Site-Specific Fracture Incidence Rates Among Patients With Type 1 Diabetes, Type 2 Diabetes, or Without Diabetes in Denmark (1997–2017)

According to the most recent estimates, ∼463 million people are living with type 1 and type 2 diabetes worldwide, with type 2 diabetes accounting for ∼90% of all patients with diabetes (1). Importantly, the number of patients with type 1 and type 2 diabetes is predicted to increase rapidly in the coming decades (1,2). Several studies have reported an increased risk of fragility fractures among patients with type 1 and type 2 diabetes (3,4), with most focusing on the total (nonspecific) fracture risks or major osteoporotic sites of the hip or vertebrae (5). Additionally, patients with type 1 and type 2 diabetes have a higher risk of postfracture complications, prolonged healing of fractures, and higher mortality after a hip fracture (6,7).

Although type 1 and type 2 diabetes share factors that influence fracture risk, including diabetic complications of the microvascular system, e.g., retinopathy, neuropathy, and nephropathy (8) that may impair bone formation (9) and increase the risk of falls (10), differences may account for variable fracture risks. Decreased bone mineral density (BMD) is the most common risk factor for an osteoporotic fracture among elderly individuals. However, there is a paradox in patients with diabetes, where patients with type 1 diabetes typically have low BMD in adulthood, whereas BMD is normal or slightly increased in patients with type 2 diabetes (4). Thus, given the differential BMD profile, other factors may influence fracture risk. For example, bone composition and skeletal fragility may play an important role, as patients with type 1 or 2 diabetes have lower bone turnover (9) and disease-specific differences in the bone microstructure (11).

Finally, the past two decades have seen substantial changes in the pharmaceutical management of type 1 and type 2 diabetes (12). Longer-acting insulins and continuous administration of insulin by pumps are now commonly used to treat type 1 diabetes (13), presumably resulting in more stable glucose levels and lower risk of hypoglycemia (13), which can lead to lower fracture risk. Furthermore, glucagon-like peptide 1 receptor agonists and sodium–glucose cotransport 2 inhibitors have been implemented as treatments of type 2 diabetes and are not associated with a lower bone quality or increased risk of fracture (14). Simultaneously, sulfonylureas are associated with hypoglycemia and increased fracture risk (5) and are becoming less commonly used in type 2 diabetes (15). Moreover, glycated hemoglobin levels have decreased over time in Denmark both in patients with type 1 and type 2 diabetes, suggesting that the patients have more well-regulated glycemia over time (13), which could cause fewer hypoglycemic episodes, reducing fracture risk (16).

While previous studies have found that major osteoporotic fracture rates are stable or even have a declining trend in the general population (17), it is unknown whether the incidence in fractures among patients with type 1 or type 2 diabetes have changed. More specifically, it is largely unknown whether the incidence of specific fractures differs among patients with type 1, type 2, and without diabetes. First, we investigated the trends in the fracture incidence rates (IRs) in the Danish population between 1997 and 2017, and second, we assessed the differences in IRs for patients with diabetes compared with patients without diabetes. Third, we calculated IR ratios (IRRs), both unadjusted and age adjusted, as a sensitivity analysis.

This observational study used data from the Danish National Patient Register (DNPR) to identify all patients in Denmark with a bone fracture between 1997 and 2017. This database is also known as the National Hospital Discharge Register, covering all inpatient contacts from 1977 onward. Since 1995, it has also included information on all outpatient contacts to hospitals, outpatient clinics, and emergency departments. Since 1994, the register has coded diagnoses using the ICD-10 system; before this, ICD-8 was used. As Denmark has universal coverage of health care, the hospital discharge register captures 100% of all hospital contacts. Previous studies have demonstrated high validity for fracture codes in this database (18). To identify medication use prior to fracture, we used the Danish Medicines Agency Register of Medicinal Products Statistics (RMPS), which is a nationwide prescription database for all medications dispensed at community pharmacies since 1995. The drugs are classified by Anatomical Therapeutic Chemical codes, and information is available on the date of prescription and dosage. Health care data in Denmark are linked based on the individual social security numbers assigned to all Danish residents, which is a personal 10-digit code.

