- Email: [email protected]
The Clinical Practice Research Datalink for Drug Safety in Pregnancy Research: an Overview
Thérapie 2014 Janvier-Février; 69 (1): 83–89 DOI: 10.2515/therapie/2014007
PHARMACOEPIDEMIOLOGY
© 2014 Société Française de Pharmacologie et de Thérapeutique
The Clinical Practice Research Datalink for Drug Safety in Pregnancy Research: an Overview Rachel Charlton, Julia Snowball, Cormac Sammon and Corinne de Vries Department of Pharmacy and Pharmacology, University of Bath, Bath, United Kingdom Text received October 8th, 2013; accepted November 18th, 2013
Keywords: electronic health records; pregnancy; pregnancy outcome; congenital abnormalities; teratogens
Abstract – Medicine use during pregnancy is common; however the safety of medicine use during pregnancy is largely unknown when a medicine comes to market. Electronic healthcare databases, including the Clinical Practice Research Datalink (CPRD), are increasingly being used for post-marketing surveillance in this field. The CPRD contains anonymised, longitudinal medical records routinely collected in primary care. Using CPRD data it is possible to identify medical records indicative of pregnancy, including pregnancy losses. Data on prescriptions issued can be used to determine maternal exposure and for about 80% of pregnancies it is possible to link the mother’s medical record to the medical record of the child. Data in the medical records of the mother and child can then be used to identify adverse pregnancy outcomes, including congenital malformations. This paper describes some of the complexities involved in using CPRD data for pregnancy related research and discusses some of its strengths and limitations.
Mots clés : dossiers de santé électroniques ; grossesse ; issues de grossesse ; anormalités congénitales ; teratogènes
Résumé – À propos de l’utilisation du Clinical Practice Reseach Datalink pour étudier la securité des médicaments pendant la grossesse. L’utilisation de médicaments pendant la grossesse est une pratique courante, néanmoins la sécurité de d’utilisation des médicaments chez la femme enceinte est largement inconnue lorsque un médicament est mis sur le marché. Les bases de données électroniques de santé, y compris la base de recherche clinique Clinical Practice Research Datalink (CPRD), sont de plus en plus utilisées pour la surveillance post-commercialisation dans ce domaine. Le CPRD contient des données médicales longitudinales anonymisées, recueillies lors des soins primaires. L’utilisation des données du CPRD, rend possible l’identification des dossiers médicaux indiquant une grossesse, y compris les interruptions de grossesse. Les données sur les ordonnances délivrées peuvent être utilisées pour déterminer l’exposition de la mère et pour environ 80 % des grossesses, il est possible de relier le dossier médical de la mère au dossier médical de l’enfant. Les données contenues dans les dossiers médicaux de la mère et de l’enfant peuvent être utilisées pour identifier les issues défavorables de la grossesse, y compris les malformations congénitales. Cet article décrit les difficultés liées à l’utilisation des données du CPRD pour la recherche liée à la grossesse et traite de certaines de ses forces et limites.
Abbreviations: see end of article.
1. Background Medication use is common in women of child bearing age. Studies have shown that between 27% and 99% of women take some form of medicine during pregnancy,[1-3] with estimates varying depending on the country and calendar time of study, the types of products included and the specific pregnancy time period of interest. Pregnant women are often excluded from clinical trials for ethical reasons and this means there is little information available on a medicine’s safety during pregnancy when a new medicine enters the
market. As a result, post-marketing surveillance is required to monitor and evaluate the safety of these products when used by pregnant women. Following the thalidomide disaster in the 1960s,[4] a number of post-marketing surveillance systems have been created to monitor the safety of medicine use during pregnancy. In recent years, the increase in routinely collected healthcare data has led to an increase in the use and number of electronic healthcare databases available for medication in pregnancy research. This paper aims to describe one electronic healthcare database, the Clinical Practice Research
Article publié par EDP Sciences
84
Charlton et al.
Datalink (CPRD) and provide an overview of the complexities, as well as the strengths and limitations, of this data source as a tool for studying the safety of medicine use during pregnancy. This paper will also provide a summary of research in this field that has been carried out using the CPRD to date.
2. The Clinical Practice Research Datalink The Clinical Practice Research Datalink, formerly the General Practice Research Database (GPRD), contains anonymised, longitudinal patient medical and prescribing records collected within United Kingdom general practice.[5] Data is entered into the database by general practice staff as part of routine patient management and includes information relating to pregnancy, symptoms and diagnoses, immunisations, consultations, tests, issued prescriptions and some information on lifestyle factors.[6] In the UK, the GP acts as the gatekeeper to services within the National health service and therefore some information relating to hospital or specialist referrals and admissions as well as outpatient and emergency visits may also be recorded. At present the CPRD captures ~8.5% of the UK population with ~5.4 million patients actively contributing data at any one time, in addition to over 7 million patients for whom historic data are available. The population captured by the CPRD is broadly representative of the UK population in terms of age and sex, although children aged 0-4 years and young adults (particularly males aged 17-30) are slightly under-represented.[7] The recording of data from each GP practice is subject to quality control checks and each practice is assigned an “up-to-standard” (UTS) date, which is the date the database provider considered the practice to have started contributing data that is of a standard suitable for the purposes of research. The CPRD group has obtained ethical approval from a Multi-Centre Research Ethics Committee (MREC) for all purely observational research using CPRD data. For each study, additional approval is required from the Independent Scientific Advisory Committee (ISAC).[8] Although the CPRD enables links to other data sources (such as hospital episode statistics and cancer registries) this paper focuses primarily on data that is captured and recorded within UK general practice.
