Fracture Risks in Patients Treated With Different Oral Anticoagulants: A Systematic Review and Meta‐Analysis

Abstract

Background

Evidence on the differences in fracture risk associated with non‐vitamin K antagonist oral anticoagulants (NOAC) and warfarin is inconsistent and inconclusive. We conducted a systematic review and meta‐analysis to assess the fracture risk associated with NOACs and warfarin.

Methods and Results

We searched PubMed, Embase, Cochrane Library, Scopus, Web of Science, and ClinicalTrials.gov from inception until May 19, 2020. We included studies presenting measurements (regardless of primary/secondary/tertiary/safety outcomes) for any fracture in both NOAC and warfarin users. Two or more reviewers independently screened relevant articles, extracted data, and performed quality assessments. Data were retrieved to synthesize the pooled relative risk (RR) of fractures associated with NOACs versus warfarin. Random‐effects models were used for data synthesis. We included 29 studies (5 cohort studies and 24 randomized controlled trials) with 388 209 patients. Patients treated with NOACs had lower risks of fracture than those treated with warfarin (pooled RR, 0.84; 95% CI, 0.77–0.91; P<0.001) with low heterogeneity (I ²=38.9%). NOACs were also associated with significantly lower risks of hip fracture than warfarin (pooled RR, 0.89; 95% CI, 0.81–0.98; P=0.023). A nonsignificant trend of lower vertebral fracture risk in NOAC users was also observed (pooled RR, 0.74; 95% CI, 0.54–1.01; P=0.061). Subgroup analyses for individual NOACs demonstrated that dabigatran, rivaroxaban, and apixaban were significantly associated with lower fracture risks. Furthermore, the data synthesis results from randomized controlled trials and real‐world cohort studies were quite consistent, indicating the robustness of our findings.

Conclusions

Compared with warfarin, NOACs are associated with lower risks of bone fracture.

Nonstandard Abbreviations and Acronyms

NOAC

non‐vitamin K antagonist oral anticoagulant

OAC

oral anticoagulant

Clinical Perspective

What Is New?

• This systematic review and meta‐analysis gathered data from 388 209 participants in 29 studies and showed that patients taking non‐vitamin K antagonist oral anticoagulants had an overall 16% lower risk of developing fractures compared with those taking warfarin.

• Subgroup analyses for individual non‐vitamin K antagonist oral anticoagulants demonstrated that dabigatran, rivaroxaban, and apixaban were significantly associated with lower fracture risks.

• The evidence from real‐world cohort studies and randomized controlled trials is quite consistent, indicating the robustness of our findings.

What Are the Clinical Implications?

• This meta‐analysis provided up‐to‐date evidence showing that non‐vitamin K antagonist oral anticoagulants may be the preferred alternatives to warfarin for lowering fracture risks in patients requiring oral anticoagulant therapy.

Oral anticoagulants (OACs) are commonly prescribed to prevent or treat thromboembolic events in patients with atrial fibrillation or venous thromboembolism. ¹ , ² Warfarin, a vitamin K antagonist, is a traditional OAC and has been a primary long‐term treatment option for patients with atrial fibrillation or venous thromboembolism for decades. Recently, non‐vitamin K antagonist oral anticoagulants (NOACs) have been approved as alternatives to vitamin K antagonists and have demonstrated similar or superior efficacy and safety compared with warfarin. ³ , ⁴ Because aging is one of the strongest risk factors for both atrial fibrillation and venous thromboembolism, ⁵ , ⁶ the prescription of OACs has gradually increased in the aging population worldwide.

