Correction factor for the analysis of the hip fracture incidence--differences between age, sex, region, and calendar year.

Several studies evaluated hip fracture incidences and its predictors and trends using hospital discharge registries. However, this source does not provide patient-related data, therefore the hospital changes or re-hospitalisations cannot be identified as "double counting". If double counting differs with age, sex, region, and time, the estimates may be biased. Aim of our study was to evaluate the magnitude of multiple counting and, in particular, its variation with age, sex, region, and calendar year. We used data of a German-wide health insurance (1.6 million members). Between 1998 and 2009, we assessed all hip fractures (ICD 9: 820, ICD 10: S.72.0-2) in individuals aged 50 years or older and calculated the probability to be a patient's "first" fracture in each calendar year. Using multiple logistic regressions, we estimated the influence of age, sex, region, and calendar year. The probabilities of a "first fracture" per patient and year varied between 86.7 % (95 % confidence interval 83.9-89.2 %, year 2003) and 93.9 % (90.9-96.2 %, year 1998). Age (odds ratio per 5 years 0.89; 95 % CI 0.86-0.92), region (East vs. West Germany: 0.65; 0.52-0.81), and calendar year (per year 0.97; 0.95-0.99) were significantly associated in the multiple regression. The probability to have multiple counting of hip fracture events varied significantly with age, region, and calendar year. It should be discussed that analyses which do not account for this may provide invalid estimates and conclusions when differences between age groups and regions or trends are analyzed.