3. Identification of pregnancies Data is largely entered into the CPRD in the form of medical Read codes and there are over 4 000 different Read codes that a GP can enter into a woman’s medical record relating to pregnancy. These codes relate to pregnancy tests, the date of the first day of the last menstrual period (LMP), the estimated date of delivery, delivery bookings, antenatal care and pregnancy outcomes, as well as neonatal and postnatal care. Pregnancy care in the UK is led primarily by midwives and obstetricians although occasionally the GP will be the primary point of contact throughout pregnancy. Most details are recorded in a paper file the woman takes with her wherever she goes
© Société Française de Pharmacologie et de Thérapeutique
in the healthcare system during her pregnancy. Therefore, the level of detail recorded in the CPRD is largely dependent on how much of the paper records is transferred into the general practice computer system; this may be more in cases where the GP is the primary point of contact or at practices where there is a more established routine of recording all details on the computer. As a result the level of detail recorded varies between patients. When carrying out research using CPRD data, most investigators identify the majority of pregnancies based on a Read code for a pregnancy outcome.[9,10] An advantage of the CPRD, compared to some other electronic healthcare databases, is that it captures all types of pregnancy outcome including live deliveries, stillbirths, induced terminations and spontaneous pregnancy losses, although very early spontaneous losses that occur before the pregnancy is clinically recognised are not captured. For some pregnancies identified in the CPRD, in addition to a medical code providing information on the date and type of pregnancy outcome, there is a medical code stating the date of conception or the date of the last menstrual period. For many pregnancies, however, accurate information on the beginning of pregnancy is not available and unlike databases such as “Évaluation chez la Femme Enceinte des MÉdicaments et de leurs RISques” (evaluation about drugs and their risks on pregnant women, EFEMERIS)[11] there is often little information on the gestational age at delivery from which this could be inferred. For these pregnancies it is therefore necessary to create an algorithm to try to estimate the date each pregnancy started by evaluating all the Read code entries within each woman’s medical record and making assumptions based on the data available. For example, if when a woman becomes pregnant she visits her GP and he enters the expected date of delivery of the pregnancy it is possible to work back from this date to determine an estimate of the date the pregnancy started. Alternatively, if a woman had a record for an antenatal scan at 12 weeks’ gestation an assumption can be made that the pregnancy started roughly 12 weeks before that date.[9] Although the creation of an algorithm can be beneficial in determining the start and end dates of a pregnancy, sometimes there is conflicting or insufficient information available in a woman’s medical record and assumptions need to be made regarding the duration of the pregnancy. For a delivery, these assumptions commonly take a defaulted pregnancy duration of between 270 and 280 days whilst a shorter duration of approximately 70 days has been taken for pregnancy losses.[9,12-15] Within the CPRD there are some deliveries where there is an indication that the pregnancy was premature (for example the Read code states “baby premature”) but there is no further information relating to the gestational age; here an alternative, shorter defaulted duration of 35 or 36 weeks may be assigned.[9] Assigning defaulted durations is one way to ensure a reasonable estimate of a pregnancy start date is established for all pregnancies, however for those that end in a premature or post-mature delivery, where there is no coded evidence to indicate the delivery did not occur at term, assigning a defaulted duration will result in an over or under-estimate of the duration of pregnancy.[16]
Thérapie 2014 Janvier-Février; 69 (1)
Drug Safety in Pregnancy Using the CPRD
4. Determining the type and timing of exposure Prescription data has been found to be reasonably complete in the CPRD and this is likely to result from the GP needing to use the computer system in order to generate a prescription.[17] The CPRD contains data on all prescriptions issued by GPs, regardless of whether the prescription was actually filled and dispensed at a pharmacy. In drug safety in pregnancy research the timing of drug exposure is critical, as different organ systems develop and are susceptible to interference at different timings of gestation. When using data from the CPRD, as with all electronic healthcare databases, it may therefore not be considered appropriate to determine exposure status based only on prescriptions issued during the first trimester of pregnancy, or the specific time period of interest. This is because, for example, a prescription with a duration of 60 days that is issued 15 days before the start of pregnancy has the potential to result in the female being exposed during the first 45 days of her pregnancy, which is a time when central nervous system and cardiac development is already taking place. When determining first trimester drug exposure investigators may choose to identify all prescriptions issued in the four to six months before pregnancy and map exposure based on the assumed duration of the prescription and the assumption that one prescription cannot start until the previous one has been completed, unless there is clear evidence of product switching.[12,13] The aforementioned difficulties in determining the gestational timing of events can also result in exposure misclassification. In general, exposure misclassification may be less of a concern for products prescribed for long-term use and chronic conditions than for products used intermittently or for short term use only.[16] Although the CPRD has the advantage of exposure data being recorded prospectively and independently by a prescriber, thereby avoiding any potential recall bias, exposure misclassification may occur as the result of non-compliance, or a woman deciding to stop taking her medicine when she is trying to become pregnant or once she discovers she is pregnant. Using CPRD data it is not possible to identify such non-compliance or discontinuation of treatment midprescription. The majority of prescriptions issued in secondary care will not be recorded in the CPRD, however repeat prescriptions issued by the GP for treatment initiated in secondary care should be captured. The CPRD does not capture medicines bought over-thecounter without a prescription, including standard dose (0.4 mg) folic acid.
5. Identifying pregnancy outcomes of interest 5.1. Major congenital malformations The most commonly studied outcome in drug safety in pregnancy research is the risk of a major congenital malformation
© Société Française de Pharmacologie et de Thérapeutique
85
(MCM) in the offspring. In order to evaluate this risk, it is necessary to link exposure data in the mother’s medical record to outcome data in the child’s medical record. This is done primarily using information on family number, as individuals within the CPRD are each assigned a practice-specific family number which is based on postal address. For each live delivery attempts are made to link the offspring to the mother, based on a birth or registration record within the same family on or around the same date as a live delivery or a pregnancy end date.[13] Care needs to be taken when there are two or more females with the same family number who both have records relating to pregnancy and also in situations when there is more than one possible child but the date of birth for each child is different. In these cases it may not be possible to reliably link the mother’s records with those of the offspring. In previous studies it has been possible to link over 80% of all live pregnancy outcomes to a child.[13] The CPRD has the advantage that there is the potential for relatively long periods of follow-up making it possible to identify MCMs that are diagnosed later on in life and not immediately following birth. For pregnancies that end in a live delivery the presence of an MCM can be identified based on Read codes in the child’s medical record. Some Read codes contain a lot of detail (e.g. coarctation of aorta) whilst others are vague (e.g. other specified congenital malformations of the heart). The nature of recording in the CPRD means that in some circumstances the presence of a single code for a particular condition may not be considered confirmation of a diagnosis. There may be situations where the GP enters a Read code as a “working diagnosis” and following further investigation the diagnosis is ruled out, but the Read code is not updated in the child’s record.[6] It is therefore often necessary to verify each of the potential MCMs identified and this can be done in a number of ways. Firstly, for some infants there may be supporting evidence of the diagnosis in the form of additional Read codes; for example there may be a code for a cleft lip and later in the record a code relating to surgical repair of the cleft lip. In these cases investigators can be confident the diagnosis was a true diagnosis. For MCMs where this is not the case it may be possible to request additional non-coded information recorded by the child’s GP. When entering Read codes into the computer system GPs have the opportunity to record additional non-coded comments, frequently referred to as “free text”. This free text can be used to record additional symptoms or detail alongside the Read code; for example the Read code description might be “coarctation of aorta” and the free text might state “has undergone coarctation repair”. Anonymised information recorded by the GP in the free text fields can be requested for all individuals in the CPRD, regardless of whether they are still registered with a GP who contributes to the database. If the patient is still registered with the GP practice a questionnaire can be sent to the GP to obtain additional information to help confirm or refute the diagnosis. For patients still registered it is also possible to request an anonymised photocopy of the patient’s part or full medical record enabling access to all referral and outpatient letters as well as correspondence from specialists.[18] All these
Thérapie 2014 Janvier-Février; 69 (1)
86
Charlton et al.
services are available on a project specific basis and at a charge via the CPRD Group. In addition to identifying MCMs in live births, it is also important to identify MCMs in pregnancies that ended in a pregnancy termination. This is particularly important for ensuring the capture of more severe congenital malformations, such as anencephaly, for which in the United Kingdom prenatal diagnosis and associated pregnancy termination is possible. For some pregnancy terminations recorded in the CPRD there is evidence of the reason for the termination, for example Read codes relating to an “unplanned” or “unwanted” pregnancy, but often this information is not available or may only have been recorded as free text. To identify the majority of MCMs in pregnancies that end in an induced termination it is necessary to request the free text comments recorded in association with the termination of pregnancy code(s) as often the reason for the termination is not evident from the Read code alone.[13] For example the Read code might just state “termination of pregnancy” and the free text comment might add that it was “for spina bifida at 16 weeks’ gestation”. 5.2. Embryonic and foetal deaths While a large number of spontaneous pregnancy losses occur in the first 8 weeks of pregnancy (embryonic deaths), many of these are unlikely to be recorded in the CPRD. Where they are recorded, they are likely to represent a selective population of planned pregnancies, for instance those where a couple is having difficulty to conceive. Drug safety in pregnancy studies carried out in the CPRD should therefore ideally be restricted to spontaneous pregnancy losses occurring from the 9th week of gestation (foetal deaths). In the UK, foetal deaths can be categorised as those occurring between gestational weeks 9 and 24 (miscarriage)[19] and those occurring after 24 weeks’ gestation (stillbirths).[20] Read codes for both of these outcomes exist. However, again assumptions may need to be made regarding the gestational age at the time of delivery or miscarriage. In doing so, one needs to be aware not all GPs will use the same definition for miscarriage and stillbirth and the magnitude of error in estimating gestational age might differ between live births, stillbirths and miscarriages. If such differential error is present, this will lead to differential misclassification of exposure and therefore biased risk estimates. Therefore, while the CPRD can be used for studies of foetal death, the potential for misclassification should be carefully considered and additional free text, questionnaire and/or linked data should be obtained for validation where possible. 5.3. Obstetric complications/delivery characteristics Read codes exist for a range of obstetric complications such as preeclampsia, obstetric haemorrhage and placental abruption and
© Société Française de Pharmacologie et de Thérapeutique
for a range of details recorded at delivery, including birth weight, mode of delivery, Apgar score, small for gestational age and preterm birth. However, as these events are managed predominantly in secondary care, they are not always recorded in a woman’s record in general practice. To our knowledge, there are no studies verifying the validity or completeness of recording of these events in the CPRD; however some investigators have compared the rates observed in the CPRD to the rates in more complete sources and concluded there is under-recording in the CPRD.[21,22] This has implications for the utility of the CPRD to study such outcomes or to include them as variables in any statistical evaluation. Where this information is considered a key study variable, linkage of CPRD data to more complete sources of secondary care data may be necessary.
