Evaluation of Fracture Risk Among Patients With Type 2 Diabetes and Nonvalvular Atrial Fibrillation Receiving Different Oral Anticoagulants

Type 2 diabetes (T2D) is a global health problem, with the prevalence estimated to be 10.5% in 2021, which is expected to rise to 12.2% by 2045 (1). It is a major cardiovascular risk factor, associated with significant cardiovascular events and mortality (2). Atrial fibrillation (AF) is the most common sustained cardiac arrhythmia, affecting 40 million people globally (3), with its prevalence expected to rise (4). T2D and AF are closely related. Approximately 30% of patients with diabetes have AF, and 15% of patients with AF have diabetes. Not only does diabetes increase the likelihood of developing AF (5) but also the presence of comorbid T2D and AF confers a worse prognosis than the presence of either one alone, including incident major adverse cardiovascular events and mortality (6). The dreaded consequence of AF are thromboembolic events, including cardioembolic stroke, of which the presence of T2D further increases the risk (7). Anticoagulation therapies, either warfarin or non–vitamin K antagonist oral anticoagulants (NOACs), have been proven to significantly reduce stroke risk in clinical trials in patients with and without diabetes (8). Therefore, anticoagulation is the core component in the management of AF in patients with T2D and should be initiated in all patients with AF and T2D (5).

There are concerns about the potential adverse impact of the use of the vitamin K antagonist warfarin on osteoporosis, as osteocalcin (the most abundant noncollagenous protein in bone) is incorporated into bone through vitamin K–dependent γ-carboxylation (9). In contrast, NOACs theoretically do not interfere with bone metabolism, as they work through mechanisms independent of the vitamin K pathway (10). Fragility fractures are the main consequences of osteoporosis. Hip and vertebral fractures are among the most devastating types of fragility fractures, leading to significant morbidity and mortality. Hence, knowledge of fracture risks associated with different anticoagulant use carries significant health implications (11). Meta-analyses of clinical trials and real-world cohort studies have shown that NOAC use is associated with lower fracture risk compared with warfarin use in general (12).

Patients with T2D are at even higher risks of major osteoporotic fractures, notably hip and vertebral fractures, compared with those without diabetes (13). In contrast to age-related osteoporosis in the general population due to accelerated bone resorption, bone fragility in T2D (also known as diabetic bone disease) represents a unique pathophysiology of attenuated bone turnover (bone formation and resorption) and impaired bone microarchitecture (14). Moreover, a range of diabetes-specific factors, such as glycemic control, antidiabetic agents (e.g., thiazolidinedione), and duration of diabetes, contribute to fracture risk in T2D (15). The presence of T2D compounds the morbidities and mortality associated with fragility fractures (14,15). Interestingly, it has been found that circulating levels of bone formation markers, including osteocalcin, are reduced in patients with T2D (16). Moreover, an animal study demonstrated that dabigatran, one of the NOACs, was associated with a weak positive site-specific effect at the tibial cortical bone (17). Hence, the comparative effect of warfarin versus NOAC on bone among patients with T2D remains to be fully elucidated. Although previous studies of the association between anticoagulants and fracture risks in patients with AF have included a minority of patients with diabetes, a dedicated evaluation among patients with T2D is lacking. Considering that various diabetes-specific risk factors may have an impact on fracture risks, such a study will provide evidence to support anticoagulation treatment recommendations based on risks and benefits for patients with T2D comorbid with AF. Hence, we performed a population-based analysis of fracture risks among patients with T2D comorbid with AF treated with NOACs versus warfarin.

Study data were retrieved from a centralized electronic database comprising all medical records of patients under the Hong Kong Hospital Authority (HA). The HA is a statutory body that provides public medical services and manages all public hospitals in Hong Kong. The majority of patients with diabetes and chronic diseases are managed under the HA because of high government subsidies (18). The HA electronic database has been commonly used to conduct population-based studies of patients with diabetes (19) and AF (20,21) in Hong Kong.

