Aims: To test a decision model for non-invasive estimation of left ventricular filling pressure (LVFP) in patients with left ventricular (LV) dysfunction and a wide range of ejection fractions (EF).
Methods and results: In patients with LV dysfunction (n = 270; EF = 42 +/- 16%), classification and regression tree (CART) analysis was used to generate a model for the prediction of elevated LVFP, defined as pulmonary capillary wedge pressure (PCWP) >15 mmHg, in a derivation cohort (n = 178). At each step of the decision tree, nodes including single or multiple criteria connected by Boolean operators were tested to achieve the best information entropy gain. Averaged mitral-to-myocardial early velocities ratio (E/e') > or =13 OR E-wave deceleration time <150 ms was closely associated with elevated LVFP. Alternatively, prediction of PCWP >15 mmHg needed the following criteria to be satisfied: (i) intermediate E/e' (13 > E/e' > 8); (ii) left atrial volume index >40 mL/m(2) OR ratio of mitral E-wave and colour M-mode propagation velocity >2 OR difference in duration of pulmonary vein and mitral flow at atrial contraction >30 ms; (iii) estimated pulmonary artery systolic pressure >35 mmHg. Patients were correctly allocated according to PCWP with an 87% sensitivity and a 90% specificity. Compared with the best single parameter estimating LVFP, a 17% relative increase in accuracy was achieved in patients with EF >50%. The model was prospectively validated in a testing group (n = 92): 80% sensitivity, 78% specificity.
Conclusion: This sequential testing is useful to non-invasively predict LVFP in patients with LV dysfunction, especially in those with preserved EF.