BackgroundStratifying residents at increased risk for fractures in long-term care facilities (LTCFs) can potentially improve awareness and facilitate the delivery of targeted interventions to reduce risk. Although several fracture risk assessment tools exist, most are not suitable for individuals entering LTCF. Moreover, existing tools do not examine risk profiles of individuals at key periods in their aged care journey, specifically at entry into LTCFs. PurposeOur objectives were to identify fracture predictors, develop a fracture risk prognostic model for new LTCF residents and compare its performance to the Fracture Risk Assessment in Long term care (FRAiL) model using the Registry of Senior Australians (ROSA) Historical National Cohort, which contains integrated health and aged care information for individuals receiving long term care services. MethodsIndividuals aged ≥65 years old who entered 2079 facilities in three Australian states between 01/01/2009 and 31/12/2016 were examined. Fractures (any) within 365 days of LTCF entry were the outcome of interest. Individual, medication, health care, facility and system-related factors were examined as predictors. A fracture prognostic model was developed using elastic nets penalised regression and Fine-Gray models. Model discrimination was examined using area under the receiver operating characteristics curve (AUC) from the 20 % testing dataset. Model performance was compared to an existing risk model (i.e., FRAiL model). ResultsOf the 238,782 individuals studied, 62.3 % (N = 148,838) were women, 49.7 % (N = 118,598) had dementia and the median age was 84 (interquartile range 79–89). Within 365 days of LTCF entry, 7.2 % (N = 17,110) of individuals experienced a fracture. The strongest fracture predictors included: complex health care rating (no vs high care needs, sub-distribution hazard ratio (sHR) = 1.52, 95 % confidence interval (CI) 1.39–1.67), nutrition rating (moderate vs worst, sHR = 1.48, 95%CI 1.38–1.59), prior fractures (sHR ranging from 1.24 to 1.41 depending on fracture site/type), one year history of general practitioner attendances (≥16 attendances vs none, sHR = 1.35, 95%CI 1.18–1.54), use of dopa and dopa derivative antiparkinsonian medications (sHR = 1.28, 95%CI 1.19–1.38), history of osteoporosis (sHR = 1.22, 95%CI 1.16–1.27), dementia (sHR = 1.22, 95%CI 1.17–1.28) and falls (sHR = 1.21, 95%CI 1.17–1.25). The model AUC in the testing cohort was 0.62 (95%CI 0.61–0.63) and performed similar to the FRAiL model (AUC = 0.61, 95%CI 0.60–0.62). ConclusionsCritical information captured during transition into LTCF can be effectively leveraged to inform fracture risk profiling. New fracture predictors including complex health care needs, recent emergency department encounters, general practitioner and consultant physician attendances, were identified.