Role of AI in the analysis of total knee arthroplasty

TL;DR

This study evaluated an AI model's ability to predict outcomes and infection risk in knee replacement surgery, finding it moderately accurate but needing improvement.

Contribution

A ChatGPT-based AI model was tested for predicting 1-year outcomes and infection risk in total knee arthroplasty patients.

Findings

The AI model underpredicted KOOS scores by 7-8 points compared to surgeon-reported outcomes.

The model's predicted infection risk was slightly higher than the observed rate with an ROC-AUC of 0.70.

The model showed moderate correlation (r = 0.45) with actual outcomes but requires fine-tuning for clinical use.

Abstract

This retrospective cohort assessed a ChatGPT-based AI model for 1-year KOOS prediction and infection risk in 98 primary total knee
replacement (TKR) patients. The model predicted based on preoperative clinical, demographic, and intra-operative data. The model under
predicted the mean knee injury and osteoarthritis Outcome Score (KOOS) by 7-8 points compared to surgeon-reported outcomes (p = 0.02)
with moderate correlation (r = 0.45).Predicted risk of infection (2.4%) was nominally higher than observed (1.8%), with an ROC-AUC of
0.70. Directionally accurate, the model requires further fine-tuning prior to clinic use.