Fracture Risk Assessment in Long-term Care (FRAiL): Development and Validation of a Prediction Model

Abstract

Background: Strategies used to predict fracture in community-dwellers may not be useful in the nursing home (NH). Our objective was to develop and validate a model (Fracture Risk Assessment in Long-term Care [FRAiL]) to predict the 2-year risk of hip fracture in NH residents using readily available clinical characteristics.

Methods: The derivation cohort consisted of 419,668 residents between May 1, 2007 and April 30, 2008 in fee-for service Medicare. Hip fractures were identified using Part A diagnostic codes. Resident characteristics were obtained using the Minimum Data Set and Part D claims. Multivariable competing risk regression was used to model 2-year risk of hip fracture. We validated the model in a remaining 1/3 sample (n = 209,834) and in a separate cohort in 2011 (n = 858,636).

Results: Mean age was 84 years (range 65-113 years) and 74.5% were female. During 1.8 years mean follow-up, 14,553 residents (3.5%) experienced a hip fracture. Fifteen characteristics in the final model were associated with an increased risk of hip fracture including dementia severity, ability to transfer and walk independently, prior falls, wandering, and diabetes. In the derivation sample, the concordance index was 0.69 in men and 0.71 in women. Calibration was excellent. Results were similar in the internal and external validation samples.

Conclusions: The FRAiL model was developed specifically to identify NH residents at greatest risk for hip fracture, and it identifies a different pattern of risk factors compared with community models. This practical model could be used to screen NH residents for fracture risk and to target intervention strategies.

Figure 1.

Diagram of the selection of long-stay nursing home residents included in the derivation sample and the internal and external validation samples of the FRAiL model. * CPS = Cognitive performance scale; MDS = Minimum data set.

Figure 2.

Receiver operating characteristic curve of (A) the FRAiL model for women, and (B) the FRAiL model for men in the derivation sample.

Figure 3.

Calibration plots comparing the observed to expected probabilities of hip fracture in long stay nursing home residents across deciles of predicted risk: (A) Women in the derivation sample, (B) Men in the derivation sample, (C) Women in the external validation sample, and (D) Men in the external validation sample.