CNS Drugs (2019) 33:383–395 https://doi.org/10.1007/s40263-019-00611-9 ORIGINAL RESEARCH ARTICLE Risk of Acute Liver Injury in Agomelatine and Other Antidepressant Users in Four European Countries: A Cohort and Nested Case–Control Study Using Automated Health Data Sources Manel Pladevall‑Vila1,2 · Anton Pottegård3 · Tania Schink4 · Johan Reutfors5 · Rosa Morros6,7,8,9 · Beatriz Poblador‑Plou10 · Antje Timmer11 · Joan Forns1 · Maja Hellfritzsch3 · Tammo Reinders4 · David Hägg5 · Maria Giner‑Soriano6,7,8 · Alexandra Prados‑Torres10 · Miguel Cainzos‑Achirica1 · Jesper Hallas3 · Lena Brandt5 · Jordi Cortés6,7,8,12 · Jaume Aguado1 · Gabriel Perlemuter13,14,15 · Bruno Falissard16 · Jordi Castellsagué1 · Emmanuelle Jacquot17 · Nicolas Deltour17 · Susana Perez‑Gutthann1 Published online: 4 March 2019 © The Author(s) 2019 Abstract Background Agomelatine is a melatonin receptor agonist and serotonin 5-HT2C receptor antagonist indicated for depression in adults. Hepatotoxic reactions like acute liver injury (ALI) are an identified risk in the European risk management plan for agomelatine. Hepatotoxic reactions have been reported for other antidepressants, but population studies quantifying these risks are scarce. Antidepressants are widely prescribed, and users often have risk factors for ALI (e.g. metabolic syndrome). Objective The goal was to estimate the risk of ALI associated with agomelatine and other antidepressants (fluoxetine, par- oxetine, sertraline, escitalopram, mirtazapine, venlafaxine, duloxetine, and amitriptyline) when compared with citalopram in routine clinical practice. Method A nested case–control study was conducted using data sources in Denmark, Germany, Spain, and Sweden (study period 2009–2014). Three ALI endpoints were defined using International Classification of Diseases (ICD) codes: primary (specific codes) and secondary (all codes) endpoints used only hospital discharge codes; the tertiary endpoint included both inpatient and outpatient settings (all codes). Validation of endpoints was implemented. Adjusted odds ratios (ORs) and 95% confidence intervals (CIs) for current use were estimated for each data source and combined. Results We evaluated 3,238,495 new antidepressant and 74,440 agomelatine users. For the primary endpoint, the OR for agomelatine versus citalopram was 0.48 (CI 0.13–1.71). Results were also < 1 when no exclusion criteria were applied (OR 0.37; CI 0.19–0.74), when all exclusion criteria except alcohol and drug abuse were applied (OR 0.47; CI 0.20–1.07), and for the secondary (OR 0.40; CI 0.05–3.11) and tertiary (OR 0.79; CI 0.50–1.25) endpoints. Regarding other antidepressants versus citalopram, most OR point estimates were also below one, although with varying widths of the 95% CIs. The result of the tertiary endpoint and the sensitivity analyses of the primary endpoint were the most precise. Conclusion In this study, using citalopram as a comparator, agomelatine was not associated with an increased risk of ALI hospitalisation. The results for agomelatine should be interpreted in the context of the European risk minimisation measures in place. Those measures may have induced selective prescribing and could explain the lower risk of ALI for agomelatine when compared with citalopram. Most other antidepressants evaluated had ORs suggesting a lower risk than citalopram, but additional studies are required to confirm or refute these results. The study protocol was registered in the European Medicines Agency electronic Register of Post-Authorisation Studies (EU PAS Register # EUPAS10446). Dr. Castellsagué is now retired. Electronic supplementary material The online version of this article (https​://doi.org/10.1007/s4026​3-019-00611​-9) contains supplementary material, which is available to authorized users. Extended author information available on the last page of the article Vol.:(0123456789) 384 M. Pladevall‑Vila et al. and drug abuse and dependence, and metabolic syndrome Key Points [10–12]. Other risk factors for ALI include older age, female sex, concurrent use of hepatotoxic medications, previous Agomelatine did not increase the risk of hospitalisation acute and chronic hepatic, biliary, and pancreatic conditions, due to acute liver injury (ALI) in the studied populations malnutrition, HIV infection, and chronic inflammatory dis- when compared to citalopram. Results were robust in eases [13, 14]. multiple sensitivity analyses. Two previous studies on the risk of ALI associated with In the populations studied, risk minimisation measures the use of duloxetine have been conducted using Ingenix were in place and may have contributed to the lower risk Research Data Mart and the Optum Research Database in found for agomelatine when compared with citalopram. the United States [15, 16], suggesting an increased risk of Thus, compliance with relevant contra-indications, ALI with duloxetine when compared with venlafaxine [15, precautions of use, and biological liver testing before and 16] and selective serotonin reuptake inhibitors [15]. during treatment are still required to prescribe agomela- The primary goal of this post-authorisation safety study tine. [17], using a nested case–control analysis, was to evaluate the risk of hospitalisation for ALI associated with agomela- Of the other antidepressants, sertraline, escitalopram, tine and eight other antidepressant drugs as used in current mirtazapine, venlafaxine, duloxetine, and amitriptyline medical practice compared with citalopram. had a lower risk of ALI hospitalisation when compared to citalopram. 2 Methods 2.1 Study Design and Study Period 1 Background In brief, we conducted a large, multinational, retrospective Agomelatine (Valdoxan, Thymanax) is a melatonin receptor longitudinal cohort and nested case–control study compar- agonist and serotonin 5-HT2C antagonist indicated for major ing new users of agomelatine (main exposure of interest) depressive episodes in adults [1]. Hepatotoxic reactions are and new users of eight other study antidepressants with new an identified risk of agomelatine included in the European users of citalopram (common reference group). Citalopram risk management plan and the drug label, which recommend was selected as a comparator because it was the most com- that aminotransferase levels are checked before treatment monly used antidepressant in three of the four countries initiation and then after 3, 6, 12, and 24 weeks and following and, according to literature reviews available at the time of a dose increase [2]. While other antidepressant drugs used writing the protocol, was among the antidepressants with in Europe do not have similar recommendations, hepato- the least potential for hepatotoxicity [3, 4]. The other anti- toxic reactions do occur with some of them. The severity and depressants were selected because they were commonly frequency of those reactions vary among antidepressants. used across all countries. The study period in each data According to studies based only on case reports, they seem source started after the launch of agomelatine in the respec- to be more common and severe with use of tricyclic antide- tive country (in 2009 or 2010) and ended with the last year pressants and monoamine oxidase inhibitors than with use for which data were available in each data source (2013 or of serotonin norepinephrine reuptake inhibitors and selective 2014). The full study protocol can be accessed at the Euro- serotonin reuptake inhibitors (0.5–1%) [3, 4]. Acute liver pean Union Electronic Register of Post-Authorisation Stud- injury (ALI) refers to the sudden appearance of liver test ies (EU PAS Register # EUPAS10446) [18]. abnormalities and encompasses a spectrum of clinical dis- eases ranging from mild biochemical abnormalities to acute 2.2 Setting liver failure [5]. When known causes of ALI have been ruled out and criteria for causality of a specific drug are met, the This study was conducted in automated health databases term drug-induced liver injury is used instead [5, 6]. Stand- in four countries: Spain (SIDIAP [Information System ardised definitions of ALI for use in epidemiological studies for Research in Primary Care] [19] and EpiChron Cohort have been proposed [5, 7], and the incidence of ALI related [EpiChron Research Group