The association of fracture risk in atrial fibrillation patients and long-term anticoagulant therapy category: a systematic review and meta-analysis

Search strategy

We searched the PubMed, Embase, and Cochrane Library databases, as well as the Clinical Trials.gov website, until January 31, 2020. The search keywords included “atrial fibrillation,” “anticoagulant,” “direct oral anticoagulant,” “vitamin K antagonist oral anticoagulants,” “non-vitamin K antagonist oral anticoagulants,” “dabigatran,” “rivaroxaban,” “apixaban,” “edoxaban,” “warfarin,” “fracture,” “osteoporosis,” “osteoporotic fractures,” and “rarefaction of bone.” We did not include any language restrictions. Our detailed search strategy is shown in File S2. The reference list for each included study was reviewed, and potentially related research from these reference lists was manually searched. Our systematic review was completed in accordance with the guidelines for the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) (Moher et al., 2009), as shown in Fig. 1.

Inclusion and exclusion criteria

The inclusion criteria for the research studies to be analyzed were the following: (1) study patients diagnosed with AF above the age of 18 years and taking oral anticoagulants due to a diagnosis of AF; (2) the study contained at least two comparison groups, one receiving warfarin therapy and the other receiving one of the NOACs (e.g., dabigatran, rivaroxaban, apixaban, edoxaban); (3) considering that the duration of the anticoagulant therapy was not uniform for each patient in the included studies, we determined that the follow-up time for the included studies must exceed 12 months to ensure that each patient received long-term anticoagulant therapy; and (4) the fracture events (e.g., all-fracture and hip fracture) were reported or documented on the Clinical Trials.gov website. The exclusion criteria were as follows: (1) for duplicated reports, the article that was published last was selected; (2) patients administered long-term oral anticoagulant therapy for reasons other than AF; (3) reports that are reviews, case reports, or unrelated studies; (4) the outcome data was unavailable.

Study selection and data extraction

All the studies were independently assessed by two reviewers (SJY and CC). After excluding duplicates, the remaining studies were read to identify potentially suitable articles by title and abstract. The reviewers independently assessed these studies for eligibility. Disagreements were resolved by discussion with a third reviewer (LLC). Data were extracted from each included article, including authors, study type, publication date, follow-up duration, study size, baseline characteristics of each study population, study interventions, and the principal summary measure (e.g., adjusted risk ratio [RR], adjusted hazard ratio [HR], and adjusted odds ratio). Relevant outcomes included all-fractures and hip fractures.

Study quality assessment

The Cochrane Collaboration tool was used to assess the risk of bias in the randomized controlled trials (RCTs) (Higgins et al., 2011). The assessment included the generation of random sequences, blinding of patients and researchers, allocation concealment, blinding of the outcome assessment, selective reporting of outcomes, completeness of the outcome data, and other threats to the validity of the results. The Newcastle Ottawa quality assessment scale (NOS) (Higgins et al., 2011) was used to evaluate the quality of the observational studies. The assessment factors included the selection of the study groups, comparability of the groups, and the ascertainment of the exposure and outcomes. A score of 7 or more was considered high-quality.

Data synthesis and analysis

Our main outcomes were all-fracture and hip fracture risks in NOAC users versus warfarin users. We used RRs and their associated 95% confidence intervals (CI) to assess the outcomes, and a P value < 0.05 was considered statistically significant. The meta-analysis heterogeneity was quantified with the I² test (low: 0–20%; moderate: 20–50%; and high: >50%). RR and their 95% confidence intervals (CIs) were calculated using random-effect models. We performed two meta-analyses and pooled the risk of all-fractures and hip fractures for NOAC versus warfarin users and stratified the results by study type (observational studies or RCTs), and duration of anticoagulation treatment (≥2 years or <2 years), respectively. We performed a subgroup analysis to evaluate the risk of all-fractures and hip fractures for each specific NOAC compared to warfarin. We also performed a sensitivity analysis to assess the impact of each study on the overall results by deleting individual articles and meta-merging the remaining ones. Publication bias within the results was displayed using a funnel plot. We used Review Manager version 5.3 (The Nordic Cochrane Centre, The Cochrane Collaboration, Copenhagen, Denmark) and Stata 14.0 (StataCorp, College Station, TX, USA) to compile our data and perform the meta-analysis.

