Atrial fibrillation and the risk of subsequent fracture

Abstract

Background:

There is conflicting evidence regarding the association between atrial fibrillation and the risk of subsequent fractures.

Methods:

We included participants aged 45 years or older from the Framingham Heart Study Offspring, Third Generation, New Offspring Spouse, Omni 1, and Omni 2 cohorts. We prespecified analyzing index age 65 years as our primary analysis; we repeated analyses for index ages 45, 55, and 75 years. The primary outcome was any incident bone fracture, except finger, toe, foot, skull, and facial fractures. We assessed the association between time-varying atrial fibrillation and subsequent fractures by an illness-death model that accounted for the competing risk of death. We estimated hazard ratios (HR) adjusted for age, sex, body mass index, smoking, diabetes, alcohol intake, and prior fracture.

Results:

We included 3,403 participants (mean age of 68 years, 53.3% female) in the analysis at index age 65 years and above. In all, 525 (15%) participants suffered incident fractures during follow-up (median 12.5 years). The HR between atrial fibrillation and subsequent fracture was 1.37 95%CI (1.06–1.79). There was no evidence of effect modification by sex (HR 1.55, 95%CI 1.06–2.26 in men; HR 1.22, 95%CI 0.84–1.77 in women; interaction p value 0.27). Results were consistent at other index ages.

Conclusion:

Atrial Fibrillation was associated with increased risk of incident fracture in the community-based Framingham Heart Study.

Introduction

In Europe and the U.S. the lifetime risk of atrial fibrillation is estimated to be about one in three in individuals of European ancestry and one in five in those of African ancestry.(1–5) The incidence of atrial fibrillation increases with advancing age.(6) Atrial fibrillation is associated with increased risk of heart failure, myocardial infarction, thromboembolism, stroke, dementia, and overall mortality.(7) Several studies observed an association between cardiovascular disease and increased risk of fractures, with heart failure, myocardial infarction, and hypertension conferring a 1.5–2 fold increase risk of fractures.(8–13)

Atrial fibrillation also may be associated with osteoporotic fractures because the hemodynamic consequences and thromboembolic potential of atrial fibrillation may lead to increased risk of falls.(14) Previous studies have reported contradictory results. Three studies that found an association between atrial fibrillation and increased risk of falls used electronic health record databases, with study populations ranging from 34,625 to 113,600 participants, and had a mean follow-up duration of 3 to 4 years.(15–17) Another analysis based on the Cardiovascular Health Study did not find evidence of an association between atrial fibrillation and risk of falls. This study had a smaller population of 4,462 participants with a longer mean follow-up period of 8 years.(18)

Osteoporotic fractures have a considerable impact on the medical system with over 850,000 fractures annually responsible for $29 billion in medical costs and $51 billion in losses due to missed work. The consequences of fractures are even more substantial in older adults, who experience increased morbidity and mortality from fractures.(19) Given the heightened costs, morbidity, and mortality associated with fractures, it is important to identify any modifiable risk factors that may predispose to risk of osteoporotic fractures.

We used data from the community-based Framingham Heart Study to assess the association between atrial fibrillation and subsequent fractures. We hypothesized that atrial fibrillation is associated with increased risk of osteoporotic fractures.

Methods

Participants

We used data from the Framingham Heart Study (FHS), a multigenerational study of cardiovascular disease in the community. We included participants aged 45 years or older from the FHS Offspring, Third Generation, New Offspring Spouse, Omni 1, and Omni 2 cohorts. We prespecified analyzing index age 65 years as our primary analysis, considering that this subset is generally at greater risk for fracture. We also analyzed index ages 45, 55, and 75 years.

