Subjects and setting
South Western Sydney Local Health District delivers hospital services to a population of approximately a million people, through five acute public hospitals that have approximately 230,000 admissions each year. A FLS has been introduced in the main teaching hospital of the local health district in early 2018 with planned phased role out to each facility. The source population of this study is patients presenting to hospital emergency departments across the district, aged 50-years or more, with a minimal-trauma-fracture, between January 1st 2003 and December 31st 2017. We have included women and men, aged 50+ years in our study, due to this being the worldwide practice among fracture liaison services [9].In this study we have only included fractures related to a fall from a standing height, to ensure minimal trauma was involved in the fracture event.
Ethical considerations
This project was considered by the South Western Sydney Local Health District Human Research Ethics Committee and was determined to meet the requirements of the National Statement on Ethical Conduct in Human Research (2007), and due to the use of routinely collected hospital separation data, the need for individual patient consent was waivered (SWSLHD HREC ref.: ETH03946).
Identification of minimal trauma fracture
Fractures were identified using the hospital clinical coding data, based on emergency presentations with ICD-10-AM codes S22-S82. Fracture of the face and skull, hands, digits, foot and toes, were excluded. In terms of fractures of the spine, only fractures at the lumbar spine were included, as fractures of the thoracic and c-spine are often associated with trauma. Details of specific ICD-10-codes are given in a Supplementary Table. To ensure only fractures related to minimal-trauma were included, fractures codes needed concurrent coding of a fall from a standing height (ICD-10-AM W00–18), and fractures with concurrent coding of malignancy were excluded (M84.5). Date of subsequent fracture or death were also obtained from hospital separations (episodes of care) data to calculate follow-up.
Statistical analysis
The primary outcome of this study was the time to subsequent fracture following an incident minimal-trauma-fracture, in the presence of the competing risk of death [14]. Initial fractures were classified as follows: hip, vertebral, major, and minor fractures. Major fractures included pelvis, distal femur, proximal tibia, ribs and sternum, and proximal humerus. Minor fractures included all remaining fractures, excluding those of the face, head or digits. Due to small numbers of hip fractures among patients aged < 60 years, and the relatively small numbers of lumbar-spine fractures, further analysis was undertaken by including hip and lumbar spine fractures with major fracture as proximal fractures, minor fractures are then referred to as distal fractures. Classification into distal and proximal groups was undertaken to follow some previous work, using these terms, as proximal fractures had previously been considered more serious in nature, compared to fractures of the distal skeleton [4].
Due to the presentation of rates of events, crude and adjusted relative risks of subsequent fracture based on sex, age, and site of initial fracture were estimated, and 95% confidence intervals (95%CI) using a Poisson error distribution [15]. The cumulative incidence of subsequent fracture in the presence of the competing risk of death, stratified by sex and initial fracture type, was estimated using the methods suggested by Kalbfliesch and Prentice [16]. This approach has two steps: [1] In the first step, Kaplan-Meier estimates are calculated of the overall survival from any event, in our case fracture and death, in other words both the event of interest and competing risk, respectively; and [2], in the second step the conditional probability of experiencing the event of interest, having avoided fracture and death, up until this point in time [17], in other words we have ensured that the risk of subsequent fracture has not been biased due to considering loss to follow-up among patients who have died, having a similar effect on the estimated absolute risk of subsequent fracture, as pateints who were alive at the end of the study period. Absolute risk based on sex, age, and site of initial fracture was estimated using the survival R package [18]. Verification of the proportional hazards assumption of the Cox models was based on a visual inspection of smoothed Schoenfeld residual plots [19].