on Chronic Diseases] [20]), to antidepressant use requiring hospitalisation has been esti- Germany (GePaRD [German Pharmacoepidemiological mated to be one to four cases per 100,000 patient-years [3]. Research Database]) [21–23] and Denmark [24–29] and Antidepressant drugs are currently among the most Sweden [29–31] (using their national registers). Character- widely used drugs in Western countries [8, 9], and often istics of the databases are described in Online Resource 1 patients using them have risk factors for ALI, such as alcohol (see the electronic supplementary material). Risk of Acute Liver Injury in New Users of Agomelatine and Other Antidepressants 385 2.3 Study Population according to the definition criteria established by an inter- national expert working group [5]. The definition criteria The study cohort included all individuals aged 18 years or are based on increases in the levels of alanine aminotrans- older at the date of the first-recorded prescription fill of any ferase (ALT), alkaline phosphatase (ALP), and bilirubin over of the study antidepressants during the study period(s) who the upper limit of normality (ULN) with less than 1 year (1) had not received a prescription fill for the same study of persistence: (≥ 5 × ULN ALT) or (≥ 2 × ULN ALP) or antidepressant within the prior 12 months (new users) and (≥ 3 × ULN ALT and > 2 × ULN bilirubin). This study was (2) had at least 12 months of continuous enrolment in the not designed to evaluate causality at the individual level to data source before the first prescription fill defining cohort ascertain whether a specific case with phenotypical ALI entry. Thus, one patient could contribute to several anti- had drug-induced liver injury or not. Rather, the study was depressant cohorts if eligibility criteria were fulfilled. For designed to examine the potential causal role of agomelatine women, an additional eligibility criterion was absence of and other antidepressants when compared with citalopram pregnancy at the start date of antidepressant use. Patients in the development of ALI at the population level. Thus, with a history of liver disease or risk factors for liver disease no formal evaluations using the Roussel Uclaf Causality (e.g. alcohol and drug abuse and dependence-related disor- Assessment Method (RUCAM) [32] were implemented at ders), chronic biliary or pancreatic disease, malignancy, or the individual case level. Potential cases were adjudicated by other life-threatening conditions (e.g. cancer, HIV infection) trained clinicians based on review of information abstracted were excluded from the study cohort (see Online Resource from medical records. This information included timing 2 in the electronic supplementary material for a detailed list and results of liver enzymes and information on presence of exclusion criteria). or absence of excluding conditions [33]. Thus, the clinical reviewers were adjudicating whether the individual potential 2.4 Endpoints Ascertainment: Selection of Cases cases met the study definition criteria of ALI and whether and Controls excluding conditions were present, not whether they met criteria for causality between antidepressant use and liver The primary endpoint was ascertained in all data sources injury. Validity of the electronic algorithms based on diag- and defined as any patient with a specific hospital discharge nosis codes used in the secondary and tertiary endpoints was diagnosis code of ALI from either the International Clas- assessed by calculating the positive predictive value (PPV), sification of Diseases, 9th Revision (ICD-9) or the Inter- defined as the probability that a patient classified as a poten- national Statistical Classification of Diseases and Related tial case was a confirmed case of ALI based on the reviewed Health Problems, 10th Revision (ICD-10). data (excluding nonevaluable cases from the denominator). The secondary endpoint was defined as validated cases In Germany, a companion external validation study of ALI (see below) identified with specific and nonspecific1 (ALIVAL) of the ICD discharge and outpatient diagnosis hospital discharge diagnosis codes and was evaluated only codes for ALI was conducted in a German hospital to esti- in Spain (EpiChron and SIDIAP) and Denmark. mate the PPV of algorithms used in the GePaRD data source The exploratory tertiary endpoint was assessed based on to identify potential cases of the primary and tertiary end- specific and nonspecific codes identified in both hospital points. In Sweden, no validation of cases was implemented. and outpatient settings and was evaluated in all data sources In the case–control analysis, all ALI cases identified regardless of whether validation was feasible. A sensitivity according to the definition of each endpoint in the study analysis restricted to validated cases was conducted in the cohort were included as cases. Controls were selected from three data sources in which validation was implemented. The the study cohort using density sampling. Up to 20 controls list of specific and nonspecific codes is included in the elec- per case were randomly selected from the risk set of each tronic supplementary material (Online Resource 3). The list case. Controls were matched to cases on age and sex, index of codes was adapted to the country-specific ICD classifica- date, and calendar year of study (cohort) entry. tion system used. The investigators in each country reviewed the codes list and made small adaptations when necessary. 2.5 Exposure Variables and Confounding Factors Additionally, local clinicians were consulted. Potential cases of ALI identified with specific and non- Time at risk was defined according to the days of supply of specific codes were confirmed by validation processes each prescription fill plus a period of 40 days. Days of sup- ply were the assumed number of days of treatment associ- ated with each prescription fill. The period of 40 days was 1 added to account for stockpiling and less than perfect adher- The terms “specific” and “nonspecific” are used here to indicate groups of codes that have in previous studies showed more (specific) ence [34]. or less (nonspecific) positive predictive values for ALI. 386 M. Pladevall‑Vila et al. In the nested case–control main analysis, use status was were tested in a backward elimination process [36] to avoid classified for each patient and each antidepressant into four problems with zero cells in the analysis. In the nested mutually exclusive categories according to time at risk of case–control analysis, presence of risk factors and con- the most recent prescription fill received on or before the founders was evaluated at any time before the index date, index date: but some variables (e.g. concurrent use of hepatotoxic drugs) were evaluated just 6 months before the index date. In • Current use: when the time at risk of the most recent Online Resource 5, we include a table providing additional prescription fill overlapped the index date details on the timing of evaluation for the different potential • Recent use: when the time at risk ended within 60 days confounders. before the index date For each of the three study endpoints, meta-analysis was • Past use: when the time at risk ended more than 60 days used to combine the adjusted OR estimates obtained from before the index date the nested case–control analysis in the different data sources. • Nonuse: when there was no prescription fill of the study Combined ORs and 95% CIs for ALI were produced first drug under consideration before the index date using random-effects models [37]. To assess heterogeneity in the meta-analysis, the I2 statistic was employed. When Confounding factors were those related to the risk of ALI results were homogeneous across databases, fixed-effects [13, 14, 35] and to exposure to agomelatine or another study models were used, and these results were presented in the antidepressant. Age, sex, and year of entrance in the cohort results tables. When the I2 test statistic was 30% or higher (through matching); acute alcohol intoxication; obesity; for any of the antidepressants, random-effects models were other components of metabolic syndrome (hypertension and presented for all antidepressants. Forest plots were created