Age at diabetes diagnosis and fracture risk in women: a cohort study from the Australian Longitudinal Study on Women’s Health

The Australian Longitudinal Study on Women’s Health (ALSWH) is a population-based longitudinal cohort study examining women’s health and health service use across the lifespan. The 1946–1951 cohort, consisting of 13,714 women born between these years, was recruited in 1996 through stratified random sampling of the Medicare database. Medicare is Australia’s universal health insurance system, covering all citizens and permanent residents. Comparisons with the 1996 Australian Census on a range of sociodemographic variables indicate that the cohort is broadly representative of Australian women of the same age [14]. This cohort has maintained a high level of participation over time, with 73.7% of eligible women responding to Survey 10 in 2022, 26 years after the initial survey.

This cohort has been surveyed every 3 years to collect data on health, lifestyle, and social factors, with additional linkage to administrative health datasets. These include the Medicare Benefits Schedule (MBS), which covers doctor visits and selected investigations and treatments, the Pharmaceutical Benefits Scheme (PBS) for prescriptions for subsidized medicines, hospital admissions, hospital emergency data, aged care records, and cause of death records. Survey data covers ages from 45 to 76 years (1996–2022), while some linked administrative datasets predate the surveys, with MBS records available from 1984, certain hospital data from 1970, and some aged care data from 1982. This integration provides robust longitudinal data essential for exploring health and aging in midlife and older Australian women. Further information on the ALSWH can be found elsewhere [15].

Only women who consented to data linkage were considered for inclusion in the analysis (n = 12,951, 94.4% of the original sample). Of these, women who completed only the first survey were excluded due to having no follow-up data (n = 642), and those with identical dates for diabetes diagnosis (i.e., exposure variable) and fracture (i.e., outcome variable) were excluded (n = 139). The final analytical sample consisted of 12,170 women.

A diagnosis of diabetes was identified using multiple data sources, including self-reported information from the ALSWH surveys and linked administrative health records. Detailed information on the data sources and eligibility criteria for defining diabetes is provided in the Supplementary Material (Supplementary Tables S1–S6). Diagnoses include both type 1 and type 2 diabetes but exclude gestational diabetes. The date of diabetes diagnosis was determined as the earliest date for a record of diabetes across all available data sources, and this was used to calculate age at diagnosis based on the participant’s date of birth.

A diagnosis of fracture was identified using multiple data sources, including self-reported information from ALSWH surveys and linked administrative health records. Detailed information on the data sources, eligibility criteria, and item codes used to define fractures is provided in the Supplementary Material (Supplementary Table S7).

BMI was included as a potential confounder, given its well-established association with both diabetes and fractures [16]. It was derived from self-reported weight and height provided at each survey and treated as a time-varying covariate to capture changes over the extended follow-up period. This approach provided a more accurate representation of BMI’s impact on outcomes than static baseline values.

BMI was modeled categorically: underweight (BMI < 18.5 kg/m²), normal weight (≥ 18.5 and < 25 kg/m²), overweight (≥ 25 and < 30 kg/m²), and obese (≥ 30 kg/m²), consistent with WHO classification [17].

Descriptive statistics were used to summarize key characteristics of the study population.

Participants were followed from age 27—the earliest age at which a diabetes diagnosis was recorded—until the occurrence of the outcome (fracture), death, or June 2022 (the latest date with reliable death data), whichever occurred first. To compare rates of fracture between women with or without diabetes over the follow-up period, a specialized Poisson regression model was employed [18, 19]. This modeling approach offers several advantages over a Cox model for time-to-event data. For the current analysis, it simplified the inclusion of time-varying covariates (e.g., age and BMI), enabled modeling of outcome rates (e.g., fracture rates per 1000 person-years), and allowed estimation of rate ratios, which are generally easier to interpret than hazard ratios.

For this method, the follow-up time for each participant is specified by a series of records for short intervals (e.g., single years). Each record includes the values of every variable to be used in the model. The data structure for this analysis is illustrated in Supplementary Table S8. For each woman, the records start when she was aged 27. Her age, BMI, and diabetes status refer to the beginning of the interval. If a woman started without a diagnosis of diabetes, her diabetes status was coded zero and her age at diagnosis was set to zero for every interval unless she was diagnosed with diabetes, at which time her diabetes status code switched to one and her age at diagnosis was set to her age at the time of diagnosis, and these values were retained for all subsequent intervals. Women who were diagnosed with diabetes at age 27 (the earliest recorded diagnosis) were classified as having diabetes from study entry. The duration of the interval for each record is included as it is used in the calculation of exposure time. Fractures were coded zero or one, with participants censored (i.e., no further records) after the time of their first fracture.