We identified all patients aged ≥18 years who experienced a fracture between 1997 and 2017 in Denmark. Eligible bone fractures were identified using ICD-10 codes as vertebral, humerus, forearm, hip, ankle, and foot fractures (Supplementary Table 1). All other fractures were excluded. We identified the first fracture occurring on or after 1 January 1997. To ensure that we captured first codes for each fracture sites, we excluded fractures identified in 1997 that had an identical code within the prior 365 days. This was done to capture the date of the fracture rather than potentially identifying codes for revisions or follow-up visits. Additionally, to avoid double counting fractures during follow-up, we applied a 365-day washout period as illustrated in Supplementary Fig. 1. Thus, for all eligible fractures occurring after 1 January 1997, we assessed whether a patient had a previous code for the same fracture in the prior 365 days. For example, if an individual had a forearm fracture recorded with ICD-10 code S52 in 2005 but had a second identical code in the prior 365 days, we would not include this as an eligible fracture.

Patients with diabetes were identified using algorithms from the Danish Health Data Authority. Patients with type 1 diabetes were those registered in DNPR with ICD-10 code E10 or ICD-8 code 249 (with subcodes) or those registered in RMPS with at least one filled prescription of insulin or insulin analogs without concomitant prescriptions for medications used for type 2 diabetes (i.e., noninsulin antidiabetic medications) (Supplementary Table 3). Patients with type 2 diabetes were identified as those with a minimum of one filled prescription in the RMPS for a noninsulin antidiabetic medication (Supplementary Table 4) or with a relevant diagnosis code in DNPR (ICD-10 codes E11, E12, E13, or E14 with subcodes; ICD-8 code 250 with subcodes). Female patients treated with metformin and with a filled prescription of clomiphene or antiandrogens in combination with estrogen or female patients treated with metformin and with a diagnosis in DNPR of polycystic ovary syndrome were excluded (Supplementary Table 5).

The baseline characteristics for the study population have been chosen as the first fracture ever within the time frame of the study, where the population is split into groups of type 1 diabetes, type 2 diabetes and without diabetes. The groups of medication with the Anatomical Therapeutic Chemical code used at baseline are shown in Supplementary Table 2.

We summarized the demographic characteristics at the time of first eligible fracture between 1997 and 2017, stratified by type 1 diabetes, type 2 diabetes, and without diabetes. We assessed the use of medication within 90 days prior to the first fracture. Means and SDs or counts and proportions were used as descriptive statistics, where appropriate.

We subsequently calculated annual IRs of fractures in the Danish population (number of fractures per 10,000 person-years [PY]) and the corresponding 95% CIs. The number of eligible fractures occurring during a calendar year was divided by the number of people alive in that calendar year, as identified from Statistics Denmark (https://www.dst.dk/en/) and a database containing all patients with diabetes in Denmark from 1977 to 2017. To identify differences over time, the median IR for the first 5 years (1997–2001) was compared with the median IR for the last 5 years (2013–2017), and the percent change was calculated. We considered a change of ≥10% in the median IRs to be significant, while <10% was deemed to be a stable trend. Furthermore, linear regression models were used to investigate whether changes in IRs over time were statistically significant, and pairwise t tests were added to check for differences in the trends between groups.

Additionally, we conducted the following secondary analyses. First, we repeated the primary analysis (annual IRs) adjusted for age and sex. Regression models with a negative binomial distribution were used for patients with type 1 and type 2 diabetes, and a Poisson distribution was used for patients without diabetes. Second, unadjusted and age-adjusted Poisson models estimated the IRRs with the corresponding 95% CIs for the risk of fracture in type 1 or type 2 diabetes compared with patients without diabetes. Age was used as a categorical variable (18–29, 30–39, 40–49, 50–59, 60–69, 70–79, ≥80 years). We considered IRRs to be significant when the CI did not span 1.0. All statistical analyses were performed using SAS Enterprise version 7.15 or R version 4.0.3 software. All results were stratified by the individual fracture sites (clinical vertebral, hip, humerus, forearm, foot, and ankle).

The data sets used in the study are not available for sharing because of Danish national regulations. The code used for the study analysis can be made available upon reasonable request.

