Artificial Intelligence-Driven Decision-Making for Knee Joint Manipulation Following Primary Total Knee Arthroplasty.

BACKGROUND

Manipulation under anesthesia (MUA) is a commonly performed procedure to address postoperative stiffness after total knee arthroplasty (TKA), yet optimal timing, motion thresholds, and outcome expectations remain poorly defined due to limited high-quality evidence and heterogeneous patient presentations.

METHODS

This study evaluated the capacity of an artificial intelligence (AI) platform to synthesize a data-driven decision-making framework for MUA following primary TKA. A systematic literature review on the range of motion (ROM) trajectory and manipulation following primary TKA was performed, and the data were used to train the AI model. Scenarios iterating variables, including time from surgery, knee flexion angle, extension deficit, preoperative ROM, and body mass index, were modeled to define the appropriateness of MUA and expected ROM gains.

RESULTS

According to our AI model, MUA is indicated for patients who have knee flexion < 80 degrees and/or extension deficits greater than 20 degrees at six weeks postoperatively, predicting mean improvements of 26 degrees in flexion and three degrees in extension. Delaying MUA beyond 90 days reduced expected flexion gains to 17 degrees. For persistent stiffness beyond three months, the model advised combining MUA with arthroscopic lysis of adhesions, particularly for flexion less than 90 degrees. A body mass index greater than 30 was associated with slightly greater flexion gains.

CONCLUSIONS

An AI-generated decision framework aligns well with published consensus on timing and patient selection. However, this tool should complement-not replace-clinical judgment, given limitations in current evidence quality and the multifactorial nature of postoperative stiffness.