Machine learning with D-dimer in the risk stratification for pulmonary embolism: a derivation and internal validation...

Area of Science:

• Medical Diagnostics
• Machine Learning in Healthcare
• Pulmonary Embolism Research

Background:

• Pulmonary embolism (PE) diagnosis requires accurate risk stratification.
• Traditional risk scores have limitations in predicting PE.
• Machine learning offers potential for improved diagnostic accuracy.

Purpose of the Study:

• To develop and validate a machine learning model for predicting PE.
• To compare the model's performance against established risk scores.
• To assess the impact of D-dimer inclusion on predictive accuracy.

Main Methods:

• A derivation and internal validation study was conducted.
• Seven machine learning techniques were evaluated, with generalized logistic regression using elastic net selected.
• Models were developed with and without D-dimer, using data from 3347 patients.

Main Results:

• The machine learning model incorporating D-dimer achieved an AUC of 0.89, significantly outperforming models without it (AUC 0.73).
• D-dimer addition improved AUC by 0.16 and decreased the Brier score by 14%.
• The model demonstrated a superior positive likelihood ratio and a lower false-negative rate compared to Wells score, revised Geneva score, and PERC score.

Conclusions:

• A machine learning model significantly outperforms traditional risk scores for PE risk stratification in emergency settings.
• The developed model shows promise for improving PE diagnosis and patient management.
• External validation is recommended to confirm the model's generalizability.