Ethics Statement

All data analyses performed in this retrospective study are based on anonymized data, which did not allow the precise identification of individual patients. The study dealt with two independent data sources, namely (a) spontaneous reports, which are publicly available from the FAERS database and (b) drug utilization data, which can be accessed by other researchers only upon request to relevant personnel. Therefore, submission to and approval by the Institutional Review Board was not required.

Pharmacovigilance data source

By virtue of its large population coverage (including all US reports and rare/serious events from European Countries) and free availability since 2004, FAERS is an attractive source to detect emerging safety signals and a valuable tool to explore the worldwide reporting pattern of rare Adverse Drug Reactions (ADRs) such as TdP [15]. As a matter of fact, approximately 800,000 reports are entered into FAERS each year, of which almost a half is submitted by non-US Countries (e.g., European areas) and consumers. In addition, interest is emerging in using FAERS to obtain reliable within-class comparisons [24].

Before performing the analysis, a number of issues have been considered to obtain a reliable dataset: drug and event codification through ad hoc and standardized (MedDRA) archives, respectively; duplicate detection; management of missing data [25]. As previously detailed [15,26], the 2004-2010 period was analyzed through a record-linkage strategy of different files: DEMO (demographic data), DRUG (the analysis was restricted to APs reported as suspect or interacting), REACTION and OUTCOME (to assess the seriousness of the event; i.e., whether or not the ADR resulted in death or life-threatening conditions).

Organizing pharmacovigilance data

In a recent study, we described the consensus process to define the torsadogenic potential of drugs and extract relevant cases of interest [27]. In summary, a wide range of clinical events with decreasing drug-attributable risk for TdP was collected: 1) TdP, 2) QT interval abnormalities (including QT prolongation), 3) Ventricular Arrhythmia (VA, including ventricular tachycardia and fibrillation) and 4) Sudden Cardiac Death (SCD). Building on this previous approach, in the present study, these clinical events were combined as follows: (a) TdP reports were analyzed with QT abnormalities as they are strongly intertwined with a high degree of drug-attributable risk (TdP is a so-called designated medical event, i.e., a rare and serious event with high drug-attributable component [28]); (b) VA reports were separately analyzed with SCD events as they are not strictly correlated with TdP and carry a reduced degree of drug-attributable risk [29].

Based on these events and considering the clinical scenario in which TdP usually occurs (i.e., the so called “reduced repolarization reserve” due to multiple risk factors such as concomitant drugs with TdP liability [30]), the following criteria/parameters (with relevant thresholds) were defined to analyze pharmacovigilance data:

1. 1. Number of cases of TdP/QT abnormalities (at least 4 cases);
2. 2. Significant Reporting Odds Ratio, ROR (i.e., Lower Limit of the 95%CI>1 and at least 4 cases) for TdP/QT abnormalities, adjusted for confounders (e.g., concomitant Class I/III antiarrhythmic drugs) in the stratum without drugs listed by Arizona CERT (effect modifiers) [27];
3. 3. Number of cases of VA/SCD (at least 4 cases);
4. 4. Significant ROR for VA/SCD, unadjusted and un-stratified;
5. 5. Significant unadjusted and un-stratified ROR that persists throughout a cumulative approach (i.e., from TdP to SCD).

In order to analyze drug utilization data in a public health perspective, the following groups were created to characterized signals in terms of strength, based on the consistency of the aforementioned criteria:

1. A. Antipsychotics with very strong torsadogenic signal (i.e., consistency among all criteria): all criteria exceeded relevant thresholds;
2. B. Antipsychotics with strong torsadogenic signal (i.e., the majority of criteria are fulfilled): at least 4 out of 5 criteria exceeded relevant thresholds;
3. C. Antipsychotics with weak/moderate torsadogenic signal (i.e., criteria not homogeneously fulfilled): at least 3 out of 5 criteria exceeded relevant thresholds;
4. D. Antipsychotics with unclear/uncertain torsadogenic signal (i.e., the majority of criteria are not fulfilled): only 2 (or less) out of 5 criteria exceeded relevant thresholds.

