EvaluatioN of ApiXaban in strOke and systemic embolism prevention in patients with non‐valvular atrial fibrillation in clinical practice Setting in France, rationale and design of the NAXOS: SNIIRAM study

Abstract Non‐vitamin K antagonists oral anticoagulants (NOACs) have recently challenged vitamin‐K antagonists (VKAs) for stroke and systemic embolism prophylaxis in patients with non‐valvular atrial fibrillation (NVAF). Nevertheless, little information is available in routine clinical practice for France. The aim of this study is to describe the effectiveness and safety of apixaban, rivaroxaban, dabigatran or VKAs in routine clinical practice in adult NVAF patients for the prevention of stroke and systemic embolism in France. The NAXOS study is a nationwide observational retrospective cohort generated from the French national healthcare insurance database (SNIIRAM—a comprehensive in‐ and outpatient healthcare consumption database), consisting of eight distinct sub‐cohorts of anticoagulant‐naive or anticoagulant‐experienced patients diagnosed with NVAF, newly initiated with either NOACs (dabigatran, rivaroxaban or apixaban) or VKAs. Patients will be included if initiating a new anticoagulant treatment for AF during the study period from 1 January 2014 to 31 December 2016. Primary effectiveness outcome will be the incidence of stroke or systemic thromboembolic events; primary safety outcome will be the incidence of major bleeding during the exposure period. The NAXOS study will provide routine clinical practice data on the effectiveness and safety profiles of apixaban vs other NOACs and VKAs in the prevention of stroke and systemic embolism in adult patients with NVAF in clinical practice conditions in France.


| INTRODUCTION
Atrial fibrillation (AF) can lead to significant mortality, morbidity, and cost. 1,2 Long-term prophylaxis with anticoagulation therapy is recommended to prevent stroke and systemic embolization in patients with AF presenting an independent risk factor for stroke. 3 Four non vitamin-K antagonists oral anticoagulants (NOACs) are currently available: the direct factor Xa inhibitors (rivaroxaban, apixaban, and edoxaban) and the direct factor IIa inhibitor (dabigatran). These four NOACs have demonstrated, in randomized clinical trials, a consistent favorable risk-benefit profile across a wide range of patients with reductions in stroke or systemic embolism, intracranial hemorrhage, and mortality but increased gastrointestinal bleeds when compared with warfarin. [4][5][6][7][8] Clinical practice data regarding routine use of the different NOACs largely mirrors the results of clinical trials. [9][10][11][12][13] In addition, a recent study using the French medico-administrative databases (SNIIRAM and PMSI), including patients with NVAF who initiated dabigatran or rivaroxaban was previously reported but did not include data on apixaban. 14  • To describe demographic and clinical characteristics.
• To estimate the risk of occurrence of a composite morbidity criterion of stroke, systemic embolism, and major bleeding.
• To estimate the risk of all-cause mortality.
• To describe treatment patterns at anticoagulant initiation and over time.
• To estimate the healthcare resource utilization.
• To estimate the off-label use of apixaban in the French valvular AF population initiating apixaban. Patients will be followed from the date of their first OAC prescription until the end of follow-up, which is defined by the occurrence of a following event, whichever comes first: 1. Switch to another anticoagulant treatment: defined as dispensation of another anticoagulant molecule recorded after initiation of the studied anticoagulant treatment. Date of switch (and end of follow-up) will be the date of the first dispensation of the other anticoagulant molecule.
2. Discontinuation: a patient who remains more than 30 days after the coverage by the last dispensation of studied anticoagulant treatment without refilling it will be considered as stopping this treatment. If a hospitalization occurs within this period, the length of the hospital stay will be deducted from the number of days without refilling of the treatment. For patients treated with VKAs, International Normalized Ratio testing realized during a hospitalization will be counted as a VKA dispensation. International Normalized Ratio testing will be used as a proxy of a VKA prescription only to extend VKA treatment exposure, but it will not be used as an index date. OACs drug coverages will be derived based on the recommended dosage, and the coverage of VKA treatment exposure will be defined as 35 days (ie, the median coverage time in days of VKA treatments, based on prior assessment on claims F I G U R E 1 Study design. AC, anticoagulant; AF, atrial fibrillation; NVAF, non-valvular atrial fibrillation database). The median number of coverage days will also be calculated in the VKA patient population after data extraction.
3. Last patient's health record: defined by the last care (ie, consultations, dispensations, medical procedures…) recorded in the database before a period of 6 months without any reimbursed care.
This includes situations like emigration and geriatric homes entry. 4. Patient's death.