The Offspring cohort (n=5,124) was enrolled in 1971 and comprised adult children of the FHS Original cohort and their spouses. The Offspring cohort was followed at 4 to 8 years with standardized FHS examinations. The Third-Generation cohort (n=4,095) was enrolled in 2002 and followed at about 6-year intervals. The multi-ethnic Omni 1 cohort and Omni 2 cohort were enrolled in 1994 and 2003, respectively. The Omni 1 cohort (n=506) was followed at 4 to 8-year intervals. The Omni 2 cohort (n=410) was followed at 6-year intervals. The New Offspring Spouse Cohort (n=103) enrollment period was October 2003- July 21, 2005 and were followed at 6-year intervals. The exclusion criteria for the present study were as follows: age<45 years, and missing participant characteristics at inclusion.

Atrial Fibrillation

We identified newly diagnosed atrial fibrillation through December 31, 2016. Atrial fibrillation was assessed from FHS examinations and participants’ outside clinical visits or hospitalizations. In addition, questions about atrial fibrillation status were included in health history updates, which staff administer about every two years. In the FHS, two medical doctors evaluated all available records and ECGs of atrial fibrillation. Atrial fibrillation was defined as atrial flutter or fibrillation. We followed each included participant until whichever came first: fracture, death, last FHS examination or health history contact, or December 31, 2016.

Fracture Outcomes

We considered three fracture outcomes. First, we considered all fractures excluding fractures of the finger, toe, foot, skull, and facial bones. Fractures at these locations were excluded because they are generally not associated with low bone mineral density. and unlikely to occur without concurrent trauma.(20) The second outcome consisted of osteoporotic fractures including fractures of the hip, wrist/forearm, clinical vertebral fractures, upper arm and pelvis. The third outcome was hip fractures only.

The FHS ascertained hip fractures via information gathered during examinations, telephone interview, and review of hospitalization and death records. Non-hip fractures were ascertained via the FHS Osteoporosis Study and Framingham Offspring Cohort questionnaires prior to 2002. After 2002, fractures were also ascertained and adjudicated using medical records from the Framingham Offspring, Generation 3, and Omni 2 Cohort exam questionnaires, interval update questionnaires collected via mail, email, telephone, or admission logs.

Covariates

We adjusted for the following covariates: age, sex, alcohol, current smoking, body mass index, diabetes, and previous fractures. We defined diabetes as fasting blood glucose ≥126 mg/dL (if fasting blood glucose was not available, non-fasting blood glucose ≥200 mg/dL was used) or treatment for diabetes.(5) In the sex-specific analyses, for women, we further adjusted for menopausal status. Menopause was defined as cessation of menses for ≥ 1 year.(21)

Statistical analysis

We performed all analyses for each of the 3 fracture outcomes and for multiple index ages: 45, 55, 65, and 75 years, respectively. We pre-specified index age 65 years as the primary analysis. We assessed the association between incident atrial fibrillation, as a time-varying variable, and the risk of fractures. We fitted 3 models: 1) adjusted for age and sex; 2) further adjusted for other covariates at entry; 3) further adjusted for time-varying covariates.

For all analyses, we used age as the time scale. In each analysis, some participants did not enter at the index age (for example 65 years), but rather at some later entry time. We implemented the counting process style of input to accommodate late entry times. Data for each individual were identified by at-risk intervals defined by age at entry, age at exit, atrial fibrillation status, and outcome (fracture, death, or censoring).(22)

Participants were subject to the semi-competing risk of death, which might hinder the probability of observing a fracture event. Ignoring competing risks can lead to biased effect estimates, especially for older adults with multimorbidities. We accounted for the competing risk of death for all analyses. The Fine and Gray model is commonly used for analyzing competing risks endpoints. However, including time-varying covariates in the Fine and Gray model can lead to bias. We used an illness-death model to model atrial fibrillation, the time-dependent covariate, as an intermediate state (Figure 1). Under such framework, we also account for participants with prevalent atrial fibrillation at study entry. To fit the illness model, we used proportional cause-specific hazards models. Results were presented as hazard ratios (HR) and 95% confidence intervals. In secondary analyses, we repeated analyses for participants without prior fractures at the index age. We also analyzed men and women separately.

Figure 1.

Illness-death model for the association between AF and incident fractures. Participants can enter the study either with or without prior AF. Those who do not have prior AF can experience an incident fracture or death; or they can first develop AF during follow-up and then reach one of the two endpoints. Those who have prevalent AF at entry can progress to fracture or death.