Drug utilization data collection

After safety signals are identified from FAERS, drug utilization data allow us to map the risk derived from exposure to antipsychotic drugs in each Country. Collection of drug utilization data on antipsychotics requires careful consideration of the available sources across Europe, including characteristics of databases (e.g., data quality, population coverage, data access), limitations and aims of the research [31]. In this study, drug utilization data were collected from administrative databases through health authorities and health insurance personnel across Europe. Consistent and reliable data were obtained from twelve European Countries, which allowed adequate estimation of European population exposure and represented differences in geography and financing of healthcare: they comprised Northern (Sweden, Norway and Scotland), Southern-Western (Spain [Catalonia], Austria, France, Italy) as well as Central-Eastern (Croatia, Serbia, Slovenia, Estonia, Lithuania) European Countries and regions. In addition, differences in the method of financing healthcare: taxation (Italy, Norway, Scotland, Spain [Catalonia], Sweden) and health insurance based (Austria, Croatia, Estonia, France, Lithuania, Serbia and Slovenia).

Total dispended data (ATC code: N05A, excluding Lithium) were expressed as defined daily doses (DDD) per 1,000 inhabitants per day (DDD/TID, now referred to as DID) for the 2005-2010 period (with some exceptions detailed in the relevant table). Despite limitations, the DDD system is a recognized tool for standardizing antipsychotic doses in drug utilization research [32], and is currently recommended by the WHO for international drug utilization studies [33]. In the analyses, we considered a detectable use of at least 0.01 DID. The mean DID value was used to estimate the actual population exposure over the period of interest, in line also with pharmacovigilance data. A time trend analysis was also carried out, where appropriate. Data were re-validated with data providers to enhance their accuracy. A structured inventory of drug consumption databases available across Europe (with relevant details on data access and characteristics) has been recently provided within the PROTECT (Pharmacoepidemiological Research on Outcomes of Therapeutics by a European Consortium) project (http://www.imi-protect.eu/frameworkRep.shtml). Moreover, Table S1 provided details of the databases used in our study. With only a few exceptions, administrative databases regards reimbursed prescriptions in ambulatory care and covered the entire population.

Pharmacovigilance analyses

Over the 7-year-period of the analysis, 37 antipsychotics received at least one report in FAERS for the events of interest; 35 of them fulfilled at least one criterium for signal characterization. Table 1 provides a synopsis of pharmacovigilance criteria and their combination for signal characterization. Group A (i.e., antipsychotics that fulfilled all criteria) included 9 agents, 6 of which are SGAs. Of these, only cyamemazine is not included in any Arizona CERT lists (as of January 4^th, 2013). Two agents were already included in List I (i.e., chlorpromazine and haloperidol), 5 SGAs in List II, amisulpride as the only compound in List III. Group B included 7 drugs (all FGAs, except droperidol), with only droperidol listed by Arizona CERT in List I. Group C comprised 13 agents, with paliperidone and pimozide recorded in Arizona CERT Lists. Eight antipsychotics were categorized in Group D.

Quetiapine (18,070), olanzapine (11,622) and clozapine (10,082) were the most frequently reported overall and also received the highest number of reports for the events of interest, namely group TdP/QT and VA/SCD (Table 2). Notably, for most antipsychotics, also European reports have been submitted to FAERS and, in most of the cases, European Countries for which drug utilization were available have largely contributed (e.g., France). Concerning Group A drugs, cyamemazine (79.8%) and amisulpride (73.5%) received the highest proportion of reports from Europe. By contrast, quetiapine (75.7%) and risperidone (71.3%) mainly received US reports. As regards Group B agents, with the exception of fluphenazine, reports were mostly submitted by European Countries. Paliperidone and aripiprazole (Group C) received the majority of reports from US, whereas Group D agents from Europe (especially pipamperone and melperone). France, Germany and UK were the most frequent European reporter Countries.

The full original data, including the results of the different disproportionality approaches (i.e., ROR values for each antipsychotic drug according to the event of interest), are provided as supplementary material (Table S2). These data have been used to characterize the strength of the pharmacovigilance signal.