5.
End of study period.

| Data source
The data source used for this study was the SNIIRAM database, which is the French national healthcare insurance system database with individual anonymous information of primary care and secondary care (hospital data from the French Hospital Discharge database [PMSI]), and it covers currently 98.8% of the country population. The SNIIRAM database contains a large number of variables, as previously described. 15 The access to the SNIIRAM is regulated and requires approval from the "Institut des Données de Santé" (IDS, Institute of health data) and the "Commission Nationale Informatique et Libertés" (CNIL, the French data protection commission).

| Study population
Patients will be included in the study if they are covered by the Patients will be excluded of the study cohorts if they had different types of anticoagulant treatment at the index date, diagnosed with a valvular condition in the 24 months before their anticoagulant initiation or treated for an indication other than stroke prevention in AF in the 6 weeks before their first anticoagulant reimbursement (including index date).
The study population will be identified in the database through consecutive steps as indicated in Figure 2 and detailed in Supporting Information Appendix 2.
Anticoagulant-naive patients will be defined as patients with no dispensation of anticoagulant treatment during the 24 months before the index date, and anticoagulant-experienced patients corresponding to the others. Patients in the experienced group will only be included at their first entry. Therefore, all patients aged ≥18 years and covered by the French national health insurance general scheme, with at least one reimbursement for OAC treatments (VKAs, apixaban, rivaroxaban, or dabigatran) between January 2014 and December 2016, and without use of the same OAC in the 24 months before the index date (ie, date of the first dispensation) will be included. Only the first dispensation meeting this criterion determined the index date, that is, a patient could be included only once in the study.
Then, patients selected were allocated to four distinct sub-cohorts of OAC-naive patients (ie, with no dispensation of any OAC during the 24 months before the index date), receiving either VKAs, apixaban, rivaroxaban, or dabigatran during the study period.
Anticoagulant-experienced patients corresponded to the others.

| Study outcomes
Primary outcomes: • Primary effectiveness outcome will be the risk of stroke or systemic thromboembolic events which includes ischemic stroke, hemorrhagic stroke, and systemic thromboembolic events.
• Primary safety outcome will be the risk of major bleeding which includes intracranial bleeding, gastric duodenal and rectum F I G U R E 2 Global flow chart of the study population. AC, anticoagulant; AF, atrial fibrillation; VKA, vitamin-K antagonist bleedings, acute posthemmorrhagic anemia, intraocular bleeding, otorrhagia, pericardic bleeding, respiratory bleeding, hemoperitoneum, intra articular bleeding, uterine and vaginal bleeding, and other major bleedings.
Variables used to capture effectiveness and safety outcomes are summarized in Table 1.
Secondary outcomes: • Risk of occurrence of a composite morbidity criterion: unadjusted incidence rate in each sub-cohort, composite morbidity criterion being defined by stroke, systemic thromboembolic events, and/or major bleeding, whichever occurs first.
• Risk of all-cause mortality: unadjusted incidence rate of all-cause death in each sub-cohort.
• Major characteristics of patients and comorbidities, which are summarized in Table 2.   from studied anticoagulant treatment and re-exposed after the switch during a 1-year period, in each sub-cohort.
• Healthcare resources utilization description: use of medical visits, nurse acts, drugs packages delivered, biology acts, medical procedures, hospital stays, and sick leaves will be reported as: The number and percentage of patients with at least one reimbursement of a such healthcare resource use during the followup period.
The mean, SD, median, percentiles 10 and 90 of the number of care, only in patients with at least one reimbursement of the healthcare during the follow-up period.
The study is registered at ClinicalTrials.gov, number NCT02640222. No other substudies have yet been planned.

| Ethics and confidentiality of study data
This study does not require review and approval by ethics committees or informed consent. The confidentiality of records that could identify patients within the database will be protected, respecting the privacy and confidentiality rules in accordance with the applicable regulatory requirements. This database analysis will use anonymous patient's data.

