Identification of patient subgroups by hierarchical cluster analysis in femoral neck fractures and surgical outcome analysis.

INTRODUCTION: Femoral neck fractures are a heterogeneous entity with wide variability in patient profiles and outcomes. Single-variable classification may not capture the multidimensional interactions that influence prognosis. This study aimed to identify clinically meaningful patient subgroups using hierarchical cluster analysis and to compare mortality, postoperative mobility, and complication rates between these subgroups.

MATERIALS AND METHODS: This retrospective cohort study included 672 patients treated for intracapsular femoral neck fractures classified as AO Foundation/Orthopaedic Trauma Association (AO/OTA) 31-B between 2009 and 2023, including 586 who underwent hemiarthroplasty (HA) and 86 who underwent total hip arthroplasty (THA). Unsupervised hierarchical clustering was performed using the Gower distance matrix and Ward's D2 linkage method based on 20 preoperative and intraoperative clinical and surgical variables, excluding outcome measures. The optimal number of clusters was determined by silhouette analysis. Mortality predictors were assessed using Cox proportional hazards regression, and predictors of good postoperative mobility were evaluated by binary logistic regression.

RESULTS: Two distinct clinical phenotypes were identified: a high-risk cluster (n = 607; mean age 78.3 years, American Society of Anesthesiologists [ASA] physical status III 86.2%, HA 96.4%, dementia 14.8%) and a low-risk cluster (n = 65; mean age 63.4 years, ASA II 81.5%, THA 98.5%, dementia 0%). Cox regression identified age (p < 0.001), male sex (p < 0.001), dementia (p = 0.029), ASA score (p = 0.010), and red-cell distribution width coefficient of variation (RDW-CV; p = 0.011) as independent mortality predictors, whereas procedure type was not significant (hazard ratio [HR] = 0.659, 95% confidence interval [CI]: 0.427-1.015, p = 0.058). In contrast, procedure type was independently associated with good postoperative mobility (odds ratio [OR] = 3.944, 95% CI: 1.516-10.260, p = 0.005).

CONCLUSIONS: Hierarchical cluster analysis identified two clinically distinct patient phenotypes with significantly different outcomes. Procedure type did not independently predict mortality, whereas THA was independently associated with good postoperative mobility, suggesting that the apparent survival advantage of THA may largely reflect indication-based selection bias. These findings support individualised, multidimensional risk assessment rather than single-variable decision-making. Because the identified phenotypes incorporate intraoperative variables, they characterise prognosis rather than guiding preoperative arthroplasty selection.