ÌÇÐÄvlog

ImportanceÌý Vitamin K antagonist (eg, warfarin) use is nowadays challenged by the non–vitamin K antagonist oral anticoagulants (NOACs) for stroke prevention in atrial fibrillation (AF). NOAC studies were based on comparisons with warfarin arms with times in therapeutic range (TTRs) of 55.2% to 64.9%, making the results less credible in health care systems with higher TTRs.

ObjectivesÌý To evaluate the efficacy and safety of well-managed warfarin therapy in patients with nonvalvular AF, the risk of complications, especially intracranial bleeding, in patients with concomitant use of aspirin, and the impact of international normalized ratio (INR) control.

Design, Setting, and ParticipantsÌý A retrospective, multicenter cohort study based on Swedish registries, especially AuriculA, a quality register for AF and oral anticoagulation, was conducted. The register contains nationwide data, including that from specialized anticoagulation clinics and primary health care centers. A total of 40 449 patients starting warfarin therapy owing to nonvalvular AF during the study period were monitored until treatment cessation, death, or the end of the study. The study was conducted from January 1, 2006, to December 31, 2011, and data were analyzed between February 1 and November 15, 2015. Associating complications with risk factors and individual INR control, we evaluated the efficacy and safety of warfarin treatment in patients with concomitant aspirin therapy and those with no additional antiplatelet medications.

ExposuresÌý Use of warfarin with and without concomitant therapy with aspirin.

Main Outcomes and MeasuresÌý Annual incidence of complications in association with individual TTR (iTTR), INR variability, and aspirin use and identification of factors indicating the probability of intracranial bleeding.

ResultsÌý Of the 40 449 patients included in the study, 16 201 (40.0%) were women; mean (SD) age of the cohort was 72.5 (10.1) years, and the mean CHA₂DS₂-VASc (cardiac failure or dysfunction, hypertension, age ≥75 years [doubled], diabetes mellitus, stroke [doubled]–vascular disease, age 65-74 years, and sex category [female]) score was 3.3 at baseline. The annual incidence, reported as percentage (95% CI) of all-cause mortality was 2.19% (2.07-2.31) and, for intracranial bleeding, 0.44% (0.39-0.49). Patients receiving concomitant aspirin had annual rates of any major bleeding of 3.07% (2.70-3.44) and thromboembolism of 4.90% (4.43-5.37), and those with renal failure were at higher risk of intracranial bleeding (hazard ratio, 2.25; 95% CI, 1.32-3.82). Annual rates of any major bleeding and any thromboembolism in iTTR less than 70% were 3.81% (3.51-4.11) and 4.41% (4.09-4.73), respectively, and, in high INR variability, were 3.04% (2.85-3.24) and 3.48% (3.27-3.69), respectively. For patients with iTTR 70% or greater, the level of INR variability did not alter event rates.

Conclusions and RelevanceÌý Well-managed warfarin therapy is associated with a low risk of complications and is still a valid alternative for prophylaxis of AF-associated stroke. Therapy should be closely monitored for patients with renal failure, concomitant aspirin use, and poor INR control.

Atrial fibrillation (AF) is a strong independent risk factor for ischemic stroke.^1,2 Vitamin K antagonist (eg, warfarin) treatment reduces the risk of stroke by 64% and all-cause mortality by 26%, compared with control in patients with AF.³ However, warfarin confers an increased risk of hemorrhage, with intracranial bleeding the most severe effect.^4-7

For warfarin-treated patients, monitoring and close control of the international normalized ratio (INR) reduce the risk of thrombosis and bleeding.^8-14 Time in therapeutic range (TTR) and INR variability measure different aspects of warfarin treatment control: the TTR assesses the intensity of warfarin treatment,¹⁵ with a high TTR correlating with a low risk of stroke or bleeding.^8-10 International normalized ratio variability is a measure of warfarin treatment stability, with low variability correlating with a low risk of adverse events.^12-14 In Sweden, a high center-based TTR (>75%) has repeatedly been demonstrated,^10,16,17 and patients achieve high individual TTR (iTTR). Overall, warfarin treatment in Sweden has also held to a high standard when measuring rates of complications in anticoagulation clinics and primary health care setting.^11,14,18

