Predicting Survival in Thai Patients After Low Impact Hip Fracture Using Flexible Parametric Modelling: A Retrospective Cohort Study

Elsevier, Journal of Clinical Densitometry, Volume 24, 1 October 2021
Atthakomol P., Manosroi W., Phinyo P., Pipanmekaporn T., Vaseenon T., Rojanasthien S.

Predictive post-hip fracture mortality models have been presented for specific time points (in-hospital, 30-days or 1-year) and most provide marginal predictions based on the patient's risk group. However, the predictive model for individual survival probability following hip fracture is not available. This study aimed to develop a flexible parametric model for predicting individual survival probability for hip fracture patients. In this retrospective study, the medical charts of 765 Thai patients admitted to hospital with a hip fracture resulting from low-impact injury from January 2014 to December 2018 were reviewed. Predictors for all-cause mortality were identified using flexible parametric survival analysis and were used to develop the predictive model. The model was calibrated using a calibration graph and discrimination performance was evaluated using the C-statistic. Internal validity was assessed using bootstrapping. The overall mortality rate of the hip fracture patients was 14%. Predictors significantly associated with survival after hip fracture were age, active malignancy, dementia or Alzheimer's disease, chronic obstructive pulmonary disorder, diabetes mellitus, hemoglobin concentration, eGFR<30 mL/min/1.73m2 and operative treatments. The model-predicted survival was similar to that actually observed in the very low survival group in the first year after hip fracture. In bootstrapping, the apparent C-statistic and the test C-statistic of the reduced model were 0.79 (95% CI 0.77–0.81) and 0.79 (95% CI 0.78–0.80), respectively. The flexible survival model provides good predictive power for individual survival probability at any given time point within the first year after hip fracture and would be an easy to use tool in clinical practice.