Author + information
- Received September 26, 2017
- Revision received December 1, 2017
- Accepted January 1, 2018
- Published online March 5, 2018.
- Antonios Douros, MD, PhDa,b,c,
- Laurent Azoulay, PhDa,b,d,
- Hui Yin, MSca,
- Samy Suissa, PhDa,b and
- Christel Renoux, MD, PhDa,b,e,∗ ()
- aCentre for Clinical Epidemiology, Lady Davis Institute, Jewish General Hospital, Montreal, Quebec, Canada
- bDepartment of Epidemiology, Biostatistics, and Occupational Health, McGill University, Montreal, Quebec, Canada
- cInstitute of Clinical Pharmacology and Toxicology, Charité–Universitätsmedizin Berlin, corporate member of Freie Universität Berlin, Humboldt-Universität zu Berlin, and Berlin Institute of Health, Berlin, Germany
- dGerald Bronfman Department of Oncology, McGill University, Montreal, Quebec, Canada
- eDepartment of Neurology and Neurosurgery, McGill University, Montreal, Quebec, Canada
- ↵∗Address for correspondence:
Dr. Christel Renoux, Centre for Clinical Epidemiology, Lady Davis Institute, Jewish General Hospital, 3755 Cote Sainte-Catherine Road, H425.1, Montreal, Quebec H3T 1E2, Canada.
Background Non–vitamin K antagonist oral anticoagulants (NOACs) are relatively new drugs used for stroke prevention in nonvalvular atrial fibrillation (NVAF). However, there are concerns that their use may be associated with hepatotoxic effects.
Objectives The purpose of this study was to determine whether the use of NOACs is associated with an increased risk of serious liver injury compared with the use of vitamin K antagonists (VKAs) in NVAF patients with and without prior liver disease.
Methods Using the administrative databases of the Canadian province of Quebec’s health insurances, the authors conducted a cohort study among patients newly diagnosed with NVAF between January 2011 and December 2014. Adjusted hazard ratios (HRs) and 95% confidence intervals (CIs) of serious liver injury (defined as either a hospitalization or related death) were estimated using time-dependent Cox proportional hazards models, comparing current use of NOACs to current use of VKAs separately among patients with or without prior liver disease.
Results The cohort comprised 51,887 patients, including 3,778 with prior liver disease. During 68,739 person-years of follow-up, 585 patients experienced a serious liver injury. Compared with current use of VKAs, current use of NOACs was not associated with an increased risk of serious liver injury in patients without or with prior liver disease (adjusted HR: 0.99; 95% CI: 0.68 to 1.45; and adjusted HR: 0.68; 95% CI: 0.33 to 1.37, respectively).
Conclusions Compared with VKAs, NOACs were not associated with an increased risk of serious liver injury irrespective of baseline liver status. Overall, these results provide reassurance regarding the hepatic safety of NOACs.
Non–vitamin K antagonist oral anticoagulants (NOACs) are used for the prevention of ischemic stroke in patients with nonvalvular atrial fibrillation (NVAF) (1). NOACs are direct inhibitors of coagulation enzymes, with dabigatran inhibiting Factor IIa, and rivaroxaban, apixaban, and edoxaban inhibiting Factor Xa.
Although the different NOACs have been shown to have similar or better bleeding risk profiles as vitamin K antagonists (VKAs) (2), there are emerging concerns from case reports and pharmacovigilance analyses of a possible hepatotoxic risk associated with the use of NOACs (3,4). Moreover, all NOACs have been associated with transaminase elevations (5–7). In addition, ximelagatran, a Factor IIa inhibitor, failed to obtain marketing authorization due to concerns over hepatic adverse events (8,9). Thus, current guidelines by the European Society of Cardiology on the use of NOACs recommend annual monitoring of liver function (10).
To date, 1 observational study has assessed the risk of liver injury associated with the use of NOACs, showing a decreased risk when compared with VKAs (11). However, it is unclear whether these findings are the result of methodological limitations, such as channeling of the comparator drug class (i.e., VKAs) to high-risk patients, or information bias (12). Moreover, because the study population comprised a mix of patients with and without prior liver disease, no inferences can be drawn regarding hepatotoxicity according to baseline liver status. Therefore, further studies are needed to address this important safety question.
Thus, the objective of this population-based study was to determine whether the use of NOACs is associated with an increased risk of serious liver injury compared with the use of VKAs in NVAF patients with and without prior liver disease.