The non–vitamin K oral anticoagulants (NOACs) are alternatives to warfarin treatment for stroke prevention in AF. Meta-analyses^19,20 in this patient group have shown that NOACs (apixaban, dabigatran etexilate mesylate, edoxaban tosylate, and rivaroxaban) are more effective than warfarin in preventing stroke and systemic embolism. The NOACs have a lower risk of intracranial bleeding compared with warfarin, but the risk of gastrointestinal tract bleeding is higher.²⁰

The NOACs have been compared with warfarin in large, global randomized trials, but the mean TTRs in these studies were 62.2% with apixaban (Apixaban for Reduction in Stroke and Other Thromboembolic Events in Atrial Fibrillation),²¹ 64.4% with dabigatran (RE-LY [Randomized Evaluation of Long-term Anticoagulant Therapy]),²² 64.9% with edoxaban (Effective Anticoagulation With Factor Xa Next Generation in Atrial Fibrillation–Thrombolysis in Myocardial Infarction 48),²³ and 55.2% with rivaroxaban (Rivaroxaban Once Daily Oral Direct Factor Xa Inhibition Compared With Vitamin K Antagonism for Prevention of Stroke and Embolism Trial in Atrial Fibrillation).²⁴ In the warfarin arm of the RE-LY trial,¹⁷ there was a significant decrease in stroke and systemic embolism, major bleeding, and mortality with higher iTTR. Patients in the NOAC arms tended to fare better in centers where warfarin therapy was better controlled with higher iTTR. In the meta-analysis by Ruff et al,²⁰ there was heterogeneity for major bleeding, which was significantly lower with high-dose NOACs in the subgroup with a center-based TTR less than 66%, but no significant difference between NOACs and warfarin in the subgroup with a center-based TTR 66% or greater.

Meta-analyses^25,26 have shown a relative risk of major bleeding of at least 2.4 when warfarin is combined with aspirin. A population-based, case-control study²⁷ confirmed the high risk of upper gastrointestinal tract bleeding in patients receiving warfarin in combination with aspirin and/or clopidogrel bisulfate.

The aim of the present study was to report the efficacy and safety of well-managed warfarin therapy in patients with nonvalvular AF. In addition, we determined the risk of complications, especially intracranial bleeding, in patients receiving concomitant aspirin, as well as the impact of iTTR and INR variability.

Box Section Ref ID

Between January 1, 2006, and December 31, 2011, we performed a retrospective, multicenter cohort study based on deidentified data from national registries in Sweden. This study was approved by the regional ethical review board in UmeÃ¥, Sweden, and conformed to the Declaration of Helsinki.²⁸

The primary data source was AuriculA,²⁹ a national quality register for AF and oral anticoagulation, which has been funded by the Swedish Association of Local Authorities and Regions since 2008. The register was started in 2006 and now includes more than 126 000 patients from 224 participating centers throughout Sweden—specialized anticoagulation clinics as well as primary health care centers. Approximately 50% of all patients receiving warfarin in Sweden are included in AuriculA. Participation in AuriculA is mostly within whole regions, with no apparent selection bias. Atrial fibrillation is the indication for warfarin therapy in approximately 64% of the patients in the register. More than 6 million INR samples are registered.²⁹ All information related to warfarin treatment for these patients that is documented in the anticoagulation centers in everyday clinical practice is transferred to the quality register automatically once every 24 hours, unless the patient has declined to participate. AuriculA also provides a clinical decision tool, using an algorithm to aid in determining the dosage of warfarin.³⁰ If certain criteria are met, the algorithm can give a dose suggestion that can be accepted or manually changed.

The Cause of Death Register³¹ includes persons with a Swedish social security number and contains their age, sex, date and cause of death, and whether autopsy was performed.

The Swedish National Patient Register (NPR)³² contains information about hospital admissions as well as visits in outpatient clinics in Sweden for all patients with a Swedish social security number. The register was launched in 1964, with complete coverage since 1987. Currently, more than 99% of all somatic and psychiatric hospital discharges are registered in the NPR. The register includes the patient’s age, sex, dates of admission and discharge, and registered primary and secondary diagnoses, as well as codes for surgical procedures, according to the International Statistical Classification of Diseases and Related Health Problems, Tenth Revision (ICD-10).

The Swedish Prescribed Drug Register³³ includes all prescriptions dispensed in Swedish pharmacies. Since 2005, the register also includes individual social security numbers.

