FormalPara Key Points

Observational studies of drug safety involving pregnant and breastfeeding women based on secondary use of health data include more study subjects and were completed faster than studies based on primary data collection.

A very low number of product labels were updated based on the respective observational studies despite the large number of completed studies.

1 Introduction

It is well known that drug labelling documents, such as Summaries of Product Characteristics (SPCs) and Patient Information Leaflets (PILs), contain limited information about safety in pregnant and nursing women exposed to medicines [1]. While regulators such as the US Food and Drug Administration (FDA) and European Medicines Agency (EMA) require that medicinal products undergo testing to determine reproductive effects in animal models, pregnant women are typically not enrolled in clinical trials to assess the safety and efficacy of new medicinal products [2, 3].

Due to this paucity of data, there is a need to identify effects on pregnant women and their children post-licensure [4]. If new evidence about safety in pregnant and breastfeeding women is found, the pharmaceutical industry usually updates the company core data sheet which serves as the global reference document to communicate the company’s position to appropriate stakeholders worldwide for the inclusion of safety information on country-specific labels such as SPC and PIL [5].

In 2019, the FDA updated the guideline for pharmaceutical industries for Post-Approval Pregnancy Safety Studies (including recommendations about the pregnancy exposure registries [PERs]) [6]. In the same year, the EMA followed with the Good Pharmacovigilance Practices guide that advocates for adequate data collection and data assessment through existing PERs/databases, product-specific registries, or even a hybrid approach to enable patients and prescribers to have relevant information to make informed decisions about using medicines during pregnancy or breastfeeding [7].

Registry-based post-marketing safety studies designed only for breastfeeding women exposed to drugs and the effects on their infants are still very rare. The risk estimation for breastfed infants might come initially from the pharmacokinetic (PK) data of the product (assessed in a clinical lactation study) [8]. Nonetheless, in the case of a medicinal product frequently used by breastfeeding women, pregnancy registries in which infants are further observed could be useful to better assess the safety profile of a drug during the lactation period [7]. Despite the setup of the PER to identify potential drug effects in pregnancy and breastfeeding, there are challenges that impact the success of conducting observational studies in the pregnancy registries. PERs that include breastfeeding assessment may require long-term follow-up of children’s outcomes [7]. Data quality, enrollment size/retention, selection bias, lack of appropriate comparator group (e.g., a valid unexposed comparator cohort) and missing information play a key role in the challenges of these observational studies [4, 6, 9,10,11,12,13,14].

The wide availability of electronic health records (EHR) has boosted interest in assessing whether these can be used to replace or complement existing systems given their ability to overcome the aforementioned PER design limitations [9]. Electronic databases such as the IADB.nl database, Clinical Practice Research Datalink (CPRD) in the United Kingdom, Danish National Birth/Prescription Registry, Norwegian Medical Birth Registry, Certificate of Delivery Assistance (CeDAP) in Italy, National Community Child Health Database (NCCHD) in Wales, and the Mother Child Protection Centre and Hospital Medical Information System Database in France have the potential to provide detailed data on prescription drug exposures and outcome data for mothers and children [11, 15]. The Innovative Medicines Initiative (IMI)-ConcePTION project was set up to assess whether different types of data (including EHR) may facilitate the generation of real-world evidence on medicines safety in pregnancy and breastfeeding [16]. In this study, we aim to describe a landscape analysis of clinical and observational studies focusing on pregnancy and breastfeeding registered in the European Union Register of Post-Authorization Studies (EU PAS register) and ClinicalTrials.gov. We also aim to assess whether the findings of those completed studies led to updates of the respective product label.

2 Methods

2.1 Data Sources and Selection Criteria

We identified observational studies focusing on the evaluation of medicines safety during pregnancy and breastfeeding from the EU PAS register and ClinicalTrials.gov.

EU pharmacovigilance legislation requires the EMA to make public the protocols and abstracts of results of non-interventional post-authorization safety studies (PASS) imposed as an obligation of marketing authorization by a competent authority in accordance with Article 10 or 10a of Regulation (EC) No 726/2004 or with Articles 21a or 22a of Directive 2001/83/EC. Annex III of the Commission Implementing Regulation (EU) No 520/2012 further specifies that the final report of imposed non-interventional PASS must provide the date of making it public (in the EU PAS Register) [17]. ClinicalTrials.gov is a web-based resource with easily accessed information on publicly and privately supported clinical and observational studies on a wide range of diseases and conditions (including pregnancy and breastfeeding); observational studies are not required to be listed [18]. Both databases are fully searchable and allow the identification of study design, data collection (DC) type (primary data collection vs secondary data collection), expected enrollment size, study duration, study outcomes and other relevant variables.