This study used the computerized databases of the Canadian Province of Quebec’s health insurance (Régie de l’assurance maladie du Québec [RAMQ]), the Maintenance et exploitation des données pour l’étude de la clientèle hospitalière (MEDÉCHO), and the Institut de la statistique du Québec (ISQ). These databases have been extensively used for research purposes in the past (13–15), including for studies on oral anticoagulation in patients with NVAF (16).
The RAMQ is responsible for administering the universal health care services for all of its residents currently living in the Canadian province of Québec. Because health care coverage is mandatory for Québec residents, this database contains several millions of beneficiaries (e.g., 7,875,427 insurants in 2013). The RAMQ has 3 computerized databases: 1) the demographic database containing the age, sex, and postal code of all individuals registered; 2) the medical services database containing information on the medical services program, including nature of the service rendered, specialty of treating and referring physician, date and location, as well as the diagnostic code of the service; and 3) the prescription database containing information on outpatient prescriptions, including name, dose and amount of drug dispensed, date, prescribed number of days of treatment, and whether it was a refill or a new prescription. This program covers all individuals 65 years of age and older, welfare recipients, and all Quebec residents who do not have access to a private medication insurance program (roughly 40% to 45% of the Quebec population).
MEDÉCHO has maintained the hospital inpatient database since 1980, and it contains data pertaining to all Québec hospitalizations including date and type of admission and discharge, type of establishment, primary and secondary diagnoses, and procedure codes (with corresponding procedure dates). Since 2006, diagnoses are coded based on the enhanced version of the International Classification of Diseases-10th Revision-for Canada (ICD-10-CA), and procedures are coded based on the Canadian Classification of Health Interventions.
Finally, the ISQ administers the cause of death database, which contains vital statistics such as the date of death, the medical code corresponding to the underlying cause of death, and the establishment where the death took place. Each of these databases contains the (pseudonymized) individual’s Health Insurance Number, which is acquired at birth or at the time of residency, remains unchanged throughout the life of the individual, and is used for record linkage within the RAMQ databases and the MEDÉCHO. The general accuracy of linkage between the prescription database and the medical services database was found to be 98% (17,18).
We identified all patients ≥40 years of age, with a first inpatient or outpatient diagnosis of atrial fibrillation (ICD-10 code: I48) between January 1, 2011, when dabigatran was the first NOAC introduced for the treatment of NVAF in Quebec (dates of rivaroxaban and apixaban approval in Quebec were January 2012 and December 2012, respectively), and December 31, 2014. Cohort entry was defined as the date of the first inpatient or outpatient diagnosis of atrial fibrillation. If the diagnosis occurred during hospitalization, cohort entry was defined as the date of hospital discharge. All cohort members were required to have RAMQ medication coverage for at least 1 year prior to cohort entry to provide sufficient baseline information on comorbidities and prior medication use. To confirm the incident nature of the diagnosis of atrial fibrillation and to only include patients with NVAF, we excluded all patients with any mention of atrial fibrillation prior to cohort entry as well as those with a history of valvular mitral or aortic heart disease, valvular repair, or hyperthyroidism. To consider only new users of oral anticoagulants, we further excluded patients with a prescription of any oral anticoagulant in the year before cohort entry. Finally, we excluded patients under antiretroviral therapy or treatment with isoniazid or amoxicillin/clavulanic acid in the 3 months before cohort entry, because these medications have been associated with a strongly increased hepatotoxic risk (19); however, the prevalence of their use before cohort entry was low (Figure 1).
Within this cohort, we identified 2 separate subcohorts based on liver status at cohort entry: 1) the subcohort with prior liver disease comprised all NVAF patients with any history of infectious hepatitis, cholangitis, Budd-Chiari syndrome, Wilson’s disease, hemochromatosis, alcoholic or other toxic liver disease, liver fibrosis, liver cirrhosis, liver failure, malign or benign hepatic tumors, or liver transplantation; and 2) the subcohort without prior liver disease included the remaining NVAF patients. All cohort members were followed until the occurrence of the study outcome (defined in the following text), the patient’s date of emigration, leaving the prescription drug program, non–liver-related death, or end of study period, (i.e., December 31, 2014), whichever occurred earlier.
