Machine learning models may predict outcomes of MPFL reconstruction
Click Here to Manage Email Alerts
Key takeaways:
- Machine learning models may accurately predict outcome measures for patients undergoing MPFL reconstruction.
- Machine learning also predicted return to sports and recurrent instability.
Published results showed machine learning models may accurately predict outcomes for patients undergoing medial patellofemoral ligament reconstruction.
Hongwei Zhan, MD, from the department of sports medicine at Honghui Hospital and Xi'an Jiaotong University in Shaanxi, China, and colleagues analyzed data from 218 patients who underwent MPFL reconstruction between January 2018 and December 2022.
Zhan and colleagues constructed 42 predictive models for seven clinical outcomes to assess the ability of six machine learning models to accurately predict patients’ likelihood of achieving a minimal clinically important difference (MCID). They also assessed the ability of the model to predict factors that may affect return to sport and recurrent instability.
Zhan and colleagues found the multilayer perception model had the best performing predictions for failure to achieve MCID in terms of accuracy, specificity and sensitivity, with 87.3% accuracy for Lysholm score, 86.2% accuracy for IKDC score and 97% accuracy for Kujala score. It also exhibited optimal performance for predicting return to pivoting sports, with an accuracy of 75.4%, according to results.
Meanwhile, results showed the support vector machine algorithm model had the best performing predictions for failure to achieve MCID of the Tegner score (accuracy = 76.8%), as well as optimal performance in predicting return to pre-injury sports (accuracy = 95.2%) and recurrent instability (accuracy = 94.9%).
Zhan and colleagues noted low preoperative Tegner scores, shorter time to surgery and absence of severe trochlear dysplasia were predictors for return to pre-injury sports. Predictors for return to pivoting sports included the absence of severe trochlear dysplasia and patellar alta . They found older patient age, female patients and low preoperative Lysholm scores were predictive of recurrent instability.
“When applying predictive models validated with large, multicenter datasets to clinical practice in the future, they should not be employed to establish eligibility for MPFL reconstruction,” Zhan and colleagues wrote in the study. “Instead, their role should be to inform treatment decisions. It is imperative to avoid any misuse of these models to deny surgical treatment.”
Perspective
Back to TopThomas Moran, MD, and Adam B. Yanke, MD, PhD
The authors provide an interesting study examining the use of machine learning models to predict certain clinical outcomes following isolated, primary MPFL reconstruction. While the ability of these models to predict several clinical outcomes was analyzed (including failure to achieve the minimal clinically important difference and return to sports), the models developed were particularly efficacious in the ability to predict recurrent patellar instability following MPFL reconstruction. The authors highlight how their modeling shows the incidence of recurrent patellar instability to be highly predicated upon the presence of greater patient age, female sex and low preoperative Lysholm score.
The dynamic capabilities of machine learning and AI have become increasingly studied throughout the orthopedic literature. The information provided by this study is relevant to surgeons, as it can help to provide more objective data when counseling patients preoperatively about expectations regarding their eventual surgical outcome. Readers should be cautioned, however, that the machine learning models provided should not be used in isolation to inform surgical decision-making. It is important that surgical discussions continue to occur via shared decision-making, with treatment being individualized to the patient.
Collapse
Read more about
Click Here to Manage Email Alerts