Atrial fibrillation in fracture patients treated with oral bisphosphonates

Authors


  • Results were presented in part as oral presentations at the European Calcified Tissue Society Meeting, Barcelona, on 26th of May 2008 and in the 24th International Conference on Pharmacoepidemiology & Therapeutic Risk Management, Copenhagen, on 20th of August 2008.

Bo Abrahamsen, MD, PhD, Department of Internal Medicine, Copenhagen University Hospital, DK-2900 Hellerup, Gentofte, Denmark. (fax: +45 3977 7121; e-mail: b.abrahamsen@physician.dk).

Abstract.

Objectives.  To determine if patients receiving oral bisphosphonates are at excess risk of atrial fibrillation (AF), stroke and myocardial infarction.

Design.  Register-based restricted cohort study.

Setting.  National Hospital Discharge Register and National Prescriptions Database (1995–2005).

Subjects.  Fracture patients beginning bisphosphonates (n = 15 795) were matched with unexposed fracture patients of the same age, sex and fracture type (n = 31 590).

Results.  Incidence rates of AF were 16.5/1000 person years in untreated fracture patients and 20.6/1000 person years in bisphosphonate users. An age- and sex-adjusted hazard ratio (HR) of 1.29 (1.17–1.41) was found for probable AF by Cox proportional hazards analysis. The effect size was reduced to HR of 1.18 (1.08–1.29) by adjustment for co-medications and comorbidity. Selective prescribing was suggested by the observation that (i) risks were increased even in patients who stopped therapy after the first packet and (ii) risks were not increased by high adherence. Bisphosphonate-exposed patients were at increased risk of hospital-treated AF [adjusted HR: 1.13 (1.01–1.26)], but the risk amongst bisphosphonate users was inversely proportional to adherence. There was no increased risk of ischaemic stroke and an increased risk of myocardial infarction was not significant after adjustment for comorbidity.

Conclusions.  The increased occurrence of AF in fracture patients who are users of oral bisphosphonates should be attributed to targeting of bisphosphonates to patients who are already at increased risk of cardiovascular events.

Introduction

Since demonstration of their anti-fracture efficacy more than a decade ago [1–5], bisphosphonates have been the mainstay in osteoporosis treatment. Recently, once yearly intravenous (i.v.) zoledronic acid was shown to reduce the incidence of vertebral and hip fractures by 70% and 41%, respectively [6]. An unexpected increased incidence of atrial fibrillation (AF) was also reported in this randomized, placebo-controlled study comprising 7805 patients with osteoporosis. More patients in the zoledronic acid group compared with the control group developed arrhythmia (6.9% vs. 5.3%, < 0.003). Also, the incidence of serious AF (leading to hospitalization or disability or judged to be life-threatening) was significantly higher in the zoledronic acid. This unexpected finding prompted Cummings et al. [7] to re-address the data from the Fracture Intervention Trial (FIT). In the 6459 patients randomized to oral alendronate or placebo, a trend towards higher incidence of serious AF was seen in the alendronate group (P = 0.07), though no increase was seen in the overall occurrence of AF (P = 0.42). No increased incidence of AF was found in a re-analysis of data from placebo-controlled trials on risedronate comprising approximately 15000 patients [8] or in a recent study on zoledronic acid in patients with hip fracture [9]. In drug licensing, the best practice approach is not to adjust for multiple testing of adverse events as it is vital to identify all potential risks of new drugs even at the cost of false-positives [10]. Could AF have been a false-positive observation? AF occurred with no correlation in time to exposure to zoledronic acid, which seems incompatible with a simple acute phase effect or with short-term changes in electrolytes. Similarly, the increase in incidence of serious AF but not in AF per se is unexplained and the disease mechanism remains unknown. Two case–control studies have recently been published, yielding conflicting conclusions [11, 12]. We therefore used national prescription and health data to conduct a large, population-based cohort study on the incidence of AF, myocardial infarction (MI) and stroke in fracture patients exposed to oral bisphosphonates and in matched controls.

Study population and methods

Study cohort

Since 1995, the National Hospital Discharge Register has maintained a record of all hospital diagnoses in the country, with diagnoses and dates linked to civil registry numbers. To establish cohorts for analysis we identified all patients with fractures in Denmark between 1 January, 1996 and 31 December, 2005 (n = 327 213), who had been treated on an inpatient or outpatient basis.