A population-based cohort of adult patients (aged ≥18 years) with both T2D and AF managed under the HA between 1 January 2005 and 31 December 2019 was identified from the centralized electronic database. Physician-diagnosed T2D was identified by International Classification of Primary Care, Second Edition, diagnosis code T90, ICD-9-CM code 250.x0 or 250.x2, or any dispensing record of oral antidiabetic medications, while patients with AF were identified by ICD-9-CM code 427.31. The detailed definitions of disease diagnosis codes are listed in Supplementary Table 1. Patients who were diagnosed as having both T2D and AF and subsequently received warfarin or any NOAC (rivaroxaban, dabigatran, apixaban, or edoxaban) were included in this study. Each patient’s index date was defined as the first date of receiving warfarin or NOAC during the study period. Patients were excluded if they 1) had a diagnosis of valvular heart disease (including valvular replacement), hyperthyroidism, bone tumors, or end-stage renal disease (estimated glomerular filtration rate [eGFR] <15 mL/min/1.73 m², on dialysis, or renal transplant) on or before the index date; 2) had only transient AF (cardiac surgery or diagnosis of myocarditis, pericarditis, or pulmonary embolism within 90 days before their first AF occurrence); 3) received both warfarin and NOAC on the index date; or 4) were exposed to any oral anticoagulants (warfarin, rivaroxaban, dabigatran, apixaban, or edoxaban) within 180 days before the index date. The study flowchart is shown in Supplementary Fig. 1. Eligible patients were observed from the index date until the occurrence of study outcomes, death, treatment crossover (i.e., patients crossing over from warfarin to NOAC or vice versa), or 31 December 2020 (study end date), whichever came first. The study protocol was approved by the institutional review board of The University of Hong Kong/Hong Kong HA West Cluster (reference No. UW 21-320).

The primary outcome is a composite of major osteoporotic fractures, including hip, clinical vertebral, proximal humerus, and wrist fractures, identified by the ICD-9-CM codes 805.x, 812.x, 813.x, 814.x, or 820.x (Supplementary Table 1). This set of diagnosis codes has been validated in a previous study using the HA centralized electronic database (22).

The secondary outcomes are 1) hip (ICD-9-CM code 820.x), 2) clinical vertebral fractures (ICD-9-CM code 805.x), and 3) upper-limb fractures (including proximal humerus and wrist fractures defined by ICD-9-CM codes 812.x, 813.x, or 814.x). All outcomes were identified in both outpatient and inpatient settings.

Baseline covariates included demographic characteristics (age and sex), clinical and laboratory measurements, medical comorbidities, and medication prescriptions in the 6 months before the index date. Clinical and laboratory parameters comprised BMI, systolic and diastolic blood pressure, duration of diabetes, glycated hemoglobin (HbA_1c), fasting glucose, LDL cholesterol, ratio of total to HDL cholesterol, triglycerides, eGFR, and albuminuria status (macroalbuminuria if urine albumin-to-creatinine ratio [UACR] >34 mg/mmol, microalbuminuria if UACR >3.4 to ≤34 mg/mmol, and normoalbuminuria if UACR ≤3.4 mg/mmol). Medical comorbidities included congestive heart failure, ischemic stroke, transient ischemic attack, chronic obstructive pulmonary disease, liver disease, osteoporosis, history of fractures, rheumatoid arthritis and other inflammatory polyarthropathies, history of falls, severe hypoglycemia, diabetic retinopathy, hyperparathyroidism, and dementia. Medication prescriptions included antidiabetes medications (insulin, metformin, sulfonylurea, thiazolidinedione, dipeptidyl peptidase 4 inhibitors, sodium–glucose cotransporter 2 [SGLT2] inhibitors, glucagon-like peptide-1 receptor agonist, and α-glucosidase inhibitors), antihypertensive medications (ACE inhibitors or angiotensin receptor blockers, β-blockers, calcium channel blockers, diuretics, and other antihypertensive drugs), lipid-lowering agents, proton pump inhibitors, antidepressants, systemic glucocorticoids, calcium and/or vitamin D supplements, hormone replacement therapy, antiosteoporosis therapy, and medications for Parkinson disease.

Patient characteristics are described using counts and proportions for categorical variables and means with SDs for continuous variables. Multiple imputation by chained equation was adopted to impute missing baseline characteristics (23,24). Each missing baseline covariate was imputed five times by using other known and imputed covariates. Five complete data sets were generated and analyzed separately, and the linear predicted results were obtained by applying the Rubin rule (25).

Inverse probability of treatment weighting based on propensity score was applied to the study results to minimize the selection bias between treatment groups. The propensity score of each patient was the predicted probability of receiving the treatment estimated from the baseline characteristics using a logistic regression model. The inverse probability of treatment weight was generated on the basis of the propensity score and the actual treatment group, and the results were estimated by the average treatment effect (26). To reduce the bias caused by extreme weights, the weights were truncated at the lowest and highest 1% level (i.e., 1st and 99th percentiles). The minimum and maximum weights after truncation were 1.027 and 9.801, respectively. The distribution of the inverse probability of treatment weights between groups is presented in Supplementary Fig. 2. To assess the balance of baseline covariates between treatment groups, absolute standardized mean differences (ASMDs) were calculated, and ASMD ≤0.1 indicated balance between groups (27).

