Effectiveness of risk minimisation measures for valproate: A drug utilisation study in Europe

Abstract Purpose The purpose of this study was to evaluate the effectiveness of the risk minimisation measures (RMMs) implemented in Europe in 2014 for valproate‐containing products to mitigate their risk during pregnancy and to characterise valproate prescribing patterns in women of childbearing potential (WCBP) before and after implementation of RMMs. Methods A multinational cohort study based on existing data sources using a pre‐/post‐ design was performed in five European countries (France, Germany, Spain, Sweden, UK) in an outpatient setting. Effectiveness of RMMs was assessed by comparing the proportion of valproate initiations as second (or subsequent) line therapy before and after implementation of RMMs (primary outcome) with an increase in this proportion indicating success of RMMs. Overall use of valproate and incidence of pregnancies in WCBP were also examined. Results The proportion of valproate initiations as second line therapy increased after implementation of RMMs in incident female users in Sweden (from 81.1%, 95% CI 79.9%‐82.3% to 84.5%, 95% CI 83.5%‐85.5%) and the UK (from 66.4%, 95% CI 64.5%‐68.3% to 72.4%, 95% CI 70.0%‐74.9%), it remained the same in Germany and Spain and decreased in France from 48.7% (95% CI 45.6%‐51.9%) to 40.6% (95% CI 37.6%‐43.7%). In Sweden and the UK, the incidence of pregnancies exposed to valproate decreased in the post‐implementation period: 8.0 vs 9.5 and 10.9 vs 16.9 per 1000 person‐years, respectively. Conclusion The results on primary outcome of this study suggest limited effectiveness of the RMMs. Additional RMMs were implemented in 2018.


| INTRODUCTION
Valproate containing medicines have been authorised for several decades across the European Union (EU) to treat epilepsy and bipolar disorder. In a few countries marketing authorisations are granted for prevention of migraine attacks. Numerous publications provide evidence on increased risks associated with valproate treatment during pregnancy. Approximately 10% of children exposed to valproate in the womb are at risk of congenital malformations. Developmental delays were reported for up to 30%-40% of those children in preschool age. [1][2][3][4][5][6][7] Furthermore, in utero exposure to valproate is associated with increased risk for autism spectrum disorder or symptoms of attention deficit/hyperactivity disorder (ADHD). 8 contraception. The PRAC also recommended that doctors who prescribe valproate provide women with full information to ensure understanding of the risks and to support their decisions." 12 The product information was amended accordingly. In addition, to improve awareness about the risks of valproate exposure during pregnancy, communication to healthcare professionals through a Direct Healthcare Professional Communication (DHPC) and educational materials (EMs) were provided to the healthcare professionals in the EU and to women prescribed valproate. 11 The EM included a prescriber guide, a patient booklet, an acknowledgment of risk information form including a checklist for prescribers and a checklist for patients or carers.
Within the context of RMMs and in compliance with the EMA Guideline on good pharmacovigilance practices (GVP) Module XVI, 13 two post-authorisation safety studies (PASS), a drug utilisation study (DUS) and a prescriber survey, were conducted to assess the effectiveness of these RMMs and to further characterise the prescribing patterns for valproate. 14 Here we present the results of the DUS in which prescribing practices and effectiveness of these measures were assessed in an outpatient setting by comparing the proportion of valproate initiations as second line therapy in females in general and specifically in women of childbearing potential (WCBP) before and after the implementation of RMMs (ie, dissemination of EMs and DHPC). The study was performed by a consortium of marketing authorization holders (MAHs) of valproate in Europe.

| Study design
This is a multinational non-interventional cohort study performed in the outpatient setting in five European countries (France, Germany, Spain, Sweden and United Kingdom) using existing data sources. The DUS used a common protocol which was approved by PRAC and registered in the European EU PAS register (EUPAS9678). Ethic and data access approvals were obtained according to local regulations. Study countries were selected based on the availability of longitudinal databases and the high number of patients exposed to valproate. A prepost design was employed to examine the changes in prescribing of valproate after implementation of RMMs. Each three-year pre-and post-implementation period was divided into a main and a transition sub-period ( Figure 1

