Bleeding After Initiation of Multiple Antithrombotic Drugs, Including Triple Therapy, in Atrial Fibrillation Patients Following Myocardial Infarction and Coronary Intervention: A Nationwide Cohort Study

The National Patient Registry has kept records of all hospital admissions since 1978. Each hospitalization is coded according to the International Classification of Diseases, 8th and 10th revisions. The registry has also included operational classification codes since 1996 (The Nordic Medical Statistics Committees Classification of Surgical Procedures). Vital status and cause of death was obtained from the civil registration system through Statistics Denmark and the National Causes of Death Register, respectively. The latter holds information about the immediate, contributory, and underlying causes of death. The Danish Registry of Medicinal Product Statistics (national prescription registry) contains information on all drug prescriptions dispensed from pharmacies in Denmark since 1995. All drugs are classified according to the international Anatomic Therapeutic Chemical System. In Denmark, all residents are provided with a unique and permanent civil registration number that enables linkage between these nationwide administrative registries.

We identified subjects diagnosed with AF who had been hospitalized for MI or PCI between January 1, 2001 and December 31, 2009. Subjects were eligible for inclusion if their first MI/PCI event in the study period was not preceded by a MI/PCI event up to 1 year before the index date. To allow prescriptions to be claimed and exclude unmeasured complications related to index hospitalization, a 7-day quarantine period was defined after discharge. Thus, the inclusion date was 7 days after discharge. Inclusion criteria were ongoing antithrombotic treatment in subjects alive, aged ≥30 years, and no registration with diagnosis of bleeding, MI or ischemic stroke during the quarantine period. Only individuals aged ≥10 years and alive January 1, 1997 were included. For an overview, see Figure 1. The PCI-only group resembles elective patients not having a MI before the procedure. The ICD codes used are available in online-only Data Supplement Table I.

Claimed prescriptions of VKA (warfarin and phenprocoumon), aspirin, and clopidogrel were used to classify the following 5 types of treatment regimen: single antiplatelet therapy (AP) with aspirin or clopidogrel; VKA monotherapy with only VKA; dual antiplatelet therapy (DAPT) with aspirin plus clopidogrel; VKA plus single AP with VKA plus aspirin or clopidogrel; or TT including all 3 drugs. For each prescription dispensed, the daily dosage was estimated for up to 7 consecutive prescriptions, enabling us to calculate the treatment regimen at any given time. This method allows for dosages and exposure status to change through time and has previously been used.⁵ Treatment regimen was set as a time-varying explanatory variable in the analyses. Ongoing treatment at inclusion for all 7 possible combinations was defined as baseline treatment. The Anatomic Therapeutic Chemical codes of the identified therapies are listed in online-only Data Supplement Table II.

Prescriptions filled up to 6 months before inclusion were recorded for the following drugs: renin-angiotensin system inhibitors; antiarrhythmic drugs (including β-blockers, calcium channel blockers, digoxin, amiodarone, and class 1C antiarrhythmic drugs); statins, glucocorticoids; nonsteroidal anti-inflammatory drugs; and proton pump inhibitors.

Previous bleeding was defined as any admission for intracranial bleeding, gastrointestinal bleeding (bleeding ulcer, hematemesis, melena, and unspecified gastrointestinal bleeding), bleeding from the respiratory or urinary tract, and anemia caused by bleeding within 5 years of inclusion.

Any previous diagnosis or operation for aortic or mitral valve disease was used to define potential valvular AF. Hypertension was defined as the collection of at least 2 different antihypertensive drugs at baseline as previously validated.¹³ A diagnosis of ischemic stroke and transient cerebral ischemia, systemic embolism, acute and chronic renal failure, peripheral arterial disease, alcohol abuse, liver disease, and malignancy within 5 years before inclusion was also determined. Diabetes mellitus was defined as treatment with glucose-lowering medication at baseline. Heart failure was defined as any admission for heart failure and use of loop diuretics as done previously.¹⁴

