Identifying Risk Groups in 401,846 Osteoarthritis Patients Undergoing Total Hip Arthroplasty: A Machine Learning...

Area of Science:

• Orthopedics
• Health Informatics
• Data Science

Background:

• Osteoarthritis (OA) is a leading cause for total hip arthroplasty (THA).
• Patient outcomes and resource utilization after THA are highly variable due to diverse comorbidity profiles.
• Understanding these profiles is crucial for optimizing patient care and hospital resource allocation.

Purpose of the Study:

• To apply unsupervised machine learning to identify distinct comorbidity clusters in OA patients undergoing THA.
• To investigate the association between identified comorbidity clusters and non-routine discharge (NRD) and length of stay (LOS).

Main Methods:

• Utilized the 2015-2021 National Inpatient Sample database.
• Included 401,846 patients with OA undergoing THA.
• Employed k-modes clustering with 49 comorbidities and clinical covariates, validated by Davies-Bouldin and Calinski-Harabasz indices.
• Assessed NRD using multivariable logistic regression and LOS using Kruskal-Wallis H testing.

Main Results:

• Identified five distinct patient clusters based on comorbidity profiles.
• Clusters with higher comorbidity burdens (e.g., renal dysfunction, cardiovascular disease, anemia, heart failure) showed significantly higher odds of NRD (up to 3.01, p < 0.001) and prolonged LOS (median up to 4 days).
• Lower-risk clusters had shorter hospitalizations (median LOS of 2 days) and higher rates of routine discharge.

Conclusions:

• Machine learning effectively delineated clinically meaningful subgroups within the OA THA population.
• Comorbidity burden is a significant determinant of postoperative outcomes, including NRD and LOS.
• This data-driven approach can enhance perioperative risk stratification and inform resource planning for THA patients.