Can right ventricular assessments improve triaging of low risk pulmonary embolism?

Abstract

Identifying right ventricle (RV) abnormalities is important to stratifying pulmonary embolism (PE) severity. Disposition decisions are influenced by concerns about early deterioration. Triaging strategies, like the Simplified Pulmonary Embolism Severity Index (sPESI), do not include RV assessments as predictors or early deterioration as outcome(s). We aimed to (1) determine if RV assessment variables add prognostic accuracy for 5-day clinical deterioration in patients classified low risk by sPESI, and (2) determine the prognostic importance of RV assessments compared to other variables and to each other. We identified low risk sPESI patients (sPESI = 0) from a prospective PE registry. From a large field of candidate variables, we developed, and compared prognostic accuracy of, full and reduced random forest models (with and without RV assessment variables, respectively) on a validation database. We reported variable importance plots from full random forest and provided odds ratios for statistical inference of importance from multivariable logistic regression. Outcomes were death, cardiac arrest, hypotension, dysrhythmia, or respiratory failure within 5 days of PE. Of 1736 patients, 610 (35.1%) were low risk by sPESI and 72 (11.8%) experienced early deterioration. Of the 610, RV abnormality was present in 157 (25.7%) by CT, 121 (19.8%) by echocardiography, 132 (21.6%) by natriuretic peptide, and 107 (17.5%) by troponin. For deterioration, the receiver operating characteristics for full and reduced random forest prognostic models were 0.80 (0.77-0.82) and 0.71 (0.68-0.73), respectively. RV assessments were the top four in the variable importance plot for the random forest model. Echocardiography and CT significantly increased predicted probability of 5-day clinical deterioration by the multivariable logistic regression. A PE triaging strategy with RV imaging assessments had superior prognostic performance at classifying low risk for 5-day clinical deterioration versus one without.