The cumulative incidence (number of events/total number of patients) and incidence rate (number of events/total person-time at risk, in terms of cases per 100 person-years) of study outcomes were calculated. The hazard ratios (HRs) and 95% CIs of study outcomes were calculated using Cox proportional hazards regression models.

Subgroup analyses were performed according to age (<75 vs. ≥75 years), sex (male vs. female), HbA_1c level (<7% vs. ≥7%), duration of diabetes (<5, 5 to <10, and ≥10 years), and history of severe hypoglycemia (yes vs. no) to assess their influence on the risk of outcomes between anticoagulant treatment groups. Further analyses were performed regarding the fracture risks associated with factor Xa inhibitors (apixaban, rivaroxaban, and edoxaban) and the direct thrombin inhibitor dabigatran compared with warfarin. A head-to-head comparison between factor Xa inhibitors and the direct thrombin inhibitor dabigatran was performed as well. To achieve this purpose, a propensity score–weighted cohort of three groups of anticoagulant users (factor Xa inhibitors, direct thrombin inhibitor, and warfarin) was built. Several sensitivity analyses were performed: 1) Patients were not censored when they switched drug treatment from warfarin to NOAC and vice versa, 2) including only patients with duration of follow-up ≥1 year, 3) including patients with at least two prescriptions of anticoagulants within 12 months, and 4) excluding fracture events that were accompanied by the diagnostic code for accidental falls.

All statistical analyses were performed using Stata 14.0 software (StataCorp LP, College Station, TX). All significance tests were two-tailed, and P < 0.05 was considered statistically significant.

A total of 15,770 patients with T2D comorbid with AF were identified (9,288 in the NOAC group and 6,482 in the warfarin group) (Supplementary Fig. 1). Their mean age was 75.7 (9.7) years with no sex preponderance (49.2% men) (Supplementary Table 2). The mean HbA_1c was 7.1 (1.4%) (54.5 [15.0] mmol/mol), with a mean duration of diabetes of 9.2 years (Supplementary Table 2). Baseline characteristics before propensity score weighting with their respective data completion rates are summarized in Supplementary Tables 2 and 3, respectively. Specifically, sociodemographic characteristics, comorbidities, and medication use were captured completely. Among the clinical parameters, eGFR and duration of diabetes were completely captured. Clinical parameters that had missing data included <5% missing for glycemic parameters and lipid profiles, 15% missing for blood pressure measurements, 34% missing for BMI, and 38% missing for UACR (Supplementary Table 3). Before propensity score weighting, compared with warfarin users, NOAC users had a greater female predominance, older age, slightly better glycemic control, and slightly lower BMI. eGFR was similar between the two groups (Supplementary Table 2). Among the NOAC users, the agents prescribed, in descending order of frequency, were apixaban (44.8%), dabigatran (35.1%), rivaroxaban (18.0%), and edoxaban (2.6%). After propensity score weighting (Table 1), all baseline characteristics were well balanced as indicated by ASMD ≤0.1. Notably, only empagliflozin and dapagliflozin were the SGLT2 inhibitors used. The median follow-up of the cohort was 20.0 (interquartile range [IQR] 11–52) months (NOAC group: median 16.0 [IQR 8–27] months; warfarin group: median 50.0 [IQR 20–88 months]). There were 2,272 patients (14.4%) who were censored because they crossed over to the other treatment group (1,955 warfarin to NOAC and 317 NOAC to warfarin).

With regard to the study’s primary outcome, 551 patients (3.5%) sustained major osteoporotic fractures (201 [2.2%] in the NOAC group, 350 [5.4%] in the warfarin group) during follow-up. The median time from the index date to the major osteoporotic fractures was 27 months. The adjusted cumulative incidence of major osteoporotic fractures is summarized in Fig. 1. The adjusted cumulative incidence was lower among NOAC users (1.02 per 100 person-years; 95% CI 0.88, 1.19) than warfarin users (1.26 per 100 person-years; 95% CI 1.11, 1.44), with an HR of 0.80 (95% CI 0.64, 0.99; P = 0.044) (Table 2).

With regard to secondary outcomes, 384 patients (2.4%), 67 patients (0.4%), and 164 patients (1.0%) sustained hip fractures, clinical vertebral fractures, and upper-limb fractures, respectively. NOAC users had a lower risk of hip fractures (HR 0.74; 95% CI 0.57, 0.96; P = 0.025) than warfarin users. On the other hand, there was no statistically significant difference in clinical vertebral fractures (HR 0.94; 95% CI 0.54, 1.65; P = 0.840) or upper-limb fractures (HR 0.87; 95% CI 0.57, 1.34; P = 0.539) between NOAC and warfarin users.