dyslipidaemia); diabetes; inflammatory bowel disease; pre- for each analysis table to display the adjusted combined and existing comorbidity measured by the Charlson Comorbid- data source–specific results. When no events were observed ity Index; acute biliary and pancreatic disease; peptic ulcer in some of the data sources, we included only the results disease; rheumatic diseases; concurrent use of hepatotoxic from data sources with observed events. drugs (list of drugs available in Online Resource 4; see the To check the robustness of the results, several planned electronic supplementary material); concurrent use of other sensitivity analyses were performed: the effect of adding antidepressants (different from study antidepressants); num- 15 days or 60 days (instead of 40 days) to the days of supply ber of liver tests performed; and health care resource utili- of the most recent prescription before the index date was zation measures were considered potential risk factors or explored; the effect of recent and past use of each study confounders. antidepressant was compared with current use of citalopram; switching and multiple current use were compared with cur- 2.6 Statistical Analyses rent single use of citalopram; and analyses restricted to cases without known causes of ALI were conducted. The imple- Crude and adjusted odds ratios (ORs) and 95% confidence mentation of those sensitivity analyses was limited in most intervals (CIs) for ALI for current use of each study antide- data sources by the limited number of cases. In addition, pressant were estimated using conditional logistic regres- two post hoc sensitivity analyses recommended by regula- sion models, with current use of citalopram as the reference tory reviewers were implemented for the primary endpoint. category (main analysis). In one, no exclusion criteria were implemented, and in the Crude ORs (and 95% CIs) were estimated including only other, all study exclusion criteria were applied except those the exposure variables in the conditional logistic regression related to alcohol use disorder and drug abuse. model. Due to the matching of cases and controls and con- ditional analyses, these crude ORs were adjusted by age, sex, and calendar year at study entry, but were referred to as 3 Results crude ORs regarding the rest of the confounders. Because of the low number of cases in some data sources, the original A total of 3,238,495 new users of antidepressants (Epi- exposure variables had to be reclassified into fewer catego- Chron, n = 185,628; SIDIAP, n = 203,101; the GePaRD, ries (i.e. current use vs. other), and in each of the separate n = 817,072; the Danish National Health Registers, logistical models comparing citalopram with each one of the n = 664,205; and the Swedish National Registers in Swe- other antidepressants, a new single exposure variable was den, n = 1,368,489) were included in the main analysis, of created to classify current use status of the two antidepres- which 74,440 were new users of agomelatine (Table 1). sants being compared. Agomelatine, the most recently launched of the studied Important risk factors and potential confounders were a antidepressants, was the least dispensed antidepressant in priori included in the models. Other potential confounders the study. A table presenting the distribution of the main Table 1  Summary of the cohort attrition for users of study antidepressants EpiChron SIDIAP GePaRD Denmark Sweden All a a a a a All ­users New users All ­users New users All ­users New users All ­users New users All ­users New users All ­usersa New users ­includedb ­includedb ­includedb ­includedb ­includedb ­includedb Citalopram 19,299 9016 (46.7%) 69,533 41,295 (59.4%) 628,785 229,895 (36.6%) 356,097 199,887 (56.1%) 1,047,829 302,719 (28.9%) 2,121,543 782,812 (36.9%) Agomelatine 10,700 8826 (82.5%) 4686 3243 (69.2%) 70,179 30,155 (43.0%) 22,131 18,032 (81.5%) 25,003 14,184 (56.7%) 132,699 74,440 (56.1%) Fluoxetine 27,406 15,379 (56.1%) 36,306 19,235 (53.0%) 110,549 38,162 (34.5%) 31,601 15,475 (49.0%) 203,924 62,318 (30.6%) 409,786 150,569 (36.7%) Paroxetine 43,800 23,570 (53.8%) 58,326 31,392 (53.8%) 82,165 27,599 (33.6%) 32,311 15,581 (48.2%) 111,950 27,453 (24.5%) 328,552 125,595 (38.2%) Sertraline 22,177 11,054 (49.8%) 37,382 21,148 (56.6%) 127,195 43,461 (34.2%) 167,973 113,861 (67.8%) 835,412 290,107 (34.7%) 1,190,139 479,631 (40.3%) Escitalopram 74,591 46,129 (61.8%) 42,911 23,107 (53.8%) 107,081 39,287 (36.7%) 99,190 44,817 (45.2%) 346,375 107,027 (30.9%) 670,148 260,367 (38.9%) Mirtazapine 34,876 19,402 (55.6%) 25,874 14,161 (54.7%) 473,284 160,771 (34.0%) 205,378 124,647 (60.7%) 674,519 237,907 (35.3%) 1,413,931 556,888 (39.4%) Risk of Acute Liver Injury in New Users of Agomelatine and Other Antidepressants Venlafaxine 22,748 10,577 (46.5%) 24,960 11,840 (47.4%) 209,438 74,850 (35.7%) 116,798 71,649 (61.3%) 399,754 102,116 (25.5%) 773,698 271,032 (35.0%) Duloxetine 32,314 20,899 (64.7%) 19,281 10,368 (53.8%) 105,947 32,357 (30.5%) 49,504 30,212 (61.0%) 223,435 64,850 (29.0%) 430,481 158,686 (36.9%) Amitriptyline 31,797 20,776 (65.3%) 42,562 27,312 (64.2%) 412,180 140,535 (34.1%) 46,261 30,044 (64.9%) 354,381 159,808 (45.1%) 887,181 378,475 (42.7%) Total 319,708 185,628 (58.1%) 361,821 203,101 (56.1%) 2,326,803 817,072 (35.1%) 1,127,244 664,205 (58.9%) 4,222,582 1,368,489 (32.4%) 8,358,158 3,238,495 (38.7%) GePaRD German Pharmacoepidemiological Research Database, SIDIAP Information System for Research in Primary Care a Total number of users before applying eligibility and exclusion criteria b Total number of users after applying eligibility and exclusion criteria. Percentages in this column are row percentages of the total number of all users 387 388 M. Pladevall‑Vila et al. Table 2  Number of new users in each antidepressant cohort and number of cases and controls by study endpoint and antidepressant cohort New users, Primary endpoint (specific hospital Secondary endpoint (specific Tertiary endpoint (specific and N (%)a discharge codes) and non-specific hospital non-specific hospital discharge and [n = 3,238,495] discharge codes—only validated outpatient codes) cases included) Number of ­casesb Number of Number of ­casesb Number of Number of ­casesb Number of (%) ­controlsc (%) (%) c ­controls (%) (%) ­controlsc (%) [n = 472] [n = 9438] [n = 178] [n = 3540] [n = 17,118] [n = 342,070] Citalopram 782,812 (24.2%) 111 (23.5%) 1790 (19.0%) 45 (25.3%) 806 (22.8%) 2667 (15.6%) 53,296 (15.6%) Agomelatine 74,440 (2.3%) 2 (0.4%) 112 (1.2%) n<5 59 (1.7%) 284 (1.7%) 4771 (1.4%) Fluoxetine 150,569 (4.6%) 9 (1.9%) 217 (2.3%) n<5 51 (1.4%) 547 (3.2%) 10,974 (3.2%) Paroxetine 125,595 (3.9%) 7 (1.5%) 201 (2.1%) 1 (0.6%) 72 (2.0%) 590 (3.4%) 10,853 (3.2%) Sertraline 479,631 (14.8%) 41 (8.7%) 805 (8.5%) 13 (7.3%) 314 (8.9%) 775 (4.5%) 15,412 (4.5%) Escitalopram 260,367 (8.0%) 25 (5.3%) 490 (5.2%) 16 (9.0%) 255 (7.2%) 603 (3.5%) 12,902 (3.8%) Mirtazapine 556,888 (17.2%) 41 (8.7%) 967 (10.2%) 18 (10.1%) 458 (12.9%) 1348 (7.9%) 23,433 (6.9%) Venlafaxine 271,032 (8.4%) 35 (7.4%) 637 (6.7%) 7 (3.9%) 254 (7.2%) 1178 (6.9%) 19,900 (5.8%) Duloxetine 158,686 (4.9%) 11 (2.3%) 322 (3.4%) 2 (1.1%) 122 (3.4%) 502 (2.9%) 9297 (2.7%) Amitriptyline 378,475 (11.7%) 22 (4.7%) 467 (4.9%) 3 (1.7%) 125 (3.5%) 1084 (6.3%) 17,644 (5.2%) Percentages in each cell were obtained from the total number of ALI cases or controls included in the column header Due to data protection policies, the exact number of cases could not be provided when the number of cases was less than 5 ALI acute liver injury a Percentages in this column are row percentages of the total number of all users b This number represents the total number of ALI cases. Cases in the body of the table are only those identified among current users c This number represents the total number of controls. Controls in the body of the table are only those with current use of the antidepressant confounder factors at cohort entry in each data source is study data sources, and the PPV was 62.7% in the external included in Online Resource 6 (see the electronic supple- validation study in Germany (ALIVAL). mentary material). Results of the case–control analyses for the current use In the main analysis, a total of 472 cases were identified of agomelatine