Data on redeemed prescriptions

Since 1995, data on all prescriptions dispensed from Danish pharmacies have been collected in the National Prescription Database. This database contains data on individual level (identified by civil registry numbers) of ATC codes, strength, number of tablets and date of redemption. Repeatable prescriptions generate a record each time the medication is dispensed.

Identification of patients exposed and not exposed to bisphosphonates

The analysis was undertaken as a restricted cohort study. This approach seeks to mimic – within an observational setting – the patient selection process and intention to treat (ITT) principle known from clinical trials [13]. We identified all patients who began bisphosphonate treatment (n = 20 741) within the fracture cohort. Fourteen patients had received prescriptions for i.v. bisphosphonates. For each oral bisphosphonate-exposed patient, we then assigned two randomly selected bisphosphonate-unexposed fracture controls, fulfilling all the following criteria: (i) same sex, (ii) born in the same year, (iii) fracture at same anatomical location (hip, spine, shoulder, forearm or ‘other’), (iv) sustained index fracture prior to matched case redeeming first bisphosphonate prescription and (v) alive on the date when matched case redeemed first bisphosphonate prescription. After first randomizing the sequence of bisphosphonate exposed fracture cases and unexposed fracture controls, a procedure of assignment and testing was reiterated until matching was achieved (Fig. 1). Successful 2 : 1 matching meeting the above eligibility criteria was possible for 15 795 of 20 727 oral bisphosphonate cases. We then excluded patients from both arms with suspicion of recent or chronic AF at baseline using the criteria described below. We did accept 1 : 1 matching where subjects were excluded due to baseline AF. WHO International Classification of Diseases (ICD10) modified Charlson co-morbidity Index [14, 15] was calculated from hospital contacts in the year preceding prescription of bisphosphonate.

Figure 1.

 The fracture cohort comprised all patients treated at any hospital for fractures in Denmark between 1 January, 1996 and 31 December, 2005 (n = 32 7213). Patients exposed to bisphosphonate were identified and two controls were matched for age, sex and fracture location for each case selected (see methods, section 3 for further details on matching). AF, atrial fibrillation; i.v., intravenous.

Data on outcome

The primary outcome for the analyses was incident probable AF, defined as the first redemption of a prescription for a cardiac glycoside and/or the first occurrence of an ICD10 hospital diagnosis code of I48.9, AF and atrial flutter. We excluded patients with a record of prior or baseline AF and all patients who had redeemed digoxin prior to the date of their first bisphosphonate prescription. Secondary outcomes were hospital-treated AF (ICD10 code I48.9), MI (I21.0–I21.9) and ischaemic stroke (I63.0–I63.6).

Ethics

The use of data from the discharge register and the prescriptions database required specific permission from the National Board of Health, and data access itself was subject to approval from the National Bureau of Statistics pursuant to guidelines approved by the Data Protection Agency. We did not have access to any information containing patient names, addresses or civil registry numbers.

Statistical analyses

Data are shown as mean ± SD or median [range]. Patient characteristics were compared by t-tests and chi-squared analysis as appropriate. We used Cox proportional hazards models, incorporating length of time from the date of first bisphosponate redemption to AF, death or end of study (31 December 2005) to obtain crude and adjusted hazard ratios (HR). In the event of a significant outcome, prespecified subgroup analyses were carried out. Medication possession ratio (MPR) was calculated by dividing the number of defined daily doses redeemed within the at-risk period with the length of this period in days. Adjustment was also made for age, sex and the use of cardiovascular medications (any anti-hypertensive drug, any anti-diabetic drug including insulin, any anti-thrombotic agent or any cholesterol-lowering agent) in the previous year. Finally, in the fully adjusted analysis, the total number of co-medications and the Charlson Index were also included.

For outcomes where event rates were higher in patients treated with bisphosphonates, we performed two additional analyses aimed at addressing selective prescribing versus tentative causality. If bisphosphonates cause AF, event rates could be higher in patients who adhere to treatment compared with those who do not. By contrast, if patients who begin bisphosphonates simply represent a high-risk group in terms of cardiovascular disease, we would expect to observe an increased risk of AF even in those patients who cease treatment almost immediately. The first analysis was carried out by entering MPR as a time-dependent covariate in patients treated with bipshosphonates and test for effects of MPR on risk and the second analysis entailed comparing risks between patients who redeemed only the first packet of a bisphosphonate (10% of patients) and those who went on to redeem additional packets. Analyses were carried out using spss version 13.0 and 14.0 (SPSS Inc., Chicago, IL, USA).