In the subgroup analyses (Table 3), the protective effect against fractures among NOAC users compared with warfarin users was consistent across sex (P for interaction = 0.729), age (P for interaction = 0.681), HbA_1c categories (P for interaction = 0.634), duration of diabetes (P for interaction = 0.721), and history of severe hypoglycemia (P for interaction = 0.707).

Further analyses of fracture risk comparison among the three anticoagulant classes showed that in general, users of factor Xa inhibitors and the direct thrombin inhibitor dabigatran had a lower HR of fracture risks (factor Xa inhibitors: HR 0.69 [95% CI 0.53, 0.91], direct thrombin inhibitor: HR 0.82 [95% CI 0.62, 1.10]) than warfarin users (Supplementary Table 4). Nevertheless, the HR reached statistical significance only for factor Xa inhibitor users (P = 0.009). Comparison of fracture risks of factor Xa inhibitor versus direct thrombin inhibitor users did not show a significant difference (HR 0.84; 95% CI 0.60, 1.18]). Baseline characteristics of the three anticoagulation groups are shown in Supplementary Table 5.

All the sensitivity analyses showed consistent results that NOAC use was associated with lower risks of major osteoporotic fractures compared with warfarin use (Supplementary Table 6). Patients were not censored when they switched drug treatment from warfarin to NOAC and vice versa (HR 0.80, P = 0.041), including only patients with duration of follow-up ≥1 year (HR 0.79, P = 0.075), including only patients with at least two prescriptions of anticoagulants within 12 months (HR 0.77, P = 0.048), and excluding fracture events that were accompanied by the diagnostic code for accidental falls (HR 0.80, P = 0.349).

This propensity score–weighted population-based observational study is the first to compare fracture risks associated with NOACs and warfarin exclusively among patients with AF and T2D. We show that NOAC use was associated with a 20% reduction in the risk of incident major osteoporotic fractures, especially hip fractures. The protective effect was consistent across age, sex, and diabetes-specific factors (glycemic control, duration of diabetes, and history of severe hypoglycemia). These results carry important clinical implications because both T2D and AF are common medical conditions associated with significant morbidities and mortality, and patients with T2D are at higher risk for fracture and have worse postfracture outcomes than those without diabetes. Our study supports the prescription of NOACs over warfarin from the perspective of bone health (an additional consideration) for patients with T2D comorbid with AF.

A recent systematic review and meta-analysis evaluated the association of NOACs versus warfarin and fracture risks among 388,209 patients with AF in 29 studies (5 cohort studies and 24 randomized controlled trials) distributed across different parts of the world, including North America, Europe, and Asia (12). It revealed that NOAC use was associated with a statistically significant 15% reduction in any fractures and 10% reduction in hip fractures compared with warfarin use and a nonsignificant 25% reduction in vertebral fractures among NOAC users (P = 0.061). Subgroup analyses revealed that all NOACs were associated with a reduction in fracture risk compared with warfarin.

Nonetheless, patients with diabetes represented only 20–40% of the cohorts in the abovementioned studies. Diabetes was ascertained using ICD coding. Subgroup analysis according to glycemic status was not available (28). Our study takes a closer look into this specific subgroup of patients with T2D. We found that NOAC use was associated with a lower incidence of major osteoporotic fractures, even in patients with T2D who had a different pathophysiology of bone fragility compared with patients with age-related osteoporosis. Moreover, consistent with the abovementioned meta-analysis, the reduction in fracture risk among NOAC users was mainly driven by a reduction in hip fractures, the main health burden of bone fragility in T2D. The key to bone fragility in T2D is exposure to hyperglycemia with the subsequent formation of advanced glycation end products and their crosslinking, leading to more brittle bone (13). It might be postulated that in the case of the already weakened bone due to long-standing exposure to hyperglycemia per se, the choice of anticoagulants might not matter. Nonetheless, our results show that the choice of anticoagulant mattered in terms of fracture risks across various levels of glycemic control and duration of diabetes. These findings support the relevance of choice of NOAC for anticoagulation in all patients with both AF and T2D from the perspective of bone health. Indeed, in addition to bone fragility, there have been concerns about the clinical efficacy and safety of warfarin use among patients with T2D and AF, including instability of international normalized ratio values and potential association with vascular calcifications (29), favoring the prescription of NOACs.