6. Information on risk factors and confounders When calculating risk estimates for adverse pregnancy outcomes following exposure to a particular medicine it is often necessary to evaluate the effect of potential risk factors and confounders. Within the CPRD it is possible to obtain information on a range of variables that are either risk factors for the outcome of interest or act as confounders within drug safety in pregnancy research. These include maternal age, body mass index, alcohol drinking status, smoking status, deprivation status, co-morbidities (e.g. diabetes) and co-prescribing. For some patients where this information is missing and not recorded during pregnancy it is possible to make assumptions, for example if someone was a non-smoker before they became pregnant it is considered unlikely that they will have taken up smoking during pregnancy. There are, however, a number of individuals who have no data recorded in their medical record relating to some or all of these variables and then their status would be classified as unknown. Information on parity and gravidity is not consistently recorded in the CPRD, however the number of previous deliveries or number of previous spontaneous pregnancy losses recorded in the CPRD might be used as a proxy for these variables.[15] Within the CPRD there is some information on family history of congenital malformations; however this is thought to be selectively recorded and therefore considered too incomplete for any useful analysis. Data relating to genetic makeup are sporadically available within the database.
7. Overview of drug safety in pregnancy research using the GPRD/CPRD Having provided an overview of the CPRD and highlighted some of the key areas for consideration when using it as a tool for evaluating the safety of medicine use during pregnancy the remainder of this paper summarises pregnancy related research that has been carried out using the GPRD/CPRD to date.
Thérapie 2014 Janvier-Février; 69 (1)
Drug Safety in Pregnancy Using the CPRD
7.1. First use of the GPRD for drug safety in pregnancy research The first studies to report on pregnancy outcome research using the GPRD were published in the late 1990s.[23,24] These evaluated exposure to anticonvulsants[23] and fluconazole[24] and the associated risk of having a child born with a congenital malformation. The publications lacked detail on the exact methods of determining first trimester exposure and identification of congenital malformations. In addition both studies only captured live born pregnancy outcomes and had small sample sizes resulting in lack of statistical power. These studies were then followed by one in 1999 investigating pregnancy outcomes after intra-uterine exposure to any of three acid suppression drugs.[25] The methods reported were more detailed than the earlier papers but given what is known about the importance of the timing of drug exposure for this kind of research, the study used a wide time window of interest to determine exposure status (30 days before through to 100 days after LMP). This study did, however, capture non-live pregnancy outcomes including terminations of pregnancy. 7.2. Methodological studies In 2004, the first study focussing on the methodological aspects of using the GPRD for drug safety in pregnancy research was published. This paper evaluated strategies for identifying pregnancies in the database with a focus on the complexity and the vast number of medical codes available for use by GPs.[26] This area of research was then also covered by Snowball and De Vries in 2007.[9] In 2010, Devine et al. published a paper continuing on the topic of identifying pregnancies in the GPRD, this paper not only reported on the creation of a computer based algorithm for identifying pregnancies but also on the results of validating this algorithm.[10] The algorithm was more successful for live births than for other types of pregnancy outcome. It was suggested this might be due to difficulties in determining pregnancy start and end dates when no child’s record exists and where often multiple records relating to a termination of pregnancy are recorded. In 2006, Hardy et al. went on to evaluate the types of medicines that women in the GPRD were prescribed during pregnancy and focussed on those pregnancies where it was possible to link the mother to her child.[27] This study captured prescriptions issued during two time periods; the 90 days before and the 70 days after the first recorded pregnancy code. Although more work was required to improve the precision of the timing of exposure, this study did highlight the large proportion of women within the GPRD who are exposed to medicines for which the safety in pregnancy was not fully known. 7.3. Congenital malformation verification In 2007, the first research was published that evaluated the accuracy and completeness of congenital malformation recording in the
© Société Française de Pharmacologie et de Thérapeutique
87
GPRD with a focus on cardiovascular defects.[28,29] The first study compared the prevalence of congenital heart defects in the GPRD with two UK population based sources (the National congenital anomaly system and the UK contributors to EUROCAT).[29] The overall prevalence of congenital heart defects in the GPRD was higher than that reported in both of the UK population based sources. The authors suggested this could be owing to differences in data collection methods and in particular the voluntary nature of reporting by healthcare professionals to the UK population based sources. The second study attempted to verify three specific heart defects identified in the electronic medical records, using additional information obtained from free text and GP questionnaires.[28] In this study, there was a 93.5% positive predictive value (PPV) between the computerised records and responses to the GP questionnaires, ranging from 90% for tetralogy of Fallot to 100% for coarctation of the aorta. Similar research was published in 2008, this time focussing on neural tube defects (NTDs).[30] Owing to the nature of this class of defect the study included diagnoses that resulted in a termination of pregnancy, in addition to those found in live and stillbirths. An algorithm was developed to identify new cases of NTDs in either the mother’s or the child’s record and the identified cases were then verified using GP questionnaires. Of the 169 questionnaires returned, 117 (71%) NTD cases were verified. The PPV of the algorithm varied by NTD diagnosis and ranged from 0.81 for anencephaly to 0.47 for spina bifida. The authors reported that the low PPV for spina bifida could have resulted from the inability of the algorithm to differentiate between cases where the mother herself had spina bifida and cases where it was present in the foetus or child. The annual prevalences of NTDs in the GPRD were comparable to those of the National congenital anomaly system in later years, but slightly higher in the early years. In 2011, a paper reporting on the sensitivity of congenital malformation recording in the GPRD and the added value of requesting photocopied medical records and free text was published.[18] This study attempted to verify a range of potentially major congenital malformations identified in a cohort of women with epilepsy in the GPRD. By using a combination of photocopied medical records and free text, 160/188 (85.1%) malformations could be verified. The percentage that could be verified was higher for those where photocopied records were requested (91.7%) than those where only free text was available because the child was no longer registered with the practice or the practice was no longer contributing data to the GPRD (77.9%). All studies that included verification of MCMs in the GPRD reported it was possible to identify the MCMs of interest using the electronic data but concluded that obtaining additional detail and information from free text comments, photocopied medical records or GP questionnaires was strongly recommended.[18,28-30] 7.4. Potential to act as a pregnancy registry system One paper reported on the mean number of annual exposures to a range of drugs during pregnancy captured in the GPRD and
Thérapie 2014 Janvier-Février; 69 (1)
88
Charlton et al.
compared this with the numbers enrolled in pregnancy exposure registries over the same time period.[31] For more prevalent conditions, such as depression, the GPRD had captured a similar mean annual number of exposures as a pregnancy registry. Charlton et al. then investigated whether the GPRD could be used to replicate the findings of the UK epilepsy and pregnancy register and whether it was possible to identify a known teratogenic association using GPRD data.[13] An increased MCM risk following first trimester polytherapy antiepileptic drug exposure compared with no antiepileptic drug exposure was found; however lower rates of MCMs were identified in the GPRD following exposure to valproate than in the UK epilepsy and pregnancy register. It was possible however to identify in the GPRD the known teratogenic association between first trimester exposure to valproate and an increased risk of spina bifida.