compared with current use of citalopram are for the ALI primary endpoint (specific hospital discharge presented in Table 3. codes), ranging from 19 (SIDIAP) to 170 (Danish National Current use of agomelatine compared with current use Health Registers). Online Resource 7 presents the number of citalopram yielded ORs below 1.00 in all data sources of cases and controls in each data source and the association with cases (no cases under current use of agomelatine were between potential confounders and the ALI primary end- identified in SIDIAP and Sweden) with imprecise estimates. point in the multivariable adjusted models, which was dif- The combined (meta-analysis) adjusted OR for current use ferent in the five data sources. Table 2 displays the number of agomelatine compared with citalopram was 0.48 (95% CI of new users in each antidepressant cohort and the number 0.13–1.71). of cases and controls by endpoint, antidepressant cohort, In the post hoc sensitivity analysis without exclusions and overall numbers. (including 4,833,774 new users of antidepressants, of which 117,240 were new users of agomelatine), more cases were identified, and therefore, the OR estimates were more precise 3.1 Primary Endpoint and in line with the main analysis. The combined adjusted OR for current use of agomelatine compared with citalo- Online Resource 8 presents the age- and sex-standardised pram was 0.37 (95% CI 0.19–0.74). The point estimates incidence rates of ALI for the primary endpoint by data were homogeneous across all data sources for the sensitiv- source and studied antidepressant. The estimates were ity analysis without exclusions. imprecise, ranged from 0 to 27 cases per 100,000 person- In the post hoc sensitivity analysis including patients with years, and were overall within the range of the ALI incidence disorders related to alcohol and drug abuse (3,531,529 new rates described previously in the literature [38]. The PPVs users of antidepressants, of which 84,210 were new users of for the specific codes used to identify the primary endpoint agomelatine), the combined adjusted OR for current use of ranged from 60.0% (SIDIAP) to 84.2% (EpiChron) in the agomelatine compared with citalopram was 0.47 (95% CI Risk of Acute Liver Injury in New Users of Agomelatine and Other Antidepressants 389 Table 3  Results for the primary endpoint and current use (agomelatine vs. citalopram) in each data source and combined in the main analysis and the two post hoc sensitivity ­analysesa EpiChron SIDIAP GePaRD Danish National Swedish National Combined OR (95% CI) OR (95% CI) OR (95% CI) Health Registers Registers OR (95% CI) OR (95% CI) OR (95% CI) Primary endpoint (specific codes, hospitalised patients) Main ­analysisb 0.82 (0.06–10.70) – 0.55 (0.06–4.72) 0.30 (0.04–2.32) – 0.48 (0.13–1.71) Sensitivity post hoc analyses No exclusion criteria 0.66 (0.06–7.16) 0.61 (0.08–4.87) 0.36 (0.10–1.39) 0.32 (0.12–0.84) 0.26 (0.01–8.17) 0.37 (0.19–0.74) applied Exclusion criteria 1.60 (0.12–21.33) 1.43 (0.15–13.84) 0.51 (0.08–3.03) 0.30 (0.09–1.00) 0.16 (0.01–3.85) 0.47 (0.20–1.07) applied, except alco- hol and drug abuse CI confidence interval, GePaRD German Pharmacoepidemiological Research Database, OR odds ratio, SIDIAP Information System for Research in Primary Care a Adjusted for confounding factors; the list of confounders differed by data source. The following confounders were included in most analyses: obesity, hyperlipidaemia and hypertriglyceridaemia, diabetes, hypertension, indication of treatment with antidepressants for major depression, indication of treatment with antidepressants for anxiety disorders, indication of treatment with antidepressants for other mental and behavioural disorders, Charlson Comorbidity Index, number of liver tests performed, concurrent use of hepatotoxic drugs, and concurrent use of other anti- depressants b“ ” – indicates that the model did not converge 0.20–1.07). The individual point estimates from the differ- Denmark only; the adjusted OR estimate for current use ent data sources were less homogeneous: both EpiChron was 0.40 (95% CI 0.05–3.11). For the other antidepressants and SIDIAP, which had the smallest agomelatine cohorts, when compared with citalopram, most combined OR point showed OR estimates above 1.00 with wide 95% CIs. estimates were less than one, except that of fluoxetine. The Figure 1 presents the combined results for current use of estimates were imprecise (Online Resource 10). each antidepressant in the main analysis and the two post hoc sensitivity analyses for the primary endpoint. For all anti- 3.3 Tertiary Endpoint depressants except fluoxetine and paroxetine, the combined point estimates were less than 1.00 in the main analysis. The Overall, there were 17,118 cases of the tertiary study end- 95% CIs were more imprecise in the main analysis than in point and 342,070 controls. The GePaRD had overall the the two post hoc sensitivity analyses, in which all antidepres- largest number of cases (11,917), followed by SIDIAP sants had ORs below 1.00 when compared with citalopram. (2826), Sweden (1099), Denmark (1088), and EpiChron Results of the planned sensitivity analyses (including the (268). The PPV of the tertiary endpoint cases was low in all ones assessing different exposure definitions and analysis data sources, but especially in SIDIAP (7.7%). The high- excluding known causes of ALI) for agomelatine and the est PPVs (47.0%) were found in Denmark and the ALIVAL other antidepressants were, in general, consistent with the external study (Germany, 45.1%). In EpiChron, the PPV was main analysis and produced combined OR point estimates 25.4%. In Sweden, no validation of cases was implemented. for agomelatine below 1.00 for current use (data not shown). For this tertiary endpoint, the combined estimate for Results for all antidepressants, combined and in each indi- agomelatine for current use was 0.79 (95% CI 0.50–1.25). vidual data source, and for the three endpoints are available The results were heterogeneous (I 2 = 71%, indicating in the electronic supplementary material (Online Resources the presence of strong heterogeneity). The individual 9–11). data source–adjusted ORs for current use of agomelatine were also below 1.00, except in the GePaRD, and ranged 3.2 Secondary Endpoint from 0.36 (95% CI 0.11–1.25) in Sweden to 0.95 (95% CI 0.33–2.75) in EpiChron. In the GePaRD, the adjusted The secondary endpoint included only cases that had been OR was 1.24 (95% CI 1.07–1.42) and contrasted with confirmed after validation, which resulted in a lower num- the ORs observed in Denmark and Sweden, which were ber of events than for the primary endpoint. A total of 178 approximately 0.5 (Online Resource 12). In Denmark, the confirmed cases (150 in Denmark, 20 in EpiChron, eight adjusted OR estimate for current use of agomelatine was in SIDIAP) and 3540 controls were identified. Confirmed 0.44 (95% CI 0.22–0.87). In the sensitivity analysis that cases during current use of agomelatine were identified in included only confirmed cases (confirmed cases available 390 M. Pladevall‑Vila et al. Drugs for Data sources depression included Cases Controls Odds ratio (95% CI) I2 111 1,790 Citalopram 418 7,096 (Reference) 168 2,870 Ep, G, Dk 2 112 0.48 (0.13-1.71) 0% Agomelatine Ep, Sd, G, Dk, Sw 7 518 0.37 (0.19-0.74) 0% Ep, Sd, G, Dk, Sw 5 202 0.47 (0.20-1.07) 0% Ep, G, Dk, Sw 9 217 1.07 (0.52-2.21) 0% Fluoxetine Ep, Sd, G, Dk, Sw 41 933 0.63 (0.32-1.25) 36% Ep, Sd, G, Dk, Sw 23 434 0.89 (0.43-1.82) 34% Ep, Sd, G, Dk 7 201 1.29 (0.65-2.57) 0% Paroxetine Ep, Sd, G, Dk, Sw 39 912 0.70 (0.44-1.12) 0% Ep, Sd, G, Dk, Sw 12 370 0.88 (0.41-1.90) 31% Ep, Sd, G, Dk, Sw 41 805 0.88 (0.58-1.34) 0% Sertraline Ep, Sd, G, Dk, Sw 137 3,375 0.65 (0.49-0.85) 0% Ep, Sd, G, Dk, Sw 69 1,361 0.82 (0.59-1.15) 0% Ep, Sd, G, Dk, Sw 25 490 0.80 (0.45-1.40) 0% Escitalopram Ep, Sd, G, Dk, Sw 110 2,180 0.80 (0.58-1.09) 0% Ep, Sd, G, Dk, Sw 44 875 0.82 (0.53-1.26) 0% Ep, Sd, G, Dk, Sw 41 967 0.69 (0.45-1.05) 0% Mirtazapine Ep, Sd, G, Dk, Sw 233 4,158 0.85 (0.61-1.20) 32% Ep, Sd, G, Dk, Sw 87 1,643 0.83 (0.60-1.14) 0% Ep, Sd, G, Dk, Sw 35 637 0.89 (0.55-1.44) 7% Venlafaxine Ep, Sd, G, Dk, Sw 132 2,603 0.78 (0.59-1.03) 0% Ep, Sd, G, Dk, Sw 66 1,121 0.81 (0.49-1.36) 28% Ep, G, Dk, Sw 