A list of studies in the EU PAS register were obtained from EMA from its inception to December 2018 [19]. Observational studies that tagged ‘pregnant women’ or ‘breastfeeding mothers’ in the ‘Other population’ field of the register were included in this study.

EU PAS register identifiers were used to retrieve the study and extract data. We created a query (see Electronic Supplementary Material [ESM] 3) using the application programming interface of ClinicalTrials.gov to retrieve the variables of interest from observational studies with ‘pregnancy’ or ‘lactation’ conditions registered up to 24 May 2020 [18, 19].

Study titles retrieved from the EU PAS register were compared with those from the ClinicalTrials.gov list to identify and remove duplicates.

After removal of duplicates, we included

  1. (i)

    studies including pregnant or breastfeeding women in the study population,

  2. (ii)

    observational studies (including drug utilization studies) and

  3. (iii)

    studies aiming to measure medicine use or outcomes related to pregnancy or breastfeeding.

We excluded studies under the following criteria:

  1. (i)

    experimental studies,

  2. (ii)

    case reports and

  3. (iii)

    observational studies whose exposure comprised only non-medicinal products (e.g., devices, procedures, behavioral) and studies focused on a medical abortion program.

2.2 Extraction of Information From Eligible Studies

A summary of the definitions of the variables that were collected are presented in Table 1. Full descriptions of the variables, as well as the location where they were extracted from, can be found in ESM 1. Information was extracted on a set of key parameters from high-level metadata to detailed data (Table1).

Table 1 Summary of the variable definitions
Full size table

This included information regarding study classification, study design, data collection and enrollment size as provided by the respective researchers of each study from Sect. 1 of the EU PAS register and from the respective ClinicalTrials.gov field. The start date (date when contract was signed or date of first register, whichever is the earliest) and end date (date of analysis for ‘ongoing studies’ or date of final report entered in the forms) of each study were collected. When only month and year were available (e.g., December 2012), we considered the day to be the 15th of the month (e.g., 15 December 2012). We described the variables for each eligible observational study listed in ESM 2.

The eligible studies involved not only European countries and the US but also study sites in other continents.

2.3 Analysis

Descriptive analyses were done to describe and compare characteristics. We compared key variables between studies based on primary data collection and secondary use of health data Emphasis>from the EU PAS Register and ClinicalTrials.gov.

Kaplan-Meier plots were generated through IBM SPSS Statistics version 26.0 to compare the study period duration of primary data collection versus secondary data collection. The duration of these eligible studies was estimated using the start date and either the date of this analysis for ‘ongoing’ studies (they are considered as ‘censored’ because we did not observe their time to completion) or the end date of the study for ‘completed’ studies.

We performed an inter-rater reliability analysis of randomly selected studies to check consistency of data collection between two reviewers. To address potential inconsistent data extracted in ESM 1, data recorded by one reviewer was checked by another reviewer for variables in the same eligible study. All eligible studies were assessed by two independent reviewers and the adjudication of the discrepancies was documented to generate the final version of ESM 2.

2.4 Impact Assessment

We considered studies in EU PAS register/ClinicalTrials.gov to be ‘completed’, when the study’s end date preceded the date of the extraction of this research. The study’s end date is defined as the date entered by the study sponsor as either the final study report in the EU PAS register or as the study completion date in ClinicalTrials.gov. We captured the dates as shown in ESM 2.

To evaluate the impact of the results of completed studies on the respective label we first located the final reports or published papers; second, we compared the different versions of the European SPCs of corresponding drugs in the Electronic Medicines Compendium (EMC) via https://www.medicines.org.uk/emc. If the conclusion in the final report (when available) from a ‘completed’ observational study was similar to the respective medicine product label, we either accessed European public assessment reports or contacted the pharmaceutical companies responsible for the label (we contacted the pharmaceutical industry representatives who are part of the IMI-ConcePTION members project).

Missing data (such as confidential fields or absence of protocol/study results) were entered as Unknown (ESM 2).

3 Results

A total of 758 observational and interventional studies were identified that included a pregnant population: 610 studies in ClinicalTrials.gov and 148 studies in the EU PAS register. After the removal of 14 duplicates, 744 studies were left for further screening. A total of 603 studies were excluded Emphasis>as they were not observational, or not eligible for other reasons as shown in the flowchart (Fig. 1). The final number of studies evaluated in detail was 141 (40 from ClinicalTrials.gov and 101 from the EU PAS register).