HAS-BLED (Hypertension, Abnormal renal/liver function, Stroke/thromboembolism, Bleeding history, Labile INR [international normalized ratio], Elderly [age >65 years], Drug consumption/alcohol abuse) scores were calculated from recorded comorbidities as previously described.¹⁵ However, data on labile INRs was not available. The use of antiplatelet therapy was not included in the score because this was an explanatory variable in the analyses. CHADS₂ score (Congestive heart failure, Hypertension, Age ≥75 years, Diabetes mellitus, Stroke/transient ischemic attack) for every patient were also calculated as done previously.¹³

Primary outcome was nonfatal or fatal bleeding defined as either first admission with a diagnosis of nonfatal bleeding registered at discharge or as death from bleeding identified through the National Causes of Death Register.^4,5 Occurrence and type of bleeding as recorded in hospital databases have shown a positive predictive value of 89% to 99%.¹⁶ If admission for nonfatal bleeding was followed by death due to bleeding within 1 week of admission, the bleeding was recorded as fatal. For secondary outcome, we used thromboembolic events of cardiovascular death or death from ischemic stroke, nonfatal MI, or nonfatal ischemic stroke. For all outcomes under investigation, only the first event was registered.

Patient characteristics according to treatment regimen at baseline are expressed as percentages or as means and standard deviations. Crude incidence rates of fatal bleeding, nonfatal bleeding, and thromboembolic events were determined within 1 year. Adjusted incidence rate ratios were calculated in a Poisson regression model. A Cox proportional hazards model was used to estimate the risk of the primary outcomes in a landmark analysis in early (0–89 days) and delayed (90–360 days) time periods, and of the secondary outcome within 1 year. For both models treatment regimen was set as a time-varying covariate. This implied that subjects were only considered at risk when taking the corresponding antithrombotic drugs. For sensitivity, a model in which the baseline treatment regimen was used as non–time-dependent exposure was also constructed. All analyses were adjusted for age, sex, year of admission, and PCI as inclusion group, whereas HAS-BLED was used for bleeding outcomes and CHADS₂ score was used for thromboembolic outcomes. Subjects were censored after 360 days from inclusion, when the follow-up period ended on December 31, 2009 or in the case of death from other causes than the outcome under investigation. The interval 180 to 360 days from inclusion was used as reference when calculating the incidence rate ratios. For the Poisson and Cox model, we tested goodness-of-fit and the proportional hazard assumption, respectively. Both models were investigated for linearity of continuous variables and any significant interactions between the available covariates. None of the assumptions were violated, and no interactions were found. A 2-sided significance level of 0.05 was used. All analyses were performed with SAS statistical package version 9.2. (SAS Institute Inc, NC) and Stata statistical package version 11.0 (StataCorp LP).

The Danish Data Protection Agency approved this study (ref: 2007-58-0015, int ref: GEH-2010-001), and data were made available such that individuals could not be identified. Retrospective register studies do not require ethical approval in Denmark.

Selection of the study population and baseline characteristics are summarized in Figure 1 and Table 1, respectively. Of 11 480 subjects in the study, 6993 (60.9%) were males. Mean age was 75.6 years (SD 10.3). The group with MI as inclusion criteria consisted of 8775 (76.4%) subjects, and hereof 1521 (17.3%) had a PCI performed within 1 week, and the solitary PCI group comprised 2705 (23.6%) subjects. Males were ≈6 years younger. Almost 10% had a history of bleeding before inclusion, and the mean HAS-BLED score was 2.1 (SD 0.9), and the mean CHADS₂ score was 1.5 (SD 1.5). All regimens containing clopidogrel were more common in patients having a PCI performed regardless of MI status.

Figure 2 demonstrates markedly raised crude incidence rates of bleeding immediately after initiation of TT, and TT bleeding rates were higher than for all other regimens regardless of time elapsed. Within 30 days, crude rates were 22.6, 20.3, and 14.3 major bleeding events per 100 person-years for TT, VKA plus single AP, and DAPT, respectively. The adjusted analysis of incidence rate ratios demonstrated that for all types of regimens, with the exception of VKA monotherapy, the risk was frontloaded (Figure 3). An adjusted landmark analysis of risk in early (0–89 days from inclusion) and delayed (90–360 days) time periods also showed continually elevated bleeding risk with TT relative to VKA plus single AP or DAPT (Figure 4). Increased hazard ratios were also seen for TT in comparison with VKA plus clopidogrel but insignificant; early (hazard ratio [HR] 1.37 [0.81;2.31] and delayed (HR 1.07 [0.66;1.73]) time periods. A cross-sectional view of the study population at days 90 and 180 from discharge showed that 989 (of 9962 subjects at risk) and 719 (of 9076 subjects at risk) were treated with ongoing TT, respectively. Hence, the proportion of subjects on ongoing TT at inclusion, day 90, and day 180 after discharge were 13.0%, 9.9%, and 7.9%, respectively.

