Personalized Medicine in Hip Preservation: Use of Machine-learning Methodology for Creation of a Preoperative Prognostic Model for Patients Undergoing Hip Arthroscopy

Phillip Rosinsky ¹ Vivian Ouyang ¹ Camille Go ¹ Jeffrey Gornbein ³ Jacob Shapira ¹ Mitchell Meghpara ^1,2 David Maldonado ¹ Ajay Lall ^1,4 Benjamin Domb ^1,4

¹Orthopedics and Sports Medicine Research, American Hip Institute Research Foundation, USA
²Orthopedics, Amita Health St. Alexius Medical Center, USA
³General Internal Medicine and Health Services Research, UCLA Department of Medicine Statistics Core, USA
⁴Orthopedics, American Hip Institute, USA

Background:
Personalized medicine models to predict outcomes of orthopedic surgery are scarce, and have often required data that can only be obtained during surgery.

Purpose:
The purpose of this study was to establish a method for predictive modeling to enable individualized prognostication and shared decision-making based on preoperative patient factors, using data from a prospective hip preservation registry.

Methods:
Preoperative data of 2,415 patients undergoing hip arthroscopy (HA) for femoroacetabular impingement syndrome between February 2008 and November 2017 were retrospectively analyzed. Two machine-learning analyses were evaluated: tree-structured survival analysis (TSSA) and Cox proportional hazards modeling for predicting time to event and for computing hazard ratios for conversion to total hip arthroplasty (THA). The Fine-Gray model was similarly used for revision HA. Variables were selected for inclusion using minimum Akaike Information Criteria (AIC). Stepwise selection was used for the Cox and Fine-Gray models. A web-based calculator was created based on the final models.

Results:
The TSSA model performed poorly, resulting in a Harrell C-statistic lower than 0.6, rendering it inaccurate and not interpretable. In contrast, the Harrell C-statistic of the Cox model calculators for THA and the Fine-Gray model for revision hip arthroscopy were 0.848 and 0.662, respectively. Using AIC, thirteen preoperative variables were identified as predictors of THA and six variables as predictors for revision HA.

Conclusion:
This study successfully created an institution-specific prognostic model forprediction of revision HA and conversion to THA in patients undergoing HA. While both models are useful in predicting outcomes, the model predicting THA was more accurate. This prognostic model may be used at other institutions after performance of an external validation. Perhaps more importantly, this study may serve as proof of concept for a methodology for development of an individualized prognostication model.