Applying a Random Forest Approach in Predicting Health Status in Patients with Carotid Artery Stenosis 30 Days Post-Stenting

Abstract

ABSTRACTBackgroundApproximately 20% of ischemic strokes in the U.S. result from carotid artery stenosis. Carotid artery stenting (CAS) can reduce stroke risk, but variability in post-stenting health outcomes and their predictors are poorly understood. We examined 30-day post-CAS health status and derived its most important clinical predictors.MethodsThe SAPPHIRE Worldwide Registry measured pre-procedural and 30-day health status for carotid artery stenosis patients undergoing transfemoral-CAS using the SF-36 or EQ-5D between 2010-2014. Four health status scores were calculated: SF-36 Mental Component Summary (MCS) and Physical Component Summary (PCS), EQ-5D Index Value and Visual Analogue Scale (VAS). Random Forest models ranked 66 pre-procedural candidate variables by relative importance in predicting 30-day post-CAS health status, stratified by symptomatic status. Variables with highest importance were used to develop predictive multivariable linear regression models. Model accuracy was assessed via Out of Bag accuracy and R-square, respectively.ResultsHealth status was assessed using the SF-36 in 3,017 patients and EQ-5D in 3,390 patients. Random Forest models demonstrated high accuracy (86.7% - 95.2%) and identified nine key predictors of post-stenting health status: pre-procedural health status (RI 100%), Modified Rankin Scale (RI 26.2-76.5%), NIH Stroke Scale (RI 12.1-28.0%), history of stroke (RI 9.2- 19.8%), congestive heart failure (RI 12.3-19.7%), spinal immobility (RI 6.7-31.0%), diabetes mellitus (RI 8.1-32.9%), severe pulmonary disease/COPD (RI 13.8-45.6%), and non-Hispanic/Latino ethnicity (RI 8.4-32.4%). Multivariable linear regression models explained ∼36- 61% of health status variance, with the asymptomatic SF-36 PCS model explaining 61%. SF-36 PCS and MCS models outperformed EQ-5D Index Value and VAS models regarding R-square and visual fit of observed vs. predicted values.ConclusionsWe successfully derived prediction models for patient-centered outcomes following CAS which partially explained 30-day post-CAS health status outcomes. Pre-procedural health status, stroke scale scores, and medical comorbidities should be considered when discussing health status benefits in pre-CAS treatment decision discussions.