Drug Utilisation analyses

In 2010, the total exposure to antipsychotics (including, for instance, dixyrazine, thioridazine and sertindole, not included in or detected by the pharmacovigilance analysis) varied by a factor of approx 2, between Countries with the lowest (5.94 DID in Estonia) and highest (12.89 DID in Spain) use. In 2009, France peaked 13.99 DID (the latest available year of data collection). In all Countries, antipsychotic use increased over the period of interest, albeit with different trends among the different Countries. The increment was greatest in Central and Eastern European countries (+3.09 DID in Lithuania, +3.03 in Croatia) and especially in Serbia (+5.61 DID, +183%). This potentially reflects more drugs being placed on the reimbursement list, whereas in Northern European countries the consumption was more stable (+0.35 in Norway, +0.43 in Sweden) (Figure 1).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 1. Trends in antipsychotic utilization according to data availability.

AUT: Austria; CRO: Croatia; EST: Estonia; FRA: France; ITA: Italy; LIT: Lithuania; NOR: Norway; SPA: Spain (Catalonia); SCO: Scotland; SER: Serbia; SLO: Slovenia; SWE: Sweden; EMA: European Medicines Agency; FDA: Food and Drug Administration; FGA: first-generation antipsychotics; SGA: second-generation antipsychotics; TdP: Torsades de Pointes. Boxes indicate regulatory safety warnings on cardiovascular risk.

https://doi.org/10.1371/journal.pone.0081208.g001

Table 3 provides an overview of drug utilization data on 37 antipsychotics arisen from the pharmacovigilance analyses and expressed in terms of mean DID over the study period. The mean exposure varied substantially: from France (12.92) and Slovenia (10.62) to Estonia and Lithuania (5.07 and 5.04, respectively). The mean use of SGAs was appreciably higher than the mean use of FGAs in all Countries, with the largest exposure to SGAs in Catalonia (Spain) (7.83).

Figure 2 provided the exposure to these 37 antipsychotics according to their characterization for torsadogenic signal. This time-trend analysis found that the overall utilization substantially increased in all Countries, especially in Serbia (+5.60) and France (+3.74). Only a limited increase in utilization was observed among Northern European Countries (+0.45 in Sweden, +0.51 in Norway). The consumption of SGAs increased in all Countries (+3.79 in France), although lower in Norway (+1.15) and Sweden (+0.67). By contrast, the use of FGAs showed a positive trend only in Serbia (+4.25). This positive trend in antipsychotic use was mainly driven by the increase in utilization of Group A agents: from +3.47 in France to +0.34 in Sweden, with substantial increment also in Croatia (+2.71), Austria (+2.32) and Slovenia (+2.24). In 2010 Spain (Catalonia) was the Country with the highest exposure to Group A agents (8.57); France ranked first in 2009 (10.25).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 2. Antipsychotic utilization according to signal characterization: 2005 (Panel A) versus 2010 (Panel B).

Countries were ranked by magnitude of 2010 antipsychotic consumption. Percentage: consumption of antipsychotics classified as group D/total antipsychotic consumption. In parenthesis changes in use of group D compounds (absolute values); + = increment in consumption; - = decrease in consumption.

https://doi.org/10.1371/journal.pone.0081208.g002

By contrast, the use of agents included in Group D was irregular and very low across Europe. In 2010, the magnitude of use was marked only in Croatia (1.62) and France (0.55). Generally, decreased utilization was recorded across all Countries, especially in Norway (-1.17) and Sweden (-0.76). An increase was found only in Lithuania (+0.06).

There was generally limited use of antipsychotics included in Group B, as compared to Group A. However, the 2010 exposure to levomepromazine was high in Serbia (3.78). Appreciable use was also seen for fluphenazine, especially in Slovenia (1.63 in 2010), Serbia (1.02) and France (0.92 in 2009). The consumption of zuclopenthixol was considerably higher among Northern European Countries as compared to the remaining areas, with a peak of 0.31 and 0.60 (Norway in Sweden, respectively). Prothypendyl could be a specific concern in Austria (0.62) and chlorprothixene in Norway (0.48).