With a type I error of 0.05, a power of 80%, a sample size of 3,403 participants, among whom 687 had atrial fibrillation prior to fracture, we estimated that the minimally detectable hazard ratio from a Cox model would be 1.13 (Table A1).

Results

Patient Characteristics

For the primary analysis we analyzed 3,403 individuals at index age 65 years, from the Offspring, Third Generation, New Offspring Spouse, Omni 1, and Omni 2 FHS cohorts. The mean age of participants was 67.8 years, 53.3% were female, 10.1% were current smokers, and 31.9% had a history of fractures (Table 1). We display the baseline characteristics at the other index ages of 45, 55, and 75 years in Appendix Table A2–A4.

Table 1:

Characteristics of participants at index age 65 years

Characteristics	Fracture during follow-up N=525	No fracture during follow-up N=2878
Age, years	67.4±2.0	67.9±2.4
Female	366 (69.7%)	1446 (50.2%)
BMI, kg/m2	27.5±5.0	28.3±5.1
SBP, mmHg	133±20	132±18
DBP, mmHg	74±10	75±10
Hypertension treatment, %	192 (36.7%)	1410 (49.1%)
Current smoking, %	75 (14.3%)	269 (9.4%)
Diabetes mellitus, %	56 (11.1%)	447 (16.0%)
Menopause, %	363 (99.2%)	1421 (98.3%)
Alcohol, drinks/week	2 [0, 7]	2 [0, 7]
Prior fractures, %	221 (42.1%)	864 (30.0%)
Prior AF, %	24 (4.6%)	175 (6.1%)

Data pertain to all fractures excluding fractures of the finger, toe, foot, skull and facial bones.

Data are mean±SD, median [q1, q3], or count (%).

Missing values: diabetes n = 105 (3.1%); BMI n= 89 (2.6%); alcohol n = 10 (0.3%); hypertension treatment n = 10 (0.3%); SBP n = 2 (0.1%); DBP n = 2 (0.1%); smoking n = 2 (0.1%)

Incidence of fracture

After the index age of 65 years, there was a total of 525 (15.4%) incident fractures excluding finger, toe, foot, skull, and facial bones during follow-up period (median 12.5 years, 25^th–75^th percentiles 7.6–21.0 years). Among the participants who experienced fracture during the follow-up period, 42.1% had a history of prior fractures. We provide the results at index ages 45, 55, and 75 years in the Appendix Tables A5–A7.

Association between atrial fibrillation and fracture

At index age 65 years,199 participants had prevalent atrial fibrillation at study entry and 488 other participants developed atrial fibrillation during follow-up. Figure 2 shows the cumulative incidence of atrial fibrillation and the cumulative incidence of fracture. The corresponding curves for index age 45, 55, and 75 years were similar (Appendix Figures A1–A3). Among 687 individuals with atrial fibrillation totaling 8,883 person-years, 72 had subsequent fracture (8.1 subsequent fracture per 1,000 person-years with atrial fibrillation). Time-varying atrial fibrillation was associated with an increased rate of subsequent fracture, with a HR of 1.36 (95% CI 1.05–1.77) in the multivariable-adjusted model with time-varying covariates (Table 2). Results at other index ages are similar and reported in Table 2.

Figure 2.

Cumulative incidence of fracture and of atrial fibrillation prior to fracture, accounting for competing risk of death, among individuals 65 years or older.

Table 2:

Association between time-varying atrial fibrillation and risk of fracture after index age 45, 55, 65, and 75 years

Index age	N	n atrial fibrillation prior to fracture	n fractures	Age- and Sex-adjusted Model	Model Adjusted for Baseline Covariates*	Model Adjusted for Time-varying Covariates†
45 years	7711	813	1328	1.27 (1.00, 1.60)	1.25 (0.98, 1.59)	1.30 (1.02, 1.66)
55 years	5468	794	908	1.31 (1.04, 1.65)	1.25 (0.97, 1.60)	1.36 (1.07, 1.73)
65 years	3403	687	525	1.33 (1.04, 1.71)	1.30 (1.01, 1.69)	1.37 (1.06, 1.79)
75 years	1421	397	220	1.17 (0.84, 1.63)	1.24 (0.88, 1.75)	1.22 (0.86, 1.72)

Data are hazard ratios and 95% confidence intervals from the illness-death model accounting for time-varying AF and competing risk of death.

