Oral Presentation ANZBMS-MEPSA-ANZORS 2022

Fracture risk assessment: does competing risk adjustment make a difference? (#135)

Thach Tran 1 2 , Tuan V Nguyen 3 4
  1. Garvan Institute of Medical Research, Darlinghurst, NSW, Australia
  2. Faculty of Medicine, University of New South Wales, Sydney, Australia
  3. School of Medicine Sydney, University of Notre Dame Australia, Sydney, Australia
  4. School of Biomedical Engineering, University of Technology Sydney, Sydney, NSW, Australia

The estimate of fracture risk can be compromised by the competing event of mortality, in which individuals die during a period of time without sustaining a fracture, resulting in the current suboptimal prediction of fracture risk and contributing to the global crisis of osteoporosis management. We hypothesized that the multistate model (MSM) could predict the risk fracture more accurately than the currently used competing risks models.

We used data from the Dubbo Osteoporosis Epidemiology Study that monitored bone health and mortality of 3,035 elderly individuals for more than 20 years. Fragility fracture was ascertained by X-ray reports. Mortality was ascertained from the NSW Registry of Births, Deaths and Marriages. We considered 4 statistical models for estimating an individual's risk of fracture: (i) standard Cox’s proportional hazards (conventional); (ii) cause-specific (CS); (iii) Fine-Gray subdistribution (FG); and (iv) MSM. These models were fitted and validated in the development (60% of the original sample) and validation (40%) cohort, respectively. Accuracy of fracture risk prediction was assessed using discrimination and calibration analysis.

During a median of 11.3 years, 628 (34.5%) individuals in the development cohort had sustained a fracture, and 630 (34.6%) had died without a fracture. All models had comparable discrimination performance (C index ~ 0.73 and 0.69 at 5 and 10-year prediction, respectively). All models had little difference in the 5-year predicted risk, though the conventional model tended to overestimate the fracture risk. For 10-year risk prediction, the MSM (calibration-in-the-large index = -0.05), CS (-0.23) and FG (-0.31) models outperformed the conventional Cox's model (-0.54) which over-estimated fracture risk (Figure).

The multistate model produces the most accurate estimate of fracture risk, especially for long-term prediction and practically superior to the currently used models. The multistate model should be considered as the optimal competing risks approach in development of a prediction model.628722170d846-MSM+and+competing+death.jpg

  1. Figure: Accuracy of the conventional and different competing risks model for predicting fracture risk