(P-026) Assessment of Predictors of Fractures in German Older Adults Using a Large Population Sample

Background: In older adults fractures are associated with mortality, disability, loss of independence and high health care costs. Knowledge of predictors of fractures can assist physicians, pharmacists and regulators to recognize the need for measures to prevent fractures in older adults.

Objectives: To assess predictors of fractures in German older adults using a large population sample

Methods: We used the German Pharmacoepidemiological Research Database (GePaRD) with health insurance claims data from ~20% of the German population. We included all persons aged 65 years and older with ≥1 year continuous insurance coverage in 2016 and no records of fracture any time prior to 1st January 2017. Potential predictors were selected based on literature and considered as far as available in claims data. Use of medication and healthcare utilization was assessed in 2016; morbidities were assessed any time prior to 1st January 2017. Fractures were assessed from January 1st to December 31st 2017. For the calculation of logistic regression and random forest models the study population was randomly divided into a training (80%) and test sample (20%). A number of models with different combinations of predictors were fitted. Analyses were stratified by age groups to provide age group specific predictors of fractures

Results: Among 2,426,286 persons (54% Female), the incidence of any fractures was 180.4 per 10,000 person years (178.7-182.1). The incidence increased with age and was higher in men than in women: it was 82.8 (men) and 131.6 (women) in 65-74 year olds, 149.7 (men) and 252.1 (women) in 75-84 year olds and 379.7 (men) and 601.0 (women) in persons aged ≥85 years. The maximum of predictive performance as measured by the area under the curve (AUC) across models was 0.65. According to Youden’s statistics, this model showed, for example, in age group 75-84 a sensitivity of 61% at a specificity of 60%. The consideration of medication in addition to diagnosis codes had only a minor impact on predictive performance.

Conclusions: Our study suggests that there is some potential of predicting fractures in older adults based on German health claims data. It seems worthwhile to explore further options (e.g. data-driven variable selection) to increase predictive power in order to realize the potential value of this data source for the development of preventive strategies