Some previous studies have suggested that warfarin may increase fracture risks via its vitamin K antagonizing effect, which impairs bone mineralization; in contrast, NOACs are independent of mechanisms associated with vitamin K antagonists. However, previous studies have provided conflicting evidence on the association between warfarin and fracture risks. ⁷ , ⁸ , ⁹ , ¹⁰ A previous cohort study published in 2017 was the first to compare the fracture risk associated with an NOAC (dabigatran) and warfarin, reporting a significantly lower fracture risk in dabigatran users. ¹¹ Another cohort study in 2017 observed no significant difference in fracture risks among patients taking NOACs (both dabigatran and factor Xa inhibitors) and warfarin. ¹² In 2018, a meta‐analysis based on 12 randomized controlled trials (RCTs) showed that patients treated with NOACs had significantly lower risks of overall fracture, but not hip or vertebral fracture, than patients treated with warfarin. ¹³ However, all trials included in that meta‐analysis considered fracture data as adverse events, and none of them were specifically designed to assess fracture risks. Moreover, the follow‐up duration was ≤12 months in over half of these trials, which might yield insufficient statistical power for evaluating fracture events and the relatively low risks of fracture reported by these trials. Recently, several population‐based cohort studies with larger sample sizes, longer follow‐up durations, and greater statistical powers were conducted to evaluate the association between different OACs and fracture risks. ¹⁴ , ¹⁵ , ¹⁶ , ¹⁷ In addition, there is a growing number of RCTs comparing NOACs with warfarin. The relevant evidence regarding the fracture risks associated with OACs continues to accumulate.

Because osteoporosis and bone fractures pose major threats to the elderly and OACs are mainly prescribed to older adults who have multiple risk factors for fractures, ¹⁷ , ¹⁸ it is critically essential to determine whether a difference in fracture risks exists between NOACs and warfarin. Therefore, we conducted a systematic review and meta‐analysis to compare the risk of bone fractures between patients treated with NOACs and those treated with warfarin. We searched for both clinical trials and observational studies to comprehensively evaluate the current evidence on this issue and compared the results from RCTs to real‐world evidence.

Methods

The authors declare that all supporting data are available within the article and its online supplementary files.

Data Sources and Literature Search

This systematic review and meta‐analysis was conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta‐Analyses guidelines. ¹⁹ We searched PubMed, Embase, Cochrane Library, Scopus, Web of Science, and ClinicalTrials.gov for studies from their inception until May 19, 2020. We applied 2 search strategies. In brief, we initially searched articles using combinations of the following terms: “NOAC,” “dabigatran,” “rivaroxaban,” “apixaban,” “edoxaban,” “warfarin,” “vitamin K antagonist,” and “fracture” (search strategy 1). This search was conducted to identify articles that evaluated the fracture risk in patients treated with NOACs versus those treated with warfarin, regardless of the study design. However, this search could not identify RCTs, which may have reported fracture data as a part of adverse events only, available on websites for clinical trial registration (eg, ClinicalTrials.gov). Therefore, we conducted another search that used the following terms: “NOAC,” “dabigatran,” “rivaroxaban,” “apixaban,” “edoxaban,” and “clinical trial,” to identify all potential eligible trials related to NOACs (search strategy 2). These terms represent a simplified search concept; the full details and the combinations of search terms used in literature search strategies 1 and 2 are described in Data S1. Additionally, we examined the reference lists from relevant review articles and meta‐analyses for additional articles to be included. The protocol registration application for this study was performed (PROSPERO [International Prospective Register of Systematic Reviews] registration number: CRD42020182607). Institutional ethical approval was not required because this was a meta‐analysis of data from published studies only.

Study Selection and Inclusion Criteria

We included studies if they (1) presented outcome measurements for any fracture in both NOAC and warfarin users, regardless of whether it was reported as primary/secondary/tertiary outcomes or referred to as an adverse/safety event in the articles or supplementary data or on online sites (such as the ClinicalTrials.gov website); and (2) compared and reported the relative risk (RR) of fracture between NOACs and warfarin, or it could be derived from the data in the study. No specific restrictions were set for study population, treatment indication, or treatment/follow‐up duration. Clinical trials, cohort studies, and case–control studies that provided sufficient data were eligible. As observational studies (cohort and case–control) should report properly adjusted estimates by considering potential confounders (age and sex, at minimum), studies in which only unadjusted estimates were available were excluded. We excluded review articles, case reports, editorials, and letters to the editor that did not report original findings, and studies conducted in a laboratory or on animals. Three reviewers (H.‐K.H., C.‐C.P., and S.‐M.L.) independently screened all titles and abstracts and evaluated the relevant articles. If a study was deemed eligible by any reviewer, that study was included for further full‐text review. Two reviewers (H.‐K.H. and C.‐C.P.) then independently assessed the full texts of the studies, and any disagreement was resolved by consensus among members of the study team. We used Covidence systematic review software (Veritas Health Innovation, Melbourne, Australia) to manage our systematic review process, coordinating each author’s work and enabling collaboration among the entire review team.

