Patient-reported satisfaction after total knee arthroplasty (TKA) is low compared to other orthopedic procedures. Although several factors have been reported to influence TKA outcomes, it is still challenging to identify patients who will experience dissatisfaction five years after surgery, thereby improving their management. Indeed, both perioperative information and follow-up questionnaires seem to lack statistical predictive power.

This study aims to demonstrate that machine learning can improve the prediction of patient satisfaction, especially when classical statistics fail to identify complex patterns that lead to dissatisfaction.

Patients who underwent primary TKA were included in a Registry that collected baseline data and clinical outcomes at different follow-ups. The patients were divided into satisfied and dissatisfied groups based on a satisfaction questionnaire administered five years after surgery. Satisfaction was predicted using linear statistical models compared to machine learning algorithms.

A total of 147 subjects were analyzed. Regarding statistics, significant differences between satisfaction levels started emerging only six months after the intervention, and the classification was close to random guessing. However, machine learning algorithms could improve the prediction by 72% soon after the intervention, and an improvement of 178% was possible when including follow-ups up to one year.

This study demonstrates the feasibility of a registry-based approach for monitoring and predicting satisfaction using ML algorithms.

III.