Background In the busy clinic, identifying individuals at high fracture (fx) risk who warrant intervention can be a challenge. There are several medical conditions that increase the risk for bone loss and falls which are recorded in the electronic medical record (EMR).
Objectives We explored whether we could exploit data available in the EMR to estimate FRAX treatment thresholds to help passively identify patients who would benefit from further bone mineral density screening and management.
Methods We studied 912 women and men, previously recruited for our bone health studies, in whom FRAX scores (with and without BMD) had been determined and comprehensive medical diagnoses were available through the medical linkage system of the Rochester Epidemiology Project. All diagnoses were categorized by the Clinical Classification Software (CCS) system whereby over 14,000 ICD-9-CM diagnoses are reduced to 568 clinically meaningful categories. If a subject had at least two diagnoses in a CCS category that were at least 30 days apart and within 5 years of their FRAX assessment, the subject was treated as having that CCS code. We used Gradient Boosting Machine (GBM) to create models that would predict the treatment thresholds for the FRAX 10-year risk for major osteoporotic (OP) fx (>20%) and hip fx (>3%), based on available diagnoses. Models were fit using age, sex and CCS category from 80% of the data, retaining 20% for validation.
Results Of the 564 (62%) women and 348 (38%) men, the mean ± SD age was 61±16 yrs. There were no significant differences in subject characteristics used for FRAX calculation or FRAX scores between the training and validation sets. The c-statistic for GBM models predicting treatment thresholds for FRAX, calculated with BMD, for major OP fx and hip fx were 0.95 and 0.96, respectively, for the training set, and 0.88 and 0.94 for the validation set. Similar results were observed for FRAX scores without BMD.
Conclusions FRAX treatment thresholds may be reasonably estimated from data available in the EMR to help identify to the clinician those at highest risk for fracture who would warrant further evaluation. Further work to implement and validate these findings in the EMR system would be necessary.
Disclosure of Interest None declared