Deviation From AI-Predicted Implant Sizes in Total Hip Arthroplasty Is Associated With Increased Complication Risk

Patient characteristics may predict implant sizes in total hip arthroplasty (THA), but the clinical value and generalizability of such models remain uncertain. This study evaluated the accuracy and relevance of an artificial intelligence (AI)-driven prediction tool for THA sizing in a diverse population.

We retrospectively reviewed 2410 primary THAs at 2 academic hospitals (2007-2021). Data included patient characteristics, surgical approach, surgeon fellowship training, and complications (periprosthetic femur fracture [PPF], aseptic loosening within 1 year). An AI model predicted femoral stem and acetabular cup sizes using age, sex, height, weight, and race/ethnicity, and these predictions were compared with implanted sizes. In a subset, predictions were compared to traditional radiographic templating. Multivariate analyses assessed associations between size deviation and complications.

AI-predicted femoral stems and cups were within one size of implanted components in 72.0% and 77.7% of cases, and within 2 sizes in 91.2% and 94.7%, respectively. Excluding complex cases (dysplasia, avascular necrosis, head collapse, conversion THA) improved accuracy (P < .0001). In subgroup analysis, AI predictions outperformed radiographic templating for stems (mean deviation 0.731 vs 1.179; P < .0001) and cups (0.673 vs 1.071; P = .036). Arthroplasty-trained surgeons implanted components significantly closer to AI-predicted sizes (P < .001). Closer alignment with AI-predicted stem size was associated with reduced aseptic loosening and PPF (P < .0001).

A free AI-assisted prediction tool (Mortho) demonstrated strong accuracy and generalizability for implant sizing in primary THA, particularly in standard cases. Predictions mirrored arthroplasty-trained surgeons' selections and outperformed radiographic templating, supporting patient characteristic-based AI as a valuable adjunct for THA planning.