All outpatient prescriptions for NOACs and VKAs approved for NVAF in Quebec during the study period and dispensed during follow-up were identified. A time-dependent exposure definition was used, in which each person-day of follow-up was classified into 1 of the following 4 mutually-exclusive categories: current use of a NOAC in monotherapy (dabigatran, rivaroxaban, or apixaban), current use of a VKA in monotherapy, current use of 2 or more oral anticoagulants, and no current use. For all categories, exposed person-time was defined by the prescription duration plus a 30-day grace period. Thus, patients were considered continuously exposed if the duration of one prescription overlapped with the date of the next prescription, using the grace period in the event of 2 nonoverlapping successive prescriptions. Because NOACs represent the alternative to VKAs (1), the reference category for all analyses consisted of current use of VKAs.
The outcome of interest was serious liver injury, defined as hospitalization with liver injury or death due to liver injury, using the respective ICD-10 codes (K71, K72, K74.6, K75.9, K76.2, K76.9, Z944) in the MEDÉCHO (anywhere in the hospitalization record) and the ISQ (as the underlying cause of death). Codes from the International Classification of Diseases classification system have been successfully used to assess adverse hepatic outcomes in the past (20,21).
All models were adjusted for risk factors known to be associated with liver injury and that might also influence the decision to initiate treatment with oral anticoagulants. Demographic characteristics were measured at cohort entry and consisted of age, sex, and year of cohort entry. Age was modeled flexibly as a continuous variable using restricted cubic splines with 5 interior knots to account for a potentially nonlinear relationship with the outcome (22). The models were also adjusted for congestive heart failure, dyslipidemia, and diabetes mellitus recorded at any time before cohort entry, because these conditions have been associated with liver disease (23,24). Moreover, we adjusted for use of acetaminophen in the year prior to cohort entry, because acetaminophen-related hepatotoxicity is the most common cause of drug-induced liver injury, and the prevalence of its use in our cohort was high (Table 1) (25). Finally, we adjusted for the number of nonanticoagulant drugs in the year prior to cohort entry as a proxy for overall health.
We used descriptive statistics to summarize the characteristics of the entire cohort and the 2 subcohorts defined by prior liver disease status. Crude incidence rates for serious liver injury with 95% confidence interval (CIs) based on the Poisson distribution were calculated for each exposure group. We used time-dependent Cox proportional hazards models to estimate hazard ratios (HRs) and 95% CIs of serious liver injury associated with current use of NOACs compared with VKAs. The models were adjusted for the potential confounders listed previously. All analyses were conducted separately in the 2 subcohorts.
We conducted 4 secondary analyses. First, we assessed whether there was a duration-response relation between current use of NOACs and risk of serious liver injury, categorizing current use according to 3 pre-defined durations (<3, 3 to 6, and >6 months). Second, we assessed whether current use of each individual NOAC was associated with an increased risk of serious liver injury compared with current use of VKAs. For this analysis, in the subcohort with prior liver disease, the Factor Xa inhibitors rivaroxaban and apixaban were grouped together due to power issues. Third, to investigate a possible effect modification by age, we assessed whether the risk of serious liver injury associated with NOACs differed between patients ≤75 years of age and patients >75 years of age. Finally, to investigate a possible effect modification by sex, we assessed whether the risk of serious liver injury associated with NOACs differed between sexes.
We conducted 8 sensitivity analyses to assess the robustness of our findings. First, to assess possible exposure misclassification, we repeated the primary analysis using grace periods of 60 days and of 180 days between successive prescriptions. Second, we used a stricter outcome definition by considering only diagnoses for hepatic failure (ICD-10 code: K72). Third, we adjusted for 16 additional covariates to assess the potential for residual confounding (Online Methods 1). Fourth, we used the disease risk score approach as an alternate means to control for confounding (26). The disease risk score was calculated using a historical cohort of patients diagnosed with NVAF between January 1, 2007, and December 31, 2010. We performed multivariate logistic regressions to assess the association between each covariate used in the primary analysis and liver injury, and subsequently estimated the baseline risk of liver injury for each member in the study cohort. The final outcome model was then stratified on disease risk score quintiles. Fifth, we conducted a propensity score-based analysis as an additional effort to control for confounding (27). We identified patients in our cohort initiating oral anticoagulation with an NOAC or a VKA, and defined cohort entry as the date of the first NOAC or VKA after the NVAF diagnosis. Next, we performed a multivariate logistic regression to estimate the probability of receiving an NOAC versus VKA, conditional on all variables listed in the paper plus “time from NVAF diagnosis to treatment initiation” as an additional covariate. We then trimmed patients with nonoverlapping propensity score distributions. The remaining patients were followed from cohort entry until they switched from NOACs to VKAs or vice versa, discontinued treatment, or experienced the outcome, whichever occurred first. Patients were allowed to switch between different NOACs (e.g., from dabigatran to rivaroxaban) or VKAs during follow-up. Finally, the HR of liver injury was estimated using a Cox proportional hazards model adjusted for propensity score quintiles, separately for patients with and without prior liver disease. Sixth, to assess the potential effect of time-varying confounding, we repeated the primary analysis using a marginal structural Cox proportional hazards model (Online Methods 2) (28) Finally, an additional analysis was conducted accounting for competing risk due to death using the subdistribution model proposed by Fine and Gray (29). All analyses were conducted with SAS version 9.4 (SAS Institute, Cary, North Carolina) and R (R Foundation for Statistical Computing, Vienna, Austria).
