Assessing Risk of Osteoporotic Fractures in Primary Care: Development and Validation of the FRA-HS Algorithm

We aimed to develop and validate the FRActure Health Search (FRA-HS) score for prediction of risk of osteoporotic fractures in primary care in Italy. We selected a cohort of patients aged 40 years between 1999 and 2002. They were followed until the occurrence of osteoporotic fracture, death, end of data registration, or end of data availability (December 31, 2012). Age, sex, history of osteoporotic fractures, secondary osteoporosis, long-term use of corticosteroids, rheumatoid arthritis, body mass index, smoking, and alcohol abuse/alcohol-related diseases, and the interaction terms sex*use of corticosteroids and age*secondary osteoporosis were entered in a competing-risk regression (Fine and Gray method) to predict the risk of hip/femur or overall major osteoporotic fractures. The coefficients were combined to obtain the FRA-HS for individual patients. Explained variance, discrimination, and calibration measures were computed to evaluate the models accuracy. The final model was tested using an independent data source. The FRA-HS explained 47.36 and 20.6% of the variation for occurrence of hip/femur and overall major osteoporotic fractures, respectively. Area Under Curve was 0.77 and 0.73, respectively. Predicted/observed ratios revealed a margin of error lower than 30% in the 80% of the population. After stratifying by sex, prediction models for hip/femur fractures confirmed acceptable accuracy in both sexes, while poor explained variance (<20%) was observed for overall major fractures. These findings indicate that FRA-HS might be implemented in primary care for risk prediction of hip/femur fractures. General practitioners could be therefore supported by this tool in clinical decision making.