Fig. 1
figure 1

Flow chart for selection of observational studies involving pregnancy and lactation. CT clinical trials, EU PAS register European Union Register of Post-Authorization Studies

Full size image

Pharmaceutical companies were the most frequent study sponsors (76%), followed by health authorities (14%) and academic research institutions (14%) (Table 2). The cohort design was the most frequently applied (80%) followed by case-control (7%) and cross-sectional (4%). Regarding data collection, 63 studies (45%) were based on primary data collection, 55 studies (39%) were based on secondary use of health data across multiple databases, while single database studies utilizing secondary data made up only 9% of the total eligible studies.

Table 2 Description of the key characteristics of the studies
Full size table

The majority of the studies (71%) were designed to assess the effects of prenatal exposure on pregnancy outcomes and a few studies (around 7%) aimed to assess the effects of medicines use during breastfeeding. We noticed that the most common risk window (18%) is ‘anytime during pregnancy’. When data is available, the most common period to follow a child was up to 1 year of age, as shown in Table 2.

A study may have more than one category of outcome, as defined in ESM 1 (e.g., general neonatal outcomes and specific maternal outcome as the primary study endpoint). General neonatal outcomes (including major and minor congenital anomalies [CAs]) were investigated in 36% of the selected studies as the primary study outcome, while general maternal outcome (38%) was the most frequent category in secondary study outcomes. Long-term outcomes in the child defined as any developmental outcomes (e.g., neurodevelopmental disorders) represented 18% and 8% of primary and secondary outcomes, respectively. About a quarter of the studies (26%) did not specify the disease or drug indication of the exposed pregnant women. When it was specified, influenza/H1N1, multiple sclerosis, rheumatoid arthritis/arthritis and depression were the most common disease indication areas (approximately 6%).

A descriptive comparison of studies based on primary data collection and secondary data collection (including studies with secondary use of data across multiple databases [MDS]) is presented in Table 3. Cohort studies are more common in secondary data studies (71%) compared with primary data (63%).

Table 3 Comparison of characteristics between studies based on primary data collection and secondary use of data
Full size table

Studies with secondary use of health data included many patients: the most frequent were between 501 and 5000 patients (32%) followed by more than 10,000 patients (31%). In contrast, 19% of studies using primary data collection enrolled < 100 patients as opposed to 7% of the studies with secondary use of data.

Major neonatal outcomes and general maternal outcomes was the most common proposed study outcome category for studies with either primary data collection or secondary use of health data (Table 3).

Ninety-one studies were ‘completed’. Survival analysis based on the time between study start and end lasted around 2.9 years as opposed to 7.5 years for duration of studies based on primary data collection (Fig. 2; completed studies with their finding [secondary data (B) × primary data (A)] over time (in days) and p-value < 0.05 long-rank test).

Fig. 2
figure 2

Completed studies with their finding (secondary use of health data × primary data) over time (in days)

Full size image

We could assess the impact on the pregnancy section of the product label for 46 studies with final reports or published papers. Nine labels had similar content in the pregnancy section as described in the final report/paper of the observational studies. Table 4 lists results of label updates.

Table 4 Label impact assessment
Full size table

4 Discussion

This study presents a landscape analysis of observational studies focusing on pregnancy and breastfeeding as registered in the ENCePP/EU PAS register and ClinicalTrials.gov. We also investigated whether ‘completed’ studies led to an update of the respective product label.

The key findings of our study are that there is about an equal number of studies that are based on primary data collection and secondary use of health data, but studies based on secondary use of health data were larger and were completed faster than studies based on primary data collection. In spite of the fact that a large number of studies were conducted, and completed, only a very low number of labels were updated based on a respective observational study. This finding is in line with an analysis conducted specifically on studies for multiple sclerosis, which concluded that “prospective, treatment-specific registries” have generally failed to deliver robust information. For this reason, other study approaches, in particular cohort studies using existing healthcare databases, should be considered for evaluating drug safety in pregnancy, including in MS [14]. Moreover, work on breastfeeding was very minimal. None of the labels assessed in this study were updated in the breastfeeding section.

More specific findings included that the period of exposure during pregnancy was not very detailed: ‘anytime during pregnancy” was most frequently chosen among the study sponsors. Automated databases often lack the collection of last menstrual period (LMP) data and gestation length, which allows estimation of the due date, approximate conception date, and gestational age at birth [20]. Thus, the study protocols should clearly define the period that women who are enrolled in the observational study are exposed to the drug at the point in gestation with the highest risk of causing congenital fetal effects, which most commonly is the first trimester [3]. Long-term follow-up information with infants is very limited.