Study Outcomes

The primary outcome was any fracture event. The secondary outcomes were events of hip and vertebral fractures (the most common and quintessential osteoporotic fractures). Comparisons were made between all NOACs and warfarin, as well as between individual NOACs and warfarin.

Data Extraction and Quality Assessment

Two authors (H.‐K.H. and C.‐C.P.) independently extracted data using a prespecified standardized form, which included author names, trial name, publication year, study design, country, treatment indication, NOAC type, follow‐up duration, mean age, sex, sample size, reported fracture sites, fracture risk estimates, and covariates adjusted in observational studies. A third author (S.‐M.L.) then independently examined the extracted data and resolved any discrepancies. For observational studies, adjusted hazard ratios (HRs) or risk ratios along with standard errors were extracted. When a study reported estimates from covariate‐adjusted models and propensity score matching/weighting models, we considered the latter less biased and preferred for inclusion in the meta‐analysis. ¹⁵ , ¹⁶ , ¹⁷ In RCTs, we extracted fracture events and patient numbers in each treatment group. The total number of fracture events was the sum of any fracture site reported on the ClinicalTrials.gov website, because we were unable to obtain the exact number of patients developing fractures from the full texts of articles and trial registration websites.

Study quality was accessed independently by 3 reviewers (H.‐K.H., R.H.‐E.C., and B.B.‐C.W.) using the Cochrane Collaboration’s Risk of Bias tool for RCTs and the Newcastle–Ottawa Scale for observational studies. ²⁰ , ²¹ Discrepancies were resolved via discussions among members of the study team.

Statistical Analysis

Risk ratios for fracture reported by RCTs and the adjusted HRs reported by observational studies were pooled together to calculate RRs in our meta‐analysis. Between‐study heterogeneity was evaluated using the I ² and τ ² statistics. ²² , ²³ The heterogeneity was considered low, moderate, and high for I ²<50%, 50% to 75%, and >75%, respectively. The τ ² was interpreted using the same units as the pooled effect (logarithm of RR). Considering the between‐study heterogeneity, we calculated the pooled RR and respective 95% CIs using the DerSimonian and Laird random effects model. ²⁴

We conducted several predefined subgroup analyses to determine if the pooled estimates were influenced by different study‐level factors. The subgroup analyses were conducted according to NOAC type (dabigatran, rivaroxaban, apixaban, or edoxaban), study design (RCT or cohort study), indication (atrial fibrillation or venous thromboembolism), mean follow‐up period, mean age, sample size, and the geographic location of participants (North America, Europe, Asia, or multiple continents). We made efforts to contact the corresponding authors when additional information was required for subgroup analyses, but we did not receive any replies from the authors of those studies. ¹¹ , ¹² , ¹⁷ Meta‐regressions were further performed if sufficient studies (n≥10) were available. Egger’s regression test and Begg’s adjusted rank correlation test were conducted to determine publication bias. ²⁵ , ²⁶ If there were ≥10 studies included in the meta‐analysis, we conducted a funnel plot analysis to assess publication bias or small study bias. A sensitivity analysis was conducted to evaluate the influence of each study on the overall pooled estimate (by omitting each study individually). All statistical tests were 2 sided, and results with P<0.05 were considered statistically significant. All statistical analyses were conducted using Stata version 15.1 (Stata Corporation, College Station, TX, USA).

Results

Search Results

With search strategy 1, a total of 1742 studies were identified. After excluding duplicates and screening the titles and abstracts, 64 potentially relevant studies were retrieved for full‐text review. Three studies, which were conducted by Lutsey et al, ¹⁵ Norby et al, ²⁷ and Bengtson et al, ²⁸ used the same databases (MarketScan Commercial Claims and Encounters and MarketScan Medicare Supplemental and Coordination of Benefits databases). The study by Lutsey et al was included in the final analysis because only this study declared fracture risks as the primary outcome and thus reported more comprehensive results. ¹⁵ In addition, Lau et al published 2 studies ¹¹ , ¹⁷ using the same database (Clinical Data Analysis and Reporting System of the Hong Kong Hospital Authority); we included the data of only the latter ¹⁷ in the meta‐analysis owing to its more comprehensive data and longer follow‐up period. We excluded a cross‐sectional study reporting the clinical signs of cranial fracture after a traumatic brain injury in patients using different anticoagulants. ²⁹ With search strategy 2, a total of 17 400 relevant studies were identified. After excluding duplicates and screening the titles and abstracts, 2428 studies were retrieved for review of the full‐text or trial registration. The search strategies for article selection are shown in Figure 1A and 1B.

