P. Jermin (Chester, GB)
The Robert Jones & Agnes Hunt Orthopaedic Hospital SurgeryPresenter Of 1 Presentation
10.2.10 - Modelling and predicting the long-term clinical outcome of Autologous Chondrocyte Implantation
Abstract
Purpose
Autologous Chondrocyte Implantation (ACI) has been used clinically for over 20 years to treat symptomatic articular cartilage defects. However, not every patient enjoys long-term benefit from ACI. Predicting long-term outcome will help decide if ACI is worthwhile. This study aims to (1) Formulate a statistical model describing long-term clinical outcome of ACI (2) Derive baseline predictors of the model parameters.
Methods and Materials
All ACI patients at our Centre provide Lysholm knee scores before ACI and annually thereafter. Consenting patients treated between 1996 and 2010 were asked if and when they had a knee replacement and had their baseline data (age, gender, number of defects, location, size and grade of defect, previous operations and need for concomitant operations) obtained from records. A multilevel model was fit, describing Lysholm score over time until final followup or knee replacement. Models were compared using likelihood ratio tests, Akaike Information Criteria and R2-values. Baseline variables were added to the overall best fit model as fixed interaction terms with the random parameters to predict variations in Lysholm pattern.
Results
One hundred seventy patients consented (mean followup 7.9 year, range 0.3-17.4). The overall best-fit function describing Lysholm scores over time had a linear first phase lasting 12 months with random slope, and a quadratic second phase with random linear and quadratic parameters (Figure; conditional R2=0.76). Three baseline variables significantly affected score pattern: baseline Lysholm, a patellar or tibial defect, and two or more defects. These three explained 34% of the variation in outcome (marginal R2=0.34).
Conclusion
Following ACI, Lysholm knee scores follow a biphasic pattern with a 12 month rehabilitation phase when the score changes linearly followed by a longer phase when the score follows a quadratic curve. Three variables explained 31% of the variation in outcome. We implemented the model in a web-based app (https://ork.app) predicting long-term knee function.