Reporting changes in right ventricular systolic pressure: insights from Classification and Regression Tree (CART) analysis.

Abstract

Background: Right ventricular systolic pressure (RVSP) is an echocardiographic metric to monitor pulmonary hypertension (PH). However, there is no recommendation on what constitutes a meaningful change. In this study, we aimed to gain insight into how physicians at our institution report significant changes in RVSP. We then aimed to quantify the analytic variability of reported RVSP using duplicate analysis.

Methods: We utilized the Stanford CardioShare Registry to identify 5,934 patients with 32,656 echocardiogram pairs with reported RVSP. Natural Language Processing was employed to categorize pairs into decrease, increase, no change, and no direct mention. Classification and Regression Tree (CART) analysis was applied to these groups to identify reporting thresholds among physicians. To assess the performance of the CART model, accuracy, precision, recall, and F1-score were reported using a stratified cross-validation method. In a separate cohort comprising 210 healthy volunteers and 208 patients with PH, two blinded core laboratory cardiologists measured the peak tricuspid regurgitation velocity. We employed a duplicate analysis method to model bias and a robust precision method for reporting RVSP and assessing analytical variability.

Results: Of the total pairs of echocardiographic studies, RVSP was reported as stable in 48.9%, increased in 12.5%, decreased in 9.9% while 28.7% did not have a direct reference to RVSP change. CART analysis revealed that physicians most commonly determine change based on an absolute threshold of 8 mmHg and whether the change occurred within or outside the reference range. On cross-validation, the accuracy and F1-score were 83% and 79% for the increase and 81% and 74% for the decrease algorithms. In the duplicate analysis cohort, the analytic precision was 8-10% with worst relative precision at lower values of RVSP. This translates into a 15% reference change value, assuming a 4-5% biological variation.

Conclusion: The study provides insights on real world practice of physician reporting in RVSP and provides directions for future recommendations regarding report changes.