Risk prediction models for hip fracture: parametric versus Cox regression

Abstract

Hip fracture is a public health burden due to high morbidity, mortality and cost. Risk prediction models can aid clinical decision-making by identifying individuals at risk. Objective: To build risk prediction model for incident hip fracture using Weibull regression and compare this with Cox regression model. Method: The Study of Osteoporosis prospectively collected risk factors were used to build a risk prediction model for first hip fracture using Threshold regression with Wiener process. Similar predictors were fitted using Cox regression for comparison. Results: There were 632 first hip fractures. Age, bone density, maternal and personal prior fractures were significant risk factors for hip fracture. Weibull had better goodness of fit, higher D-statistic and R-squared values than the exponential. Models did not differ in c-index and ten-fold cross validation showed similar areas under the ROC curves. Conclusion: Parametric and Cox models were comparable. External validation of the prediction model is required.