Source of data and study population
General practitioners (GPs) play an essential role in the public health care system of Spain, as they are responsible for primary health care, long-term prescriptions and specialist and hospital referrals. The Spanish public health care system covers more than 98% of the population. The data in this study was obtained from the SIDIAP Database, comprised of electronic medical records of a representative sample of patients attended by GPs in Catalonia (North-East Spain), covering a population of more than 5.8 million patients (about 80% of the total of 7.5 million population of Catalonia) from 274 primary care practices with 3,414 participating GPs. The SIDIAP data comprises the clinical and referral events registered by primary care health professionals (GPs and nurses) and administrative staff in electronic medical records, comprehensive demographic information, prescription and corresponding pharmacy invoicing data, specialist referrals, primary care laboratory test results, and hospital admissions and their major outcomes. Health professionals gather this information using ICD-10 codes, and structured forms designed for the collection of variables relevant for primary care clinical management, such as country of origin, sex, age, height, weight, body mass index, tobacco and alcohol use, blood pressure measurements, blood and urine test results. Only GPs who meet quality control standards can participate in the SIDIAP database . Encoding personal and clinic identifiers ensures the confidentiality of the information in the SIDIAP Database. Recent reports have shown the SIDIAP data to be useful for epidemiological research .
All patients aged ≥50 years registered in the database in 2009 were eligible for the current study.
Ascertainment of fractures
Fractures registered in 2009 in the SIDIAP database were identified using medical codes for a list of skeletal sites of fracture (Additional file 1), based on the ICD-10 classification. Fractures considered for these analyses were those defined by Center and Eisman  as major fractures based on their associated mortality (hip, clinical spine, pelvic, multiple rib and proximal humerus), and the most prevalent minor osteoporotic fracture in our data (wrist/forearm). We could not tease out high impact fractures, as we had no access to free text contained in medical records for confidentiality reasons.
Validation of fractures in the SIDIAP database
To assess the completeness and accuracy of the fractures coded in the SIDIAP database, we linked and compared our data to ARTPER data, a population-based prospective cohort study that has been ongoing in 28 primary care centres of Catalonia for the last 4 years [19, 20]. The ARTPER study included a random sample of 3,786 individuals aged >49 years, and was powered to estimate the prevalence of peripheral arterial disease in the general population and to study its association with cardiovascular outcomes. Further, a question on the occurrence of hip, spine and wrist/forearm fractures (based on the EPOS study questionnaire ) was asked of all ARTPER participants in the 4-year follow-up phone questionnaire in an effort to investigate a potential association between cardiovascular disease and fragility fractures. A total of 3,775 patients (99.7% of the initially recruited) answered the question “Since participating in the baseline survey, have you fractured a bone?”. If the response was “yes”, they were asked to report the date of fracture and to identify which bone(s) they broke, choosing from the following categories: spine, hip/femur, wrist/forearm, and other. Of these reports, 3,402 (90.1%) could be verified and linked to the SIDIAP database. A new dataset with anonymized ID and data on patient-reported (from ARTPER) and physician-registered (in SIDIAP records) fractures was constructed for our analysis of the sensitivity, specificity, and positive and negative predictive value of a SIDIAP code for fracture compared to the ARTPER cohort study. In addition, hip fractures in the SIDIAP database were validated by linkage to the 2009 regional hospital admission database (CMBD), which was considered as a gold standard for comparison. We considered hospitalization for hip fracture only when one of the corresponding ICD-9 codes appeared as the primary diagnosis in the hospital admission database.
Age-specific fracture incidence rates for each fracture site were calculated separately for males and females by dividing the number of patients with a fracture by the total person-years of follow-up and plotting the result against age (incidence estimates are available from the corresponding author). The 95% confidence intervals were estimated using the delta method, as proposed by Kirkwood et al. .
For the validation of fracture codes in the SIDIAP database, crosstabs were used, with rows indicating SIDIAP fracture (yes/no) and columns indicating either ARTPER fracture (yes/no) or CMBD fracture (yes/no). Specificity, sensitivity and predictive values were estimated for each of the fracture sites studied.
All these analyses were carried out using Microsoft Excel 2008 for Mac, SPSS for Mac version 18.0 and R for Mac version 2.9.1.
SIDIAP provided purely observational data for this study. It obtained approval from the SIDIAP Scientific Committee, responsible for reviewing protocols for scientific quality. The ARTPER cohort study was approved by the local Ethics Committee (IDIAP Jordi Gol i Gurina). Informed consent was obtained from all the participants, and the recommendations of the World Medical Association Declaration of Helsinki were followed throughout the study.