We identified 1,782,916 individuals with a fracture between 1997 and 2017 from the DNPR database (Fig. 1). Following exclusions, a total of 838,307 patients were included, of whom 3,411 (0.4%) had type 1 diabetes, 51,874 (6.2%) had type 2 diabetes, and 783,022 (93.4%) without diabetes.

Baseline characteristics of the study population were summarized in Table 1, stratified by diabetes status. Mean age was 52.9 (SD 19.7) years for patients with type 1 diabetes, 70.2 (SD 14.0) years for patients with type 2 diabetes, and 57.2 (SD 21.1) years for patients without diabetes. For patients with type 1 diabetes, 51.5% of the fractures occurred in men, whereas in patients with type 2 diabetes and without diabetes, 39.2% and 38.5% of fractures occurred in men, respectively. On average, patients with type 1 diabetes had a slightly longer diabetes duration at time of first fracture (9.5 [SD 7.8] years) compared with those with type 2 diabetes (7.0 [SD 8.3] years). When looking at the use of antidiabetic medications in the 90 days prior to the first fracture, 73.6% of patients with type 1 diabetes had a recent insulin prescription filled, while 56.9% of patients with type 2 diabetes had a prescription of either biguanide (31.8%) or sulfonylurea (25.1%). Among patients with type 2 diabetes, 26.1% had received insulins and analogs within the 90 days prior to the first fracture.

Figure 2 illustrates the IRs per 10,000 PY trends over time for the included fracture sites: clinical vertebral fractures (Fig. 2A), hip fractures (Fig. 2B), humerus fractures (Fig. 2C), forearm fractures (Fig. 2D), foot fractures (Fig. 2E), and ankle fractures (Fig. 2F). The annual counts, IRs, and 95% CIs by respective fracture site and diabetes status are provided in Supplementary Tables 6–11.

Clinical vertebral, hip, humerus, and forearm fracture IRs were higher in patients with type 1 or 2 diabetes during the study period (Fig. 2A–D). Patients with type 2 diabetes had higher IRs for clinical vertebral (Fig. 2A), hip (Fig. 2B), and humerus (Fig. 2C) fractures compared with patients with type 1 diabetes and patients without diabetes during the observation period.

Similar to the osteoporotic fracture sites, the IRs of ankle fractures in patients with type 1 and type 2 diabetes were slightly higher than in patients without diabetes (Fig. 2F). Ankle fractures were similar to osteoporotic sites, with patients with type 1 and type 2 diabetes showing slightly higher IRs than patients without diabetes (Fig. 2F).

Time trends were observed in most fracture sites. The descriptive percent change between the median IRs for the first 5 years and the last 5 years for each fracture site, stratified by diabetes status, is provided in Table 2. For osteoporotic fracture sites of the vertebra, hip, humerus, and forearm, consistent trends were observed for all groups: clinical vertebral fractures increased (between 14.8 and 38.9%), while hip (between 23.4 and 47.0%), humerus (between 1.2 and 25.0%), and forearm (between 3.9 and 23.8%) fractures decreased (Table 2). While the magnitude differed, similar decreasing trends were observed for humerus and forearm fractures. Results from the linear regression identified significant trends for clinical vertebral, hip, and humerus fractures among patients with type 2 diabetes and patients without diabetes (Fig. 2A–C). However, only the decline in hip fractures was significant among patients with type 1 diabetes (Fig. 2B). When comparing differences across groups, we found significant differences among all groups for clinical vertebral, hip, and humerus fractures. Among forearm fractures, only the difference between patients with type 1 diabetes and patients without diabetes was significant (Fig. 2A–D).

For foot and ankle fractures that generally are not considered to be associated with osteoporosis, the trends differed among the groups. Among patients with type 1 diabetes, the median IRs for both foot and ankle fractures were considered stable over time with only modest increases (1.3 and 8.7%, respectively), while for patients with type 2 diabetes, the median IRs decreased at both sites but with a stable trend of 2.5% decrease for foot fractures and a substantial 31.0% decrease for ankle fractures (Table 2). In patients without diabetes, we found contrasting results, where foot fractures had a stable trend of 9% increase, while ankle fractures declined 11.3% (Table 2). Results from the linear regression only identified significant trends among the patients without diabetes (Fig. 2E and F). When comparing groups, the trends for foot fracture were significantly different among all groups, while for ankle fractures, the differences between patients with type 1 or type 2 diabetes were significantly different from patients without diabetes, but no differences between the diabetes groups were observed.