All patients in AuriculA who started warfarin treatment due to AF from January 1, 2006, to December 31, 2011, were initially included. Patients already receiving warfarin were not included. Individuals younger than 18 years were excluded to avoid bias (n = 1). Of the remaining 40 909 patients, 460 had, in addition to AF, valve malfunction (mitral stenosis [n = 82] or mechanical prosthetic valves [n = 378]) and were therefore excluded. A final cohort of 40 449 patients, all with warfarin treatment initiated for nonvalvular AF, was included in the final analysis and followed until treatment cessation, death, or end of study period on December 31, 2011.

Between February 1 and November 15, 2015, data from AuriculA were linked to NPR, Cause of Death Register, and Prescribed Drug Register. Complications (mortality, bleeding, and thromboembolic events) and risk factors were analyzed in association with anticoagulation treatment quality as measured by iTTR and INR variability. Baseline characteristics, risk scores, determined using CHADS₂ (congestive heart failure, hypertension, age ≥75 years, diabetes mellitus, and stroke [doubled]) and CHA₂DS₂-VASc (cardiac failure or dysfunction, hypertension, age ≥75 years [doubled], diabetes mellitus, and stroke [doubled]–vascular disease, age 65-74 years, and sex category [female]) and complications were retrieved from the NPR, using ICD-10 codes (eTable 1 in the Supplement).

Major bleeding was defined according to the International Society on Thrombosis and Hemostasis criteria; however, information on hemoglobin reduction of 2.0 g/dL (to convert to grams per liter, multiply by 10) or transfusion of at least 2 U of blood is not included in the NPR. Thus, intracranial, gastrointestinal tract, or other bleeding events requiring in-hospital care were categorized as major bleeding.^34,35Thromboembolic events were defined as diagnosed stroke, transient ischemic attack, and peripheral emboli (arterial); venous thromboembolism; or myocardial infarction (ICD-10 codes reported in eTable 1 in the Supplement). To reduce the risk of overreporting, only the first complication of every subtype was included for each treatment period. For the same reason, only the main diagnosis in the NPR was used for cerebral hemorrhage or infarction as well as venous thromboembolism; however, both primary and secondary diagnoses were used to define other bleeding complications, myocardial infarctions, or baseline characteristics (eTable 1 in the Supplement). For venous thromboembolism and cerebral infarction, follow-up contact soon after the index event is common and might cause double reporting, leading to the diagnosis being addressed as a complication rather than indication for treatment. To avoid this possibility, a blanking period of 2 weeks was applied for ICD-10 codes identical to those of the index event in these patients.

Date of death was retrieved from Cause of Death Register. Death within 1 day from treatment cessation was included in all-cause mortality. For all other complications, events were counted when occurring within the treatment period. The treatment period was defined as time from start day until stop day of warfarin treatment or December 31, 2011, with information derived from AuriculA.

Time in therapeutic range, calculated according to Rosendaal et al,¹⁵ was retrieved from AuriculA, where INR and exact treatment time are registered. If there was a gap longer than 90 days between 2 INR values, this period was excluded from the patient’s TTR calculation. The cohort was divided in 2 groups regarding iTTR, with a cutoff value of 70%. An iTTR of 70% or greater was considered good anticoagulation control according to European guidelines.³⁶

Information on simultaneously used antiplatelet drugs was retrieved from the Prescribed Drug Register. Study patients with antiplatelet drugs dispensed during their warfarin treatment period (minus the last 7 days in the treatment period) were defined as patients receiving concomitant medication.

International normalized ratio variability was calculated as estimated SD of INR values around the regression line after transformation to the normal distribution, as proposed by Lind et al.¹⁴ Mean INR variability was then used as the cutoff point, dividing the cohort in halves, with INR variability lower than the mean indicating stable anticoagulation and INR variability higher than the mean indicating unstable anticoagulation.

Baseline characteristics are presented descriptively. The annualized incidence of complications was calculated as events per treatment year, with the results expressed as percentages. Cox regression analysis was used for calculating indicators of intracranial bleeding and differences in bleeding risks between patients with additional aspirin and those without additional antiplatelet therapy. Findings are reported as mean (SD) and percentage or hazard ratio (HR) (95% CI). Data were analyzed using SPSS, version 21 (SPSS Inc), and R, version 3.0.0 (R Foundation for Statistical Computing).