Results

Exposure

Table 1 shows the baseline characteristics of fracture patients exposed to bisphosphonates (38 615 patient years, mean follow-up time 2.7 years) and unexposed matched fracture controls (77 574 patient years, mean follow-up time 2.7 years). The number of co-medications was higher in bisphosphonate-exposed cases (8.4 ± 6.0 vs. 6.4 ± 5.2, < 0.01) in part due to more widespread use of anti-hypertensives (< 0.01), anti-diabetics (< 0.01) and hormone therapy (P = 0.05). Alendronate (62.2%) and etidronate (30.8%) were by far the most widely used bisphosphonates (Table 2).

Table 1.   Baseline characteristics of bisphosphonate-exposed fracture cases and nonexposed fracture controls in the 2 : 1 matched analysis of outcome rates
 Bisphosphonate exposed, n = 14 302Matched controls, n = 28 731P-value
Age74.3 ± 8.874.3 ± 8.8Matched
Male sex10.9%10.9%
Hip fracture22784575
Spine fracture9521912
Humerus fracture14682951
Forearm fracture41958423
Other fracture540910 870
Anti-thrombotics23.9%23.3%0.13
Anti-hypertensives51.1%48.3%<0.01
Anti-diabetics4.6%6.2%<0.01
Cholesterol-lowering drugs6.8%6.9%0.88
Hormone therapy (in female population)13.6%12.9%0.05
Number of co-medications8.4 ± 6.06.4 ± 5.2<0.01
Table 2.   Distribution of oral bisphosphonates in the treated cohort
ATC code, bisphosphonate usedn%
M05BA04, Alendronate982162.2
M05BA01, Etidronate487030.8
M05BB01, Etidronate/calcium7564.8
M05BA07, Risedronate3071.9
M05BA02, Clodronate320.2
M05BA06, Ibandronate90.1
Total15 795100.0

Characteristics of patients who developed AF

Incidence rates for probable AF were 16.5/1000 person years in untreated fracture patients and 20.6/1000 person years in fracture patients exposed to bisphosphonates. The duration of exposure and cumulative dose of bisphosphonate before probable AF varied widely with a median exposure time of 1.8 years (range: 0.0–9.4 years) and a median cumulative dose of 280 (WHO defined daily dose; range: 14–2572). Within the bisphosphonate-exposed cohort, patients who developed AF were older, used a larger number of co-medications and were more likely to use anti-hypertensives or anti-thrombotic drugs (Table 3). By contrast, fewer patients used cholesterol-lowering drugs. A multivariate Cox regression analysis identified the following factors as associated with increased risk of incident probable AF after exposure to bisphosphonates: male sex (HR: 1.35 [1.01–1.80], < 0.05), increasing age (each year, HR: 1.06 [1.05–1.07], < 0.001), number of co-mediciations (for each co-medication, HR: 1.04 [1.02–1.05], < 0.001) and use of anti-hypertensive drugs (HR: 1.35 [1.12–1.62], < 0.01).

Table 3.   Characteristics for bisphosphonate-exposed fracture patients who did or did not develop atrial fibrillation (AF)
 Bisphosphonate-exposed fracture patients who developed AF, n = 797Bisphosphonate-exposed fracture patients who did not develop AF, n = 13505P-value
  1. See methods section for definition of probable AF.

Age (years)75.2 ± 8.170.6 ± 9.1<0.001
Male sex (%)11.010.40.30
Hip fracture (%)18.115.1<0.05
Spine fracture (%)8.86.3
Humerus fracture (%)9.910.4
Forearm fracture (%)25.830.2
Other fracture (%)37.438.0
Anti-thrombotics (%)28.021.7<0.001
Anti-hypertensives (%)60.747.0<0.001
Anti-diabetics (%)4.64.10.47
Cholesterol-lowering drugs (%)4.46.5<0.05
Hormone therapy (in female population) (%)12.313.90.32
Number of co-medications9.3 ± 5.77.9 ± 5.8<0.001

Main analysis

The unadjusted risk of all four outcomes was increased in patients exposed to oral bisphosphonates (BP) (Table 4).