KEY POINTS
• This study provides insights to valproate prescribing patterns in five EU countries (France, Germany, Spain, Sweden and the UK) in the period before and after implementation of risk minimisation measures (RMMs) in 2014.
• The change in the proportion of valproate initiations as second (or subsequent) line therapy varied across countries and suggests limited effectiveness of the RMMs on this specific measure.
• The number of valproate prescriptions in all female patients and specifically in women of childbearing potential decreased after implementation of RMMs.
• A decrease in the incidence rates of pregnancies exposed to valproate suggests a positive effect of RMMs.

| Population
The study population included all female patients receiving at least one prescription of an oral formulation of valproate in the outpatient setting during the study period in the selected databases. The main analysis unit was the prescription. Prescriptions for women aged 13 to 49 years at the prescription date were included in the WCBP subgroup.
Valproate prescriptions issued during the study period were defined as "incident" if there was no use of valproate within 12 months prior to the prescription date; "first-ever" valproate prescriptions were defined as no use of valproate documented during the patient's entire medical history. To avoid misclassification due to short medical history, classification as "incident" or "first-ever" prescription required medical history available in the database for at least 12 months prior to the valproate prescription date.

| Primary outcome
Given the safety concern that led to the prescription condition "Val-

| Pregnancy
Changes in the proportion of pregnancies were analysed in WCBP for the entire 36-month pre-and post-implementation periods. Of note, for Sweden data was available for only 23 months of the postimplementation period for this analysis. Pregnancies were identified via ICD-10 codes. The overall number of pregnancies as well as the number of pregnancies exposed to valproate were reported. A pregnancy F I G U R E 1 Study periods was considered as "exposed" if at least one valproate prescription was recorded during pregnancy. In Sweden, pregnancy start and end date were mainly derived from last menstrual period (LMP) or delivery date recorded in the Medical Birth register. In the CPRD Pregnancy Register in UK, estimated start and end date of pregnancy were available based on an algorithm described elsewhere. 17,18 In the physician panels used for France, Germany and Spain, pregnancies could mainly be identified in case of medical visits due to pregnancy complications or adverse events; information on start/end date was usually not available. If information on the pregnancy start/end date was not available in the data source, the pregnancy was considered as "exposed" if at least one valproate prescription was issued between 90 days before and 180 days after the first record related to pregnancy.

| Statistical analysis
Data were analysed at the prescription level by country and, in Germany and Spain, by physician panel (PCP and neurologists/psychiatrists) separately. Number of valproate prescriptions per study period was reported. Age was assessed at the date of each prescription and patient age was defined as age at the first prescription during the study period. Changes in prescribing behaviour before and after implementation of RMMs were primarily assessed using data from the two main study periods. The main study periods were used to ascertain the effectiveness of RMMs regardless of transition periods around their implementation. The entire 36-month pre-and post-implementation periods were used to analyse the incidence of pregnancies. All analyses were performed by descriptive statistical methods.
The results of the primary outcome were calculated as proportions (%) with 95% confidence intervals (CI). Non-overlapping 95% CIs indicated a significant difference between pre-and post-implementation periods. The incidence of pregnancies in WCBP was estimated per 1000 person-years. Data were analysed using SAS System 9.3 or higher (SAS Inc., Cary/NC, US).