Inclusion of studies and quality assessment

Our database search yielded 1,484 research articles and clinical trials. After removing duplicates, reviews, case reports, unrelated studies, and those with unavailable data, 10 articles and 12 datasets were remained to include in our meta-analysis. Five of the articles were observational studies, and the other five were RCT studies with double-blind designs (Table 1). The overall process is illustrated in Fig. 1. The five observational studies were all retrospective cohort studies that directly compared the fracture risk for NOAC users with warfarin users. None of the RCT studies reported fracture events in the articles themselves, but they did document the outcomes on the Clinical Trials.gov Website. We counted the fracture events from each of these studies and calculated the RR values. The outcome data and RR values of each RCT study are documented in Tables S1 and S2 (File S4). The baseline characteristics of the patients are documented in Table 2. The mean age of the patients in each study ranged from 67 to 74 years. Regarding the total number of patients reported on in these studies, 12.4–63.4% of the study population had a history of stroke or transient ischemic attack. Diabetes and heart failure were common in the included studies, accounting for 13.8–40.4% and 17.3–62.3% of the study patients, respectively. Additionally, 30.5–40.5% of the study patients received aspirin antiplatelet therapy. All five of the RCT studies were considered high-quality because of their low risk of bias. Of the five observational studies, four were considered high-quality for their representativeness of the exposed cohort, sufficient follow-up for reliable outcomes, and low risk of bias, while the remaining one was considered a medium-quality study due to an insufficient baseline in the study population, significant clinical heterogeneity, and a lack of adjustment for confounding variables. The results of the quality assessment for the studies is shown in Table 1. For additional details, please refer to Tables S3 and S4 (File S4).

Results of the meta-analysis

The 326,846 subjects from the included studies were pooled together for the all-fracture risk meta-analysis. Patients who were treated with NOACs did not have a significant risk of fracture compared with those taking warfarin (RR=0.91, 95% CI [0.81–1.01], I² = 60%, Fig. 2A). Due to the underlying heterogeneity of the article types, however, we analyzed the data in subgroups. Compared with RCTs, those with NOACs therapy had significant lower all fracture risk in retrospective cohort studies (RR = 0.83, 95% CI [0.71–0.97], P = 0.02, I² = 58%, forest plot shown in Fig. 2 [A]). For the hip fracture risk meta-analysis, data from four RCT studies and three retrospective cohort studies (298,439 patients) were pooled together. In this grouping, there were no differences in the hip fracture risk observed between the NOAC and warfarin groups in these studies, but the patients in the retrospective cohort studies showed a statistically significant decrease in the risk of hip fracture compared to the patients in the RCT studies (RR = 0.89, 95% CI [0.80–0.99], P = 0.03, I² = 0%, forest plot shown in Fig. 2B).

To investigate the influence of anticoagulant treatment duration on the risk of all-fractures and hip fractures, we regrouped the studies based on whether the anticoagulant therapies were administered for <2 years or ≥2 years. The results showed the NOACs users have a lower all-fracture risk than warfarin users when the anticoagulation therapy duration was ≥2 years (4 RCTs and 2 observational studies, RR = 0.88, 95% CI [0.81–0.96], P = 0.004, I² = 27%, forest plot shown in Fig. 3A), while no differences in hip fracture risk.

We also performed another subgroup analysis in which we compared the all-fracture and hip fracture risks for each of the individual NOACs (e.g., dabigatran, rivaroxaban, apixaban, and edoxaban) to warfarin. We included 2 of the observational studies and 2 RCTs in this analysis. There was insufficient evidence to prove that dabigatran, apixaban, and edoxaban were associated with lower all-fracture or hip fracture risks compared with warfarin (see the forest plot shown in Figs. 4A, 4B, 4E, 4F, 4G, 4H). However, rivaroxaban anticoagulant therapy patients did show a significantly lower risk of all-fractures than warfarin patients (two observational studies and two RCTs, RR = 0.81, 95% CI [0.76–0.86], I² = 0%, Fig. 4C), but rivaroxaban anticoagulant therapy failed to significantly decrease the hip fracture risk (RR = 0.89, 95% CI [0.76–1.04], I² = 0%, Fig. 4D).

Finally, we performed a sensitivity analysis to test the robustness of our results by orderly elimination of each included study and meta merged the rest studies. Our results are statistically reliable, as shown in Figs. 5 and 6. The robust results from this analysis of each NOAC is shown in Fig. S1. The publication bias was then assessed using a funnel plot. As shown in Figs. 7 and 8, there was no evidence of obvious publication bias in our study, but a slight bias brought by some grey literature cannot be completely excluded in our study.

Study limitations

Our study is the largest review and meta-analysis to investigate the risk of fractures in AF patients by comparing warfarin administration with other treatments. However, some limitations should be considered. First, five of the studies evaluated were observational studies. Although a propensity score-matched analysis is a good method for reducing selection bias, it is inevitable that significant clinical heterogeneity still exists. Although common fracture risk factors (history of fractures and osteoporosis) were propensity matched between the NOAC and warfarin groups in some of the analyzed studies, other important factors, such as bone mineral density, serum calcium, and vitamin D levels were not investigated in any of the reports. Future studies should include these indicators to enhance the strength of the studies. Second, lifelong anticoagulation is the standard treatment for AF management. However, the longest follow-up in the included studies was only 2.8 years. The effects of vitamin K antagonists on bone metabolism are cumulative, and the progression from osteoporosis to fractures is a long process. Therefore, extending the follow- up period in future research is necessary. Third, because the research comparing specific NOACs (dabigatran, rivaroxaban, apixaban, and edoxaban) with warfarin is inadequate, we did not perform a subgroup analysis across the study types (observational study or RCT). This also led to high heterogeneity in several of our comparisons.