Among the 48,109 patients without prior liver disease, 319 were hospitalized with or died due to liver injury (crude incidence rate: 4.9 events per 1,000 persons/year). During follow-up, 9,137 patients initiated treatment with NOACs, and 13,899 patients initiated treatment with VKAs. Among NOAC initiators, 489 (5.4%) switched to VKAs, whereas 3,128 (22.5%) VKA initiators switched to NOACs. Moreover, there were 266 hospitalizations with or deaths due to liver injury among 3,778 patients with prior liver disease (crude incidence rate: 69.4 events per 1,000 persons/year). During follow-up, 347 patients initiated treatment with NOACs and 859 patients initiated treatment with VKAs. Among NOAC initiators, 14 (4.0%) switched to VKAs, whereas 167 (19.4%) VKA initiators switched to NOACs.
Table 1 presents the characteristics of the NVAF patients who received NOACs versus VKAs at cohort entry according to prior liver disease status. In both subcohorts, NOAC users were younger; were less likely to have a history of congestive heart failure, dyslipidemia, or diabetes mellitus; and were less likely to have used acetaminophen compared with users of VKAs. Moreover, they were more likely to have entered the cohort in the last 2 years of the study period.
Table 2 shows the results related to NOAC use and the risk of serious liver injury in the subcohort without prior liver disease. Current use of NOACs was not associated with a higher risk of serious liver injury compared with current use of VKAs overall (crude incidence rates: 3.9 vs. 4.5 per 1,000 persons/year; adjusted HR: 0.99; 95% CI: 0.68 to 1.45), and for the 3 pre-specified categories of duration of use. Furthermore, drug-specific analyses revealed similar risks for dabigatran, rivaroxaban, and apixaban. Age and sex did not modify the association between NOAC use and risk of serious liver injury (Online Tables 1 and 2). The results of the sensitivity analyses are summarized in Central Illustration and presented in detail in Online Tables 3 to 10. Overall, the results remained consistent with those of the primary analysis, except for the sensitivity analysis, where only hepatic failure was considered for the outcome definition (adjusted HR: 0.59; 95% CI: 0.30 to 1.16) (Online Table 5). However, this stricter outcome definition analysis was based on few exposed events.
Table 3 presents the results related to NOAC use in the subcohort with prior liver disease. Current use of NOACs was associated with a numerically decreased but not statistically significant risk of serious liver injury compared with current use of VKAs overall (crude incidence rates: 24.5 vs. 44.8 per 1,000 persons/year; adjusted HR: 0.68; 95% CI: 0.33 to 1.37), and for durations of use <3 and 3 to 6 months, but not for durations of use >6 months. Numerically decreased but not statistically significant risks were also observed in the drug-specific analyses for dabigatran and the 2 Factor Xa inhibitors, rivaroxaban and apixaban. However, the analyses on duration-response relation and drug-specific effects were based on very few exposed events. Age and sex did not modify the association between NOAC use and risk of serious liver injury (Online Tables 11 and 12). The results of the sensitivity analyses remained consistent with those of the primary analysis (summarized in the Central Illustration, presented in detail in Online Tables 13 to 20).
Our study assessed the risk of serious liver injury associated with the use of NOACs in the distinct populations of NVAF patients without prior liver disease and NVAF patients with prior liver disease. Compared with the use of VKAs, the use of NOACs was not associated with an increased risk of serious liver injury irrespective of liver status at baseline, either overall, by duration of use, or by individual agents. The findings of the primary analyses remained consistent in several sensitivity analyses.