11 322 0.60 (0.31-1.16) 0% Duloxetine Ep, Sd, G, Dk, Sw 62 1,433 0.71 (0.39-1.28) 42% Ep, G, Dk, Sw 31 563 0.80 (0.48-1.33) 2% Ep, Sd, G, Dk, Sw 22 467 0.55 (0.22-1.41) 54% Amitriptyline Ep, Sd, G, Dk, Sw 95 1,847 0.83 (0.50-1.38) 46% Ep, Sd, G, Dk, Sw 30 673 0.71 (0.28-1.76) 68% 0.0 0.5 1.0 1.5 2.0 2.5 3.0 Main analyses Ep = EpiChron Sd = SIDIAP Favours other study Favours comparator No exclusion criteria drugs for depression (citalopram) applied G = GePaRD Exclusion criteria applied, Dk = Denmark except alcohol and drugs Sw = Sweden Risk of Acute Liver Injury in New Users of Agomelatine and Other Antidepressants 391 ◂Fig. 1  Current use combined adjusted estimates for all antidepres- reuptake inhibitors (adjusted hazard ratio 1.07; 95% CI sants (primary endpoint main and two sensitivity post hoc analyses 0.51–2.23) [39]. removing exclusion conditions). Odds ratio estimates were adjusted for confounding factors. The list of confounders differed by data In our study, analyses of most study antidepressants source and type of analysis (main vs. sensitive analyses). The follow- yielded ORs of ALI hospitalisation lower than 1.00 when ing confounders were included in most analyses: obesity, hyperlipi- compared with citalopram in the combined analyses for all daemia and hypertriglyceridaemia, diabetes, hypertension, indication endpoints. Since we had no pre-suspicion that citalopram of treatment with antidepressants for major depression, indication of treatment with antidepressants for anxiety disorders, indication of would carry a particularly high risk of ALI, the consistency treatment with antidepressants for other mental and behavioural dis- in the direction of this association across antidepressants orders, Charlson Comorbidity Index, number of liver tests performed, was unexpected. However, the limitations discussed here- concurrent use of hepatotoxic drugs, and concurrent use of other anti- after, particularly those related to the low number of cases, depressants. CI confidence interval, GePaRD German Pharmacoepi- demiological Research Database, SIDIAP Information System for preclude drawing definite conclusions. Citalopram is, in Research in Primary Care many countries, one of the first-line treatment options and is considered one of the safest antidepressants and could be selectively prescribed to those cases at the highest risk only in Denmark; see Online Resources 13) of the tertiary of ALI, which could potentially result in confounding by endpoint, the OR was 0.75 (95% CI 0.17–3.22). indication, but other antidepressants in this study share a For the other antidepressants (Online Resource 11), similar drug prescription pattern. Although citalopram has all combined OR estimates were between 0.94 and 1.11. been considered safe when compared with other antidepres- There was an indication of heterogeneity for mirtazapine, sants in analyses of spontaneous reports of adverse drug duloxetine, and amitriptyline (I2 > 50%, indicating at least reactions [40, 41] and in reviews of published clinical data moderate heterogeneity). The combined OR 95% CIs of [3], two epidemiological studies found an increased risk of the tertiary endpoint were more precise than for the other drug-induced liver injury associated with citalopram use two endpoints. In the sensitivity analysis that included only [42, 43]. Of note, the French study [39] did not find any confirmed cases (Online Resource 13), all combined OR relevant differences in risk between selective serotonin reup- estimates were below one except for duloxetine (OR 1.03; take inhibitors, serotonin and noradrenaline reuptake inhibi- 95% CI 0.60–1.79). tors, and other antidepressants (e.g. agomelatine). As in any study in automated data, misclassification of exposure status, occurrence of incident events, and the 4 Discussion covariates to be included in the multivariable models is possible. In this study, this would likely have resulted in Use of agomelatine was not associated with higher risk of nondifferential misclassification of the endpoints, potentially ALI hospitalisation compared with use of citalopram in a biasing the estimates towards unity. However, small differ- large cohort comprising 3.2 million new users of antide- ences in misclassification of the exposure or the disease may pressants in five populations from four countries: Spain, result in bias towards or away from unity in an unpredictable Denmark, Sweden, and Germany. Although precision of the manner [44]. combined risk estimates was low for the primary endpoint, To minimise misclassification of endpoints, specific the results were similar and more precise in the unrestricted codes were used for the primary endpoint, and validation sensitivity analyses and consistent both with the other sen- of the secondary endpoint that included nonspecific codes sitivity analyses and with the results of the other two end- was implemented. Validation of potential cases was imple- points considered. In the combined analysis, no increase in mented in a sensitivity analysis of the tertiary endpoint. risk was observed in populations including alcoholic patients The sources and the quantity of information available dif- or patients with other various risk factors. For the other anti- fered across data sources and the underlying health care depressants, most presented ORs below 1, except for fluox- systems (e.g. body mass index was available in data sources etine and paroxetine. Further, as for agomelatine, similar with access to primary care data), which may explain some results were obtained when using other outcome definitions, of the differences observed in the prevalence of some clini- that is, the secondary and tertiary endpoints. cal features, as well as the differences observed across the The estimates of risk associated with agomelatine use study data sources in the PPVs and some OR estimates. in this study are consistent with those from a recent cohort For the primary endpoint, the PPVs of the specific codes study funded by the French National Agency for Medi- ranged from 60% in SIDIAP to 84% in EpiChron and were cines and Health Products Safety and conducted using the consistent or higher than the estimates from previous stud- French health insurance database. This study did not find ies [45–52]. In SIDIAP, access to hospital medical records an increased risk of severe liver injury associated with the was not available, and PPVs would have probably been use of agomelatine compared with use of selective serotonin higher otherwise. Some PPV estimates were imprecise, 392 M. Pladevall‑Vila et al. especially in SIDIAP and in the ALIVAL study, and dif- increase was found for the tertiary endpoint, which was in ferences between data sources could be also explained by principle more sensitive to surveillance bias and misclassi- random error. Importantly, however, the results of the sec- fication than the primary and secondary endpoints. Surveil- ondary endpoint that included only confirmed cases were lance bias would have resulted in ORs above 1.00. However, consistent with the results of the primary endpoint. in the context of low PPVs, nondifferential misclassification As already mentioned, the number of identified events would produce bias towards seeing no association. of the primary and secondary endpoint was limited among The possibility of residual confounding cannot be dis- agomelatine users and users of other antidepressants. For carded. As mentioned, the amount of information on some the tertiary endpoint, the number of cases was much higher potential confounders (e.g. obesity) was limited in some of (especially in the GePaRD and SIDIAP), yet the low PPVs the data sources. Moreover, the limited number of identified observed for this endpoint definition limit the interpretation of ALI cases for the primary and secondary endpoints impacted these results in the analysis that included unconfirmed cases. the multivariable logistic regression strategy. To ensure a Patients taking agomelatine undergo routine liver enzyme sufficient case-to-covariate ratio, the number of covariates monitoring as a risk minimisation measure. Therefore, and number of categories for categorical covariates included patients at known risk of developing ALI may not have been in the