^*adjusted for age, sex, diabetes, BMI, SBP, DBP, hypertension treatment, smoking status, alcohol drinking, and prior fractures at baseline

^†adjusted for age, sex, time-varying diabetes, time-varying BMI, time-varying SBP, time-varying DBP, time-varying hypertension treatment, time-varying smoking status, time-varying alcohol drinking, and time-varying prior fractures

Subgroup and sensitivity analysis

There was evidence that atrial fibrillation was associated with osteoporotic fractures when adjusting for time-varying covariates (HR 1.39, 95% CI 1.02–1.90). However, overall, atrial fibrillation was not associated with hip fracture (HR 0.95, 95% CI 0.55–1.62). However, among men, there was an association between atrial fibrillation and increased risk of subsequent hip fractures (HR 1.53, 95% CI 1.05–2.23). Among women, there was little evidence of an association between atrial fibrillation and risk of subsequent hip fractures (Table 3). However, there was no significant effect modification of the association between atrial fibrillation and subsequent fracture by sex (interaction test p value = 0.27).

Table 3:

Subgroup and sensitivity analyses at index age 65 years

	n participants	n atrial fibrillation prior to fracture	n fractures	Age- and Sex-adjusted Model	Model Adjusted for Baseline Covariates*	Model Adjusted for Time-varying Covariates†
Osteoporotic Fractures	3403	728	349	1.35 (1.00, 1.82)	1.35 (0.99, 1.84)	1.41 (1.03, 1.92)
Hip Fractures	3403	764	122	0.96 (0.57, 1.59)	0.91 (0.54, 1.52)	0.95 (0.55, 1.64)
Male	1591	433	159	1.47 (1.01, 2.12)	1.48 (1.02, 2.16)	1.55 (1.06, 2.26)
Female	1812	254	366	1.23 (0.87, 1.74)	1.19 (0.83, 1.71)	1.22 (0.84, 1.77)
No Prior Fractures	2318	484	304	1.32 (0.95, 1.85)	1.35 (0.96, 1.90)	1.32 (0.92, 1.88)
Male	1119	309	99	1.38 (0.85, 2.23)	1.46 (0.89, 2.37)	1.44 (0.88, 2.36)
Female	1199	175	205	1.27 (0.80, 2.02)	1.28 (0.80, 2.06)	1.19 (0.70, 2.00)

Data are hazard ratios and 95% confidence intervals from the illness-death model accounting for time varying AF and competing risk of death.

^*adjusted for age, sex, diabetes, BMI, SBP, DBP, hypertension treatment, smoking status, alcohol drinking, and prior fractures at baseline; further adjusted for menopause among females in secondary analysis

^†adjusted for age, sex, time-varying diabetes, time-varying BMI, time-varying SBP, time-varying DBP, time-varying hypertension treatment, time-varying smoking status, time-varying alcohol drinking, and time-varying prior fractures; further adjusted for time-varying menopause among females in secondary analysis

Out of 3,403 participants for the analysis at index age 65 years, 2318 (68%) individuals had no history of prior fracture (median follow-up x months, 25^th–75^th percentiles; x% women). In this subgroup, atrial fibrillation was no longer significantly associated with fracture during follow-up period (HR 1.31, 95% CI 0.92–1.87). In men and women without history of prior fracture, the results were consistent, the HR estimates were 1.46 (95%CI 0.89–2.37) and 1.28 (95%CI 0.80–2.06), respectively.

Discussion

In our community-based analysis using FHS Cohorts, we observed that atrial fibrillation was associated with the risk of subsequent fracture. In our subgroup analysis, atrial fibrillation was also associated with increased risk of osteoporotic fractures. There was no evidence of effect modification of the association by sex.

