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