WP2 - Central Database and Software Development
Work Undertaken (Deliverables)
- EDMP software (EDMP, LDMP and DCMP) and manuals ready (Deliverable 7)
- Fully documented EUROmediCAT database ready for future pharmacovigilance studies (Dataset A, see Table 2.1) (Deliverable 8)
- Development of Signal Detection methodology (Deliverable 9)
- Investigation of Signal Detection results (Deliverable 10)
|
WP2 Results and Foreground
Description of EUROCAT data made available for EUROmediCAT Central Database (Dataset A)
EUROCAT is a network of population-based registries for the epidemiological surveillance of congenital anomalies (CA) established in 1979. Each registry send anonymised, uniformly coded data on cases of CA registered in the local population. Registries use the EUROCAT Data Management Program (EDMP) to transmit standardised data to a central database. Cases of congenital anomaly are all major structural congenital and chromosomal anomalies diagnosed prenatally or postnatally. These include livebirths, fetal deaths from 20 weeks gestation age (GA) and terminations of pregnancy for fetal anomaly. All registries in the network use a standardised methodology ie. Standard variables, definitions, coding instructions (EUROCAT Guide 1.4, http://www.eurocat-network.eu/content/EUROCAT-Guide-1.4-Full-Guide.pdf), data dictionary.
Common variables include:
- Baby: type of birth, sex, survival, gestational age
- Diagnosis: 8 malformations and syndromes coded in ICD10 with BPA extension
- Maternal illness exposure (both acute and chronic)
- Medication exposure
- Family history
- Possible confounders such as maternal age, parity, and in some registries: BMI, socioeconomic status, maternal education
EUROCAT regroups coded congenital anomalies into 89 standard subgroups, and has a standard list of minor anomalies for exclusion (EUROCAT Guide 1.4).
Design and software development
WP2 developed the EUROmediCAT central database (Dataset A, Table 2.1) to collate uniformly-coded, validated data on cases of congenital anomaly, with information on maternal medication exposure during the first trimester of pregnancy, for 15 EUROCAT registries (Figure 2.1).
Table 2.1 Overview of datasets generated by EUROmediCAT
Figure 2.1 Map of EUROmediCAT Member Registries
The EUROmediCAT database built on the existing EUROCAT central registry database and the EDMP already used by local registries to transmit standardised anonymised data on congenital anomalies according to EUROCAT guidelines. Software developments were implemented in the central database and EDMP to improve coding of medication exposure data, including the provision of additional drug and maternal illness variables, as well as enabling 7 digit ATC drug codes to be entered for cases born before 2005 (Deliverable 7). Two new subgroups which group together aetiologically related congenital anomalies (laterality defects and neural crest defects) according to their ICD10 codes were approved by the EUROCAT Coding & Classification Committee and implemented in Dataset A.
A Prescription Linkage Program (LDMP) was developed and implemented in 5 registries with access to local prescription or administrative databases with information on prescriptions (Deliverable 7). Congenital anomaly data, extracted from Database A, was imported to LDMP which was then linked with local prescription/administrative databases. This distributed database (Dataset B, see Table 2.1) enabled validation of registry medication data, and generated standard pre-defined tables on medication use associated with congenital anomalies which could then be combined for analysis. See Workpackage 3 for further description. A Diabetic Cohort Management Program (DCMP) was developed and implemented in 7 registries participating in a cohort study of diabetic pregnancies. Data relating to the mother and her offspring including maternal insulin exposure and glycaemic control was entered/ imported to DCMP by the participating registries (Database C). Malformed cases in DCMP were linked to cases in Dataset A (Deliverable 7). These software developments were essential for WP 3, 4 and 5. A data manual with instructions on how to run the new software was produced and uploaded to the website: “The EUROmediCAT Manual: User Instructions Relating to Work Packages 2-4”.
Whereas the original intention had been to create an additional dataset, “dataset D”, from the population cohort linkage studies, it became evident that due to data protection considerations, the population cohort linkage studies in Workpackage 5 would have to proceed by independent linkage and analysis of databases, followed by meta-analysis. No Dataset D was therefore created.
Dataset A - establishment and documentation
Fifteen EUROmediCAT registries covering a total population of 7.2 million births in the period 1995-2012 contributed to Dataset A (Table 2.3, Figure 2.1).
Table 2.2 EUROmediCAT Registries contributing to Dataset A
There were 160,221 CA cases, defined as all major structural congenital and chromosomal anomalies diagnosed prenatally or postnatally (prevalence = 221.9 per 10,000 births) of which 89% were livebirths (LB), 1% were fetal deaths from 20 weeks gestationsl age and 9% were terminations of pregnancy for fetal anomaly (TOPFA). Of the 160,221 CA, 36,726 had ATC-coded medication exposure information excluding Vitamins/Minerals/Folic Acid during the first trimester of pregnancy, which equates to a medication exposure prevalence of 229.2 per 1,000 CA cases (Table 2.2, Figure 2.1). Figure 2.2 shows the distribution of CA types.