The results for the annual IRs after adjusting for age and sex are provided in Supplementary Tables 12 and 13. For hip fractures, the trend of a significant decrease over time observed in the primary analysis remained after adjusting for age and sex. The overall trends for other fracture sites remained consistent with the primary analysis. For example, the adjusted IR for clinical vertebral fractures increased across all groups, while humerus fractures decreased. However, because of low counts, we observed increasing variance, which is reflected in wider CIs from the primary analysis.

The age-adjusted IRRs for all six fracture sites are provided in Table 3 (vertebral, hip, and humerus fractures) and Table 4 (forearm, foot, and ankle fractures), while the unadjusted IRRs are presented in the Supplementary Tables 14 and 15. The age-adjusted IRRs of hip and humerus fractures were significantly elevated in patients with type 1 diabetes throughout the study period, while the IRRs of clinical vertebral fractures were rarely significantly elevated in type 1 diabetes (Table 3). For patients with type 2 diabetes, the incidence of forearm fractures was significantly lower compared with patients without diabetes (Table 3), whereas the incidence for the clinical vertebral, hip, and humerus fractures was similar to patients without diabetes (Table 4).

For the nonosteoporotic fracture sites, the age-adjusted analyses identified that foot and ankle fractures were higher among patients with type 1 diabetes compared with patients without diabetes over the study period (Table 4). The incidence of foot fractures decreased over the years and became significantly lower in patients with type 2 diabetes compared with patients without diabetes since 2010. Ankle fractures were elevated in patients with type 2 diabetes compared with patients without diabetes (Table 4). However, from 2007 onward, the difference in incidence was nonsignificant between patients with type 2 diabetes and patients without diabetes.

In this study, we investigated the trends over time in IRs of multiple fracture sites in patients with type 1 diabetes, type 2 diabetes, and without diabetes in Denmark between 1997 and 2017. Patients with type 1 and type 2 diabetes had higher fracture IRs for all fracture sites except the foot, where fracture IRs were lower in patients with type 2 diabetes than those with type 1 diabetes or without diabetes. During the 21 years of observation, the incidence of hip, humerus, and forearm fracture decreased in all groups. By contrast, the IRs of clinical vertebral fractures increased in all groups. The time trends were predominantly consistent after adjusting for age and sex. Compared with patients without diabetes, fractures remained elevated in patients with type 1 diabetes after adjustment for age but not in patients with type 2 diabetes after 2010.

Time trends in fracture rates in the general population have been reported (17), but changes in patients with type 1 diabetes have not been reported. Our analysis shows higher IRs of osteoporotic fractures and ankle fractures in patients with diabetes, which is in line with other studies of fracture risk in patients with diabetes (19,20). A meta-analysis by Wang et al. (19) reported that hip fractures are the most frequently studied in diabetes populations. In addition, hip fracture was reported to be the type with the highest risk among patients with diabetes (type 1 diabetes relative risk [RR] 4.35 [95% CI 2.91–6.49; P < 0.001], type 2 diabetes RR 1.27 [95% CI 1.16–1.39; P < 0.001]). Increased risk of upper arm (type 1 diabetes RR 1.83 [95% CI 1.41–2.39; P < 0.001], type 2 diabetes RR 1.54 [95% CI 1.19–1.99; P = 0.001]) and ankle (type 1 diabetes RR 1.97 [95% CI 1.24–3.14; P = 0.004], type 2 diabetes RR 1.15 [95% CI 1.01–1.31; P = 0.029]) fractures were also reported. By contrast, the analysis did not identify associations between diabetes and risks of distal forearm (type 1 diabetes RR 1.09 [95% CI 0.43–2.75; P = 0.861], type 2 diabetes RR 0.97 [95% CI 0.66–1.09; P = 0.573]) or vertebral (type 2 diabetes RR 1.74 [95% CI 0.96–3.16; P = 0.070]) fractures. Still, evidence at individual sites was limited, minimizing the ability to identify differences between diabetes types. Differences between the meta-analysis and our study may be explained by changes over time, number and types of factors included in the analyses, and definitions of type 1 and type 2 diabetes.