Figure 1. Flow diagram of literature search and article selection. (A) Search strategy 1. (B) Search strategy 2.

*In search strategy 2, we searched only for RCTs of NOACs; if the titles/abstracts were enough to help judge that the publications were not RCTs of NOACs or on an irrelevant topic, they were excluded in the title/abstract screening stage. NOAC indicates non‐vitamin K antagonist oral anticoagulant; and RCT, randomized controlled trial.

After careful review, 29 studies met the eligibility criteria, including 5 cohort studies ¹² , ¹⁴ , ¹⁵ , ¹⁶ , ¹⁷ and 24 RCTs ³⁰ , ³¹ , ³² , ³³ , ³⁴ , ³⁵ , ³⁶ , ³⁷ , ³⁸ , ³⁹ , ⁴⁰ , ⁴¹ , ⁴² , ⁴³ , ⁴⁴ , ⁴⁵ , ⁴⁶ , ⁴⁷ , ⁴⁸ , ⁴⁹ , ⁵⁰ , ⁵¹ , ⁵² , ⁵³ from search strategies 1 and 2, respectively. For cohort studies, we obtained the fracture data from the full‐text articles. For RCTs, fracture data were obtained from the ClinicalTrial.gov website, because fracture was reported as an adverse event only in the included trials; no trial reported fracture events as a primary or secondary outcome.

In total, 388 209 participants were included in this meta‐analysis. Except for RCTs, all included cohort studies had a large‐scale population‐based design. Additional information, such as age, sex, sample size, treatment indication, follow‐up duration, NOAC type, and reported fracture sites, are summarized in Tables S1 and S2 (for cohort studies and RCTs, respectively). The results of the quality assessments are summarized in Table S3 (for cohort studies) and Table S4 (for clinical trials). The adjusted covariates for each cohort study are shown in Table S5.

Risk of Any Fracture Associated with All NOACs Versus Warfarin

All 29 studies reported at least 1 fracture site and were included in the analyses for any fracture. Patients treated with NOACs experienced a lower risk of any fracture than those treated with warfarin (pooled RR, 0.84; 95% CI, 0.77–0.91; P<0.001) with low heterogeneity (I ²=38.9%) (Figure 2; Table 1). No evidence of publication bias was detected according to Egger’s test (P=0.149) and Begg’s test (P=0.955). The associated funnel plot is shown in Figure S1.

Figure 2. Forest plot of the relative risk of any fracture associated with NOACs compared with warfarin. ¹² , ¹⁴ , ¹⁵ , ¹⁶ , ¹⁷ , ⁵³ .

NOAC indicates non‐vitamin K antagonist oral anticoagulant; and RR, relative risk.

Table 1.

The Pooled Relative Risks of Any Fracture, Hip Fracture, and Vertebral Fracture in Patients Treated with NOACs as Opposed to Warfarin

Fracture Site	No. of Studies	Pooled RR (95%CI)	P Value	I 2 (%)
Any fracture
NOAC overall	29	0.84 (0.77–0.91)	<0.001	38.9
Dabigatran	13	0.86 (0.75–0.99)	0.041	40.2
Rivaroxaban	8	0.78 (0.69–0.87)	<0.001	35.8
Apixaban	6	0.76 (0.65–0.89)	0.001	35.1
Edoxaban	7	0.89 (0.77–1.03)	0.122	0.0
Hip fracture
NOAC overall	17	0.89 (0.81–0.98)	0.023	0.0
Dabigatran	7	1.00 (0.86–1.16)	0.958	0.0
Rivaroxaban	6	0.89 (0.76–1.03)	0.124	0.0
Apixaban	4	0.68 (0.52–0.89)	0.006	0.0
Edoxaban	3	0.89 (0.63–1.27)	0.535	0.0
Vertebral fracture
NOAC overall	11	0.74 (0.54–1.01)	0.061	38.8
Dabigatran	3	0.82 (0.29–2.33)	0.711	63.9
Rivaroxaban	5	0.73 (0.63–0.85)	<0.001	0.0
Apixaban	3	0.47 (0.23–0.95)	0.035	49.1
Edoxaban	2	0.86 (0.60–1.23)	0.414	0.0

A pooled RR<1 indicates that NOAC is associated with a lower fracture risk than warfarin.

