WP3 - Prescription Data Linkage
Work Undertaken (Deliverables)
- Linkage of prescription data to CA data (Deliverable 11)
- Review of literature on internet medication purchasing
- Prescription linkage report (Deliverable 12)
- Scientific paper on methodology and results of prescription data linkage (Deliverable 13)
|
The FP7 EUROmediCAT Research Programme piloted and validated data linkage. This involved 26,552 linked cases from 5 registries:
- Wales (years 1998-2010); n=5322 linked cases
- Norway (years 2004-2010); n=13,474 linked cases
- Odense (years 1998-2010); n=2006 linked cases
- Emilia Romagna (years 2004-2010); n=3034 linked cases
- Tuscany (years 2003-2010); n=2716 linked cases
WP3 Results and Foreground
The main objective of WP3 was to link electronic prescription data to congenital malformation registers to improve the information on medication exposure in the malformation registers. The linked data will be utilised in WP5.
Prescription data linkage (Dataset B)
Due to data confidentiality regulations, registries in Wales and Denmark were not allowed to send the linked case data outside their protected digital environment (server). Therefore, we applied a distributed database model in which the linked datasets were kept at the individual registries. This distributed database model replaced the intended central dataset B. CA cases born between 1998 (or first year available in the prescription database) and 2010 were included in data linkage. Linkage was performed by matching personal identification numbers or maternal characteristics. In the deliverable 11 report, the methodology of the linkage, the structure and the contents of the linked datasets are described.
The linked datasets consist of a master table with data on the malformed cases derived from the CA register and transformed to the EUROCAT format and a table with prescription data. Additional variables were added to the master file to indicate if the mother was identified in the prescription database. The table with prescription data is a relational table; therefore there is no maximum on the number of prescriptions per case. Required variables from the prescription data were ATC code of the prescribed medication and date of prescription. Optional variables included name of the medication, amount prescribed and daily dose prescribed.
Software development
An Access-based software module, the Linkage Data Management Program (LDMP), was developed to ensure a validated dataset B. The LDMP includes functions to import and export data, validate data and generate tables for evaluation and analyses. The LDMP builds further on the EUROCAT Data Management Program (EDMP), which was used by all registries to ensure a validated CA dataset B. The use of the LDMP ensured the compatibility of anomaly subgroups and medication groups among participating registries. First trimester exposure was defined as a prescription with the date of prescription between 0-97 days after LMP. For certain analyses the period for first trimester exposure was extended to include the 31 days before LMP. The functions of the LDMP are described in the EUROmediCAT manual. Case selection and predefined tables were generated through the LDMP. The tables were sent to the coordinator of the WP to be analysed centrally.
Odense, Denmark was not able to use the LDMP as the use of special software is not allowed on the server of Statistics Denmark. In addition, only the four medication groups of main interest in the EUROmediCAT project were included in the linkage in Odense because of the costs associated with using the data. The data from the CA register in Odense was validated through the EDMP. The prescription data was validated manually by an experienced statistician and medication subgroups and exposure periods were created according to the definitions in the linkage protocol. The required tables were generated manually, using the same selection criteria and definitions as in the LDMP. The results of the prescription data linkage are extensively described in the report on prescription data linkage (deliverable 12).
Linkage rate
A total of 52,619 cases of CA were present in the five CA databases. Linkage was possible for 65.6% of these cases. The linkage success varied per registry, from (almost) 100% in Odense and Norway to 32% in Wales. In Wales linkage was not possible for all pregnancies, as the primary care practices that supply the prescription data cover only 40% of the population. Therefore the prescription database did not cover 100% of the population. Emilia Romagna could not include TOPFA cases in the linkage, since the identification number used for linkage is only given at birth. The CA registry in Tuscany could recover the ID number for 52% of the mothers. Foreign born women were less likely to have their ID number recovered.
Maternal and case characteristics between linked and non-linked cases
For the registries that could not link all CA cases to the prescription database (Wales, Norway, ER and Tuscany), possible differences in characteristics between linked and non-linked CA were examined. There was a correlation between year of birth and linkage for all registries. Type of birth, maternal age and certain types of defects differed significantly between linked and non-linked cases for Tuscany and ER. Maternal age differed also between linked and non-linked cases in Wales and Norway.
First trimester exposure rates
It should be noted that the CA registers already contained some medication exposuredata pre-data linkage to prescription databases. In general, first trimester exposure rates for antiepileptic medication, insulin analogues and antiasthma medication were comparable between the prescription databases and the pre-linkage CA data.
Differences in prescription rates between registries can be attributed to prescribing differences between regions. For the SSRIs the first trimester exposure rate according to the prescription data was 3 times higher than according to the CA register data in Wales and Tuscany. The exposure rates for antibiotics according to the prescription data were much higher than the exposure rates according to the CA register data for all databases. Except for Wales, the exposure rates for gonadotropins were higher in the prescription data than in the CA register data.
Agreement between medication recorded in CA registers and prescription databases
Despite comparable first trimester exposure rates among CA registers and prescription databases, agreement between these two data sources may be low as the exposures can be recorded in different pregnancies. For medicines that are taken for chronic conditions such as epilepsy and diabetes, agreement according to the prescription data is high, above 60% (except for antiepileptic medication in Emilia Romagna), meaning that >60% of the cases who received a prescription for antiepileptic medication or insulin analogue in the first trimester also were exposed according to the CA register data. The agreement, both according to prescription data and CA register data for asthma medication is lower than for antiepileptic medication. A possible explanation is that certain asthma medication is only taken when necessary, so that a prescribed medicine can be used even months after prescription. For SSRIs, the agreement according to the prescription data is lower than according to the CA data. This means that if exposure to an SSRI is recorded in the CA data, usually a prescription in the first trimester is recorded in the prescription data, whereas if a prescription in first trimester is recorded in the prescription data, the exposure is not always recorded in the CA data. For antibiotics and gonadotropins the agreement for both data sources is low.
Extending the first trimester definition in the prescription data to include the month before the start of the pregnancy (-31 days to 97 days) did not influence the agreement for antiepileptic medication and insulin analogues, but did increase the agreement according to the CA data for antiasthma medication and SSRIs.
Actual use of prescribed medication
As part of this WP we performed a cross-sectional study embedded in the Northern Netherlands CA registry. In this study we investigated to what extent prescription data reflect the actual use of medication, prescribed during pregnancy. Compliance rates were calculated for several medication groups as the number of mothers that had taken the medication divided by the number of mothers to which the medication had been prescribed. Compliant use was defined as the mother confirming she took the medication, even when she received multiple prescriptions from the same medication group and took only one. In addition, we determined for each prescription that was taken, whether the medication administered conformed to prescribed dosage and duration. During the first trimester the compliance rate ranged from 0.83 for medication for chronic diseases and 0.92 for medication for pregnancy related symptoms. Most of the actually used medications were used according the dosage and duration prescribed, and if not, the used dosage and duration was lower than prescribed. Therefore we conclude that studies using prescription data will most likely overestimate exposure. However, this overestimation seems minimal which makes prescription records a reliable source in research on associations between medication use in pregnancy and congenital malformations.