Two Symptoms to Triage Acute Concussions: Using Decision Tree Modeling to Predict Prolonged Recovery After a Concussion.

Abstract

The objective was to examine the 22 variables from the Sport Concussion Assessment Tool's 5th Edition Symptom Evaluation using a decision tree analysis to identify those most likely to predict prolonged recovery after a sport-related concussion. A cross-sectional design was used in this study. A total of 273 patients (52% men; mean age, 21 ± 7.6 yrs) initially assessed by either an emergency medicine or sport medicine physician within 14 days of concussion (mean, 6 ± 4 days) were included. The 22 symptoms from the Sport Concussion Assessment Tool's 5th Edition were included in a decision tree analysis performed using RStudio and the R package rpart. The decision tree was generated using a complexity parameter of 0.045, post hoc pruning was conducted with rpart, and the package carat was used to assess the final decision tree's accuracy, sensitivity and specificity. Of the 22 variables, only 2 contributed toward the predictive splits: Feeling like "in a fog" and Sadness. The confusion matrix yielded a statistically significant accuracy of 0.7636 (P [accuracy > no information rate] = 0.00009678), sensitivity of 0.6429, specificity of 0.8889, positive predictive value of 0.8571, and negative predictive value of 0.7059. Decision tree analysis yielded a statistically significant decision tree model that can be used clinically to identify patients at initial presentation who are at a higher risk of having prolonged symptoms lasting 28 days or more postconcussion.