Table 4.   Risk of primary and secondary outcomes: any oral bisphosphonate
  Rate in treated versus untreated group (1000 person years)UnadjustedAdjusted for sex, age, use of cardiovascular medicationsa, number of co-medications and Charlson Index
  1. aCardiovascular medications were defined as any anti-hypertensive, anti-thrombotic, anti-diabetic or cholesterol-lowering drug redeemed in the year preceding initiation of bisphosphonate.

Probable atrial fibrillation20.6 vs. 16.51.29 (1.17–1.41)1.18 (1.08–1.29)
 Men1.57 (1.20–2.06)1.39 (1.04–1.86)
 Women1.26 (1.15–1.32)1.16 (1.06–1.28)
 ≤75 years1.32 (1.13–1.54)1.13 (0.97–1.33)
 >75 years1.26 (1.13–1.40)1.20 (1.07–1.34)
Hospital-treated atrial fibrillation13.7 vs. 11.21.22 (1.09–1.36) 1.13 (1.01–1.26)
 Men1.33 (0.96–1.86)1.21 (0.85–1.72)
 Women1.21 (1.08–1.36)1.12 (0.99–1.26)
 ≤75 years1.16 (0.97–1.40)1.04 (0.86–1.26)
 >75 years1.23 (1.08–1.41)1.17 (1.02–1.34)
Ischaemic stroke1.3 vs. 1.11.14 (0.80–1.62)1.06 (0.74–1.52)
 Men0.66 (0.25–1.81)0.62 (0.22–1.79)
 Women1.25 (0.86–1.82)1.16 (0.78–1.70)
 ≤75 years1.14 (0.69–1.88)0.95 (0.57–1.59)
 >75 years1.12 (0.68–1.85)1.16 (0.70–1.92)
Myocardial infarction8.3 vs. 7.11.17 (1.02–1.34) 1.06 (0.92–1.22)
 Men1.51 (1.08–2.10)1.25 (0.88–1.79)
 Women1.12 (0.97–1.31)1.03 (0.88–1.20)
 ≤75 years1.33 (1.07–1.66)1.14 (0.91–1.44)
 >75 years1.06 (0.89–1.23)1.00 (0.84–1.20)

Probable AF.  An unadjusted HR of 1.29 (1.17–1.41) was found for probable AF in the Cox’ proportional hazards model. When adjusted for co-medications and Charlson Index, the HR remained significantly increased at 1.18 (1.08–1.29). Male sex (HR: 1.32 [1.11–1.58]), increasing age (HR: 1.07 [1.06–1.07]), use of anti-hypertensive (HR: 1.48 [1.32–1.66]) and anti-diabetic drugs (HR: 1.26 [1.02–1.56]) and the number of co-medications (HR: 1.04 [1.03–1.05]), all indicated increased risk of AF. However, patients who stopped therapy after the first pack had a higher risk of probable AF than patients who remained on therapy (Fig. 2). MPR showed a significant, inverse relationship with risk in the bisphosphonate-treated patients (HR: 0.80 [0.69–0.94], < 0.01, Fig. 3).

Figure 2.

 Hazard ratios for atrial fibrillation in fracture patients exposed to oral bisphosphonates (n = 14 302) and matched untreated fracture patients (n = 28 731). Adjusted estimates are shown for the total population and for subgroups. Results adjusted for age, sex, use of anti-diabetic, anti-hypertensive, cholesterol-lowering or antithrombotic drugs, modified Charlson comorbidity score and number of co-medications. Men: 1497 treated and 3077 untreated. Women: 12 805 treated and 25 654 untreated. Alendronate: 8871 treated and 17 742 untreated. Etidronate: 5116 treated and 10 232 untreated. Redeemed one packet only: 1772 treated and 3566 untreated.

Figure 3.

 Relationship between medication possession ratio (MPR) and risk of probable atrial fibrillation. Solid line indicates regression spline; dotted lines, 95% CI; and individual datapoints, the mean hazard ratio effects for percentiles of MPR. Cox proportional hazards model with MPR as a time-dependent covariate.

Hospital-treated AF.  The risk of hospital-treated AF was higher in BP users (unadjusted HR: 1.22 [1.09–1.36], fully adjusted HR: 1.13 [1.01–1.26]). In addition, increasing age and use of anti-thrombotic agents were significant indicators of greater risk. There was no influence of other medications or Charlson Index. The risk was higher in patients who ceased therapy after redeeming one packet than in patients who continued therapy (Fig. 2) and the risk declined with increasing MPR (HR: 0.82 [0.66–0.99]).