BMI was included as a time-varying categorical variable. As BMI data were only available from age 45–50 years onward, missing values for earlier ages were imputed using self-reported height and weight at Survey 1, along with recalled highest and lowest prior weights. These prior weights were collected at Survey 1, where participants were asked to recall their highest weight at any previous age and their lowest weight since age 18. Assumptions were made about the timing of these prior weights, as detailed in Supplementary Table S9.

To explore the individual effects of key covariates, two initial models were fitted. The first included age and diabetes status, and the second included age and BMI (categorized as described above). These minimally adjusted models allowed the independent association of each variable (i.e., diabetes and BMI) with fracture to be assessed prior to their inclusion in the more fully specified models. Age was modeled using natural splines with six knots, placed to ensure an approximately equal number of outcome events between them.

Two further models were then developed. The base model included the time-dependent effects of age, diabetes status, and BMI. The age-at-diagnosis model built on the base model by including age at diabetes diagnosis as an additional covariate.

As a sensitivity analysis, we excluded women likely to have type 1 diabetes—defined as those diagnosed before age 50 years and treated only with insulin—an approach based on similar age-and-treatment criteria used in previous epidemiological studies to identify probable type 1 cases when definitive classification is unavailable [20]. As a further sensitivity analysis, we repeated the primary models using weight (kg) as a time-varying covariate in place of BMI. Nonlinearity was assessed by comparing restricted cubic splines with a linear term.

All statistical analyses were conducted using R version 4.3.0 (2023–04–21), utilizing the survival, Epi, and popEpi packages.

The baseline characteristics of the study cohort are shown in Table 1. The number of women remaining in the cohort, the prevalence of diabetes, distribution of BMI, and the proportion of women who had sustained at least one fracture by ages 30, 40, 50, 60, and 70 years are shown in Table 2. Over the analysis period, 883 (7.3%) of the women died. In total, 2251 (18.5%) women had a report of diabetes during the study period. The earliest age of a diabetes report was 27 years, with half of the reports occurring between the ages of 52 and 63 years. Overweight and obesity prevalence increased with age.

A total of 3761 women (30.9%) experienced at least one fracture during the period. The age at first fracture ranged from 33.1 to 76.7 years, with a median age of 62.8 years (Q1: 56.9, Q3: 68.6).

Age-specific fracture rates are shown in Fig. 1, based on minimally adjusted models (adjusted for age only), and stratified by BMI category (Fig. 1a) and diabetes status (Fig. 1b). These plots illustrate the increasing rate of fractures with age and depict the higher rates observed in women who were underweight (Fig. 1a) and those with diabetes (Fig. 1b). In the corresponding regression models, adjusted only for age, diabetes was associated with a modest increase in fracture risk (RR 1.12, 95% CI 1.02–1.24), while no association was observed between BMI category and fracture risk (Table 3).

Bild vergrößern

In the base model, which included age, BMI, and diabetes status, the association between diabetes and fracture risk remained (RR 1.13, 95% CI 1.02–1.24).

The age-at-diagnosis model revealed that the excess fracture risk in individuals with diabetes diminished as age at diagnosis increased (Table 3). For example, women diagnosed with diabetes at age 35 years had a 61% higher fracture risk (RR 1.61, 95% CI 1.23–2.12) than those without diabetes. The RR declined to 1.12 (95% CI 1.01–1.24) for those diagnosed with diabetes at 55 years, and beyond this age, the confidence interval included the null value of 1.

Overall, the RR of fracture in individuals with diabetes relative to those without diabetes decreased as age at diabetes diagnosis increased, transitioning from a significantly elevated risk at younger ages to no evident association at older ages. Figure 2 illustrates this trend, showing that individuals diagnosed with diabetes later in life had progressively lower excess fracture rate, with estimates approaching those of those without diabetes.

Bild vergrößern

In sensitivity analyses excluding women classified as likely to have type 1 diabetes (n = 20), results were essentially unchanged, with relative risks and age-at-diagnosis patterns remaining virtually identical to the main analysis. Results were also unchanged when weight (kg) was used in place of BMI as the adjustment variable. In the age-at-diagnosis model, weight itself was not associated with fracture risk (RR per 5-kg increase = 1.00, 95% CI: 0.99–1.01), and there was no evidence of nonlinearity when comparing splines with a linear term (p = 0.90).