To date, conflicting results exist regarding the hepatotoxic potential of NOACs (3,4,11). Case reports and analyses of pharmacovigilance data have suggested an increased risk for NOACs, especially rivaroxaban (3,4). However, issues including general under-reporting of adverse events, selective increased reporting for recently approved drugs such as NOACs, and the often incomplete data contained in pharmacovigilance databases may limit the validity of these findings (30).
Recently, an observational study showed a 43%, 12%, and 30% decreased risk of liver injury associated with dabigatran, rivaroxaban, and apixaban, respectively, when compared with VKAs in patients with NVAF (11). However, the authors used a population that comprised a mix of patients with and without prior liver disease (approximately 5% of the patients had been previously diagnosed with liver disease), excluding only patients with a hospitalization due to liver injury in the 3 months before cohort entry. Therefore, effect modification due to underlying liver disease or channeling of VKAs to patients with underlying liver disease and thus at higher risk of developing liver injury could have accounted for these results (12). Moreover, the use of an intention-to-treat exposure definition may have introduced misclassification of exposure. The intention-to treat approach generally leads to a nondifferential misclassification of exposure resulting in biased risk estimates toward a null effect. However, because switching from VKAs to NOACs is observed more often than switching from NOACs to VKAs (31,32), a differential misclassification of exposure resulting in biased risk estimates away from a null effect cannot be excluded. Finally, defining liver injury based on symptoms with low specificity such as jaundice, and potentially extrahepatic entities such as biliary tract disorders, may have introduced misclassification of the outcome, another potential source of bias in observational studies.
Our study assessed the risk of serious liver injury associated with the use of NOACs in NVAF patients with and without prior liver disease separately. Therefore, the results of each subcohort analysis can be generalized to the respective populations. Moreover, we used a time-dependent exposure definition, thus mitigating misclassification of exposure. However, to account for the possibility of liver injury occurring or diagnosed after drug discontinuation, we also used an extended grace period of 60 and 180 days in sensitivity analyses. Finally, our definition of serious liver injury was based solely on liver specific entities leading to hospitalization or death, thus reducing misclassification of the outcome.
In NVAF patients without prior liver disease, we observed that NOACs were not associated with an increased risk of serious liver injury compared with VKAs. In NVAF patients with prior liver disease, we observed a numerically decreased but not statistically significant risk of serious liver injury associated with the use of NOACs. Given the small size of this subcohort leading to lower precision with relatively wide 95% CIs, a firm conclusion regarding the potential safety of NOACs compared with VKAs cannot be drawn. A decreased risk of liver injury for NOACs could result from their lower degree of hepatic metabolism and elimination compared with VKAs (3), because drugs with less hepatic involvement in their pharmacokinetics have been associated with reduced hepatotoxicity (33). However, the aforementioned channeling of VKAs to high-risk patients could also have accounted for the results. Indeed, given the recent concerns over the potential hepatotoxicity of NOACs as opposed to the well-known safety profile of VKAs, which have only rarely been associated with hepatic adverse events (34,35), physicians may preferentially prescribe VKAs to patients with a higher perceived risk. Moreover, patients who are newly prescribed NOACs are generally younger and healthier than those who are newly prescribed VKAs (36,37), which was also the case in our study.
Study strengths and limitations
Our study has a number of strengths. First, we assessed the risk of serious liver injury associated with the use of NOACs in NVAF patients with and without prior liver disease separately. Thus, while providing evidence that the hepatotoxic risk of NOACs in patients with healthy liver function is comparable to VKAs, our results also present a valuable contribution on the risk profile of NOACs in patients with impaired liver function, given the scarcity of drug safety data in the setting of underlying hepatic disease (38). Second, the population-based setting and the few exclusion criteria applied during the formation of our study cohort make the results of this study highly generalizable.
Our study also has some limitations. First, because the RAMQ does not contain laboratory values, our outcome definition was based only on ICD-10 codes and we could not consider liver enzyme elevations. Although this may have reduced the sensitivity of our outcome definition, we included only cases of liver injury requiring hospitalization or leading to death, which are more likely to be valid in databases (39). Therefore, we do not expect any substantial misclassification of the outcome. Second, given the observational nature of the study, residual confounding is possible. In particular, channeling of high-risk patients to VKAs could be a source of bias. Indeed, VKA patients were older and had a higher prevalence of hepatotoxic risk factors at baseline. To mitigate this potential limitation, we stratified patients based on liver status at baseline and adjusted for well-established potential confounders. Moreover, we conducted several sensitivity analyses, such as adjusting for 16 additional covariates, using disease risk score and propensity score as alternate means to control for confounding, and using a marginal structural Cox proportional hazards model to assess potential time-varying confounding; these analyses led to results consistent with those of the primary analysis. Finally, due to the relatively small size of the subcohort with prior liver disease, we cannot exclude a decreased risk of serious liver injury associated with the use of NOACs in this population. Moreover, the effect estimates for specific NOACs, and especially for the 2 Factor Xa inhibitors rivaroxaban and apixaban, require replication due to the low number of exposed events.