7.5. Drug utilisation and risk assessment studies In recent years the GPRD/CPRD has been used to evaluate the utilisation of asthma medicines by pregnant women[12] as well as the uptake of pandemic influenza vaccination during pregnancy.[14] Data from the CPRD has also been used to carry out risk assessment studies to evaluate the risk of congenital anomalies following exposure during pregnancy to topical corticosteroids,[22] first trimester exposure to antihypertensive drugs,[32] antidepressants[33] and asthma medicines.[34] Studies have also been conducted to evaluate the risk of cardiac malformations following exposure to selective serotonin reuptake inhibitors,[35] the risk of pregnancy losses in women with diabetes[36] and the hazard of foetal death following H1N1 influenza vaccination.[15]
8. Conclusion The CPRD has a number of key strengths for drug safety in pregnancy research when compared with some other data sources. These include the representative nature of the population captured, the availability of the data, the ability to identify many types of pregnancy outcome, the potential for large sample sizes and the presence of a denominator population. The capability of the CPRD is likely to be restricted, however, by small sample sizes when evaluating products used to treat less prevalent conditions, the small study population available for evaluating rare birth defects, the inaccuracies associated with determining the pregnancy duration, establishing exposure based on the issue of a prescription and a lack of information on some potential confounding factors such as over-the-counter folic acid exposure. The on-going expansion of the CPRD, in terms of the number of patients captured and the ability to link to secondary and tertiary care datasets, will increase sample sizes and have the potential to improve the precision and accuracy of drug safety in pregnancy studies using CPRD data. The CPRD therefore has the
© Société Française de Pharmacologie et de Thérapeutique
potential to complement the work of other surveillance systems including pregnancy exposure registries and case-control systems.
Acknowledgements Some of the work presented in this paper formed part of the following PhD thesis: RA Charlton. (2012) The General Practice Research Database as an alternative to registries for studying drug safety in pregnancy: anticonvulsants as a case study. University of Bath, Bath, UK. Conflicts of interests. The authors have no conflicts of interests to declare. The authors did not receive any funding specific to the writing of this manuscript. Abbreviations. CPRD : Clinical Practice Research Datalink; EFEMERIS: Évaluation chez la Femme Enceinte des MÉdicaments et de leurs RISques (evaluation about drugs and their risks on pregnant women); GP: general practitioner; GPRD: General Practice Research Database; ISAC: Independent Scientific Advisory Committee; LMP: last menstrual period; MCM: major congenital malformation; MREC: Multi-Centre Research Ethics Committee; NTD: neural tube defects; PPV: positive predictive value; UTS: up-to-standard.
References 1.
Daw JR, Hanley GE, Greyson DL, et al. Prescription drug use during pregnancy in developed countries: a systematic review. Pharmacoepidemiol Drug Saf 2011; 20(9): 895-902
2.
Cleary BJ, Butt H, Strawbridge JD, et al. Medication use in early pregnancyprevalence and determinants of use in a prospective cohort of women. Pharmacoepidemiol Drug Saf 2010; 19(4): 408-17
3.
Lacroix I, Damase-Michel C, Lapeyre-Mestre M, et al. Prescription of drugs during pregnancy in France. Lancet 2000; 356(9243): 1735-6
4.
McBride W. Thalidomide and congenital abnormalities. Lancet 1961 (Dec 16); 2: 1358
5.
The Clinical Practice Research Datalink. Medicines and healthcare products regulatory agency; 2013 http://www.cprd.com Accessed November 18th, 2013
6.
Charlton RA. The General Practice Research Database as an alternative to registries for studying drug safety in pregnancy: anticonvulsants as a case study. University of Bath, Bath, UK. 2012: PhD Thesis
7.
.Campbell J, Dedman J, Easton S, et al. Is the GPRD GOLD population comparable to the UK population? Pharmacoepidemiol Drug Saf 2013 (22); S1: 1-521 http://onlinelibrary.wiley.com/doi/10.1002/pds.3512/pdf Accessed November 18th, 2013 [Abstract 567]
8.
Independent Scientific Advisory Committe for MHRA database research. http://www.cprd.com/ISAC/ Accessed November 18th, 2013
Thérapie 2014 Janvier-Février; 69 (1)
Drug Safety in Pregnancy Using the CPRD
9.
Snowball JM, de Vries CS. Determination of pregnancy on the General Practice Research Database. Pharmacoepidemiol Drug Saf 2007; 16: S1–274 DOI: 10.1002/pds [Abstract 249]
10. Devine S, West S, Andrews E, et al. The identification of pregnancies within the General Practice Research Database. Pharmacoepidemiol Drug Saf 2010; 19(1): 45-50 11. Lacroix I, Hurault C, Sarramon MF, et al. Prescription of drugs during pregnancy: a study using EFEMERIS, the new French database. Eur J Clin Pharmacol 2009; 65(8): 839-46 12. Charlton RA, Hutchison A, Davis KJ, et al. Asthma management in pregnancy. PLoS ONE 2013; 8(4): e60247
89
24. Jick SS. Pregnancy outcomes after maternal exposure to fluconazole. Pharmaotherapy 1999; 19(2): 221-2 25. Ruigomez A, Garcia Rodriguez LA, Cattaruzzi C, et al. Use of cimetidine, omeprazole and ranitidine in pregnant women and pregnancy outcomes. Am J Epidemiol 1999; 150: 476-81 26. Hardy JR, Holford TR, Hall GC, et al. Strategies for identifying pregnancies in the automated medical records of the General Practice Research Database. Pharmacoepidemiol Drug Saf 2004; 13: 749-59 27. Hardy JR, Leaderer BP, Holford TR, et al. Safety of medications prescribed before and during early pregnancy in a cohort of 81 975 mothers from the UK General Practice Research Database. Pharmacoepidemiol Drug Saf 2006; 15: 555-64
13. Charlton RA, Weil JG, Cunnington M, et al. Comparing the General Practice Research Database and the UK epilepsy and pregnancy register as tools for postmarketing teratogen surveillance: anticonvulsants and the risk of major congenital malformations. Drug Saf 2011; 34(2): 157-71
28. Wurst KE, Ephross SA, Loehr J, et al. The utility of the General Practice Research Database to examine selected congenital heart defects: a validation study. Pharmacoepidemiol Drug Saf 2007; 16: 867-77
14. Sammon CJ, McGrogan A, Snowball JM, et al. Pandemic influenza vaccination during pregnancy: an investigation of vaccine uptake during the 2009/ 10 pandemic vaccination campaign in Great Britain. Hum Vaccin Immunother 2013; 9(4): 917-23
29. Wurst KE, Ephross SA, Loehr J, et al. Evaluation of the General Practice Research Database congenital heart defects prevalence: Comparison to United Kingdom national systems. Birth Defects Res A Clin Mol Teratol 2007; 79: 309-16
15. Sammon CJ, Snowball J, McGrogan A, et al. Evaluating the hazard of foetal death following H1N1 influenza vaccination: a population based cohort study in the UK GPRD. PLoS ONE 2012; 7(12): e51734
30. Devine S, West SL, Andrews E, et al. Validation of neural tube defects in the full featured - General Practice Research Database. Pharmacoepidemiol Drug Saf 2008; 17: 434-44