At the start of the study, a cohort of 40 449 patients with AF was included. The mean (SD) age was 72.5 (10.1) years, and 16 201 (40.0%) were women. The total of treatment years was 65 424. For 11 303 (27.9%) patients, no diagnosis was recorded in the NPR, indicating that warfarin treatment was initiated in primary health care settings with no prior hospital visit, and background data were therefore not available. For the remaining 29 146 (72.1%) patients with available background data, baseline characteristics are presented in Table 1.

A total of 5075 patients were receiving concomitant aspirin treatment; of these, 764 individuals were receiving dual or triple antiplatelet therapy prescribed during warfarin treatment, resulting in 4311 patients receiving only aspirin in addition to warfarin, of whom 3552 had available background data presented as baseline characteristics in Table 1. There were a total of 1287 patients receiving antiplatelet therapy in addition to warfarin that was not aspirin; of these, 1121 had available background data. In general, patients with additional aspirin treatment had more cardiovascular comorbidities, especially previous myocardial infarction, and consequently had higher mean CHA₂DS₂-VASc scores compared with patients without additional antiplatelet treatment (mean [SD], 3.9 [1.7] vs 3.2 [1.5]) (Table 1). Percutaneous coronary intervention was documented for 763 (17.7%) patients in the aspirin group and 1566 (6.4%) in the group that receiving no additional antiplatelet agent. Coronary stenting was performed during or within 12 months prior to warfarin treatment in 113 (2.6%) patients in the aspirin group compared with 149 (0.6%) in the group with no antiplatelet therapy. Cardioversion of AF was planned or performed for 8192 (20.3%) of all treated patients.

Patients with an iTTR 70% or greater had a higher prevalence of stroke or transient ischemic attack at baseline compared with patients with an iTTR less than 70%, with almost similar mean CHA₂DS₂-VASc scores in the 2 groups owing to more comorbidities in the latter subgroup (Table 1). The mean iTTR for the overall cohort was 68.6% (22.6%), with this calculated after excluding 1561 patients lacking an iTTR, since no or only 1 INR value was available. A high mean TTR with low variation between calendar years in the study was seen for the whole cohort (eFigure in the Supplement). The annual incidence of all-cause mortality was 2.19% (95% CI, 2.07-2.31) and, for intracranial bleeding, 0.44% (95% CI, 0.39-0.49) (Table 2).

Additional treatment with aspirin resulted in significantly higher incidences of any major bleeding, gastrointestinal tract bleeding, and other major bleeding compared with no additional antiplatelet therapy. No significant difference was seen between the groups of patients in risk for intracranial bleeding (Table 2). The results were consistent, even after adjustment for age, sex, and other risk factors. After adjustment, the risk for gastrointestinal tract bleeding was increased by 59% (HR, 1.59, CI, 1.24-2.02) for patients receiving aspirin compared with those without concomitant antiplatelet therapy (eTable 2 in the Supplement).

Patients with an iTTR 70% or greater had a significantly lower incidence of treatment complications overall, compared with those with an iTTR less than 70% (Table 3). For patients with low INR variability, defined as below the mean value of 0.83, significantly lower complication rates were seen for all events, except intracranial bleeding and venous thrombosis, compared with patients with high INR variability (Table 3). For patients with an iTTR 70% or greater, no statistical significance was seen in the cumulative incidence of complications when comparing groups regarding INR variability (Figure).

On Cox regression analysis (reported as HR [95% CI]), significant indicators for intracranial bleeding were renal failure, stroke, hypertension, male sex, and age at treatment initiation (Table 4). Patients with renal failure had more than a 2-fold increased risk of intracranial bleeding (2.25, [1.32-3.82]). Those with previous stroke had a 58% higher risk of intracranial bleeding (1.58 [1.19-2.10]), and, for patients with hypertension, this risk was 37% (1.37 [1.04-1.81]). The risk of intracranial bleeding increased significantly with age. For every year, the risk increased by 3% (1.03 [1.01-1.04]). Women had lower risk than men for developing an intracranial bleeding event (0.71 [0.53-0.93]) (Table 4).