This study found that diabetes was associated with an increased risk of fracture, and age at diabetes diagnosis modified this association, with younger age at diagnosis corresponding to a higher rate of fracture.

The relative risk was highest for those diagnosed at younger ages and gradually declined with increasing age at diagnosis. For women diagnosed with diabetes at age 60 or later, there was little evidence of increased fracture risk compared to women without diabetes.

Previous studies have consistently reported an association between diabetes and increased fracture risk, with meta-analyses estimating relative risks between 1.05 and 1.50 [6, 7, 9, 21]. Our result (RR 1.13, 95% CI: 1.02–1.25) falls within this range, reinforcing existing evidence.

Some research has examined diabetes duration, generally finding that longer duration is associated with higher fracture risk [22‐25]. However, studies report differing patterns—some indicating threshold effects [24], while others describe dose–response trends (based on point estimates, though these are not always statistically significant at earlier time points) [12, 26, 27] —highlighting that duration may not fully capture the complexity associated with the timing of diabetes diagnosis.

To our knowledge, no previous studies have conducted an in-depth examination of age at diabetes diagnosis in relation to fracture risk. By focusing on age at diagnosis, our study offers a novel perspective on risk variation among individuals with diabetes. While conceptual overlap exists between age at diagnosis and disease duration, these are not interchangeable. Prior research has focused on duration because it fits conventional models of cumulative disease burden. In contrast, age at diagnosis may capture important differences in disease phenotype and trajectory, offering an alternative lens through which to understand fracture risk in diabetes.

While diabetes is increasingly recognized as a risk factor for fractures, our results suggest that fracture risk is not uniform across all individuals with diabetes and may depend on when the disease develops. These findings highlight the importance of considering the timing of onset, rather than treating diabetes as a single, homogenous risk factor.

The higher relative rate of fracture observed in individuals diagnosed with diabetes at younger ages may be partly explained by a more severe metabolic phenotype. Type 1 diabetes, which is more common among younger individuals, is associated with greater fracture risk than type 2 diabetes [28], likely due to absolute insulin deficiency, lower bone mineral density, and impaired bone quality [29‐31]. Although often considered a childhood-onset disease, a substantial proportion of type 1 diabetes cases are diagnosed in adulthood—over one-third occur after age 30 [32]. In our study, a sensitivity analysis excluding women classified as likely to have type 1 diabetes showed negligible impact on the findings, suggesting that the observed higher risk in younger-diagnosis groups is not solely attributable to type 1 diabetes. Nonetheless, precise differentiation between diabetes types was not possible with our data, and some misclassification is likely.

In addition to type 1 diabetes, early-onset type 2 diabetes may also contribute to increased fracture risk. Individuals diagnosed with type 2 diabetes at younger ages often experience a more aggressive disease course, with poorer glycaemic control, earlier complications, and a greater burden of cardiovascular disease, neuropathy, and nephropathy [33]. This more severe metabolic profile may extend to skeletal health, predisposing individuals to increased fracture risk. Further research is needed to clarify the mechanisms involved.

Another important factor is the longer lifetime exposure to diabetes-related metabolic and vascular effects in those diagnosed at younger ages. Prolonged hyperglycaemia may impair bone quality through mechanisms such as increased advanced glycation end-products and reduced bone turnover [10, 11]. Longer disease duration is also associated with a higher prevalence of complications such as neuropathy, visual impairment, and insulin use—all of which contribute to fall risk, with insulin use in particular linked to hypoglycaemia-related falls [34, 35].

While disease severity and lifetime exposure likely explain much of the increased fracture risk in younger-onset diabetes, other factors may also contribute. Differences in lifestyle—such as physical activity, diet, or other health behaviors—were not assessed in this study but could influence bone health across age groups. Treatment differences may also play a role; individuals diagnosed at younger ages may have earlier insulin initiation or distinct medication regimens, which could affect bone metabolism or fall risk. Although treatment effects were not examined, variation in management may contribute to the observed risk patterns.

The findings of this study have important implications for fracture risk assessment in individuals with diabetes. Widely used tools such as FRAX [36] treat diabetes as a binary risk factor—limited to type 1—and do not account for the timing of diagnosis. As prediction models evolve to incorporate more individualized factors, age at diabetes diagnosis could be considered to better reflect fracture risk variation among individuals with diabetes.