16. Toh S, Mitchell AA, Werler MM, et al. Sensitivity and specificity of computerized algorithms to classify gestational geriods in the absence of information on date of conception. Am J Epidemiol 2008; 167(6): 633-40
31. Charlton RA, Cunnington MC, de Vries CS, et al. Data resources for investigating drug exposure during pregnancy and associated outcomes: the General Practice Research Database (GPRD) as an alternative to pregnancy registries. Drug Saf 2008; 31(1): 39-51
17. Walley T, Mantgani A. The UK General Practice Research Database. Lancet 1997; 350(9084): 1097-9 18. Charlton RA, Weil JG, Cunnington M, et al. Identifying major congenital malformations on the General Practice Research Database: a study reporting on the sensitivity and added value of photocopied records and free text. Drug Saf 2010; 33(9): 741-50 19. Royal College of obstetricians and gynaecologists. The management of early pregnancy loss. 2006 http://www.rcog.org.uk/files/rcog-corp/uploaded-files/ GT25ManagementofEarlyPregnancyLoss2006.pdf Accessed November 18th, 2013 (18 pages) 20. Royal College of obstetricians and gynecologists. Registration of stillbirths and certification for pregnancy loss before 24 weeks of gestation. 2005 http://www.rcog.org.uk/files/rcog-corp/uploaded-files/ GoodPractice4RegistrationStillbirth2005.pdf Accessed November 18th, 2013 (4 pages) 21. Howard LM, Goss CLAU, Leese MORV, et al. Medical outcome of pregnancy in women with psychotic disorders and their infants in the first year after birth. Br J Psychiatry 2003; 182(1): 63-7 22. Chi CC, Mayon-White RT, Wojnarowska FT. Safety of topical corticosteroids in pregnancy: a population-based cohort study. J Invest Dermatol 2011; 131(4): 884-91 23. Jick S, Terris BZ. Anticonvulsants and congenital malformations. Pharmacotherapy 1997; 17(3): 561-4
© Société Française de Pharmacologie et de Thérapeutique
32. Vasilakis-Scaramozza C, Aschengrau A, Cabral HJ, et al. Antihypertensive drugs and the risk of congenital anomalies. Pharmacotherapy 2013; 33(5): 476-82 33. Vasilakis-Scaramozza C, Aschengrau A, Cabral H, et al. Antidepressant use during early pregnancy and the risk of congenital anomalies. Pharmacotherapy 2013; 33(7): 693-700 34. Vasilakis-Scaramozza C, Aschengrau A, Cabral HJ, et al. Asthma drugs and the risk of congenital anomalies. Pharmacotherapy 2013; 33(4): 363-8 35. Margulis AV, Abou-Ali A, Strazzeri MM, et al. Use of selective serotonin reuptake inhibitors in pregnancy and cardiac malformations: a propensityscore matched cohort in CPRD. Pharmacoepidemiol Drug Saf 2013; 22(9): 942-51 36. McGrogan A, Snowball JM, de Vries CS. Pregnancy losses in women with type 1 or type 2 diabetes in the UK: an investigation using primary care records: pregnancy losses in women with diabetes. Diabetic Medicine 2013; Sep 30: doi: 10.1111/dme.12332
Correspondence and offprints: Rachel Charlton, Department of Pharmacy and Pharmacology, University of Bath, Claverton Down, Bath, BA2 7AY, United Kingdom. E-mail: [email protected]
Thérapie 2014 Janvier-Février; 69 (1)
PHARMACOEPIDEMIOLOGY
© 2014 Société Française de Pharmacologie et de Thérapeutique
The Clinical Practice Research Datalink for Drug Safety in Pregnancy Research: an Overview Rachel Charlton, Julia Snowball, Cormac Sammon and Corinne de Vries Department of Pharmacy and Pharmacology, University of Bath, Bath, United Kingdom Text received October 8th, 2013; accepted November 18th, 2013
Keywords: electronic health records; pregnancy; pregnancy outcome; congenital abnormalities; teratogens
Abstract – Medicine use during pregnancy is common; however the safety of medicine use during pregnancy is largely unknown when a medicine comes to market. Electronic healthcare databases, including the Clinical Practice Research Datalink (CPRD), are increasingly being used for post-marketing surveillance in this field. The CPRD contains anonymised, longitudinal medical records routinely collected in primary care. Using CPRD data it is possible to identify medical records indicative of pregnancy, including pregnancy losses. Data on prescriptions issued can be used to determine maternal exposure and for about 80% of pregnancies it is possible to link the mother’s medical record to the medical record of the child. Data in the medical records of the mother and child can then be used to identify adverse pregnancy outcomes, including congenital malformations. This paper describes some of the complexities involved in using CPRD data for pregnancy related research and discusses some of its strengths and limitations.
Mots clés : dossiers de santé électroniques ; grossesse ; issues de grossesse ; anormalités congénitales ; teratogènes
Résumé – À propos de l’utilisation du Clinical Practice Reseach Datalink pour étudier la securité des médicaments pendant la grossesse. L’utilisation de médicaments pendant la grossesse est une pratique courante, néanmoins la sécurité de d’utilisation des médicaments chez la femme enceinte est largement inconnue lorsque un médicament est mis sur le marché. Les bases de données électroniques de santé, y compris la base de recherche clinique Clinical Practice Research Datalink (CPRD), sont de plus en plus utilisées pour la surveillance post-commercialisation dans ce domaine. Le CPRD contient des données médicales longitudinales anonymisées, recueillies lors des soins primaires. L’utilisation des données du CPRD, rend possible l’identification des dossiers médicaux indiquant une grossesse, y compris les interruptions de grossesse. Les données sur les ordonnances délivrées peuvent être utilisées pour déterminer l’exposition de la mère et pour environ 80 % des grossesses, il est possible de relier le dossier médical de la mère au dossier médical de l’enfant. Les données contenues dans les dossiers médicaux de la mère et de l’enfant peuvent être utilisées pour identifier les issues défavorables de la grossesse, y compris les malformations congénitales. Cet article décrit les difficultés liées à l’utilisation des données du CPRD pour la recherche liée à la grossesse et traite de certaines de ses forces et limites.
Abbreviations: see end of article.
1. Background Medication use is common in women of child bearing age. Studies have shown that between 27% and 99% of women take some form of medicine during pregnancy,[1-3] with estimates varying depending on the country and calendar time of study, the types of products included and the specific pregnancy time period of interest. Pregnant women are often excluded from clinical trials for ethical reasons and this means there is little information available on a medicine’s safety during pregnancy when a new medicine enters the
market. As a result, post-marketing surveillance is required to monitor and evaluate the safety of these products when used by pregnant women. Following the thalidomide disaster in the 1960s,[4] a number of post-marketing surveillance systems have been created to monitor the safety of medicine use during pregnancy. In recent years, the increase in routinely collected healthcare data has led to an increase in the use and number of electronic healthcare databases available for medication in pregnancy research. This paper aims to describe one electronic healthcare database, the Clinical Practice Research
Article publié par EDP Sciences
84
Charlton et al.
