Abstract
Predicting length of stay (LOS) for revision total hip arthroplasty (rTHA) remains
challenging due to the complexity of cases and increased complication rates compared
with primary total hip arthroplasty. Artificial intelligence may play a role in predicting
LOS after rTHA. This study aims to develop a machine learning model to predict LOS
in rTHA patients based on preoperative and intraoperative factors. The American College
of Surgeons National Surgical Quality Improvement Program database was used to identify
patients undergoing a rTHA between 2012 and 2018. The data were used to train both
random forest machine and logistic regression machine learning models to predict short
LOS (≤ 1 day) or long LOS (≥ 2 days). Statistical analysis was performed to analyze
differences between short and long LOS groups. Logistic regression analysis was used
to calculate odds ratios associated with long LOS. A total of 4,228 patients were
identified in this analysis with a mean postoperative LOS of 3.69 days. Preoperative
features associated with a short LOS included male sex, noninfectious revision, spinal
anesthesia, later year of operation, younger age, smoking history, and lower body
mass index. Area under the receiver operating characteristic curve was calculated
to measure model performance for the random forest model and logistic regression model
as 0.76 and 0.78, respectively. The machine learning model presented here was able
to reasonably predict LOS for revision hip arthroplasty showing promise for use as
patient selection tool to identify short LOS patients.
Keywords
machine learning - clinical decision support - patient selection - revision total
hip arthroplasty - length of stay