During a mean follow-up of 288.1 days, 728 (6.3%) subjects experienced nonfatal or fatal bleeding (Table 2), and a total of 2534 (22.1%) events of the combined end point of cardiovascular death, MI, and ischemic stroke were observed. Nine of 10 fatal bleedings were intracranial or gastrointestinal of origin. Table 3 summarizes characteristics of bleeding events for each combination of therapy at 360 days of follow-up. Figure 5 shows crude incidence rates and number of thromboembolic and bleeding events. The highest bleeding rate was observed for TT (14.2 events per 100 person-years), whereas the rate of the combined end points of cardiovascular death, MI, and ischemic stroke was similar for TT and VKA plus single AP with rates of 20.1 and 19.4, respectively. The overall trend was that rates of bleeding increased, whereas rates of thromboembolic events decreased with intensity of antithrombotic treatment regimen. The adjusted Cox model showed significant increased risk of bleeding for TT relative to VKA plus single AP at 1-year (HR 1.41 [1.10;1.81]), whereas the estimates for the combined end points of cardiovascular death, MI, and ischemic stroke showed no significant benefit with TT use (HR 1.15 [0.95;1.40]).

Our main results of increased early bleeding risk for all regimens were shown to be consistent when baseline treatment regimen was used as non–time-dependent exposure (online-only Data Supplement Figure I). We calculated adjusted incidence rate ratios of subgroups consisting of male/female, elderly (aged ≥75 years/<75 years), inclusion status (MI or PCI) and CHADS₂ score ≥2 (online-only Data Supplement Figure IIA and IIB). No interactions were found, ie, the risk associated with each treatment regimen was not significantly different between the various subgroups. For patients with a definite theoretical indication for VKA therapy (ie, 7240 patients with a CHADS₂ score ≥2) crude incidence rates of bleeding were similar for all antithrombotic combinations (Table 3).

The study includes nationwide data, independent of socioeconomic status, race, employment status, and participation in health insurance programs (minimizing selection bias). We did not have clinical information to define the severity of the bleeding episode, but by using a definition of bleeding based on admissions, we presume that this event of bleeding is clinically severe enough to warrant hospitalization. Last, minor bleeding not causing hospitalization or death could not be identified, and therefore our results could underestimate the true risk associated with antithrombotic therapy. However, our study population reflects real-world patients, in contrast to patients enrolled in a randomized setting who are often highly selected with respect to low bleeding risk, and standardized management protocols are rigorously applied and may not mirror everyday practice.²⁴

Data in the registries have previously been validated, including diagnosis of MI, stroke, and AF.^25–27 Several studies demonstrate that choice of antithrombotic treatment shows wide variability,^28,29 and, in this present observational study, the primary limitations consist of confounding by indication, ie, we had no knowledge of reasons affecting the physicians' choice of antithrombotic therapy and hence may confound outcomes. This bias is expected to affect our results on bleeding risk conservatively, ie, more intensive antithrombotic therapy is most likely to be prescribed to healthier patients at a perceived lesser risk of bleeding. Furthermore, we adjusted for prognostic factors for bleeding by the validated HAS-BLED score. Aspirin is available as over the counter and hence not recorded, but chronic aspirin users collect financial reimbursement when prescribed, ensuring high persistence.⁵ We applied sensitivity analyses to ensure the robustness of our findings. To assess an effect of presumed physician's choice of treatment during admission, baseline treatment regimen was used as non–time-dependent exposure, and we also found a similar early risk of bleeding in patients with a theoretical indication of VKA therapy (CHADS₂ score ≥2). Unmeasured confounders, eg, INR levels, have not been determined, and quality of treatment in a real-life setting is difficult to establish. In a controlled environment, time in therapeutic range in Denmark is high,³⁰ but the bias of residual confounding remains.