| Primary effectiveness and safety outcomes
For primary effectiveness outcome, estimates by anticoagulant treatment and for both populations (anticoagulant-naive and anticoagulant-experienced patients) of: • The number and percentage of patients presenting at least a stroke and/or a systemic embolic event during the 12 months after index date, and during the overall follow-up period. This description will be also conducted by diagnoses sub-division presented in Table 1.
• The unadjusted incidence rate: number of first event of stroke and/or systemic embolism by 100 person-years and during the overall follow-up period will be estimated by anticoagulant treatment. Person-years are defined with the sum of follow-up durations of the at-risk population. The corresponding two-sided 95% CI will also be represented using Poisson distribution.
• The time-to-event of first occurrence of stroke and/or systemic thromboembolism will be estimated and plotted using cumulative incidence function (the cumulative probability of failure from a specific cause over time) accounting for differences in follow-up time and the competing risk of mortality (if mortality is greater than 10%). In the presence of competing risks of mortality, informative censoring of the observation time at death will be used, as mortality will be considered a competing risk for observing stroke and systemic thromboembolic events since patients who die before an event is observed cannot go on to have an event. In the absence of competing risks, the differences in follow-up time will be accounted for through non-informative censoring of the observation time at the end of follow-up for each patient. Overall, the causes of censoring of each treatment group will be reported once the results will become available. Cumulative incidence curves will present the event risk over time (within first 12 months) for each newly initiated anticoagulant treatment while accounting for the competing risk of mortality (if needed) and differences in followup time.
• Median duration to occurrence of first event of stroke and/or systemic thromboembolism: median duration (in days) between the index date and the date of the first studied event will be computed for patients with the event.
For primary safety endpoint, estimates by anticoagulant treatment and for both populations (anticoagulant-naive and anticoagulantexperienced patients) will be addressed by using the same methods as described for the primary effectiveness endpoint.
In order to account for confounding in the comparative analyses, several methods will be performed to compare groups of patients with the similar characteristics including: adjustment on propensity score (estimation of the average treatment effect in the entire population if the treatment might be offered to every member of the population, ATE), matching on propensity score (estimation of the average treatment effect of matched subjects who ultimately received the treatment, ATT), and two other comparative approaches as sensitivity analyses (using high dimensional propensity scores, and adjustment on confounders).
The main method will consist in propensity score adjustment to compare groups of patients with the same characteristics. The propensity score is the probability of treatment assignment conditional on observed baseline characteristics. Propensity score is used when selection bias due to non-random treatment assignment is likely, because of observational data. Three different propensity scores will be performed according to the comparison (apixaban vs dabigatran, apixaban vs rivaroxaban or apixaban vs VKAs). The propensity score conventional method will be used, which is to derive a score based on all the confounding factors previously identified. The probability to be treated with a given therapeutic combination will be estimated using a logistic regression model (with all confounding factors and treatment as dependent variable). Several checks will be performed to ensure a good balance of propensity score and of covariates between apixaban and comparison groups: • First, the treatment group propensity score distribution will be analyzed with a graphical representation. If no overlap between distributions is observed, patients will be defined as different, so comparisons would not be performed.
• Then, the balance of propensity score across treatment and comparison groups will be checked, with comparison of means. The balance of covariates across treatment and comparison groups will be checked using standardized difference. 17 If no balance, a new propensity score model will be specified (deletion of variables or transforms). 17 For all other descriptive objectives, the analyses will be performed in the eight sub-cohorts.

| Secondary outcomes
Estimation by anticoagulant treatment and for both populations (anticoagulant-naive and anticoagulant-experienced patients) as follows: • Quantitative variables will be described with the sample size, mean, standard deviation, median, interquartile range, minimum and maximum.
• Qualitative and ordinal variables will be described with the sample size and frequencies.
Analyses in anticoagulant-naive and anticoagulant-experienced populations as follows: • The major characteristics of patients will be described for each anticoagulant treatment.
• Treatment patterns at anticoagulant initiation, over time, and concomitant treatment will be tabulated by anticoagulant treatment.
• Time-to-discontinuation will be estimated and plotted using cumulative incidence function by anticoagulant treatment.
• Healthcare resources utilization will be described by anticoagulant treatment.
For each endpoint of interest in the study, only the first evidence of its event will be considered in the analyses. Cox models have been chosen to align with the methodology used in other OAC publications.
As this study has been conducted to address a specific request from French health authorities, the comparability of the methods with other OAC publication was an important contributor of the analytical choices made a priori.
Cox models or Fine Gray models are planned to be used in the analyses. The final choice of model will be based on the observed death rates, that is, in case of a high likelihood of competing risk of death, Fine Gray models will be used rather than Cox models.

| DISCUSSION
Registration clinical trials, such as RE-LY, 4 ROCKET-AF, 5 or ARIS- while it is reassuring that the body of emerging clinical practice data on NOACs in other countries largely mirrors that in clinical trials, [9][10][11][12][13] only little data is available for France. To the best of our knowledge, the NAXOS study will be the first nationwide clinical practice study in  analyses of this study could also include complementary comparative approaches, such as using IPTW, which is another and probably better method to estimate an average treatment effect in the entire population if the treatment might be offered to every member of the population (ATE).

| CONCLUSION
The NAXOS study will allow the generation of new data regarding the characteristics and management of NVAF patients and unique data on the effectiveness and the safety of NOACs in a clinical practice setting in France.

ACKNOWLEDGMENT
The NAXOS study is supported by Bristol-Myers Squibb and Pfizer.