Constraint implant decision-making in robot-assisted total knee arthroplasty: A comparative study of surgeon...

Constraint selection during total knee arthroplasty (TKA) remains a subjective intraoperative decision, with both over-constraint and under-constraint posing risks to implant longevity and patient outcomes. With the growing adoption of robot-assisted TKA (RA-TKA), there is increasing interest in whether intraoperative robotic assessment can support more consistent and conservative constraint decision-making. This study evaluated the degree of agreement between surgeon preoperative predictions and intraoperative decisions regarding constrained implant use (varus-valgus constrained [VVC] or posterior-stabilised plus [PS-Plus]) in RA-TKA, and examined whether intraoperative robotic assessment was associated with more conservative implant selection.

In this retrospective observational study, six orthopaedic surgeons of varying seniority independently reviewed 100 robot-assisted TKA cases using standardised clinical and radiographic data, predicting whether a constrained implant would be required. Predictions were compared with actual intraoperative implant selection to characterise prediction-decision agreement. Sensitivity, specificity, positive predictive value, and negative predictive value were calculated as descriptive measures of concordance, and inter-rater agreement was assessed using Fleiss' kappa.

Surgeons demonstrated moderate specificity but lower sensitivity, with performance varying substantially across individuals. Among 18 cases where the majority of surgeons (≥4 out of 6) predicted constraint, only 10 ultimately received a constrained implant intraoperatively. Of the 8 cases with unanimous prediction, 2 were nonetheless managed with a primary implant. These findings correspond to a prediction-decision discordance rate of 25.0 to 44.4% across cases with majority or unanimous agreement, and inter-rater agreement was moderate (κ = 0.51).

These findings demonstrate a notable degree of prediction-decision discordance, particularly in cases with high preoperative consensus, suggesting that intraoperative assessment facilitated by robotic systems may have informed more conservative implant choices than preoperative data alone would have indicated. Further prospective studies are warranted to evaluate long-term outcomes and to explore the integration of intraoperative data into predictive decision-support tools.