Our results provide reassurance regarding the hepatic safety of NOACs. Compared with VKAs, NOACs were not associated with an increased risk of serious liver injury in NVAF patients with or without previously diagnosed liver disease. Thus, liver status should not be a central part of physician decision making regarding the appropriate oral anticoagulation for stroke prevention in NVAF.
COMPETENCY IN MEDICAL KNOWLEDGE: In patients with newly identified atrial fibrillation, target-specific oral anticoagulants are associated with no greater risk of liver injury than VKAs, independent of antecedent liver disease.
TRANSLATIONAL OUTLOOK: Larger studies are needed to compare the effects of the individual target-specific oral anticoagulants on hepatic function.
This work was supported by an infrastructure grant from the Canadian Foundation for Innovation, and the database was acquired thanks to unrestricted funding from Bayer Pharma AG. The sponsors had no other role in the study. Dr. Douros has received a research fellowship from the German Research Foundation (Deutsche Forschungsgemeinschaft). Dr. Azoulay has received a Chercheur-Boursier Award from the Fonds de recherche du Québec–Santé and a William Dawson Scholar award from McGill University. Dr. Suissa has received the James McGill Professorship award; has received research grants from Bayer Pharma, Boehringer Ingelheim, and Bristol-Myers Squibb; and has participated in advisory board meetings or as a speaker for AstraZeneca, Boehringer Ingelheim, and Novartis. All other authors have reported that they have no relationships relevant to the contents of this paper to disclose.
- Abbreviations and Acronyms
- confidence interval
- hazard ratio
- International Classification of Diseases-10th Revision, for Canada
- Institut de la statistique du Québec
- Maintenance et exploitation des données pour l’étude de la clientèle hospitalière
- non–vitamin K antagonist oral anticoagulant
- nonvalvular atrial fibrillation
- Régie de l’assurance maladie du Québec (Canadian Province of Quebec’s health insurance)
- vitamin K antagonist
- Received September 26, 2017.
- Revision received December 1, 2017.
- Accepted January 1, 2018.
- 2018 American College of Cardiology Foundation
- ↵EMA. Xarelto: Annex I: summary of product characteristics. 2017. Available at: http://www.ema.europa.eu/docs/en_GB/document_library/EPAR_-_Product_Information/human/000944/WC500057108.pdf. Accessed January 15, 2018.
- ↵EMA. Eliquis: Annex I: summary of product characteristics. 2017. Available at: http://www.ema.europa.eu/docs/en_GB/document_library/EPAR_-_Product_Information/human/002148/WC500107728.pdf. Accessed January 15, 2018.
- ↵EMA. Pradaxa: Annex I: summary of product characteristics. 2017. Available at: http://www.ema.europa.eu/docs/en_GB/document_library/EPAR_-_Product_Information/human/000829/WC500041059.pdf. Accessed January 15, 2018.
- ↵EMA. AstraZeneca withdraws its application for Ximelagatran 36-mg film-coated tablets. 2006. Available at: http://www.ema.europa.eu/docs/en_GB/document_library/Press_release/2010/02/WC500074073.pdf. Accessed January 15, 2018.
- Alonso A.,
- MacLehose R.F.,
- Chen L.Y.,
- et al.
- Raschi E.,
- Ponti F.D.
- Renoux C.,
- Coulombe J.,
- Suissa S.
- Hussaini S.H.,
- Farrington E.A.
- Bui C.L.,
- Kaye J.A.,
- Castellsague J.,
- et al.
- Lamberts M.,
- Staerk L.,
- Olesen J.B.,
- et al.
- Loo S.Y.,
- Dell'Aniello S.,
- Huiart L.,
- Renoux C.
- Yao X.,
- Abraham N.S.,
- Alexander G.C.,
- et al.
- Teschke R.,
- Danan G.