| RESULTS
The  Germany (neurologists/psychiatrists panel, see Figure 3 and Table S1 in Data S1).  Table S2 in Data S1). In France the proportion was lower in the post-implementation period: 7.4% vs 19.7%. Under consideration of non-overlapping 95% CIs, the differences between two periods were statistically significant in Sweden and France.
In the subgroup "bipolar disorder" the proportion of valproate ini-  Table S2 in Data S1).
The results for first-ever valproate users were consistent with those for incident users in both "epilepsy" and "bipolar disorder" subgroups (Table S2 in Data S1). However, the proportion was generally higher for first-ever users than for all incident users. Based on the non-overlapping 95% CIs, the difference between the two main study periods was statistically significant in both indication subgroups in Sweden. Duration of entire pre-and post-implementation periods was 36 mo; in Swedenentire post-implementation period -35 mo. b Change of prescription number in the entire post-implementation period in comparison to the entire pre-implementation period.

| Pregnancy
In total, 451 of 923 pregnancies (48.9%) were exposed to valproate in the entire pre-implementation period and 182 of 350 pregnancies (52.0%) in the entire post-implementation period in all five target countries (Table S3 in Data S1).
The most comprehensive data on pregnancy was available in the data sources for Sweden and the UK. In Sweden, 179 of  and 85 of 177 (48.0%) exposed pregnancies in entire pre-and postimplementation periods, respectively.

The number of pregnancies identified in the EMR databases in
France and Spain was low (3 to 51); only one pregnancy was identified in Germany in the entire pre-implementation period.
The incidence rate of pregnancies (overall and exposed to valproate) in WCBP was lower in the entire post-implementation period than in the entire pre-implementation period in all analysed countries (Table 5).

| DISCUSSION
This DUS provides insights in the prescribing practices of valproate before and after implementation of RMMs and the effectiveness of these RMMs in the outpatient setting in five European countries. The most comprehensive data on pregnancy was available in the National Medical Birth Register in Sweden and in the UK, where the CPRD Pregnancy Register is a valid database for analysis of pregnancy. 18 The observed decrease in incidence of pregnancies in Sweden and the UK might indicate a positive impact of the RMMs.
Despite the decrease in pregnancy incidence, a number of pregnancies were still exposed to valproate during the post-implementation period. This finding supported the relevance of the more recent additional RMMs, including pregnancy prevention programs, introduced in 2018, in order to further reduce the number of exposed pregnancies. A limitation of this study was the inclusion of all female patients, irrespective of their fertility status. As the proportion of infertile/sterile women is expected to be low in the population and is unlikely to vary between the pre-and post-implementation periods, this limitation was considered minor with no impact on the results of comparative analyses.
On the other hand, this DUS has some major strengths. This study included extensive numbers of patients from large longitudinal databases in five European countries receiving valproate in the realworld setting with a lengthy observation period of up to 6 years. No exclusion criteria potentially introducing selection bias or affecting the external validity of results were applied in the study.
Overall, this study indicates that the effectiveness of RMMs was limited with regard to the valproate initiations as a second line therapy. However, the decrease in the overall number of valproate initiations and incidence of pregnancies in Sweden and the UK (the countries with the most interpretable data on pregnancy) suggest a reduction of the total exposure and exposure in pregnancy.
In 2018, the EMA introduced additional RMMs to strengthen the previous restrictions on valproate use. 23 Further studies are underway to evaluate effectiveness of these measures. An extension of this DUS is in preparation to further monitor the use of valproate in WCBP and particularly the occurrence of pregnancies exposed to valproate.
T A B L E 5 Incidence rate of pregnancies in overall and exposed to valproate (per 1000 person-years) in WCBP (age group 13 to 49 y); entire pre-and post-implementation periods Incidence per 1000 person-years (n pregnancies) Entire preimplementation period Entire postimplementation period a France All pregnancies 3.8 (14) 0.7 (2) Pregnancies exposed to valproate 2.7 (10) 0.4 (1)

ETHICS STATEMENT
All procedures performed in this study involving human participants were in accordance with the ethical standards of the applicable institutional review board and ethics committees and with the 1964 Declaration of Helsinki and its later amendments or comparable ethical standards. The protocol of the study was approved by the EMA and registered in the EU PAS register prior to the start of data collection.