NA indicates not applicable; NOAC, non‐vitamin K antagonist oral anticoagulants; and RR, relative risk.

Risk of Any Fracture Associated with Individual NOACs Versus Warfarin

Compared with warfarin, a significantly lower risk of any fracture was found in patients taking dabigatran (pooled RR, 0.86; 95% CI, 0.75–0.99; P=0.041), rivaroxaban (pooled RR, 0.78; 95% CI, 0.69–0.87; P<0.001), and apixaban (pooled RR, 0.76; 95% CI, 0.65–0.89; P=0.001). A lower fracture risk was observed in patients taking edoxaban than in those taking warfarin, but the difference was not statistically significant (pooled RR, 0.89; 95% CI, 0.77–1.03; P=0.122) (Table 1).

Subgroup Analyses and Sensitivity Analyses for Any Fracture Event

All subgroup analyses consistently revealed lower risks of any fracture in patients taking NOACs than in those taking warfarin, irrespective of study design, treatment indications, mean follow‐up period, mean age, sample size, and geographic location. Meta‐regressions suggested no significant differences in the protective effects of NOACs between the subgroups (Table 2). The sensitivity analysis after omitting each study, in turn, demonstrated a robust pooled RR with only negligible differences (Figure S2).

Table 2.

Subgroup and Heterogeneity Analyses of Pooled Relative Risks of Any Fracture in Patients Treated With NOACs as Opposed to Warfarin

Subgroups	No. of Studies	Pooled RR (95% CI)	P Value	I 2 (%)	Results of Meta‐Regression
					τ 2	I 2 Residual (%)	P Value
Overall	29	0.84 (0.77–0.91)	<0.001	38.9	0.014	NA	NA
Study design					0.018	40.2	0.989
Cohort study	5	0.83 (0.74–0.94)	0.004	79.9
Randomized controlled trial	24	0.84 (0.74–0.95)	0.008	9.0
Indication					0.016	40.7	0.754
Atrial fibrillation	20	0.84 (0.77–0.92)	<0.001	46.4
Venous thromboembolism	9	0.76 (0.52–1.11)	0.150	20.3
Mean follow‐up period					0.015	40.9	0.592
<1 y	16	0.90 (0.65–1.24)	0.502	4.4
≥1 y	13	0.83 (0.76–0.91)	<0.001	60.0
Mean age					0.015	41.0	0.971
<65 y	13	0.83 (0.58–1.17)	0.276	10.2
≥65 y	16	0.83 (0.76–0.92)	<0.001	53.7
Sample size					0.014	39.9	0.527
n<5000	18	0.75 (0.54–1.04)	0.084	0.0
n ≥5000	11	0.84 (0.77–0.92)	<0.001	64.4
Geographic location					0.019	29.0
North America	2	0.46 (0.06–3.51)	0.455	62.9			NA*
Europe	2	0.87 (0.77–0.99)	0.030	0.0			0.955
Asia	5	0.74 (0.57–0.96)	0.025	59.9			0.263
Multiple continents	20	0.84 (0.73–0.96)	0.012	12.5			0.610

A pooled RR<1 indicates that NOAC is associated with a lower fracture risk than warfarin.

NA indicates not applicable; NOAC, non‐vitamin K antagonist oral anticoagulant; and RR, relative risk.

^*Dummy variables were created for geographic location when performing meta‐regression using North America as the reference group.

We summarized the detailed results for each design subgroup, including the effect of each NOAC compared with warfarin. Overall, the evidence from real‐world cohort studies and those from adverse events reported by RCTs are comparable (Table 3). The forest plots of the RR of any fracture associated with NOAC versus warfarin are shown in Figures S3 (for cohort studies) and S4 (for RCTs).

Table 3.