Stroke.  Rates of ischaemic stroke were low and did not differ significantly between bisphosphonate exposed and unexposed patients (Tables 4 and 5 and Fig. 4).

Table 5.   Risk of primary and secondary outcomes: stratification by type of bisphosphonate
  Rate in treated versus untreated group (1000 person years)UnadjustedAdjusted for sex, age, use of cardiovascular medicationsa, number of co-medications and Charlson Index
  1. aCardiovascular medications were defined as any anti-hypertensive, anti-thrombotic, anti-diabetic or cholesterol-lowering drug redeemed in the year preceding initiation of bisphosphonate.

  2. No significant difference was found between patients exposed to alendronate and etidronate except for the outcome hospital treated atrial fibrillation (see results section).

Probable atrial fibrillation
 Alendronate16.7 vs. 12.01.39 (1.22–1.58)1.30 (1.14–1.48)
 Etidronate28.5 vs. 23.81.20 (1.04–1.37)1.08 (0.94–1.24)
Hospital-treated atrial fibrillation
 Alendronate11.2 vs. 8.01.40 (1.20–1.62)1.30 (1.11–1.53)
 Etidronate13.1 vs. 11.91.10 (0.95–1.26)1.01 (0.87–1.16)
Ischaemic stroke
 Alendronate1.0 vs. 0.91.04 (0.63–1.72)0.89 (0.53–1.50)
 Etidronate1.3 vs. 1.21.11 (0.71–1.73)1.10 (0.69–1.73)
Myocardial infarction
 Alendronate9.9 vs. 7.61.30 (1.07–1.59)1.18 (0.97–1.45)
 Etidronate8.2 vs. 7.51.09 (0.91–1.30)1.02 (0.85–1.22)
Figure 4.

 Hazard ratios for myocardial infarction and stroke in fracture patients exposed to oral bisphosphonates (n = 14 302) and matched, untreated fracture patients (n = 28 731). Please see legend of Fig. 2 for details.

Myocardial infarction.  The risk of MI was significantly increased in patients exposed to bisphosphonates (unadjusted HR: 1.17 [1.02–1.34]), but this was not significant after adjustment for co-medications and comorbidity (HR: 1.06 [0.92–1.22]). Male sex (HR: 3.33 [1.10–10.10]) and increasing age (HR: 1.11 [1.05–1.18]) added significantly to the risk. The relationship between MPR and risk was inverse, with patients showing good adherence being at lower risk of MI (HR: 0.77 [0.60–0.99]).

Subgroup analyses

Bisphosphonate type.  Alendronate use was indicative of significantly increased risk of AF after adjustment for comorbidity and co-medications (Table 5), whereas this was not the case for etidronate. Entering type of bisphosphonate (alendronate vs. etidronate as reference) into the survival analysis yielded the following HRs: probable AF (HR: 1.10; 95% CI: 0.96–1.28; P = 0.17), hospital-treated AF (HR: 1.23; 95% CI: 1.03–1.47; < 0.04), MI (HR: 1.10; 95% CI: 0.89–1.39; P = 0.38) and ischaemic stroke (HR: 0.99; 95% CI: 0.55–1.76; P = 0.95). Tests for interaction with year of first prescription were negative.

Age above or below 75 years.  In patients below 75 years (Table 4), bisphosphonate use was not significantly associated with the risk of any of the four outcomes after adjustment for covariates. In patients aged 75+ years, bisphosphonate use was significantly associated with the risk of both AF outcomes.

Sex.  Men accounted for only 10% of the cohorts. Nevertheless, when analyses were restricted to men, the association with probable AF – by far the most common event in the study – remained statistically significant (HR: 1.39 [1.04–1.86]) even after full adjustment. No increased risk could be shown for the other three outcomes when men were analysed separately.

Discussion

This National Cohort Study showed a moderately increased incidence of AF in fracture patients prescribed oral bisphosphonates compared with matched fracture patients who had not received bisphosphonates. The risk was greater in users of alendronate than in users of the less potent bisphosphonate etidronate. This possible difference across compounds is interesting as a significant borderline increase in the occurrence of AF was reported from a clinical trial with alendronate [7] but ruled out in trials with risedronate [8].