models had to be minimised. This resulted in more prescribed agents (such as agomelatine) that are thought to be statistically stable models, but it may have increased the risk associated with an increased risk of ALI. Moreover, detection of residual confounding. Nevertheless, the restrictive cohort of liver enzyme elevations may be more likely in this group inclusion criteria implemented likely excluded most of the and prevent patients from starting treatment with agomelatine, key potential confounders associated with ALI. Moreover, or if agomelatine treatment has been started, treatment may be the post hoc analysis that did not impose any exclusion stopped earlier or cases of liver injury may be detected more criteria resulted in a much larger number of new users of frequently and earlier than if liver enzyme monitoring had not agomelatine (117,240) and of other antidepressants (4.8 mil- been conducted. In the context of observational studies using lion overall) and yielded more precise OR estimates for the data from routine clinical practice, these scenarios could lead primary endpoint that were consistent with those obtained in to selective prescribing, surveillance bias, or both. Selective the main analysis. Also, control of confounding via multivar- prescribing, i.e. the less frequent prescription of agomelatine iable models in those analyses did not have the limitations to patients with known risk factors for ALI or with evidence of encountered in the main analysis of the primary endpoint. existing liver damage measured by liver function tests, would The study had important strengths, first of which is that it shift the results in favour of agomelatine and lead to an appar- evaluated more than 3 million (almost 5 million in one of the ently lower risk of ALI among agomelatine new users when sensitivity analyses) new users of antidepressants and is the compared with new users of citalopram. The comparison of first study of this size to include also a validation component the different antidepressant cohorts in the main analysis and in of the cases identified. Second, inclusion of multiple inde- the sensitivity analysis including patients with known risk fac- pendent data sources and populations from different countries tors (data not shown) do not seem to indicate that agomelatine allowed evaluation of the consistency of the findings across is less prescribed among patients with known risk factors (e.g. five different, heterogeneous, automated, health care data chronic liver conditions). Moreover, the results of the sensitiv- sources. Finally, including three different endpoints with vari- ity analysis are consistent with the results of the main analysis, ous degrees of PPV and yielding a varying number of cases the latter excluding patients at risk of ALI, and show an OR created different perspectives for interpreting the study results. estimate also below one and with larger precision. However, because of the risk minimisation measures in place, selective prescribing could still take place (e.g. agomelatine would not 5 Conclusion be prescribed to those patients with abnormal liver tests or those with contraindications and precautions of use accord- The results of this study do not suggest that risk of hospital- ing to approved label), and this cannot be assessed with the ised ALI with use of agomelatine (compared with use of cit- data available. Thus, it is plausible that selective prescribing alopram) constitutes a public health problem, at least among may have influenced the results of the study, which need to be patient populations in health care systems with prescription interpreted in the context of the risk minimisation measures in patterns and risk minimisation measures similar to those in place for agomelatine. In fact, our results suggest that the risk this study. Thus, it is important to keep in place and comply minimization measures in place for agomelatine are effective. with the existing risk minimisation measures for agomela- Similarly, surveillance bias is unlikely to have had a large tine, which our results suggest are effective in preventing impact on the reported estimates for the primary and sec- ALI among users of agomelatine. When compared with ondary endpoints, as these included only hospitalised and citalopram, most antidepressants had OR point estimates thereby more severe cases. In the combined results, no risk less than 1.00 for hospitalised ALI. However, uncontrolled Risk of Acute Liver Injury in New Users of Agomelatine and Other Antidepressants 393 sources of bias could explain the results, and specific studies funded by pharmaceutical companies in their institution that were not to investigate this potential association of citalopram with related to this study and without personal profit. Tania Schink and Tammo Reinders, as employees of the Leibniz Institute for Prevention ALI are needed. Research and Epidemiology – BIPS, worked on projects funded by pharmaceutical companies unrelated to this study and without personal Acknowledgements The authors would like to acknowledge the fol- profit. Anton Pottegård reports participation in research projects fund- lowing people: Estel Plana, MSc (Epidemiology, RTI Health Solu- ed by Alcon, Almirall, Astellas, Astra-Zeneca, Novo Nordisk, LEO tions, Barcelona, Spain), for her support and expertise during the Pharma, and Servier, all with funds paid to the institution where he was analysis; Carla Franzoni, BSc (Epidemiology, RTI Health Solu- employed (no personal fees) and with no relation to the work reported tions, Barcelona, Spain), for her administrative support; Dr. Niklas in this paper. Jesper Hallas has participated in research projects funded Schmedt (InGef - Institute for Applied Health Research, Berlin, Ger- by Novartis, Pfizer, Menarini, MSD, Nycomed, LEO Pharma, Almi- many), for his involvement in planning the analysis at the beginning rall, Servier, Astellas, and Alkabello, with grants paid to the institution of the study; Marieke Niemeyer, MSc (Leibniz Institute for Prevention where he was employed. He has personally received fees for teaching Research and Epidemiology – BIPS, Bremen, Germany) and Sandra or consulting from the Danish Association of Pharmaceutical Manu- Ulrich, MSc (Leibniz Institute for Prevention Research and Epidemiol- facturers and from Pfizer and Menarini. Maja Hellfritzsch has received ogy – BIPS, Bremen, Germany), for their contribution to the program- speaker honorarium fees from Bristol-Myers Squibb and Pfizer and a ming of the analysis datasets; Morten Olesen (Clinical Pharmacology travel grant from LEO Pharma. Johan Reutfors, David Hägg, and Lena and Pharmacy, Department of Public Health, University of Southern Brandt are employees of the Centre for Pharmacoepidemiology, which Denmark, Odense, Denmark) and Martin Thomsen Ernst (Clinical receives grants from several entities (pharmaceutical companies, regu- Pharmacology and Pharmacy, Department of Public Health, Univer- latory authorities, and contract research organisations) for the perfor- sity of Southern Denmark, Odense, Denmark; OPEN, Odense Patient mance of drug safety and drug utilisation studies. Gabriel Perlemuter data Explorative Network, Odense University Hospital, Odense Den- reports participation in research projects funded by Biocodex, Servier, mark), for their help with data management; Javier Marta, MD [Epi- Physiogenex, Gilead, and Pileje as an external expert consultant. He Chron Research Group, IACS, IIS Aragon; REDISSEC ISCIII, Miguel received royalties from Elsevier-Masson, Solar, and John-Libbey and Servet University Hospital, Zaragoza, Spain] and Blanca Obón, MD reports travel and participation in meetings funded by Gilead, Abbvie, [EpiChron Research Group, IACS, IIS Aragon; REDISSEC ISCIII, and Servier. Bruno Falissard has been consultant for E. Lilly, BMS, Lozano Blesa University Clinic Hospital, Zaragoza, Spain], for their Servier, Sanofi, GSK, HRA, Roche, Boeringer Ingelheim, Bayer, assistance with the clinical validation activities. The authors would Almirall, Allergan, Stallergene, Genzyme, Pierre Fabre, AstraZeneca, also like to thank the German statutory health insurance providers that Novartis, Janssen, Astellas, Biotronik, Daiichi-Sankyo, Gilead, MSD, provided