Regarding the choice of NOAC, results from our subgroup analysis support the use of either factor Xa inhibitors or the direct thrombin inhibitor dabigatran, as both were associated with lower fracture risk compared with warfarin. Our results are in line with the findings of a systematic review and meta-analysis of studies in patients with AF in general (12). Although an animal study suggested that dabigatran may have positive effects on bone, the dosage of dabigatran used in that animal study was much higher than that used clinically for AF (17). Our head-to-head comparison between NOAC classes showed no preferential protective effect on fractures, which is consistent with recent meta-analyses performed in patients with AF in general (30,31). The mechanism for the positive effect of dabigatran on bone is postulated to be due to the inhibition of thrombin, where thrombin plays a role in the modulation of the RANKL/osteoprotegerin ratio implicated in osteoclastic activation and bone degradation. On the other hand, factor Xa inhibitors act on the upstream pathway involved in the conversion of prothrombin to thrombin and, therefore, can be expected to have similar bone effects as direct thrombin inhibitors, explaining the comparable fracture risks upon head-to-head comparison.

In this study, we defined major osteoporotic fractures as the sites considered in the FRAX tool (i.e., hip, clinical vertebral, proximal humerus, and wrist fractures), as these sites account for ∼80% of the fracture burden and for considerably more disutility and economic burden (32). Ideally, the fracture events accompanied by traumatic events, especially motor vehicle accidents (ICD-9-CM codes E800–E848), should be excluded, but this information was not available in our database. Nonetheless, it is expected that traumatic fractures only explain the minority of all the fractures captured in this analysis (33). Furthermore, the mean age of the cohort was 75 years, making the fractures occurring at the typical sites of major osteoporotic fractures more likely to be due to underlying osteoporosis than trauma. Indeed, sensitivity analyses according to the age-groups ≥75 years and <75 years showed consistent HRs without significant interaction (P for interaction = 0.681) (Table 3).

As falls are one of the main mechanisms of fragility fractures, we looked these as the cause of fractures in our cohort. In the propensity score–weighted cohort, history of falls among patients treated with warfarin was similar to those treated with NOACs. Besides, in the sensitivity analysis excluding fractures that were accompanied by the diagnostic code for accidental fall, the result was consistent with the main analysis. These results are consistent with the potential beneficial effect of NOACs on the bone compared with warfarin.

A strength of our study is the well-characterized population-based cohort of patients with T2D, which allowed for subgroup analyses according to diabetes-specific factors. Nevertheless, our study results should be interpreted while bearing in mind certain limitations. First, given the observational nature of the study, there may still be unmeasured confounders, such as bone mineral density. However, bone mineral density is not the typical determinant of eligibility of anticoagulation and is not expected to cause confounding by indication. Second, it is possible that silent vertebral fractures remained undetected. Nonetheless, the underdetection was expected to be similar among NOAC and warfarin users, and should this occur, it would tend to bias the results toward the null. Third, some information was not available in this cohort, including time in therapeutic range for warfarin users and information regarding traumatic events, such as motor vehicle accidents. Fourth, similar to all large-scale pharmacoepidemiological studies using electronic medical record databases, treatment adherence could not be ascertained. In addition, we could not completely rule out the issues of over- or underreporting. For instance, only 1.2% of the patients received diagnostic coding of osteoporosis in the electronic health record, which could represent underreporting. This could be further complicated by the fact that the patients with T2D might have had higher bone mineral density than those without diabetes. Nonetheless, 7.0% of the patients had a history of fractures at sites typical of major osteoporotic fractures, meaning that the proportion of patients with bone fragility was not negligible. Finally, this study was performed among Chinese patients with T2D. The results may not be generalizable to patients of other ethnicities.

In conclusion, NOAC use was associated with a lower risk of major osteoporotic fractures than warfarin use among patients with T2D comorbid with AF. NOACs may be the preferred anticoagulant from the perspective of bone health.

Acknowledgments. The authors thank the Hong Kong HA for data provision.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. D.T.W.L., E.H.M.T., and C.K.H.W. drafted the manuscript. D.T.W.L., K.C.B.T., K.S.L.L., and C.K.H.W. conceived the research idea. E.H.M.T., I.C.H.A., T.W., and C.K.H.W. conducted the data analyses and interpreted the results. E.H.M.T., I.C.H.A., and C.K.H.W. accessed and verified the data. T.W., C.H.L., C.K.W., C.Y.Y.C., C.H.Y.F., W.S.C., Y.C.W., K.C.B.T., and K.S.L.L. reviewed the manuscript and gave critical input for revision. K.C.B.T., K.S.L.L., and C.K.H.W. supervised the study. All authors were responsible for the decision to submit the manuscript. C.K.H.W. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Data Sharing. The data that support the findings of this study were provided by the Hong Kong HA. Restrictions apply to the availability of these data, which were used under license for this study.