Is it plausible that bisphosphonates could cause atherosclerosis or arrhythmia? Animal studies indicate that bisphosphonates may accumulate in the arterial wall and inhibit arteriosclerosis [16] and affect arterial contraction [17]. Bisphosphonates may also increase the occurrence of inflammation and rupture of atherosclerotic plaques in the apolipoprotein-E knock-out mouse model of atherosclerosis [18]. In the present study, an increased overall risk of MI disappeared after adjustment for comorbidity and there was no increased risk of stroke.

In observational studies such as this factors associated with choice of therapy may also be risk factors for study outcome (confounding by indication) and careful weighing of evidence is called for. Two key findings of this study certainly suggest that anti-osteoporotic drugs are targeted preferentially to patients who are already at increased risk of AF and that bisphosphonates may not further increase this risk. First, the effect size was diminished with increasing adherence. An inverse relationship between degree of exposure and event rate argues strongly against causality. Second, the HR was above 1.0 even amongst patients who had only redeemed one packet of an oral bisphosphonate. This implies that the effect is either immediate – an interpretation that is not borne out by the observations with zoledronate [6] – or that patients were already at increased risk when treatment was initiated. Interestingly, a similar discussion exists regarding beta-blocker use and reduced risk of hip fracture. Using the GPRD and RLS health databases, deVries [19] recently demonstrated that patients who use beta-blockers are at decreased risk of hip fracture, but that the risk reduction is independent of cumulative dose and significant even in patients who just started treatment.

This work was undertaken after the Horizon trial reported a HR for serious AF of 2.6 with zoledronic acid [6] and data from the FIT trial [7] had shown a trend towards increased relative risk of 1.3 for serious AF in patients on alendronate. Neither randomized trial found increases in probable (including nonserious) AF. The Horizon (3876 patients in active arm) and FIT (1841 in active arm) trials had limited power to assess rarer events such as the risk of stroke and MI. After the unexpected finding of increased AF, Horizon cases were adjudicated and a subsample of 559 patients had an electrocardiogram taken before and 9–11 days after the third infusion. In these participants, no increased occurrence of AF was seen. By including all Danish patients with a fracture treated at hospitals on and in-patient or out-patient basis within a 10-year period who began oral bisphosphonates, we were able to set up restricted cohorts and undertake an analysis at the national level. In order to produce a risk re-estimate that could be compared with the results from standard ITT trials, all patients who received at least one dose of oral bisphosphonate were counted as BP exposed even if therapy ceased after the first packet. We also provided analyses specific to persistent users and tracked the influence of MPR on risk. The matching procedure was rigourous to the extent of matching each 70 year-old man with a hip fracture who had begun a bisphosphonate to two 70-year old men with hip fractures who were unexposed to bisphosphonates. Though matched for fracture history, groups still differed with respect to co-morbidity and co-medications and we adjusted analyses for age, sex, co-medications and Charlson Index. We also specifically took into account the use of drugs suggesting cardio- or cerebrovascular disease.

The secondary end-point hospital-treated AF was defined on the basis of ICD10-diagnosis assigned at hospital discharge or in the outpatient clinic. Validity of data from the National Hospital Discharge Register is high. Though no studies have been performed to validate the specific diagnosis of AF in this register, the registered diagnosis could be confirmed from case records in 97% and 93% regarding ischaemic stroke [20] and MI [21], respectively.

However, mild AF may not result in hospital contact. Thus, in order to capture incident AF whether treated by the hospital specialist, the primary care physician or the private practitioning cardiologist, we defined probable AF as the first redemption of a prescription for a cardiac glycoside and/or the first occurrence of a hospital diagnosis code of I48.9. We aimed for a sensitive rather than a specific end-point. Though glycosides may be prescribed for other indications, in particular heart failure, their use is chiefly in AF or atrial flutter. We excluded patients with a record of recent or chronic AF and all patients who had redeemed digoxin prior to the date of their first bisphosphonate prescription as patients with probable prevalent AF. This led to exclusion of 9.5% vs. 9.1% in the two arms. We considered exclusion also of patients who received warfarin and anti-arrhythmics at baseline as potential candidates for chronic AF. We decided against this on the grounds that many patients receive these classes of drugs for diseases other than AF (pulmonary embolism, mechanical valve prostheses, stroke and even hypertension). We were concerned that these could be the very patients who would develop AF if challenged with oral bisphosphonates and that excluding them could induce bias leading to inappropriate exoneration of bisphosphonates. Cardiovascular disease is a risk factor for fracture [22] with 1 of 3 hip fracture patients having heart disease [23]. Accordingly, the prevalence of AF at the time of initiation of bisphosphonates was higher in our tertiary prevention setting than the prevalence reported in men and women of the same age in the latest assessment in the Copenhagen City Heart study [24].