Decreasing trends in hip fractures have been identified in the general population (17,21–23); however, trends in patients with diabetes are not described, despite knowledge of an increased risk of fractures among patients with diabetes, as previously stated. In our study, the decline in hip fracture rates among all groups mirrors the study by Abtahi et al. (17), which showed a similar trend in hip fractures between 1995 and 2010 among Danish residents aged >50 years. Interestingly, when adjusted for age, differences in hip fracture rates between patients with type 2 diabetes and patients without diabetes were not observed after 2004. Conversely, the IRRs remained higher in patients with type 1 diabetes, which may indicate that more recent advances in the management of type 2, but not type 1, diabetes may help to mitigate fracture risk either by improving glucose control or through a direct effect on bone metabolism.

We observed that the IR of vertebral fractures increased in all groups and the incidence among patients with type 1 diabetes and type 2 diabetes was significantly higher than among patients without diabetes. While the increase was highest in patients without diabetes, IRs were more elevated in patients with diabetes. The increase in vertebral fracture incidence particularly after 2010 may be explained by increasing awareness of the risk of vertebral fractures in the general and specific populations, including patients with diabetes and osteoporosis. However, we note that this was consistent among patients with diabetes, and the elevated IRRs of vertebral fractures were not observed after adjustment for age, which may, in part, be explained by the fact that clinical vertebral fractures are rare.

The incidence in forearm and humerus fractures declined in patients with both types of diabetes but not in patients without diabetes. The decreasing fracture incidence in type 1 and type 2 diabetes may indicate improved clinical management and awareness of fracture risk. However, when adjusting for age, we found differences between diabetes types. Among patients with type 2 diabetes, the incidence of humerus fractures was similar to patients without diabetes since 2010, while the incidence of forearm fractures was lower throughout the observation period. Conversely, the incidence of both fracture sites was significantly elevated among patients with type 1 diabetes compared with patients without diabetes, thereby indicating that while the incidence of these fracture sites is decreasing in patients with diabetes, there remains an elevated risk in patients with type 1 diabetes. As the forearm has a higher prevalence of cortical bone, this could suggest that the mechanism behind forearm fractures differs between patients with type 1 and type 2 diabetes. Despite both hip and forearm being considered cortical sites, hip fractures are declining faster than forearm fractures, which could be a result of forearm fractures being more frequent after falls and that forearm fractures might occur earlier than hip fractures. Thus, hip fractures may be prevented by detection of poor bone health in patients after forearm fractures.

Finally, foot and ankle fractures are reported to be associated with diabetes (24). Accordingly, we found that the incidence of ankle fractures was higher in patients with diabetes compared with patients without diabetes. While the incidence remained stable over time among patients with type 1 diabetes, it decreased substantially in patients with type 2 diabetes. Adjustment for age showed that the incidence of ankle fractures remained significantly elevated among patients with type 1 diabetes compared with patients without diabetes. However, since 2007, the incidence of ankle fractures in patients with type 2 diabetes was comparable to the incidence in patients without diabetes. While others have reported that both type 1 diabetes (25) and type 2 diabetes (24) are associated with an increased risk of foot fractures, our results suggest that the highest risk is seen in patients with type 1 diabetes. Detection of low-impact foot or ankle fractures might be an early indication of impaired bone health. Additionally, patients with type 1 and type 2 diabetes have an increased prevalence of neuropathy and Charcot foot, which are both risk factors for foot and ankle fractures. Peripheral neuropathy could be associated with reduced pain from low-impact fractures, and the prevalence of neuropathy is higher in patients with type 2 diabetes than those with type 1 diabetes, which might lead to less detection of foot and ankle fractures in patients with type 2 diabetes.

The study is based on databases that capture almost the entire Danish population and has a long observation period that includes a high number of fractures, which strengthens our analysis. Additionally, we used a washout period to minimize double counting (or overestimating) fractures, as we cannot distinguish between the original fracture code and a follow-up visit. However, there can be situations where certain fractures will be calculated more than once, for example hip fractures, when the healing process and follow-up visits exceed 365 days.