The consumption of antipsychotics included in Group C was mainly driven by increasing utilization of aripiprazole, especially in Austria (0.75) and Italy (0.69) in 2010. Flupentixol (0.28 in Norway and France; 2010 and 2009, respectively) and perphenazine (0.59 and 0.44 in Norway and Sweden, in 2010) were appreciably used among Northern European Countries, melperone among in Central and Eastern European countries (peak of in 0.37 in Estonia in 2009). Sulpiride was used consistently across Europe, especially among Central and Eastern European Countries (0.41 in Croatia in 2010, where the consumption of promazine was also appreciable at 1.62).

Figure 3 provides a drug-by-drug consumption of antipsychotics included in Group A. The use of these agents grew in all 12 Countries. In Serbia, the increment was caused by risperidone (+1.09); in France by olanzapine and risperidone (+1.84 and +1.47, respectively), whereas in 9 of the remaining 11 Countries, the trend was mostly driven by quetiapine (from +0.50 in Sweden to +1.40 in Austria). Olanzapine was consistently used across all Countries except in Serbia (0.07 in 2010), reaching 2.96 DID in Norway (2010) and showing increasing consumption in all Countries (+1.84 in France). Cyamemazine was selectively used only in France with large extent (2.81 in 2009).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 3. Country-by-Country Comparison in antipsychotic utilization included in Group A: 2005 versus 2010.

Abscissa: DID. OLA: olanzapine; ZIP: ziprasidone; RIS: risperidone; QUE: quetiapine; HAL: haloperidol; CLO: clozapine; CHL: chlorpromazine; AMI: amisulpride; CYA: cyamemazine (shown only in France).

https://doi.org/10.1371/journal.pone.0081208.g003

Limitations of the study

The clinical implications of this study should be carefully evaluated in the light of limitations affecting both pharmacovigilance and drug consumption data.

The analysis of spontaneous reports does not allow establish causality and is affected by the under-reporting; thus, no incidence can be calculated. In addition, there are several external factors that may influence the reporting [54-56]. In this respect, a number of safety warnings on the cardiovascular risk have been posted by the FDA that may have stimulated reporting: in 2005, SGAs received an alert on the risk of death in elderly patients with behavioral disturbances (http://www.fda.gov/Drugs/DrugSafety/PostmarketDrugSafetyInformationforPatientsandProviders/DrugSafetyInformationforHeathcareProfessionals/PublicHealthAdvisories/ucm053171.htm), extended in June 2008 (FDA) (http://www.fda.gov/Drugs/DrugSafety/PostmarketDrugSafetyInformationforPatientsandProviders/ucm107211.htm) and November 2008 (EMA) (http://www.ema.europa.eu/docs/en_GB/document_library/Report/2010/01/WC500054057.pdf) to FGAs; in 2007, haloperidol received a specific FDA alert on the risk of TdP/QT prolongation with relevant label changes (http://www.fda.gov/Drugs/DrugSafety/PostmarketDrugSafetyInformationforPatientsandProviders/DrugSafetyInformationforHeathcareProfessionals/ucm085203.htm).

Although also serious and rare European reports are submitted to the FDA, our data may be affected by specific drug marketing penetration (e.g., drugs only marketed in Europe). In addition, the characterization of signals is dynamic so that a given antipsychotic may change groups depending on the availability of new pharmacovigilance data over time.

This study attempted to overcome the lack of information on exposed population in spontaneous reporting sources, by linking FAERS findings with drug consumption. These two data sources are independent and provide different perspectives, although they could be theoretically related and influenced by each other. As a matter of fact, it should be acknowledged that antipsychotics included in Group A are also largely and consistently used across Europe.

Although pharmacovigilance and drug utilization data sources did not cover the same geographical area (only Europe for drug utilization and theoretically all Countries for pharmacovigilance), we used these two dataset for different purposes: (a) FAERS, the largest public pharmacovigilance database, to accurately classify and characterize torsadogenic signal by antipsychotics, and (b) European drug utilization data to map the overall antipsychotic use (i.e., estimate the European population exposure).

Finally, with the exception of Scandinavian Countries [57] and now Scotland, record-linkage analyses with actual therapeutic indications, diagnoses and other clinical variables are challenging through reimbursement prescription databases. Nonetheless, we did not aim to assess whether differences arise among labeled and unlicensed indications or evaluate patients’ characteristics.