Participants were from the Geelong Osteoporosis Study [8], a longitudinal cohort study situated in south-eastern Australia. Data for this study were drawn from the baseline assessment for women (1993–1997). At baseline, 1494 women aged 20–94 years participated. There were 1053 women aged 40–90 years, and of these, 1026 had ultra-distal forearm BMD measured. The age range of 40–90 years was selected in this study because fractures are most common in this group and is the age range used by the FRAX algorithm for fracture risk prediction.

BMD at the non-dominant ultra-distal forearm (BMDUDforearm), 33% forearm (BMD33%forearm), femoral neck (BMDhip), and lumbar spine (BMDspine) were measured using a Lunar DPX-L (Lunar; Madison, WI, USA). T-scores were also calculated for each skeletal site using the young normal reference range developed by the Geelong Osteoporosis Study for use in an Australian setting [9, 10].

Weight and height were measured to the nearest 0.1 kg and 0.001 m, respectively. Body mass index (BMI) was calculated as weight(kg)/height(m)2. Biochemical data for serum albumin, serum calcium, and vitamin D were also obtained by analysis of blood samples collected after an overnight fast.

The majority of other measures included in this study were selected as they are included in the FRAX algorithm [7] and are well-known risk factors for fracture. Prior fractures were self-reported and excluded those of the face, skull, digits, and those occurring from high trauma. Radiological reports were used to confirm fractures where possible. The remaining measures were self-reported. Participants reported whether their parents had previously sustained a hip fracture. Smoking status was classified as current or not. Alcohol consumption was documented by self-report. High alcohol consumption was categorised as ≥ 30 g of alcohol per day. Secondary osteoporosis included type 1 diabetes, osteogenesis imperfecta, hyperthyroidism, premature menopause (< 45 years), chronic malnutrition, malabsorption, and chronic liver disease. These were self-reported except for malnutrition, which was classified as BMI < 18.5 kg/m2, following a previously published method [11]. Participants also reported if they had fallen during the past 12 months and which medications they used, including glucocorticoids, bisphosphonates, hormone therapy, and calcium or vitamin D supplements. FRAX 10-year probability risk estimates without BMD for major osteoporotic fracture (FRAXMOF) and hip fracture (FRAXhip) were also calculated for each participant using the Australian version of FRAX [7].

Mortality was identified by data linkage with the National Deaths Index.

Incident fractures were verified by examination of radiological reports from imaging centres across the region. Fractures of the skull, face, and digits were excluded. Those occurring by high trauma, such as a motor vehicle accident, were also excluded. Participants who sustained a distal radius fracture were also identified.

Continuous variables were presented using means and standard deviations (SD) or medians with interquartile range (IQR) as appropriate. Categorical variables were presented as n (%). Differences for participants with and without incident fracture were identified using two-sample t tests or Mann–Whitney tests for continuous variables and chi-square tests for categorical variables.

Scatterplots were generated to visualise the relationship between age and BMDUDforearm T-scores. This was also performed for T-scores at the other skeletal sites (BMD33%forearm, BMDhip, BMDspine). Another set of scatterplots was also generated to examine the relationship between T-score for BMDUDforearm and T-scores at the other skeletal sites.

Participants were followed from baseline to the date of the first fracture, date of death, or the end of the study period (31 December 2016), whichever occurred first. Cox proportional hazard models were used for multivariable (adjusted) survival analysis. The following variables were tested in the models and were retained if p < 0.05: age, weight, height, prior fracture, parental hip fracture, smoking, alcohol consumption, secondary osteoporosis, rheumatoid arthritis, falls, glucocorticoids, bisphosphonates, hormone replacement therapy, calcium supplements, vitamin D supplements, serum albumin, calcium, and vitamin D concentration. Models for FRAXMOF and FRAXhip were adjusted only for falls, bisphosphonates, hormone replacement therapy, calcium supplements, vitamin D supplements, serum albumin, calcium, and vitamin D concentration because the other variables are already accounted for in FRAX. The analyses were performed for BMDUDforearm as a continuous variable, and also as a categorical variable, employing osteopenia (T-score <  − 1.0 and ≥  − 2.5) and osteoporosis cut points (T-score <  − 2.5). For continuous BMD, hazard ratios from the Cox proportional hazards modelling were calculated to show the increase in fracture risk with a one-unit decrease in BMD T-score.

Additionally, areas under receiver operating characteristics (AUROC) curves were calculated for BMDUDforearm, BMD33%forearm, BMDhip, BMDspine, FRAXMOF, and FRAXhip. In categorical analyses, cutpoints of ≥ 20% and ≥ 3% were used for FRAXMOF and FRAXhip, respectively, according to US National Osteoporosis Foundation guidelines [12].

All analyses were conducted with two different incident fracture outcomes: (1) any fracture and (2) distal radius fractures only.

Analyses were completed using Stata (Version 17. StataCorp. 2017. Stata Statistical Software: Release 17. College Station, TX: StataCorp LLC) and Minitab (Minitab, version 19, State College, PA, USA).