Several preoperative measurements yielded high predictive value for TSA
Click Here to Manage Email Alerts
A machine learning analysis showed preoperative self-assessments of pain and objective measurements of active range of motion and strength had the highest predictive value in total shoulder arthroplasty, according to results.
Christopher P. Roche, MSE, MBA, and colleagues created predictive algorithms for the American Shoulder and Elbow Surgeons, Constant and University of California, Los Angeles (UCLA) scores using clinical data from 2,790 patients with 3,229 postoperative follow-up visits for primary anatomic TSA and primary reverse TSA.
To identify and rank the preoperative input model features based on the predictive value, researchers used the F-score and reciprocal fusion rank score to analyze the ASES, Constant and UCLA predictive algorithms. Researchers also objectively assessed the predictive value of the features by comparing the F-score and reciprocal fusion rank scores associated with each outcome measure to the top 20 feature inputs in the overall database used by each predictive algorithm.
In his presentation at the Orthopaedic Research Society Annual Meeting, Roche noted the VAS pain score had the highest predictive F-score value in the ASES model.
“All the other 10 questions are much lower in value, and all of those questions are activities of daily living questions,” Roche said. “So, that is one of the findings of the study: Activities of daily living questions have little predictive value.”
Roche added the aggregate ASES score was found in the top 20 feature inputs in the overall database. Active abduction and active forward elevation were identified as the most significant F-score values for the Constant score, followed by VAS pain score and strength, according to Roche.
“When you look at these 23 questions in the Constant relative to the top 20 questions driving the Constant model, you only see two of the 23 questions in the top 20,” Roche said. “Also, the aggregate Constant score was found to be in the top 20 as well.”
Roche noted the measurement of active forward elevation had the most significant F-score values for the UCLA score, followed closely by functional score and VAS pain score. He added three of the four questions in the UCLA score, as well as the aggregate UCLA score were found in the top 20 features driving the UCLA model.
“The UCLA questions had the highest predictive value,” Roche said. “The Constant [questions] were not as good as UCLA, but they are also better than the ASES.”
Collapse
Click Here to Manage Email Alerts