Datalink (CPRD) and provide an overview of the complexities, as well as the strengths and limitations, of this data source as a tool for studying the safety of medicine use during pregnancy. This paper will also provide a summary of research in this field that has been carried out using the CPRD to date.
2. The Clinical Practice Research Datalink The Clinical Practice Research Datalink, formerly the General Practice Research Database (GPRD), contains anonymised, longitudinal patient medical and prescribing records collected within United Kingdom general practice.[5] Data is entered into the database by general practice staff as part of routine patient management and includes information relating to pregnancy, symptoms and diagnoses, immunisations, consultations, tests, issued prescriptions and some information on lifestyle factors.[6] In the UK, the GP acts as the gatekeeper to services within the National health service and therefore some information relating to hospital or specialist referrals and admissions as well as outpatient and emergency visits may also be recorded. At present the CPRD captures ~8.5% of the UK population with ~5.4 million patients actively contributing data at any one time, in addition to over 7 million patients for whom historic data are available. The population captured by the CPRD is broadly representative of the UK population in terms of age and sex, although children aged 0-4 years and young adults (particularly males aged 17-30) are slightly under-represented.[7] The recording of data from each GP practice is subject to quality control checks and each practice is assigned an “up-to-standard” (UTS) date, which is the date the database provider considered the practice to have started contributing data that is of a standard suitable for the purposes of research. The CPRD group has obtained ethical approval from a Multi-Centre Research Ethics Committee (MREC) for all purely observational research using CPRD data. For each study, additional approval is required from the Independent Scientific Advisory Committee (ISAC).[8] Although the CPRD enables links to other data sources (such as hospital episode statistics and cancer registries) this paper focuses primarily on data that is captured and recorded within UK general practice.
3. Identification of pregnancies Data is largely entered into the CPRD in the form of medical Read codes and there are over 4 000 different Read codes that a GP can enter into a woman’s medical record relating to pregnancy. These codes relate to pregnancy tests, the date of the first day of the last menstrual period (LMP), the estimated date of delivery, delivery bookings, antenatal care and pregnancy outcomes, as well as neonatal and postnatal care. Pregnancy care in the UK is led primarily by midwives and obstetricians although occasionally the GP will be the primary point of contact throughout pregnancy. Most details are recorded in a paper file the woman takes with her wherever she goes
© Société Française de Pharmacologie et de Thérapeutique
in the healthcare system during her pregnancy. Therefore, the level of detail recorded in the CPRD is largely dependent on how much of the paper records is transferred into the general practice computer system; this may be more in cases where the GP is the primary point of contact or at practices where there is a more established routine of recording all details on the computer. As a result the level of detail recorded varies between patients. When carrying out research using CPRD data, most investigators identify the majority of pregnancies based on a Read code for a pregnancy outcome.[9,10] An advantage of the CPRD, compared to some other electronic healthcare databases, is that it captures all types of pregnancy outcome including live deliveries, stillbirths, induced terminations and spontaneous pregnancy losses, although very early spontaneous losses that occur before the pregnancy is clinically recognised are not captured. For some pregnancies identified in the CPRD, in addition to a medical code providing information on the date and type of pregnancy outcome, there is a medical code stating the date of conception or the date of the last menstrual period. For many pregnancies, however, accurate information on the beginning of pregnancy is not available and unlike databases such as “Évaluation chez la Femme Enceinte des MÉdicaments et de leurs RISques” (evaluation about drugs and their risks on pregnant women, EFEMERIS)[11] there is often little information on the gestational age at delivery from which this could be inferred. For these pregnancies it is therefore necessary to create an algorithm to try to estimate the date each pregnancy started by evaluating all the Read code entries within each woman’s medical record and making assumptions based on the data available. For example, if when a woman becomes pregnant she visits her GP and he enters the expected date of delivery of the pregnancy it is possible to work back from this date to determine an estimate of the date the pregnancy started. Alternatively, if a woman had a record for an antenatal scan at 12 weeks’ gestation an assumption can be made that the pregnancy started roughly 12 weeks before that date.[9] Although the creation of an algorithm can be beneficial in determining the start and end dates of a pregnancy, sometimes there is conflicting or insufficient information available in a woman’s medical record and assumptions need to be made regarding the duration of the pregnancy. For a delivery, these assumptions commonly take a defaulted pregnancy duration of between 270 and 280 days whilst a shorter duration of approximately 70 days has been taken for pregnancy losses.[9,12-15] Within the CPRD there are some deliveries where there is an indication that the pregnancy was premature (for example the Read code states “baby premature”) but there is no further information relating to the gestational age; here an alternative, shorter defaulted duration of 35 or 36 weeks may be assigned.[9] Assigning defaulted durations is one way to ensure a reasonable estimate of a pregnancy start date is established for all pregnancies, however for those that end in a premature or post-mature delivery, where there is no coded evidence to indicate the delivery did not occur at term, assigning a defaulted duration will result in an over or under-estimate of the duration of pregnancy.[16]
Thérapie 2014 Janvier-Février; 69 (1)
Drug Safety in Pregnancy Using the CPRD
4. Determining the type and timing of exposure Prescription data has been found to be reasonably complete in the CPRD and this is likely to result from the GP needing to use the computer system in order to generate a prescription.[17] The CPRD contains data on all prescriptions issued by GPs, regardless of whether the prescription was actually filled and dispensed at a pharmacy. In drug safety in pregnancy research the timing of drug exposure is critical, as different organ systems develop and are susceptible to interference at different timings of gestation. When using data from the CPRD, as with all electronic healthcare databases, it may therefore not be considered appropriate to determine exposure status based only on prescriptions issued during the first trimester of pregnancy, or the specific time period of interest. This is because, for example, a prescription with a duration of 60 days that is issued 15 days before the start of pregnancy has the potential to result in the female being exposed during the first 45 days of her pregnancy, which is a time when central nervous system and cardiac development is already taking place. When determining first trimester drug exposure investigators may choose to identify all prescriptions issued in the four to six months before pregnancy and map exposure based on the assumed duration of the prescription and the assumption that one prescription cannot start until the previous one has been completed, unless there is clear evidence of product switching.[12,13] The aforementioned difficulties in determining the gestational timing of events can also result in exposure misclassification. In general, exposure misclassification may be less of a concern for products prescribed for long-term use and chronic conditions than for products used intermittently or for short term use only.[16] Although the CPRD has the advantage of exposure data being recorded prospectively and independently by a prescriber, thereby avoiding any potential recall bias, exposure misclassification may occur as the result of non-compliance, or a woman deciding to stop taking her medicine when she is trying to become pregnant or once she discovers she is pregnant. Using CPRD data it is not possible to identify such non-compliance or discontinuation of treatment midprescription. The majority of prescriptions issued in secondary care will not be recorded in the CPRD, however repeat prescriptions issued by the GP for treatment initiated in secondary care should be captured. The CPRD does not capture medicines bought over-thecounter without a prescription, including standard dose (0.4 mg) folic acid.
