Use of Machine Learning to Predict Improvement After Hip Arthroscopy 

The management of expectations is crucial when counseling patients undergoing treatment for a musculoskeletal injury or condition. In hip arthroscopy, this is especially critical when discussing with patients—including athletes seeking to return to play—their anticipated outcomes following surgical treatment for femoroacetabular impingement syndrome (FAIS).

In the latest issue of JBJS, Kunze et al. report on their investigative efforts to develop and internally validate machine learning algorithms that can yield patient-specific predictions of which athletes will reach clinically relevant improvement in function after arthroscopy for FAIS.

A total of 1,118 athletes, identified through a retrospective review of clinical registry data, met the inclusion criteria. The primary outcome was attaining the minimal clinically important difference (MCID) in the Hip Outcome Score-Sports Subscale (HOS-SS) at a minimum of 2 years postoperatively. Six machine learning algorithm models were tested.

The authors found that 23.1% of the athletes did not achieve the MCID for the HOS-SS. Six variables optimized algorithm performance, with the following cutoffs found to decrease the likelihood of achieving the MCID:

  • Preoperative HOS-SS score of ≥58.3
  • Tönnis grade of 1 (early osteoarthritis)
  • Alpha angle of ≥67.1° on anteroposterior radiograph
  • Body mass index (BMI) of >26.6 kg/m2
  • Tönnis angle of >9.7° (indicating subtle instability or dysplasia)
  • Patient age of >40 years

The elastic-net penalized logistic regression (ENPLR) model was the most accurate model in this study.

The findings suggest that patient selection is paramount to the ability to achieve clinically relevant improvements in outcomes for patients treated with arthroscopy for FAIS. Multiple studies have demonstrated that increasing arthritis level and age, along with BMI, are associated with inferior patient-reported outcomes. In addition, hip instability and increased Tönnis angle have been shown to be associated with worse outcomes following hip arthroscopy. A greater alpha angle indicates a larger “deformity” and thus the potential for more damage at the time of surgery that cannot be completely addressed with today’s surgical techniques. “Higher” preoperative HOS-SS (although on a scale of 0 to 100, 58 is not that high) may make it more difficult for a patient to achieve enough of an improvement in their outcome score to be considered as having attained the MCID.

The ENPLR  model was converted into an open-source application, although as Kunze et al. point out, external validation is necessary before wider adoption of the application. Nonetheless, the model demonstrates the potential to help hip surgeons better educate our patients on expected outcomes and to assist with proper patient selection for the ever-evolving treatment of FAIS.

Matthew R. Schmitz, MD
JBJS Deputy Editor for Social Media

Co-author Kyle N. Kunze, MD discusses this study in an “Author Insights” video, found here.

One thought on “Use of Machine Learning to Predict Improvement After Hip Arthroscopy 

Leave a Reply

Related Posts

Discover more from OrthoBuzz

Subscribe now to keep reading and get access to the full archive.

Continue reading