Figure 2.2 Distrubition of congenital anomalies registered in 15 EUROmediCAT registries
*An anomaly can occur in more than one organ group; however cases with chromosomal/genetic syndromes are only included in the chromosomal and genetic subgroups
All registries except Poland and Zagreb have an overall CA prevalence of between 2 and 3% but there are marked differences between registries in the proportion of CA cases with medication exposure information (Figure 2.3).
Figure 2.3 Prevalence of congenital anomalies per 10,000 births and proportion of congenital anomaly cases with medication exposure, per EUROmediCAT registry, 1995-2012
A profile of each registry contributing medication exposure data to Dataset A was produced detailing information on how medication exposure information is collated within the registry. All registries except one accessed maternal records eg. Obstetric, midwife, gynaecologist, delivery units, pregnancy pass; 9 registries accessed additional child health records eg. paediatrician, geneticist, neonatologist, paediatric cardiologist/neurologist/surgeon; 1 registry conducted maternal interview by clinician after birth; 1 registry accessed a prescription database and 1 was linked to the national prescription database. Consideration of data sources helps interpretation of differences in prevalence of medication exposure between registries (Deliverable 8).
Dataset A was used to conduct case malformed control studies in relation to medication exposure for new antiepileptic medications (WP4), Asthma medications (WP5) and SSRIs/antidepressants/psycholeptics (WP5). It was also used for systematic signal detection (WP2). Dataset A was linked to the Diabetic cohort (WP4) and to prescription databases (WP3).
Pharmacovigilance data summary and signal detection
WP2 developed EUROmediCAT interactive tables showing the frequency and prevalence of medications recorded in the EUROmediCAT central database by registry, year of birth and type of birth. The tables are restricted to the members-only section of the website for network use. The user can choose the level at which ATC codes are viewed up to the full fifth level for individual medicines. The feasibility of conducting studies of rare medication exposures of current interest can be determined from these tables as they provide a quick summary of rare medication exposures in the dataset.
Figure 2.4 Prevalence of medication exposures in Dataset A by type of medication
Dataset A was used for general pharmacovigilance (signal detection and signal evaluation) to identify potential new teratogens based on population surveillance. A pilot study based on a 25% sample of cases exposed to medications extracted from Dataset A determined that a False Discovery Rate (FDR) of 50% was appropriate for EUROmediCAT systematic signal detection. Following the pilot study, the signal detection process was further refined for use as a routine system, as recommended by EMA (Committee for Medical Product for Human Use. Guideline on the exposure to medicinal products during pregnancy: need for Post-authorisation data. EMEA/CHMP/313666/2005), and the methodology documented (Deliverable 9). Analysis, comparing the odds of exposure of a specific CA and drug was compared to the odds of exposure to the same drug in the remainder of the dataset using a 1-sided Fisher’s exact test.
Following exclusions, the final pharmacovigilance dataset consisted of 14,950 malformed babies. Analysis was conducted on 59 CA subgroups and 836 unique ATC drug exposures, which generated 77 potential signals (Figure 2.5). Eighteen of these were related to groups of ATC codes relating to teratogenic mechanisms of action, and 35 were related to new anti-epileptic drugs, Insulin analogues, anti-asthmatics and SSRIs (anti-depressants) and were forwarded to the relevant WP4 and WP5 leaders responsible for conducting risk assessment studies related to these medications. The remaining 24 potential signals were subject to a detailed investigation.
Figure 2.5 Flowchart of CA-medication exposure combinations in pharmacovigilance signal detection
Of the 24 original ‘other’ potential signals, 9 were discarded following verification of the data and detailed follow-up investigations, which left 15 validated potential CA-drug exposure signals related to gastrointestinal drugs (4), antihypertensives (n=2), female sex hormones (n=3), drugs used in infertility treatments (n=2), antiretrovirals (n=2), selective serotonin (5HT1) agonists (n=1) and detoxifying agents for anti-neoplastic treatment (n=1). An extensive literature review was conducted, using also specialist reproductive toxicity databases. Eight of these signals had no prior supporting evidence and require confirmation in an independent dataset. A further 7 signals had some prior supporting evidence and should be prioritised for further investigation before being further evaluated in relation to clinical decision making. Afull report of the methodology and the generated signals was produced (Deliverable 10) and will be disseminated as two scientific papers.