5. Identifying pregnancy outcomes of interest 5.1. Major congenital malformations The most commonly studied outcome in drug safety in pregnancy research is the risk of a major congenital malformation
© Société Française de Pharmacologie et de Thérapeutique
85
(MCM) in the offspring. In order to evaluate this risk, it is necessary to link exposure data in the mother’s medical record to outcome data in the child’s medical record. This is done primarily using information on family number, as individuals within the CPRD are each assigned a practice-specific family number which is based on postal address. For each live delivery attempts are made to link the offspring to the mother, based on a birth or registration record within the same family on or around the same date as a live delivery or a pregnancy end date.[13] Care needs to be taken when there are two or more females with the same family number who both have records relating to pregnancy and also in situations when there is more than one possible child but the date of birth for each child is different. In these cases it may not be possible to reliably link the mother’s records with those of the offspring. In previous studies it has been possible to link over 80% of all live pregnancy outcomes to a child.[13] The CPRD has the advantage that there is the potential for relatively long periods of follow-up making it possible to identify MCMs that are diagnosed later on in life and not immediately following birth. For pregnancies that end in a live delivery the presence of an MCM can be identified based on Read codes in the child’s medical record. Some Read codes contain a lot of detail (e.g. coarctation of aorta) whilst others are vague (e.g. other specified congenital malformations of the heart). The nature of recording in the CPRD means that in some circumstances the presence of a single code for a particular condition may not be considered confirmation of a diagnosis. There may be situations where the GP enters a Read code as a “working diagnosis” and following further investigation the diagnosis is ruled out, but the Read code is not updated in the child’s record.[6] It is therefore often necessary to verify each of the potential MCMs identified and this can be done in a number of ways. Firstly, for some infants there may be supporting evidence of the diagnosis in the form of additional Read codes; for example there may be a code for a cleft lip and later in the record a code relating to surgical repair of the cleft lip. In these cases investigators can be confident the diagnosis was a true diagnosis. For MCMs where this is not the case it may be possible to request additional non-coded information recorded by the child’s GP. When entering Read codes into the computer system GPs have the opportunity to record additional non-coded comments, frequently referred to as “free text”. This free text can be used to record additional symptoms or detail alongside the Read code; for example the Read code description might be “coarctation of aorta” and the free text might state “has undergone coarctation repair”. Anonymised information recorded by the GP in the free text fields can be requested for all individuals in the CPRD, regardless of whether they are still registered with a GP who contributes to the database. If the patient is still registered with the GP practice a questionnaire can be sent to the GP to obtain additional information to help confirm or refute the diagnosis. For patients still registered it is also possible to request an anonymised photocopy of the patient’s part or full medical record enabling access to all referral and outpatient letters as well as correspondence from specialists.[18] All these
Thérapie 2014 Janvier-Février; 69 (1)
86
Charlton et al.
services are available on a project specific basis and at a charge via the CPRD Group. In addition to identifying MCMs in live births, it is also important to identify MCMs in pregnancies that ended in a pregnancy termination. This is particularly important for ensuring the capture of more severe congenital malformations, such as anencephaly, for which in the United Kingdom prenatal diagnosis and associated pregnancy termination is possible. For some pregnancy terminations recorded in the CPRD there is evidence of the reason for the termination, for example Read codes relating to an “unplanned” or “unwanted” pregnancy, but often this information is not available or may only have been recorded as free text. To identify the majority of MCMs in pregnancies that end in an induced termination it is necessary to request the free text comments recorded in association with the termination of pregnancy code(s) as often the reason for the termination is not evident from the Read code alone.[13] For example the Read code might just state “termination of pregnancy” and the free text comment might add that it was “for spina bifida at 16 weeks’ gestation”. 5.2. Embryonic and foetal deaths While a large number of spontaneous pregnancy losses occur in the first 8 weeks of pregnancy (embryonic deaths), many of these are unlikely to be recorded in the CPRD. Where they are recorded, they are likely to represent a selective population of planned pregnancies, for instance those where a couple is having difficulty to conceive. Drug safety in pregnancy studies carried out in the CPRD should therefore ideally be restricted to spontaneous pregnancy losses occurring from the 9th week of gestation (foetal deaths). In the UK, foetal deaths can be categorised as those occurring between gestational weeks 9 and 24 (miscarriage)[19] and those occurring after 24 weeks’ gestation (stillbirths).[20] Read codes for both of these outcomes exist. However, again assumptions may need to be made regarding the gestational age at the time of delivery or miscarriage. In doing so, one needs to be aware not all GPs will use the same definition for miscarriage and stillbirth and the magnitude of error in estimating gestational age might differ between live births, stillbirths and miscarriages. If such differential error is present, this will lead to differential misclassification of exposure and therefore biased risk estimates. Therefore, while the CPRD can be used for studies of foetal death, the potential for misclassification should be carefully considered and additional free text, questionnaire and/or linked data should be obtained for validation where possible. 5.3. Obstetric complications/delivery characteristics Read codes exist for a range of obstetric complications such as preeclampsia, obstetric haemorrhage and placental abruption and
© Société Française de Pharmacologie et de Thérapeutique
for a range of details recorded at delivery, including birth weight, mode of delivery, Apgar score, small for gestational age and preterm birth. However, as these events are managed predominantly in secondary care, they are not always recorded in a woman’s record in general practice. To our knowledge, there are no studies verifying the validity or completeness of recording of these events in the CPRD; however some investigators have compared the rates observed in the CPRD to the rates in more complete sources and concluded there is under-recording in the CPRD.[21,22] This has implications for the utility of the CPRD to study such outcomes or to include them as variables in any statistical evaluation. Where this information is considered a key study variable, linkage of CPRD data to more complete sources of secondary care data may be necessary.
6. Information on risk factors and confounders When calculating risk estimates for adverse pregnancy outcomes following exposure to a particular medicine it is often necessary to evaluate the effect of potential risk factors and confounders. Within the CPRD it is possible to obtain information on a range of variables that are either risk factors for the outcome of interest or act as confounders within drug safety in pregnancy research. These include maternal age, body mass index, alcohol drinking status, smoking status, deprivation status, co-morbidities (e.g. diabetes) and co-prescribing. For some patients where this information is missing and not recorded during pregnancy it is possible to make assumptions, for example if someone was a non-smoker before they became pregnant it is considered unlikely that they will have taken up smoking during pregnancy. There are, however, a number of individuals who have no data recorded in their medical record relating to some or all of these variables and then their status would be classified as unknown. Information on parity and gravidity is not consistently recorded in the CPRD, however the number of previous deliveries or number of previous spontaneous pregnancy losses recorded in the CPRD might be used as a proxy for these variables.[15] Within the CPRD there is some information on family history of congenital malformations; however this is thought to be selectively recorded and therefore considered too incomplete for any useful analysis. Data relating to genetic makeup are sporadically available within the database.
7. Overview of drug safety in pregnancy research using the GPRD/CPRD Having provided an overview of the CPRD and highlighted some of the key areas for consideration when using it as a tool for evaluating the safety of medicine use during pregnancy the remainder of this paper summarises pregnancy related research that has been carried out using the GPRD/CPRD to date.
