Routine Laboratory Tests Predict 72-h Fatality in Patients With D-Dimer Levels ≥ 2 μg/mL: A Retrospective Cohort...

Area of Science:

• Clinical diagnostics
• Biomedical data analysis
• Prognostic modeling

Background:

• D-dimer is a prognostic marker, but short-term fatality prediction across diverse conditions is under-researched.
• Limited clinical information often accompanies D-dimer measurements.

Purpose of the Study:

• To develop and compare predictive models for 72-hour fatality in patients with D-dimer ≥ 2 μg/mL.
• To utilize routine laboratory variables for risk stratification.

Main Methods:

• Retrospective analysis of 5158 patients, with external validation on 5550 patients.
• Comparison of multivariate logistic regression analysis (MLRA) with four machine learning (ML) models: Prediction One, LightGBM, XGBoost, and CatBoost.
• Utilized 40 routine laboratory tests, age, and sex as predictors.

Main Results:

• The 72-hour fatality rate was 4.67%, with intracranial disease, malignancy, and sepsis as major causes of death.
• MLRA identified age, total protein, cholesterol, aspartate aminotransferase, and D-dimer as key predictors (AUC 0.829).
• LightGBM achieved the highest performance (AUC 0.987), significantly outperforming other ML models and MLRA.

Conclusions:

• Machine learning models, especially LightGBM, effectively predict short-term fatality in patients with elevated D-dimer.
• These models enable timely risk stratification and decision-making using readily available laboratory data.
• Routine laboratory tests can be leveraged for early identification of high-risk patients when clinical data is scarce.