of 0.13 in external validation.

Conclusions:

Key PPTIC risk factors include prolonged APTT, PT, INR; low levels of hemoglobin, HCT, RBC, calcium, and sodium; low DBP; high ALT, AST, and HR; and the need for emergency surgery and transfusion. PPTIC is associated with severe complications and extended hospital stays. Among the ML models, the RF model was the most effective predictor. Clinical Trial: Ethical approvals were granted by the respective ethics committees: Medical Research Ethics (2023) No. 261, 2023 Ethical Review (48), KY 2023037, and Record ID: 63295742. As this was a retrospective study, patient informed consent was waived by the ethics committees. It was registered with the Chinese Clinical Trial Registry (ChiCTR2300078097) adhered to the Declaration of Helsinki.