data for the study in GePaRD, namely the AOK Bremen/ Lundbeck, Stallergene, Actelion, UCB, Otsuka, Grunenthal, and ViiV. Bremerhaven, the DAK-Gesundheit, and Die Techniker (TK). Prof. Antje Timmer participates in a project funded by a consortium of pharmaceutical companies not related to this study and without per- Author Contributions Authors MP, JCa, EJ, ND, and SP planned the sonal benefit. Nicolas Deltour and Emmanuelle Jacquot are employees study. MP, JCa, and SP conducted the feasibility evaluation and drafted of Servier. The statistical analysis was performed by the different re- the study protocol. All authors made contributions to the final design search partners in each data source. In EpiChron, the main contributor and approved the final version of the protocol. Authors AP, TS, JR, BP, to the analysis was BP; in SIDIAP, it was JC; in Denmark, it was AP; JF, MH, TR, DH, MG, AP, JH, LB, and JC undertook the statistical in GePaRD, the main contributors were TS and TR; and in Sweden, analysis of the different data sources. AT completed the analysis of the the main contributors were DH and LB. The study protocol is publicly external validation study in Germany. JA, MP, JF, and SP completed available at the European Union Electronic Register of Post-Authorisa- the analysis and interpretation of the combined (meta-analysis) results tion Studies [EU PAS Register # EUPAS10446] [18]. with contributions from all authors. MP, JF, MC, and SP wrote the first draft of the manuscript. All authors contributed to and have approved Ethical Approval All procedures performed in studies involving human the final manuscript. participants were in accordance with the ethical standards of the insti- tutional and/or national research committee and with the 1964 Helsinki declaration and its later amendments or comparable ethical standards. Compliance with Ethical Standards For this type of study, formal consent is not required. Funding This study was funded by Servier under a contract granting independent publication rights to the research team. Servier co-authors Open Access This article is distributed under the terms of the Crea- of this manuscript, Emmanuelle Jacquot and Nicolas Deltour, provided tive Commons Attribution-NonCommercial 4.0 International License feedback and contributed to the design of the study. The sponsor had (http://creativecommons.org/licenses/by-nc/4.0/), which permits any the opportunity to review the report and contribute to the dissemination noncommercial use, distribution, and reproduction in any medium, of the results. The open access fee was paid by RTI-HS with general provided you give appropriate credit to the original author(s) and the funds from the study account, which was funded by Servier. source, provide a link to the Creative Commons license, and indicate if changes were made. Conflict of Interest Manel Pladevall, Joan Forns, Miguel Cainzos- Achirica, Jaume Aguado, Jordi Castellsagué, and Susana Perez-Gut- thann are employees of RTI Health Solutions, a unit of RTI Interna- tional, a nonprofit organisation that conducts work for government, References public, and private organisations, including pharmaceutical companies like Servier. Alexandra Prados-Torres and Beatriz Poblador-Plou are members of the EpiChron Research Group on Chronic Diseases of 1. EMA. EPAR summary for the public. Valdoxan (agomelatine). the Aragon Health Sciences Institute (IACS), ascribed to IIS Aragón, London: European Medicines Agency 2013. http://www.ema. and do not have any conflict of interest with this project. Maria Gin- europ​a.eu/docs/en_GB/docum​ent_libra​r y/EPAR_-_Summa​r y_ er-Soriano, Rosa Morros, and Jordi Cortés worked on other projects for_the_publi​c/human​/00091​5/WC500​04622​4.pdf. Accessed 11 Feb 2019. 394 M. Pladevall‑Vila et al. 2. Les Laboratoires Servier. Valdoxan summary of product char- 21. Jobski K, Schmedt N, Kollhorst B, Krappweis J, Schink T, Garbe acteristics. 2016. http://www.ema.europ​a.eu/docs/en_GB/docum​ E. Characteristics and drug use patterns of older antidepressant ent_libra ​ r y/EPAR_-_Produ ​ c t_Infor ​ m atio ​ n /human ​ / 00091 ​ 5 / initiators in Germany. Eur J Clin Pharmacol. 2017;73(1):105–13. WC500​04622​7.pdf. Accessed 11 Feb 2019. 22. Jobski K, Kollhorst B, Garbe E, Schink T. The risk of ischemic 3. Voican CS, Corruble E, Naveau S, Perlemuter G. Antidepres- cardio- and cerebrovascular events associated with oxycodone- sant-induced liver injury: a review for clinicians. Am J Psychia- naloxone and other extended-release high-potency opioids: a try. 2014;171(4):404–15. nested case-control study. Drug Saf. 2017;40(6):505–15. 4. Park SH, Ishino R. Liver injury associated with antidepressants. 23. Pigeot I, Ahrens W. Establishment of a pharmacoepidemiologi- Curr Drug Saf. 2013;8(3):207–23. cal database in Germany: methodological potential, scientific 5. Aithal GP, Watkins PB, Andrade RJ, Larrey D, Molokhia value and practical limitations. Pharmacoepidemiol Drug Saf. M, Takikawa H, et al. Case definition and phenotype stand- 2008;17(3):215–23. ardization in drug-induced liver injury. Clin Pharmacol Ther. 24. Schmidt M, Pedersen L, Sorensen HT. The Danish Civil Reg- 2011;89(6):806–15. istration System as a tool in epidemiology. Eur J Epidemiol. 6. Hussaini SH, Farrington EA. Idiosyncratic drug-induced liver 2014;29(8):541–9. injury: an update on the 2007 overview. Expert Opin Drug Saf. 25. Lynge E, Sandegaard JL, Rebolj M. The Danish National Patient 2014;13(1):67–81. Register. Scand J Public Health. 2011;39(7 Suppl):30–3. 7. Katz AJ, Ryan PB, Racoosin JA, Stang PE. Assessment of case 26. Schmidt M, Schmidt SA, Sandegaard JL, Ehrenstein V, Pedersen definitions for identifying acute liver injury in large observa- L, Sorensen HT. The Danish National Patient Registry: a review tional databases. Drug Saf. 2013;36(8):651–61. of content, data quality, and research potential. Clin Epidemiol. 8. Lewer D, O’Reilly C, Mojtabai R, Evans-Lacko S. Antidepres- 2015;7:449–90. sant use in 27 European countries: associations with sociode- 27. Pottegard A, Schmidt SA, Wallach-Kildemoes H, Sorensen HT, mographic, cultural and economic factors. Br J Psychiatry. Hallas J, Schmidt M. Data resource profile: the Danish National 2015;207(3):221–6. Prescription Registry. Int J Epidemiol. 2017;46(3):798-f. 9. Pratt LA, Brody DJ, Gu Q. Antidepressant use among persons 28. Johannesdottir SA, Horváth-Puhó E, Ehrenstein V, Schmidt M, aged 12 and over: United States, 2011–2014. NCHS Data Brief. Pedersen L, Sørensen HT. Existing data sources for clinical epide- 2017;283:1–8. miology: the Danish National Database of Reimbursed Prescrip- 10. Breslow RA, Dong C, White A. Prevalence of alcohol-interac- tions. Clin Epidemiol. 2012;4:303–13. tive prescription medication use among current drinkers: United 29. Furu K, Wettermark B, Andersen M, Martikainen JE, Almarsdottir States, 1999 to 2010. Alcohol Clin Exp Res. 2015;39(2):371–9. AB, Sørensen HT. The Nordic countries as a cohort for phar- 11. Du Y, Wolf IK, Knopf H. Psychotropic drug use and alcohol macoepidemiological research. Basic Clin Pharmacol Toxicol. consumption among older adults in Germany: results of the 2010;106(2):86–94. German Health Interview and Examination Survey for Adults 30. Ludvigsson JF, Andersson E, Ekbom A, Feychting M, Kim JL, 2008–2011. BMJ Open. 2016;6(10):e012182. Reuterwall C, et al. External review and validation of the Swedish 12. Rethorst CD, Bernstein I, Trivedi MH. Inflammation, obe- national inpatient register. BMC Public Health. 2011;11:450. sity, and metabolic syndrome in depression: analysis of the 31. Wettermark B, Hammar N, Fored CM, Leimanis A, Otterblad 2009–2010 National Health and Nutrition Examination Survey Olausson P, Bergman U, et al. The new Swedish Prescribed Drug (NHANES). J Clin Psychiatry. 2014;75(12):e1428–32. Register—opportunities for pharmacoepidemiological research 13. Chen M, Suzuki A, Borlak J, Andrade RJ, Isabel Lucena M. and experience from the first six months. Pharmacoepidemiol Drug-induced liver injury: interactions between drug properties Drug Saf. 2007;16(7):726–35. and host factors. J Hepatol. 2015;63(2):503–14. 