In this study, our principal findings were that well-managed warfarin therapy is associated with a low risk of complications. With the results representing real-life data with no exclusions, warfarin treatment safety seems to be relatively good, with an annual incidence of 2.19% for all-cause mortality and 0.44%for intracranial bleeding—both lower than comparable results in the pivotal NOAC studies.^21-24 Therapy should be closely monitored for patients with renal failure, concomitant aspirin use, and an iTTR less than 70% or high INR variability, given the increased risk of complications.

The studied cohort of 40 449 patients with nonvalvular AF and a mean CHADS₂ score of 2.1 is somewhat comparable with the smaller cohorts in the pivotal NOAC studies.^21-24 However, our patients receive treatment in well-managed warfarin settings with good results previously shown for both center-based TTR and complication rates.¹¹ With a mean iTTR of 68.6% (compared with 55.2%-64.9% for the NOAC studies^21-24), it might be unsurprising that bleeding complication rates are lower in our cohort than in control groups in the randomized trials.

For intracranial bleeding, the most feared treatment complication for anticoagulants, we found an annual incidence of 0.44%, compared with the warfarin control groups in the randomized trials (0.70%-0.85%)^21-24 and the rate with NOACs (approximately 0.3%). The annual incidence of all-cause mortality in our cohort (2.19%) is lower than those seen in the NOAC studies, compared with both the warfarin cohorts and NOAC cohorts (3.9%-4.9% vs 3.5%-4.5%).^21-24

Our results support previous findings^26,27 of higher risk of major bleeding when warfarin is combined with aspirin, with a significantly higher annual incidence of any major bleeding, gastrointestinal tract bleeding, and other major bleeding compared with when no additional antiplatelet drugs are used with warfarin. The baseline data showed overrepresentation of cardiovascular disease in the aspirin group compared with the group receiving only warfarin, and our findings of a high incidence of thromboembolic events in the aspirin group are, therefore, as expected. As shown in the Prevention of Thromboembolic Events–European Registry in Atrial Fibrillation registry,³⁷ one could question the use of additional aspirin in most cases of our cohort, when only 2.6% of patients receiving additional aspirin had a clear indication for this therapy, with coronary stenting performed during or within 12 months before the initiation of warfarin treatment.

After adjustments for age, sex, and potential risk variables, the additional risk for bleeding associated with aspirin was 36% for any major bleeding, 59% for gastrointestinal tract bleeding, and 33% for other major bleeding. No significant additional independent aspirin effect was seen in the risk for intracranial bleeding, which contradicts previous findings.²⁵ This difference, compared with our results, may be explained by cohort variation. We excluded patients with AF and concomitant valve malfunction, but the meta-analysis conducted by Hart et al²⁵ included 4 of 6 studies based on patients with prosthetic valves and trial cohorts.

In analysis of the cohort as 2 groups regarding iTTR, with a cutoff level of 70%, most patients were in the good anticoagulation control category (≥70%).³⁶ These patients had a significantly lower risk for every analyzed event compared with patients with an iTTR less than 70%. The annual risk for intracerebral bleeding was low (0.34%) in patients with an iTTR ≥70%.

When analyzing the effect of anticoagulation stability (measured as INR variability) on warfarin treatment outcome, the cohort was divided into 2 subgroups: low INR variability (lower than mean INR variability, indicating stable anticoagulation) and high INR variability (higher than mean INR variability, indicating unstable anticoagulation). Patients with low INR variability had a generally lower risk for complications compared with those who had high INR variability. These results support previous findings^8-10,12-14 and emphasize the importance of good INR control for patients with AF who receive warfarin.

Compared with patients with an iTTR less than 70%, those with an iTTR 70% or more had the lowest incidence of general complications. In the iTTR 70% or more group, the INR variability did not seem to affect the results. Indeed, iTTR indicates the probability of events more strongly than INR variability, which contradicts previous findings of Lind et al.¹⁴

The strongest indicator of the probability for intracranial bleeding in our study was renal failure, which is a well-known risk factor for bleeding events in general³⁸ but, to our knowledge, has not previously been linked to intracranial bleeding during oral anticoagulation therapy. Other significant factors, such as advanced age, hypertension, and prior stroke, have been well recognized as clinical indicators for intracranial bleeding.³⁹ Male sex is a known risk factor for intracerebral hemorrhage in general,⁴⁰ which, together with subdural hematomas, accounts for most intracranial bleeding events.

