Abstract Body (Do not enter title and authors here): Background: Despite advancements in pulmonary arterial hypertension (PAH) management both in-hospital and long-term mortality remain high for those admitted to the ICU.

Aims: We sought to identify echocardiographic predictors of outcomes and to develop a straightforward risk prediction score to predict one-year mortality in ICU-admitted PAH patients.

Methods. This single-center, retrospective study examines adult PAH patients hospitalized in the JHH ICU from January 2010 to December 2020. Echocardiographic indices were collected and associated with outcomes using Cox proportional hazards (CPH) modeling to predict one-year mortality. Using these indices as well as clinical and lab data, a machine learning (ML) prediction model, Random Survival Forest (RSF), was trained and evaluated with a time-varying area under the curve (AUC).

Results. Out of 102 PAH patients, 77 had complete echocardiograms at index hospitalization. Baseline characteristics are presented in Table 1. Right ventricular (RV) dilatation, RV contractile dysfunction (impaired TAPSE, FAC, and RVFWLS), more positive RA conduit strain, and RV-pulmonary artery (PA) uncoupling were associated with increased one-year mortality after adjusting for age, sex, race, and functional class (Figure1A). Kaplan-Meier survival curves revealed that TAPSE, TAPSE/PASP, RVFWLS, and RVFWLS/PASP have significant prognostic discriminative ability at optimal thresholds of 13 mm, 15 mm/mmHg, -9.5%, and 15%/mmHg, respectively. Our best ML RSF model achieved a mean time-varying AUC of 0.73 (Figure1B), with the top five predictors being estimated 4-chamber LV end-diastolic volume, FAC, echo-derived PASP, RVFWLS, and echo-estimated pulmonary vascular resistance (Figure1C-D).

Conclusion. Echocardiographic right heart functional indices are associated with one-year mortality in ICU-admitted PAH patients. Employing these parameters in an ML model was predictive of adverse clinical outcomes.