Nevertheless, there are limitations that should be noted when interpreting our results. First, the data are from hospital records, including all inpatient and outpatient contacts to hospitals, outpatient clinics, and emergency departments. Thus, we only capture fractures that come to clinical attention. Low-impact fractures can go unnoticed, which occur even in patients with established osteoporosis. For example, vertebral fractures might not be noticed by the patient or physician as a potential fracture (26) and, therefore, are often not detected immediately. While it is possible that this may lead to an underestimation of clinical vertebral fractures, we also note that the increasing IR of clinical vertebral fractures since 2012 may be due to increased screening and detection, which increases the likelihood of detecting older and previously unrecognized fractures. Second, our classification of type 1 diabetes and type 2 diabetes, which used prescription and diagnostic records, may have resulted in misclassification of a small number of patients identified as those without diabetes. Few patients having only one prescription of noninsulin antidiabetic medication shortly before the fracture, and without a diagnosis code, have been identified as those without diabetes, but they could be patients with newly diagnosed type 2 diabetes. However, we expect this to be the minority of cases of patients without diabetes, and with >750,000 patients without diabetes, we do not expect that this misclassification would significantly bias the observed trends. Third, obesity may influence fracture risk, as some fractures are more common in obese patients and others less common. Compared with patients with type 1 diabetes, a larger proportion of patients with type 2 diabetes are overweight. Additionally, the proportion of overweight has increased in patients with type 2 diabetes but not in those with type 1 diabetes (13). While ICD-8 (277.99) and ICD-10 (E66) codes that reflect obesity were available in the registry, they are not used consistently by clinicians, precluding us from studying the impact of changes in weight in the population. Fourth, in Denmark, almost all patients with type 1 diabetes are treated at public hospitals, while the younger age-groups with type 2 diabetes and patients with type 2 diabetes and complications are treated at hospitals. This leaves only a few patients with type 2 diabetes who might not show up in that database with an ICD code for diabetes, but most patients are taking medications and should therefore still be identified as patients with diabetes in the study. Finally, rare cases (i.e., low number of events) can introduce large amounts of random variation. Adjustments on low number of events are generally advised against, since the results can be statistically unreliable. Thus, as the annual counts for some fracture sites were low, particularly among patients with type 1 diabetes, this may explain some variability in the adjusted analyses.

The present investigation confirmed that fracture rates are higher in patients with type 1 diabetes compared with patients without diabetes. Apart from clinical vertebral fractures, the incidence of major osteoporotic fractures has declined or appears to be stable in patients with type 1 and type 2 diabetes, as well as in patients without diabetes. The declining trends may be due to better diabetes management, including antidiabetic treatments, fracture prevention measures, and awareness of increased fracture risk in diabetes. The higher incidence of major osteoporotic fractures, particularly hip fractures, in type 1 diabetes requires further attention, including studies of fracture prevention in diabetes and implementation of fracture prevention measures in clinical practice. Furthermore, fracture rates in subgroups of type 2 diabetes need to be investigated.

Funding. This project received funding from the European Union’s Horizon 2020 research and innovation program under the H2020 Marie Skłodowska-Curie Actions grant 860898.

Duality of Interest. M.F. reported research grants from Novo Nordisk Foundation, chairmanship of the Expert Committee on Treatment of Rare Bone Diseases of the Danish Medicines Council, consulting fees from Novo Nordisk AS, and receipt of drug and placebo free of charge from Novo Nordisk AS for an investigator-initiated trial. P.V. reported consulting fees from Novo Nordisk AS. The professorship of A.M.B. was partially endowed by the ETH Foundation and pharmaSuisse, but there was no connection to the current study. No other potential conflicts to interest relevant to this article were reported.

Author Contributions. A.V.K. accessed and analyzed the data and was responsible for the data visualization. A.V.K. and M.I.N. led the scientific writing of the manuscript. A.V.K., M.I.N., P.V., M.F., and A.M.B. contributed to designing the study, interpreting the data, and critically reviewing the manuscript. P.V., M.F., and A.M.B. initiated the conceptualization of the study. All authors have read and agreed to the published version of the manuscript. P.V. and A.M.B. are the guarantors of this work and, as such, had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

Prior Presentation. Parts of this study were presented in abstract form at The American Society of Bone and Mineral Research 2021 Annual Meeting, San Diego, CA, 1–4 October 2021.