Artificial Neural Networks Outperform Linear Regression in Estimating Patient-Reported Outcomes Following Upper Extremity Fractures With a Larger Number of Included Variables