The Pooled Relative Risks of Any Fracture, Hip Fracture, and Vertebral Fracture in Patients Treated with NOACs as Opposed to Warfarin Among Different Study Designs

Fracture Site	Cohort Studies				Randomized Controlled Trials
	No. of Studies	Pooled RR (95%CI)	P Value	I 2 (%)	No. of Studies	Pooled RR (95%CI)	P Value	I 2 (%)
Any fracture
NOAC overall	5	0.83 (0.74–0.94)	0.004	79.9	24	0.84 (0.73–0.95)	0.006	6.3
Dabigatran	4	0.88 (0.77–1.00)	0.054	62.0	9	0.69 (0.41–1.15)	0.155	33.1
Rivaroxaban	3	0.77 (0.67–0.89)	<0.001	69.6	5	0.79 (0.60–1.04)	0.096	6.3
Apixaban	3	0.75 (0.60–0.92)	0.007	54.5	3	0.76 (0.40–1.44)	0.393	38.0
Edoxaban	NA	NA	NA	NA	7	0.90 (0.78–1.03)	0.131	0.0
Hip fracture
NOAC overall	3	0.89 (0.80–0.99)	0.036	0.0	14	0.90 (0.71–1.14)	0.403	0.0
Dabigatran	2	0.99 (0.85–1.15)	0.883	0.0	5	1.21 (0.52–2.82)	0.656	6.8
Rivaroxaban	2	0.89 (0.76–1.05)	0.156	0.0	4	0.88 (0.47–1.62)	0.672	3.2
Apixaban	2	0.62 (0.45–0.86)	0.004	0.0	2	0.85 (0.52–1.40)	0.528	0.0
Edoxaban	NA	NA	NA	NA	3	0.90 (0.63–1.27)	0.535	0.0
Vertebral fracture
NOAC overall	1	0.75 (0.65–0.86)	<0.001	NA	10	0.74 (0.45–1.19)	0.226	43.6
Dabigatran	1	0.81 (0.68–0.95)	0.011	NA	2	0.66 (0.04–10.38)	0.764	79.4
Rivaroxaban	1	0.73 (0.62–0.85)	<0.001	NA	4	0.81 (0.38–1.73)	0.582	11.8
Apixaban	1	0.60 (0.41–0.88)	0.009	NA	2	0.37 (0.10–1.37)	0.137	41.8
Edoxaban	NA	NA	NA	NA	2	0.74 (0.46–1.20)	0.411	0.0

A pooled RR<1 indicates that NOAC is associated with a lower fracture risk than warfarin.

NA indicates not applicable; NOAC, non‐vitamin K antagonist oral anticoagulant; and RR, relative risk.

Risk of Hip Fractures

Seventeen studies (3 cohort studies and 14 trials) reported hip fracture events for data synthesis. Overall, patients on NOACs had a lower risk of hip fracture than those on warfarin (pooled RR, 0.89; 95% CI, 0.81–0.98; P=0.023) with minimal heterogeneity (I ²=0%) (Figure 3; Table 1). Egger’s test (P=0.997) and Begg’s test (P=0.592) demonstrated no evidence of publication bias, which is supported by the funnel plot shown in Figure S5. The subgroup analyses of individual NOACs demonstrated that the risk of hip fracture was significantly lower in patients taking apixaban (pooled RR, 0.68; 95% CI, 0.52–0.89; P=0.006) but not in patients taking other NOACs (Table 1).

Figure 3. Forest plot of the relative risk of hip fracture associated with NOACs compared with warfarin. ¹⁴ , ¹⁵ , ¹⁶ , ⁴¹ , ⁴² , ⁴³ , ⁴⁴ , ⁴⁵ , ⁴⁷ , ⁴⁸ , ⁵⁰ .

NOAC indicates non‐vitamin K antagonist oral anticoagulant; and RR, relative risk.

Risk of Vertebral Fractures

Eleven studies (1 cohort study and 10 trials) reported vertebral fracture events for data synthesis. A lower, but not significant, risk of vertebral fracture was observed in NOAC users (pooled RR, 0.74; 95% CI, 0.54–1.01; P=0.061) with low heterogeneity (I ²=38.8%) (Figure 4; Table 1). No evidence of publication bias was found according to Egger’s test (P=0.923) and Begg’s test (P=0.640), and this was supported by the funnel plot shown in Figure S6. The subgroup analyses of individual NOACs demonstrated a significantly lower risk of vertebral fracture in patients treated with rivaroxaban (pooled RR, 0.73; 95% CI, 0.63–0.85; P<0.001) and apixaban (pooled RR, 0.47; 95% CI, 0.23–0.95; P=0.035) (Table 1).

Figure 4. Forest plot of the relative risk of vertebral fracture associated with NOACs compared with warfarin. ¹⁶ , ⁵⁰ .