The most recent EMA guidelines suggested that long-term pregnancy outcomes might be obtained by combining data from existing registries/databases and studies with primary data collection [6].

Size matters when studying rare outcomes such as CAs [13]. The desired sample size for the registry will depend on study design, frequency of that particular risk in both comparator group and exposed group as well as exposure frequency in the study population [9]. In general, pregnancy registries often fail to enroll and retain sufficient numbers of exposed pregnancies to detect rare CA [13, 21]. Tomson et al. classified registries that enroll pregnant women exposed to anti-epileptic drugs (AEDs) into categories: national registries, independent academic registries and pharmaceutical company registries [22]. The study highlights that independent academic registries enroll much larger numbers of prospective AED pregnancies compared with other studies [22]. One of the academic registries successfully recruited controls among the friends and family members of enrolled women [22].The two observational studies that impacted their respective product labels (Table 4) enrolled more than 10,000 patients through secondary use (MDS). We consider that a successful observational study should carefully project a realistic enrollment plan linked to study objective, continuous re-assessment of enrollment, and retention methods, in agreement with Sinclair et al. [23].

Another relevant finding of this study is that the studies with primary data collection lasted 5 years longer than studies with secondary data.

As health authorities and pharmaceutical companies agreed that the information from these observational studies was sub-optimal or insufficient, the question is raised whether only few studies generated meaningful evidence or whether the current product label process is ineffective in determining what constitutes adequate information to be included in the product label.

Given the tremendous lack of adequate information to guide pregnant women about the safety of medicines during pregnancy in the product labels, we would expect that study results will be included in such labels. From the 46 ‘completed’ studies, labels were updated for only two products whose study used secondary data collection. The breastfeeding sections of product labels have not been updated as completed studies focused only on prenatal exposure. Most of the labels that were not impacted state that there is either no data or limited data on the use of drugs in pregnant women. Limited data is deemed inadequate to determine an association between a drug and the undesirable neonatal/maternal outcomes. Some of the labels state that there are no sufficiently controlled studies in pregnant women. The lack of power of the study to find a potential association was also observed by Galperin’s research [13], which identified only one PER in their review as being properly designed primarily to assess the overall risk of major CA. It is acknowledged that the SPCs require several industry processes to validate their final content, which are ultimately approved by regulatory agencies, becoming an ‘official’ medicines information source [24]. Arguello et al. also identified that the majority of the SPCs assessed in their research stated there was no clinical experience of the use of the medicine during pregnancy and more than 60% of the SPCs indicated that excretion of the drug in human milk was unknown [24].

The sponsor should annually report on the number of pregnancies and maternal and infant outcomes from routine and additional pharmacovigilance activities (such as existing PERs and EHR) as part of the risk management plan. This approach combined with enhanced processes to obtain follow-up information on pregnancy case reports could help to overcome the challenges to generate meaningful evidence of the benefit–risk balance of medicines in pregnancy and breastfeeding. Although we used two reviewers to extract information and a structured format for such extractions, our research is limited by the fact that we use information from the EU PAS register and ClinicalTrials.gov. This information contained a significant number of ‘unknown’ variables, as presented in Tables 2 and 3, and they may not represent what was actually done. Another limitation is the completeness of the EU PAS register and ClinicalTrials.gov forms. Data were extracted mainly from these forms as filled out by the study investigator or sponsor and there is no policy established in the EU PAS register/ClinicalTrias.gov to check whether entry data is accurate. Alongside creating a policy, both databases should consider generating a unique identification number to avoid duplicates of the same study being entered either in the same database or in both the EU PAS register and ClinicalTrials.gov.

5 Conclusions

Our study aimed to review the observational studies involving pregnancy/breastfeeding entered in EU and US databases. It is clear that these studies are very heterogeneous and, especially when based on primary data collection, take a long time to reach the threshold of having adequate data. Despite completed studies having a limited impact on the product labels, we showed that two studies with secondary use of health data led to the respective product labels being updated. The current product label process should facilitate inclusion of new data on safety of medicines during pregnancy and breastfeeding. We also recommend making better use of existing health data across multiple databases. Ongoing consortiums such ConcePTION investigate how the power of health care sources on top of primary data collection can be best harnessed to improve the current information gap in knowledge about safety of medicines in pregnancy and breastfeeding to enhance our understanding of medication use and safety during pregnancy [16].