Prediction of 30-day and long-term mortality in acute pulmonary embolism based on Pulmonary Embolism Severity Index...

Accuracy of the Pulmonary Embolism Severity Index (PESI)-based prediction of 30-day mortality in acute pulmonary embolism (PE) is suboptimal. Long-term mortality prediction in this setting remains incompletely characterized.

We aimed to develop and internally validate models for 30-day and long-term mortality in acute PE.

We used logistic and Cox proportional hazard regression to develop and internally validate (1000 bootstrap samples) models for prediction of 30-day and long-term mortality based on characteristics recorded at presentation in a prospective cohort of 792 consecutive adults with objectively confirmed acute PE.

Thirty-day mortality was 14.3 %. Compared to PESI, a model including PESI, red cell distribution width (RDW), lymphocyte and neutrophil counts had better optimism-corrected performance (c-statistic 0.916 vs. 0.881, estimated calibration index 0.026 vs. 0.090, Brier score 0.071 vs. 0.085, sensitivity 92.6 % vs. 90.9 %, specificity 73.2 % vs. 69.4 %). For 679 patients who survived the initial 30 days, administrative censoring was set at 3.5 years later (1666 patient-years, 244 deaths). Baseline risk was low (1.4 %). Optimism-corrected performance for a model based on PESI, age, cancer, pre-existing cardio-/cerebrovascular conditions and psychiatric diagnoses, RDW, mean platelet volume, hemoglobin, and lymphocyte count was good: area under receiver operating characteristics 0.866 (3.5 years), Uno's c 0.810, mean and weak calibration for fixed time point and time range 0.998 and 0.978, and 0.945 and 0.951, respectively, integrated calibration index (fixed time point) 0.036, scaled Brier score 0.369.

The developed models showed promising predictive performance regarding 30-day and long-term mortality in acute PE patients.