NOAC indicates non‐vitamin K antagonist oral anticoagulant; and RR, relative risk.

Discussion

In this large‐scale meta‐analysis involving 388 209 participants from 29 studies (with follow‐up time ranging from 3 to 36 months), we found that patients treated with NOACs had an overall 16% and 11% lower relative risk of any fracture and hip fracture, respectively, than those treated with warfarin. The subgroup analyses showed that dabigatran, rivaroxaban, and apixaban were significantly associated with a lower fracture risk. The results from RCTs and real‐world population‐based cohort studies were quite consistent, indicating the robustness of our findings. We undertook separate meta‐analyses for each type of NOAC as well as for different fractures sites, providing a comprehensive evaluation to address the knowledge gap.

Comparisons with Previous Studies

One previous meta‐analysis, synthesizing data from 12 RCTs published before 2017, evaluated the differences in the risk of fracture associated with NOAC and warfarin. ¹³ All studies included in that meta‐analysis reported fracture data as adverse events only online; none of them were specifically designed to assess fracture risks. That previous meta‐analysis observed that patients taking NOACs showed a lower overall fracture risk than those taking warfarin (RR, 0.82; 95% CI, 0.73–0.93) but failed to demonstrate significant differences between NOAC and warfarin in the risk of hip fracture (RR, 0.99; 95% CI, 0.72–1.34) or of vertebral fracture (RR, 0.79; 95% CI, 0.59–1.06). In recent years, several real‐world population‐based cohort studies have provided data collected from routine clinical practice, with longer follow‐up durations and higher statistical power, for evaluating the difference in fracture risk associated with NOACs and warfarin. ¹¹ , ¹² , ¹⁴ , ¹⁵ , ¹⁶ , ¹⁷ Additionally, a growing number of RCTs evaluating NOACs and warfarin have reported fractures as adverse events. Another recent meta‐analysis, evaluating only hip fracture risk, showed that patients taking NOACs had a lower risk of hip fracture than those taking warfarin (HR, 0.89; 95% CI, 0.80–0.99); ⁵⁴ however, this meta‐analysis included only 3 observational studies. Our present meta‐analysis used updated evidence, including 29 studies (5 large‐scale cohort studies and 24 RCTs), and found a lower risk of any fracture in NOAC users (RR, 0.84; 95% CI, 0.77–0.91). The subgroup analyses identified a significantly lower risk of hip fracture (RR, 0.89; 95% CI, 0.81–0.98) in patients taking NOACs, by pooling the data of 17 studies. Pooled data from 11 studies also showed a lower, and almost significant, vertebral fracture risk with NOACs (RR, 0.74; 95% CI, 0.54–1.01). Furthermore, the present study provides novel information by evaluating individual NOACs (dabigatran, rivaroxaban, apixaban, and edoxaban) for all, hip, and vertebral fracture risks. Our results also showed that the risk estimates from real‐world population‐based cohort studies and RCTs (Table 3; Figures S3 and S4) were very similar, with a low between‐study design heterogeneity. This indicates that the results of our meta‐analyses were quite robust, and real‐world evidence from cohort studies strengthens the evidence from RCTs on the protective effects of NOACs.

Possible Underlying Mechanisms

Although precise underlying mechanisms are still unclear, some factors might explain why NOACs are associated with a lower fracture risk than warfarin. Warfarin, a vitamin K antagonist, may interfere with bone formation. ⁵⁵ Warfarin antagonizes vitamin K‐dependent processes and impairs the γ‐carboxylation of osteocalcin and other bone matrix proteins, which are important in bone mineralization and formation. ⁹ , ⁵⁵ NOACs are independent of the mechanisms related to antagonizing vitamin K and do not interfere with bone metabolism. ⁹ Previous animal studies revealed that NOACs have positive effects on bone biology, such as increased bone volume, decreased trabecular separation, and reduced bone turnover rate, increased bone mineral density of the fracture zone, and improved fracture healing, compared with those in the warfarin‐treated or control groups. ⁵⁶ , ⁵⁷ , ⁵⁸ Furthermore, possible positive effects of NOACs on bone health and the prevention of falls have been proposed recently. ⁵⁹ Additional studies are needed to evaluate the underlying mechanisms of NOACs on bone health and fracture risks.