Our analysis helps to expand on the existing literature regarding the risk of atrial fibrillation and incident fractures (Table 4). To date, there were four studies that examined the relationship between atrial fibrillation and incident fractures, three found an association and one did not. Lai et al. used claims data from 34,625 individuals in Taiwan. Over mean follow-up of 3.6 years, the authors found that non-traffic accident-related fractures of bones requiring hospitalization were more frequent in patients with atrial fibrillation than in matched controls (adjusted HR 1.85, 95%CI 1.50–2.30).(16) Lai et al. included skull fractures whereas we excluded them from our analysis. Kim et al. also analyzed health insurance data to select Korean patients with osteoporosis. The study included 31,778 patients among which 3.9% had atrial fibrillation. During median follow-up of 4 years, patients with osteoporosis and atrial fibrillation had increased risk of non-trauma-related bone fracture (1.21, 95%CI 1.02–1.41) compared to propensity-matched referents.(15) The population studied were over 50 years with known osteoporosis, which is different from our analysis that did not exclude those without osteoporosis. Wong et al. analyzed data from 113,600 patients from a tertiary referral center and teaching hospital in Australia.(17) Over a mean follow-up of 3 years, the authors found that history of atrial fibrillation was associated with increased risk of hip fracture (HR 1.97, 95%CI 1.61–2.42 in men; HR 2.08, 95%CI 1.80–2.39 in women). This study looked specifically at hip fractures, while our analysis included more fracture locations. In summary, these three studies from Taiwan, Korean, and Australian populations and focusing on non-traffic accident and non-trauma related fractures found similar results to ours. In contrast, Wallace et al. examined the association of incident atrial fibrillation with the risk of subsequent osteoporotic fracture in the Cardiovascular Health Study. Among 4,462 participants followed over a mean follow-up of 8.8 years, participants with incident atrial fibrillation were not at increased risk of subsequent osteoporotic fracture (HR 0.97, 95%CI 0.77–1.22).(18) Wallace et al. had a slightly larger population size but the follow-up period was shorter than our analysis. Most importantly, Wallace et al. analyzed atrial fibrillation as a time-varying variable, similar to our analysis, but did not account for the competing risk of death.

Table 4:

evidence in context

Evidence before this study

We searched MEDLINE up to August 2019 for observational studies that assessed the association between AF and incident fracture (search equation: (atrial fibrillation OR “Atrial Fibrillation”[Mesh]) AND (fracture OR “Fractures, Bone”[Mesh])). We identified 4 studies that respectively focused on non-traffic accident-related fractures, non-trauma-related bone fractures among patients with osteoporosis, osteoporotic fractures, and hip fractures. Three studies based on claims data found evidence of an association: Lai et al. reported an adjusted HR of 1.85, 95% CI 1.50–2.30;(14) Kim et al. found an adjusted HR of 1.21, 95%CI 1.02–1.41;(13) and Wong et al. found an adjusted HR of 1.97 (95%CI 1.61–2.42).(16) Another analysis based on the Cardiovascular Health Study cohort did not find evidence of an association: Wallace et al. reported an adjusted HR of 0.97, 95%CI 0.77–1.22.(15) None of the studies accounted for the competing risk of death.

Added value of our study

Based on data from the Framingham Heart Study, we found an association between atrial fibrillation and the risk of incidence fracture among men and women aged 65 years and older after adjustment for time-varying risk factors and for the competing risk of death. Results were consistent at index ages 45, 55, and 75 years.