The study has limitations. In observational studies, treatment is not allocated randomly but decided upon by the treating physician and influenced to a varying degree by reimbursement limitations and professional guidelines as well as by expectations and preferences of the patient. Despite these limitations, properly conducted observational studies often provide effect estimates that very closely match those of randomised controlled trials [13, 25]. The restricted cohort design has been developed specifically to adopt key principles of clinical trials including an intended ITT type analysis, a time scale for exposure and use of inclusion and exclusion criteria similar to those of clinical trials. Despite this, there may be differences in unmeasured confounders such as smoking or alcohol consumption that are risk factors for both AF [26] and fracture [27]. Similarly, shared genetic factors may predispose to both cardiovascular disease, stroke [20] and osteoporosis [28]. Higher Bone Mineral Density (BMD), another unmeasured confounder in health register studies, could be one reason why a fracture patient is not prescribed bisphosphonates. BMD could act as an indicator of general health and also reflect differences in smoking, alcohol intake and exercise, factors strongly associated with the risk of fractures and cardiovascular disease [29]. Finally, it is important to recognize that these results are specific to patients who begin bisphosphonates after a fracture. It is unknown if these results would apply to patients who begin bisphosphonates after a densitometric diagnosis of osteoporosis, in the absence of a prior fracture.

The strength of this study is that it is population-based, that it is large enough to capture rare events and that patients are not lost to follow-up. In Denmark, almost all hospitals are public and provide treatment free of charge. The risk of selection bias in identifying fracture cases for inclusion in our study is therefore small. We used 2 : 1 matching to maximize study power. However, the matching algorithm was very strict and led to exclusion of about 20% of potential study subjects because two controls could not be provided. This slightly reduced the number of elderly patients with hip fractures in the analysis. Thus, the initial study population had a mean age of 76.5 years and the final study population a mean age of 74.3 years. In the initial population, 19.7% had hip fracture and 9.9% had spine fracture. In the final population, this was 15.9% and 6.7%, respectively. A greedier algorithm would have provided more cases for analysis along with slightly better external validity, though afforded poorer confounder control. Ascertainment bias is likely because patients who receive medications for a chronic disease such as osteoporosis may be prone to be diagnosed with other ailments, including AF. Of the studies addressing this question so far, this study included the largest number of bisphosphonate-exposed patients and was the only study to pursue not only AF but also potentially related end-points. In the past, two case–control studies addressed the association between alendronate and AF, reaching conflicting conclusions. Heckbert et al. [12] found an increased risk of AF in ever-users of alendronate in a clinical practice setting, with risk estimates that overlapped those in this study. By contrast, a regional population-based Danish case–control study numbering 3862 bisphosphonate-exposed cases found no evidence of increased risk of AF [11] (hospital-treated AF, adjusted relative risk: 0.95, 95% CI: 0.84–1.07 vs. HR of 1.13, 95% CI: 1.01–1.26 in this study). Compared with these case–control studies, our matched cohort design allowed calculation of absolute incidence rates for multiple outcomes which can be weighed against the expected reductions in absolute fracture incidence.

We conclude that fracture patients who are prescribed oral bisphosphonates are at moderately increased risk of AF. This is particularly pronounced for users of alendronate. However, the risk declines in persistent users and may be caused by selective targeting of bisphosphonates to patients who are already at increased risk of AF at baseline.

Conflicts of interest

Dr Abrahamsen receives consultancy fees from Nycomed and Novartis, research grants from Roche and speaker’s fees from Servier, Eli Lilly and MSD. Dr Eiken receives speaker fees from Nycomed, Roche and Servier. Dr Brixen receives consultancy fees from Servier, Novartis, Eli Lilly, Nycomed and Osteologix as well as speaker’s fees from Eli Lilly & Co., Novartis, Roche, MSD and Nycomed.

Acknowledgements

The financial support of Kaptajnløjnant Harald Jensen og Hustrus Fond is gratefully acknowledged.

Ancillary