Summary: Failure to account for competing mortality gave higher estimates of 10-year fracture probability than if appropriate adjustment is made for competing mortality, particularly among subgroups with higher mortality. A modified Kaplan-Meier method is easy to implement and provides an alternative approach to existing methods for competing mortality risk adjustment.
Introduction: A unique feature of FRAX(®) is that 10-year fracture probability accounts for mortality as a competing risk. We compared the effect of competing mortality adjustment on nonparametric and parametric methods of fracture probability estimation.
Methods: The Manitoba Bone Mineral Density (BMD) database was used to identify men and women age ≥50 years with FRAX probabilities calculated using femoral neck BMD (N = 39,063). Fractures were assessed from administrative data (N = 2,543 with a major osteoporotic fracture, N = 549 with a hip fracture during mean 5.3 years follow-up).
Results: The following subgroups with higher mortality were identified: men, age >80 years, high fracture probability, and presence of diabetes. Failure to account for competing mortality in these subgroups overestimated fracture probability by 16-56 % with the standard nonparametric (Kaplan-Meier) method and 15-29 % with the standard parametric (Cox) model. When the outcome was hip fractures, failure to account for competing mortality overestimated hip fracture probability by 18-36 % and 17-35 %, respectively. A simple modified Kaplan-Meier method showed very close agreement with methods that adjusted for competing mortality (within 2 %).
Conclusions: Failure to account for competing mortality risk gives considerably higher estimates of 10-year fracture probability than if adjustment is made for this competing risk.