Although an observational study cannot establish the pathophysiological mechanisms underlying a connection between atrial fibrillation and fracture, there are many plausible mechanisms including frailty, hemodynamic consequence, thromboembolic potential, and anticoagulation treatment associated with atrial fibrillation (Figure 3). There is a high prevalence of frailty among individuals with atrial fibrillation. Increased frailty is associated with increased osteoporosis and fracture.(23) Hemodynamic consequences result from the beat-to-beat blood pressure variability and cardiac output reduction. These can cause inadequate forward perfusion leading to central nervous system disturbances and impaired equilibrium triggering falls.(18) Moreover, atrial fibrillation increases the risk of embolic events.(6) Emboli to the brain can also lead to transient or permanent central nervous system deficits that can cause loss of postural equilibrium leading to falls. Additionally, emboli to the microvasculature of the bone leading to impaired osteogenesis predisposing to osteoporosis leading to fracture.(16)

Figure 3.

Possible pathophysiological mechanisms for the association between AF and subsequent fractures

The anticoagulation agents used in atrial fibrillation could represent another pathophysiological mechanism between atrial fibrillation and fracture. Vitamin K is an essential factor in gamma carboxylation of osteocalcin, which is involved in osteogenesis, and low vitamin K has been associated with reduced bone mineral density.(24) Vitamin K antagonists have long been used as oral anticoagulants and patients with long-term use have reduction in bone mineral density, which in turn increases the risk of fracture.(25) Lastly, multiple studies have shown that individuals with atrial fibrillation have a greater propensity for falls.(14, 26, 27), which could explain the higher relative rates of fracture we observed in individuals with atrial fibrillation.

In addition, there are risk factors for atrial fibrillation that also predispose to fracture. Relationships have been established between conditions like heart failure and myocardial infarction, which are risk factors for both atrial fibrillation and fracture.(9, 10) One proposed mechanism for this relationship include increased calcium excretion observed in heart failure due to elevated levels of aldosterone leading to secondary hyperparathyroidism. Also, heart failure patients have lower levels of vitamin D as compared to those without heart failure, which can result in low bone density and is an established risk factor for future cardiac events.

Our study has several strengths. A major strength is the use of longitudinally, routinely collected data from the FHS, a large community-based study with multiple cohorts followed at regular intervals. The follow-up duration in our analysis is considerably longer than in previous studies. Another strength is the protocol of atrial fibrillation and fracture ascertainments in FHS which relied on multiple data sources and careful adjudication, in contrast with many of the previous studies that used electronic health record databases. In addition, we analyzed multiple fracture sites and performed analyses at multiple index ages. Finally, we used an illness-death model to account for both atrial fibrillation as an intermediate state and for the competing risk of death.

Our study also has limitations. First, we cannot establish causal relations because residual and unmeasured confounders are possible in our observational study and both atrial fibrillation and fractures increase with advancing age. We adjusted analyses on major risk factors for both atrial fibrillation and fractures, and on changes in these variables during follow-up; we also accounted for biological age by defining age as the time scale. Second, there may be misclassification – some atrial fibrillation cases may not have been identified, given that atrial fibrillation may be transient or clinically unrecognized. FHS participants are under routine surveillance for incident atrial fibrillation, but they do not undergo continuous cardiac rhythm monitoring. Moreover, the ascertainment of non-hip fractures prior to 2002 was not confirmed via medical records. Third, atrial fibrillation was not recorded as paroxysmal or persistent, which could potentially have different effects on the risk of incident fracture given the different physiological and anatomical deviations. Fourth, statistical power was limited in subgroup analyses. Finally, the FHS participants are mostly of European ancestry. We included the Omni 1 and Omni 2 cohorts that included more diverse populations, but generalizability of our findings to individuals of non-European ancestry or other regions and countries is uncertain.

In conclusion, our study adds to the existing literature regarding atrial fibrillation and its many health outcomes. In particular, our study helps shine some light on conflicting evidence regarding atrial fibrillation and risk of subsequent fractures. In our U.S. community-based study, we found evidence that atrial fibrillation was associated with increased risk of subsequent fractures.

Clinical significance.

• Among participants of a community-based cohort study in the U.S., atrial fibrillation was associated with higher risk of subsequent bone fracture, except finger, toe, foot, skull, and facial fractures.

• There was no evidence of effect modification of the association by sex.

• There was also evidence of an association between atrial fibrillation and increased risk of osteoporotic fractures, specifically.