Development and internal validation of a Cox proportional hazards model for predicting 1-year mortality in patients conservatively managed for hip fracture.

Abstract

Background: Hip fractures in older adults are associated with high mortality and functional decline, particularly in conservatively managed patients. However, prognostic models specific to conservatively managed patients remain scarce. We aimed to develop and internally validate a Cox proportional hazards (PH) model to predict 1-year mortality based on admission data.

Methods: We conducted a retrospective cohort study at a community hospital in Japan, including 76 patients aged ≥65 years who sustained femoral neck or trochanteric fractures and received conservative treatment between April 2018 and April 2020. The primary outcome was 1-year all-cause mortality. Variables included demographics, comorbidities, cognitive function, fracture type, and nutritional status. Missing data were imputed using a random forest algorithm. Univariable and multivariable Cox PH models were used. Internal validation was performed with bootstrap resampling (1000 iterations). Model discrimination was assessed using Harrell's C-index, and calibration was assessed using calibration plots.

Results: The Cox PH regression analysis yielded: h (t∣age, body mass index [BMI], fracture type) = h₀(t) exp (0.005 × age - 0.274 × BMI - 1.870 × fracture type) (Fracture type: 0 = femoral neck, 1 = trochanteric). Lower BMI (hazard ratio [HR] = 0.760; 95 % confidence interval [CI]: 0.637-0.908; p = 0.002) and trochanteric fractures (HR = 0.154; 95 % CI: 0.058-0.411; p < 0.001) were significant predictors of increased mortality. The model demonstrated good discrimination (Harrell's C-index: 0.774; optimism-adjusted: 0.762). Calibration was poor at early timepoints (90-270 days); however, it improved at 365 days (slope = 1.03; C-statistic = 0.83). Decision curve analysis confirmed clinical utility at threshold probabilities above 10 %.

Conclusions: We developed a Cox PH regression model with good discrimination and acceptable calibration at 1 year to predict mortality in patients with conservatively managed hip fractures. This model may assist clinicians in early risk stratification and individualized care planning.