Another important clinical consideration is the need for earlier screening and intervention strategies for bone health in individuals diagnosed with diabetes at younger ages. Our findings suggest that fracture risk may be elevated well before older age, reinforcing the importance of early fracture risk assessment and targeted prevention in this group.

This study has several notable strengths. Firstly, it draws on a large, high-quality longitudinal dataset from a population-based cohort, with follow-up extending from early adulthood into the 1970s. The combination of repeated surveys and linked administrative records enabled extended outcome tracking and improved diabetes ascertainment. Low attrition in the survey data further strengthened reliability and generalizability.

Secondly, we employed a specialized Poisson regression model with several methodological advantages. This framework allowed fracture rates to be modeled across the life course and incorporated time-varying covariates such as diabetes status and BMI, ensuring that exposure classifications reflected changes over time rather than relying on static baseline values. Few prior studies have modelled diabetes status as a time-varying exposure, underscoring the analytical strength of this approach.

Thirdly, this study offers novel insight by examining age at diabetes diagnosis as a modifier of fracture risk—an underexplored aspect in previous research. While most studies focus on diabetes duration, we examined the timing of disease onset, treating diabetes as a dynamic exposure across a wide age range.

Several limitations should also be considered. Firstly, for some participants—particularly those diagnosed before midlife, before the start of survey data collection—diabetes onset was determined using administrative data alone. As administrative records are less complete in earlier years, the timing of diagnosis may be less reliable for these cases, and diagnoses occurring in childhood or early adulthood may not be systematically captured.

Secondly, due to the indolent nature of type 2 diabetes, some individuals may have had undiagnosed hyperglycaemia for years before clinical recognition. Skeletal changes may therefore have developed before formal diagnosis—an inherent limitation in diabetes research.

Thirdly, distinguishing between type 1 and type 2 diabetes with certainty was not possible in this dataset, given the absence of consistent clinical classification across all data sources. While we applied an age-at-diagnosis and treatment-based definition in a sensitivity analysis, the small number of women identified as likely type 1 diabetes and the negligible impact on results suggest our findings are unlikely to be materially biased. Nevertheless, some misclassification between types is probable, and the relative contribution of each subtype to fracture risk at different diagnosis ages remains uncertain.

Fourthly, BMI was self-reported, introducing potential measurement error. As is commonly observed in self-reported data, participants may have underestimated their weight or overestimated their height, leading to BMI misclassification. Early-life BMI estimates were imputed using recalled weight histories and assumptions about timing, which may not reflect individual weight trajectories. However, any misclassification is unlikely to have substantially influenced the results, given the robustness of the overall findings. As with all observational studies, residual confounding cannot be excluded. Although we adjusted for age and BMI, other factors such as sociodemographic or lifestyle characteristics may also have influenced the observed associations.

Finally, generalizability should be considered. This study included only women; fracture risk patterns may differ in men. The cohort was also drawn from Australia, so findings may not fully apply to populations with different healthcare systems, diabetes prevalence, or fracture risk profiles. Additionally, follow-up was censored at age 76, limiting insight into fracture risk at older ages when diabetes-related fractures may be even more pronounced.

This study is the first to directly examine the modifying effect of age at diabetes diagnosis on fracture risk. While the findings offer important insights, further research is needed to confirm them in other populations and study designs. Larger samples of those diagnosed before midlife are needed to improve the precision of risk estimates.

Future studies should aim to more reliably differentiate between type 1 and type 2 diabetes, as their differing effects on bone health may influence fracture risk patterns. Achieving this will require datasets with more detailed clinical information to improve classification and sufficient sample sizes, particularly for type 1 diabetes, which is less common in population-based cohorts.

Extending follow-up beyond age 76 is another priority, given that fracture risk continues to rise with age; research is needed to determine whether the patterns observed here persist or change beyond age 80.

FRAX, the most widely used fracture risk prediction tool, currently incorporates only type 1 diabetes as a binary risk factor. It has been widely acknowledged that FRAX underestimates fracture risk in individuals with type 2 diabetes [37], and efforts are underway to address this limitation. As FRAX evolves, there may be opportunities to incorporate diabetes-specific variables that better reflect fracture risk heterogeneity. Age at diabetes diagnosis could be a useful addition, particularly given the differential risks observed in this study. Further work is needed to assess its predictive value relative to disease duration.

In the longer term, it may be valuable to investigate whether individuals diagnosed with diabetes at younger ages would benefit from earlier or more-targeted fracture prevention.