Clinical Implications

The lower risk of fractures in patients taking NOACs is an important finding. Osteoporotic fractures, especially hip and vertebral fractures, that occur more frequently in elderly people, cause significant morbidity, mortality, and socioeconomic burden. Previous studies have reported that atrial fibrillation, the most common indication for OAC treatment, is an important risk factor for osteoporotic fractures. ⁶⁰ , ⁶¹ Many risk factors for osteoporotic fractures, such as old age and a history of diabetes mellitus and cardiovascular diseases, coexist in patients with atrial fibrillation. ¹¹ , ⁶² Venous thromboembolism and fractures also share similar important risk factors, such as old age, immobilization, smoking, previous fracture, and malignancy. ⁶³ , ⁶⁴ , ⁶⁵ As patients taking OACs are often at a higher risk of fractures, evidence regarding the differences in fracture risks associated with the use of different OACs is clinically useful. The use of anticoagulants also poses a challenge to anticoagulation during the surgical treatment of fractures. ⁶⁶ , ⁶⁷ This meta‐analysis provides updated clinical evidence for the association between NOACs and lower fracture risk. Therefore, we suggest that when prescribing OAC treatment, the risk of fractures to patients should be carefully evaluated, and NOACs may be preferred over warfarin to lower fracture risks if both OAC types could be indicated. However, treatment decisions should consider all risks and benefits of NOACs versus warfarin for an individual patient.

Study Limitations

In this systematic review and meta‐analysis, we provided comprehensive and updated evidence on the protective effects of NOACs on fracture compared with warfarin. However, there are several limitations worth addressing. First, similar to the meta‐analysis mentioned previously, ¹³ the data of fracture events from included RCTs were reported as only one of the adverse events in ClinicalTrials.gov. None of the included trials were explicitly designed to assess fracture risks; therefore, detailed assessment methods for identifying fracture events remain unclear. In addition, the follow‐up duration was relatively short (≤12 months) in over half of the included trials; the proportion of fracture events to the number of patients was also low. Furthermore, we calculated the number of any fracture events by summing up the events of each fracture site, which may not be equal to the exact number of patients with fractures because a patient might have experienced more than 1 fracture. However, this method of calculating outcome events was not different between the NOAC and warfarin groups; thus, the bias in RRs we obtained from these RCTs was likely minimal. Second, despite the considerably larger sample size of real‐world observational studies than that of RCTs, the real‐world observational data may be biased owing to unknown or unmeasured confounders. The claim‐based retrospective cohort studies may have a problem in accurately capturing diseases with codes, as medical information/histories are not adjudicated or captured systematically. The mean follow‐up time of the included real‐world cohort studies was also short (range from<12 to 29.2 months; Table S1). However, in our meta‐analysis, the results from studies with 2 different study designs were comparable, indicating that such a bias was likely not of great concern. Third, the results from some of our subgroup analyses were not statistically significant (eg, the analysis of edoxaban), although their point estimates were similar to those of statistical significance (Table 1). This may be because of insufficient statistical power in the subgroup of different NOACs, especially in the analyses of specific fracture sites (hip and vertebral). It should also be noted that none of the subgroups of individual NOACs or fracture sites reached statistical significance in the analyses focusing on only RCTs (Table 3). The statistical power of these subgroup analyses was low, because these RCTs were not designed for evaluating fracture events and tended to have a relatively short follow‐up duration. More studies, especially RCTs and high‐quality prospective studies with longer follow‐up, are needed to evaluate the effect of individual NOACs on each specific fracture site.

Conclusions

This systematic review and meta‐analysis gathered data of 388 209 participants involved in 29 studies and revealed that patients taking NOACs had a 16% lower risk of developing fractures than those taking warfarin. The subgroup analyses demonstrated a similar effect on lower fracture risk for individual NOACs (dabigatran, rivaroxaban, and apixaban) than for warfarin. Based on current evidence, NOACs may be preferred over warfarin to lower fracture risks in patients with indications for OAC therapy. Future studies are necessary to investigate the mechanisms underlying the associations between different OACs and bone health.

Sources of Funding

None.

Disclosures

None.

Supporting information

Data S1. Detailed Search Strategies

Tables S1–S5

Figures S1–S6

(J Am Heart Assoc. 2021;10:e019618. DOI: 10.1161/JAHA.120.019618.)