32. Danan G, Benichou C. Causality assessment of adverse reactions 14. Leise MD, Poterucha JJ, Talwalkar JA. Drug-induced liver to drugs—I. A novel method based on the conclusions of inter- injury. Mayo Clin Proc. 2014;89(1):95–106. national consensus meetings: application to drug-induced liver 15. Lin ND, Norman H, Regev A, Perahia DG, Li H, Chang CL, injuries. J Clin Epidemiol. 1993;46(11):1323–30. et al. Hepatic outcomes among adults taking duloxetine: a retro- 33. Forns J, Cainzos-Achirica M, Hellfritzsch M, Giner-Soriano M, spective cohort study in a US health care claims database. BMC Poblador-Plou B, Hallas J, et al. How valid are the codes used Gastroenterol. 2015;15:134. to identify acute liver injury (ALI)? A study in 3 European data 16. Xue F, Strombom I, Turnbull B, Zhu S, Seeger JD. Duloxetine sources. Pharmacoepidemiol Drug Saf. 2018;27(Suppl 2):371 for depression and the incidence of hepatic events in adults. J (abstract # 811). Clin Psychopharmacol. 2011;31(4):517–22. 34. Sikka R, Xia F, Aubert RE. Estimating medication persis- 17. Pladevall M. Post-authorisation safety study of agomela- tency using administrative claims data. Am J Manag Care. tine and the risk of hospitalisation for acute liver injury 2005;11(7):449–57. 16/01/2018 2018. http://www.encep​p .eu/encep​p /viewR​e sour​ 35. Kaye JA, Castellsague J, Bui CL, Calingaert B, McQuay LJ, ce.htm?id=12730​. Accessed 2 Mar 2018. Riera-Guardia N, et al. Risk of acute liver injury associated 18. Pladevall M, Rebordosa C, Castellsague J, Perez-Gutthann S, with the use of moxifloxacin and other oral antimicrobials: a Hellfritzsch M, Hallas J, et al. Post-authorisation safety study of retrospective, population-based cohort study. Pharmacotherapy. agomelatine and the risk of hospitalisation for acute liver injury 2014;34(4):336–49. 18 May 2017 (Version 2.2) 2017. http://www.encep​p.eu/encep​ 36. Weng HY, Hsueh YH, Messam LL, Hertz-Picciotto I. Methods p/openA​ttach​ment/fullP​rotoc​olLat​est/19796​. Accessed 2 Mar of covariate selection: directed acyclic graphs and the change-in- 2018. estimate procedure. Am J Epidemiol. 2009;169(10):1182–90. 19. SIDIAP. Database. General details. 2019. https:​ //www.sidiap​ .org/ 37. Higgins JPT, Green S, editors. Cochrane handbook for system- index​.php/datab​ase/gener​al-detai​ls. Accessed 11 Feb 2019. atic reviews of interventions, version 5.1.0, 2011. http://handb​ 20. Prados-Torres A, Poblador-Plou B, Gimeno-Miguel A, Calderon- ook.cochr​ane.org/. Accessed 11 Feb 2019. Larranaga A, Poncel-Falco A, Gimeno-Feliu LA, et al. Cohort 38. Ruigomez A, Brauer R, Rodriguez LA, Huerta C, Requena G, Gil profile: the epidemiology of chronic diseases and multimorbidity. M, et al. Ascertainment of acute liver injury in two European pri- The EpiChron Cohort Study. Int J Epidemiol. 2018;47(2):382–4. mary care databases. Eur J Clin Pharmacol. 2014;70(10):1227–35. Risk of Acute Liver Injury in New Users of Agomelatine and Other Antidepressants 395 39. Billioti de Gage S, Collin C, Le-Tri T, Pariente A, Begaud B, Ver- 46. Maggini M, Raschetti R, Agostinis L, Cattaruzzi C, Troncon MG, doux H, et al. Antidepressants and hepatotoxicity: a cohort study Simon G. Use of amoxicillin and amoxicillin-clavulanic acid among 5 million individuals registered in the French National and hospitalization for acute liver injury. Ann Ist Super Sanita. Health Insurance Database. CNS Drugs. 2018;32(7):673–84. 1999;35(3):429–33. 40. Montastruc F, Scotto S, Vaz IR, Guerra LN, Escudero A, Sainz 47. Traversa G, Bianchi C, Da Cas R, Abraha I, Menniti-Ippolito M, et al. Hepatotoxicity related to agomelatine and other new F, Venegoni M. Cohort study of hepatotoxicity associated with antidepressants: a case/noncase approach with information from nimesulide and other non-steroidal anti-inflammatory drugs. BMJ. the Portuguese, French, Spanish, and Italian pharmacovigilance 2003;327(7405):18–22. systems. J Clin Psychopharmacol. 2014;34(3):327–30. 48. Shin J, Hunt CM, Suzuki A, Papay JI, Beach KJ, Cheetham TC. 41. Gahr M, Zeiss R, Lang D, Connemann BJ, Hiemke C, Schonfeldt- Characterizing phenotypes and outcomes of drug-associated liver Lecuona C. Drug-induced liver injury associated with antidepres- injury using electronic medical record data. Pharmacoepidemiol sive psychopharmacotherapy: an explorative assessment based on Drug Saf. 2013;22(2):190–8. quantitative signal detection using different MedDRA terms. J 49. Bui CL, Kaye JA, Castellsague J, Calingaert B, McQuay LJ, Clin Pharmacol. 2016;56(6):769–78. Riera-Guardia N, et al. Validation of acute liver injury cases in a 42. Douros A, Bronder E, Andersohn F, Klimpel A, Thomae M, population-based cohort study of oral antimicrobial users. Curr Sarganas G, et al. Drug-induced liver injury: results from the Drug Saf. 2014;9(1):23–8. hospital-based Berlin Case-Control Surveillance Study. Br J Clin 50. Lo Re V 3rd, Carbonari DM, Forde KA, Goldberg D, Lewis JD, Pharmacol. 2015;79(6):988–99. Haynes K, et al. Validity of diagnostic codes and laboratory tests 43. Ferrajolo C, Scavone C, Donati M, Bortolami O, Stoppa G, of liver dysfunction to identify acute liver failure events. Pharma- Motola D, et al. Antidepressant-induced acute liver injury: a coepidemiol Drug Saf. 2015;24(7):676–83. case–control study in an Italian inpatient population. Drug Saf. 51. Udo R, Maitland-van der Zee AH, Egberts TC, den Breeijen JH, 2018;41(1):95–102. Leufkens HG, van Solinge WW, et al. Validity of diagnostic codes 44. Jurek AM, Greenland S, Maldonado G. How far from non-dif- and laboratory measurements to identify patients with idiopathic ferential does exposure or disease misclassification have to be to acute liver injury in a hospital database. Pharmacoepidemiol Drug bias measures of association away from the null? Int J Epidemiol. Saf. 2016;25(Suppl 1):21–8. 2008;37(2):382–5. 52. Jinjuvadia K, Kwan W, Fontana RJ. Searching for a needle in a 45. Lo Re V 3rd, Haynes K, Goldberg D, Forde KA, Carbonari DM, haystack: use of ICD-9-CM codes in drug-induced liver injury. Leidl KB, et al. Validity of diagnostic codes to identify cases of Am J Gastroenterol. 2007;102(11):2437–43. severe acute liver injury in the US Food and Drug Administra- tion’s Mini-Sentinel Distributed Database. Pharmacoepidemiol Drug Saf. 2013;22(8):861–72. Affiliations Manel Pladevall‑Vila1,2 · Anton Pottegård3 · Tania Schink4 · Johan Reutfors5 · Rosa Morros6,7,8,9 · Beatriz Poblador‑Plou10 · Antje Timmer11 · Joan Forns1 · Maja Hellfritzsch3 · Tammo Reinders4 · David Hägg5 · Maria Giner‑Soriano6,7,8 · Alexandra Prados‑Torres10 · Miguel Cainzos‑Achirica1 · Jesper Hallas3 · Lena Brandt5 · Jordi Cortés6,7,8,12 · Jaume Aguado1 · Gabriel Perlemuter13,14,15 · Bruno Falissard16 · Jordi Castellsagué1 · Emmanuelle Jacquot17 · Nicolas Deltour17 · Susana Perez‑Gutthann1 9 * Manel Pladevall‑Vila Plataforma SCReN, UICEC IDIAP Jordi Gol, Barcelona, mpladevall@rti.org Spain 10 1 EpiChron Research Group, Aragon Health Sciences Institute Epidemiology, RTI Health Solutions, Av. Diagonal 605, 9‑1, (IACS), IIS Aragon, REDISSEC ISCIII, Miguel Servet 08028 Barcelona, Spain University Hospital, Zaragoza, Spain 2 The Center for Health Policy and Health Services Research, 11 Division of Epidemiology and Biometry, Medical Faculty, Henry Ford Health System, Detroit, MI, USA Carl von Ossietzky University, Oldenburg, Germany 3 Clinical Pharmacology and Pharmacy, Department of Public 12 Universitat Politècnica de Catalunya, Departament Health, University of Southern Denmark, Odense, Denmark d’Estadística i Investigació Operativa, Barcelona, Spain 4 Leibniz Institute for Prevention Research 13 AP-HP, Hôpital Antoine Béclère, Service and Epidemiology-BIPS, Bremen, Germany d’Hépato-gastroentérologie, 92140 Clamart, France 5 Centre for Pharmacoepidemiology, Department of Medicine, 14 Univ Paris-Sud/Paris-Saclay, Faculté de Médecine, Solna, Karolinska Institutet, Karolinska University Hospital, Paris-Sud, 94270 Kremlin‑Bicêtre, France Stockholm, Sweden 15 6 INSERM U996, Clamart 92140, France Institut Universitari d’Investigació en Atenció Primària Jordi 16 Gol (IDIAPJGol), Barcelona, Spain CESP INSERM U1018, Université Paris-Saclay, Université 7 Paris-Sud, UVSQ, APHP, Paris, France Universitat Autònoma de Barcelona, 17 Bellaterra (Cerdanyola del Vallès), Spain Pharmacoepidemiology Department, Servier, Suresnes, 8 France Institut Català de la Salut, Barcelona, Spain