Development and internal validation of machine learning algorithms to predict patient satisfaction after total hip...

Area of Science:

• Orthopedic Surgery
• Artificial Intelligence in Medicine
• Patient-Reported Outcomes

Background:

• Patient satisfaction is a critical outcome measure following total hip arthroplasty (THA).
• Predicting postoperative satisfaction is essential for optimizing patient outcomes and managing expectations.
• Machine learning (ML) offers a novel approach to analyze complex patient data for predictive insights.

Purpose of the Study:

• To evaluate the efficacy of supervised ML algorithms in predicting patient satisfaction after THA.
• To identify key preoperative factors influencing patient satisfaction 2 years post-surgery.

Main Methods:

• Utilized prospectively collected data from 1508 primary THAs.
• Developed and trained supervised ML models (Random Forest, XGBoost, SVM, Logistic LASSO) using demographics, comorbidities, and PROMs (SF-36, WOMAC, OHS).
• Validated model performance on an independent test set to predict 2-year postoperative satisfaction.

Main Results:

• ML models demonstrated fair discriminative ability, with the LASSO model achieving an AUC of 0.76.
• Identified patient age, preoperative WOMAC scores, number of comorbidities, preoperative mental component summary (MCS), prior lumbar surgery, and low BMI as significant predictors of dissatisfaction.
• Permutation importance analysis highlighted these factors' influence on predictive accuracy.

Conclusions:

• Supervised ML algorithms show promise in predicting THA patient satisfaction.
• Identified modifiable and non-modifiable predictors can enhance preoperative patient counseling and health optimization strategies.
• These findings can guide personalized patient management to improve overall satisfaction after THA.