This study had a retrospective observational design based on data retrieved from medical records and registries. Although such sources are limited by the accuracy and completeness of documentation, the validity of the NPR is good: in somatic care, the NPR lacks information about primary diagnosis in only 0.5% to 0.9% of hospital admissions. The overall positive predictive value for diagnosis is high (85%-95%) and, although the sensitivity for more severe diagnoses, such as stroke and myocardial infarction is high (>90%), it is low for less severe diagnoses, such as hypertension (8.8%-13.7%).⁴¹ Underdiagnosis of hypertension in our cohort is therefore likely, rendering falsely lower CHADS₂ and CHA₂DS₂-VASc scores.

There was a lack of baseline characteristics on patients treated solely in primary health care settings and therefore not included in the NPR. If these individuals are healthier than others, underestimation of the incidence of complications is unlikely, since events registered in this study equate to diseases requiring hospital admission or leading to death.

The definition of concomitant medication in this study did not include information on the duration of additional medication, which might limit the generalizability of the results on aspirin. Patients who were receiving warfarin with extended concomitant aspirin treatment are probably at higher risk of bleeding events than are those with shorter additional aspirin treatment.

Our study cohort is derived from Sweden with patients having a white European background. Sweden is a high-income country where stroke incidence is decreasing in comparison with low-income countries where the incidence is increasing.⁴² Thus, generalization of our findings might be difficult, especially for low-income countries or other ethnic groups.

This observational study included only patients treated with warfarin and, since it did not include a comparator group receiving other treatment, such as NOACs, one cannot draw direct conclusions from its findings on whether to use warfarin or NOACs for treatment of AF in a well-managed warfarin situation. Furthermore, comparison of results (event rates) with other studies, such as NOAC trials, is problematic due to differences in populations, data collection methods, and event definitions.

Well-managed warfarin treatment is a valid alternative in patients with AF who require anticoagulant treatments, with relatively low complication rates and low all-cause mortality. Therapy should be closely monitored in those with renal failure, concomitant aspirin use, and an iTTR less than 70% or a high INR variability. The iTTR is a strong indicator of probability for both bleeding and thromboembolic events and should be maintained at 70% or greater.

Corresponding Author: Fredrik Björck, MD, Department of Public Health and Clinical Medicine, Umeå University, SE85643 Sundsvall, Sweden (fredrik.bjorck@lvn.se).

Accepted for Publication: February 5, 2016.

Published Online: April 20, 2016. doi:10.1001/jamacardio.2016.0199.

Author Contributions: Drs Björck and Själander had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

Study concept and design: Björck, Wester, ³§Âáä±ô²¹²Ô»å±ð°ù.

Acquisition, analysis, or interpretation of data: All authors.

Drafting of the manuscript: Björck, ³§Âáä±ô²¹²Ô»å±ð°ù.

Critical revision of the manuscript for important intellectual content: All authors.

Statistical analysis: Björck, Renlund, ³§Âáä±ô²¹²Ô»å±ð°ù.

Obtained funding: ³§Âáä±ô²¹²Ô»å±ð°ù.

Administrative, technical, or material support: Svensson.

Study supervision: Lip, Svensson, ³§Âáä±ô²¹²Ô»å±ð°ù.

Conflict of Interest Disclosures: All authors have completed and submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest. Dr Lip reported having served as a paid consultant for Bayer, Astellas, Merck, Sanofi, Pfizer/Bristol-Myers Squibb, Daiichi-Sankyo, Biotronik, Medtronic, Portola, and Boehringer Ingelheim and having been a paid member of the speaker’s bureau for Bayer, Pfizer/Bristol-Myers Squibb, Boehringer Ingelheim, Daiichi-Sankyo, Medtronic, and Aventis. Dr Själander reported receiving speaker fees from Bayer, Boehringer Ingelheim, Behring, Takeda, BMS/Pfizer, Octapharma, and Novartis. No other disclosures were reported.

Funding/Support: This study was supported by the Department of Public Health and Clinical Medicine, Umeå University, and the Department of Research and Development, County Council of Vasternorrland (grants LVNFOU385111 and LVNFOU481841).

Role of the Funder/Sponsor: The funding organizations had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.