Thérapie 2014 Janvier-Février; 69 (1)
Drug Safety in Pregnancy Using the CPRD
7.1. First use of the GPRD for drug safety in pregnancy research The first studies to report on pregnancy outcome research using the GPRD were published in the late 1990s.[23,24] These evaluated exposure to anticonvulsants[23] and fluconazole[24] and the associated risk of having a child born with a congenital malformation. The publications lacked detail on the exact methods of determining first trimester exposure and identification of congenital malformations. In addition both studies only captured live born pregnancy outcomes and had small sample sizes resulting in lack of statistical power. These studies were then followed by one in 1999 investigating pregnancy outcomes after intra-uterine exposure to any of three acid suppression drugs.[25] The methods reported were more detailed than the earlier papers but given what is known about the importance of the timing of drug exposure for this kind of research, the study used a wide time window of interest to determine exposure status (30 days before through to 100 days after LMP). This study did, however, capture non-live pregnancy outcomes including terminations of pregnancy. 7.2. Methodological studies In 2004, the first study focussing on the methodological aspects of using the GPRD for drug safety in pregnancy research was published. This paper evaluated strategies for identifying pregnancies in the database with a focus on the complexity and the vast number of medical codes available for use by GPs.[26] This area of research was then also covered by Snowball and De Vries in 2007.[9] In 2010, Devine et al. published a paper continuing on the topic of identifying pregnancies in the GPRD, this paper not only reported on the creation of a computer based algorithm for identifying pregnancies but also on the results of validating this algorithm.[10] The algorithm was more successful for live births than for other types of pregnancy outcome. It was suggested this might be due to difficulties in determining pregnancy start and end dates when no child’s record exists and where often multiple records relating to a termination of pregnancy are recorded. In 2006, Hardy et al. went on to evaluate the types of medicines that women in the GPRD were prescribed during pregnancy and focussed on those pregnancies where it was possible to link the mother to her child.[27] This study captured prescriptions issued during two time periods; the 90 days before and the 70 days after the first recorded pregnancy code. Although more work was required to improve the precision of the timing of exposure, this study did highlight the large proportion of women within the GPRD who are exposed to medicines for which the safety in pregnancy was not fully known. 7.3. Congenital malformation verification In 2007, the first research was published that evaluated the accuracy and completeness of congenital malformation recording in the
© Société Française de Pharmacologie et de Thérapeutique
87
GPRD with a focus on cardiovascular defects.[28,29] The first study compared the prevalence of congenital heart defects in the GPRD with two UK population based sources (the National congenital anomaly system and the UK contributors to EUROCAT).[29] The overall prevalence of congenital heart defects in the GPRD was higher than that reported in both of the UK population based sources. The authors suggested this could be owing to differences in data collection methods and in particular the voluntary nature of reporting by healthcare professionals to the UK population based sources. The second study attempted to verify three specific heart defects identified in the electronic medical records, using additional information obtained from free text and GP questionnaires.[28] In this study, there was a 93.5% positive predictive value (PPV) between the computerised records and responses to the GP questionnaires, ranging from 90% for tetralogy of Fallot to 100% for coarctation of the aorta. Similar research was published in 2008, this time focussing on neural tube defects (NTDs).[30] Owing to the nature of this class of defect the study included diagnoses that resulted in a termination of pregnancy, in addition to those found in live and stillbirths. An algorithm was developed to identify new cases of NTDs in either the mother’s or the child’s record and the identified cases were then verified using GP questionnaires. Of the 169 questionnaires returned, 117 (71%) NTD cases were verified. The PPV of the algorithm varied by NTD diagnosis and ranged from 0.81 for anencephaly to 0.47 for spina bifida. The authors reported that the low PPV for spina bifida could have resulted from the inability of the algorithm to differentiate between cases where the mother herself had spina bifida and cases where it was present in the foetus or child. The annual prevalences of NTDs in the GPRD were comparable to those of the National congenital anomaly system in later years, but slightly higher in the early years. In 2011, a paper reporting on the sensitivity of congenital malformation recording in the GPRD and the added value of requesting photocopied medical records and free text was published.[18] This study attempted to verify a range of potentially major congenital malformations identified in a cohort of women with epilepsy in the GPRD. By using a combination of photocopied medical records and free text, 160/188 (85.1%) malformations could be verified. The percentage that could be verified was higher for those where photocopied records were requested (91.7%) than those where only free text was available because the child was no longer registered with the practice or the practice was no longer contributing data to the GPRD (77.9%). All studies that included verification of MCMs in the GPRD reported it was possible to identify the MCMs of interest using the electronic data but concluded that obtaining additional detail and information from free text comments, photocopied medical records or GP questionnaires was strongly recommended.[18,28-30] 7.4. Potential to act as a pregnancy registry system One paper reported on the mean number of annual exposures to a range of drugs during pregnancy captured in the GPRD and
Thérapie 2014 Janvier-Février; 69 (1)
88
Charlton et al.
compared this with the numbers enrolled in pregnancy exposure registries over the same time period.[31] For more prevalent conditions, such as depression, the GPRD had captured a similar mean annual number of exposures as a pregnancy registry. Charlton et al. then investigated whether the GPRD could be used to replicate the findings of the UK epilepsy and pregnancy register and whether it was possible to identify a known teratogenic association using GPRD data.[13] An increased MCM risk following first trimester polytherapy antiepileptic drug exposure compared with no antiepileptic drug exposure was found; however lower rates of MCMs were identified in the GPRD following exposure to valproate than in the UK epilepsy and pregnancy register. It was possible however to identify in the GPRD the known teratogenic association between first trimester exposure to valproate and an increased risk of spina bifida.
7.5. Drug utilisation and risk assessment studies In recent years the GPRD/CPRD has been used to evaluate the utilisation of asthma medicines by pregnant women[12] as well as the uptake of pandemic influenza vaccination during pregnancy.[14] Data from the CPRD has also been used to carry out risk assessment studies to evaluate the risk of congenital anomalies following exposure during pregnancy to topical corticosteroids,[22] first trimester exposure to antihypertensive drugs,[32] antidepressants[33] and asthma medicines.[34] Studies have also been conducted to evaluate the risk of cardiac malformations following exposure to selective serotonin reuptake inhibitors,[35] the risk of pregnancy losses in women with diabetes[36] and the hazard of foetal death following H1N1 influenza vaccination.[15]
8. Conclusion The CPRD has a number of key strengths for drug safety in pregnancy research when compared with some other data sources. These include the representative nature of the population captured, the availability of the data, the ability to identify many types of pregnancy outcome, the potential for large sample sizes and the presence of a denominator population. The capability of the CPRD is likely to be restricted, however, by small sample sizes when evaluating products used to treat less prevalent conditions, the small study population available for evaluating rare birth defects, the inaccuracies associated with determining the pregnancy duration, establishing exposure based on the issue of a prescription and a lack of information on some potential confounding factors such as over-the-counter folic acid exposure. The on-going expansion of the CPRD, in terms of the number of patients captured and the ability to link to secondary and tertiary care datasets, will increase sample sizes and have the potential to improve the precision and accuracy of drug safety in pregnancy studies using CPRD data. The CPRD therefore has the
© Société Française de Pharmacologie et de Thérapeutique
potential to complement the work of other surveillance systems including pregnancy exposure registries and case-control systems.
Acknowledgements Some of the work presented in this paper formed part of the following PhD thesis: RA Charlton. (2012) The General Practice Research Database as an alternative to registries for studying drug safety in pregnancy: anticonvulsants as a case study. University of Bath, Bath, UK. Conflicts of interests. The authors have no conflicts of interests to declare. The authors did not receive any funding specific to the writing of this manuscript. Abbreviations. CPRD : Clinical Practice Research Datalink; EFEMERIS: Évaluation chez la Femme Enceinte des MÉdicaments et de leurs RISques (evaluation about drugs and their risks on pregnant women); GP: general practitioner; GPRD: General Practice Research Database; ISAC: Independent Scientific Advisory Committee; LMP: last menstrual period; MCM: major congenital malformation; MREC: Multi-Centre Research Ethics Committee; NTD: neural tube defects; PPV: positive predictive value; UTS: up-to-standard.
References 1.
Daw JR, Hanley GE, Greyson DL, et al. Prescription drug use during pregnancy in developed countries: a systematic review. Pharmacoepidemiol Drug Saf 2011; 20(9): 895-902
2.
Cleary BJ, Butt H, Strawbridge JD, et al. Medication use in early pregnancyprevalence and determinants of use in a prospective cohort of women. Pharmacoepidemiol Drug Saf 2010; 19(4): 408-17
3.
Lacroix I, Damase-Michel C, Lapeyre-Mestre M, et al. Prescription of drugs during pregnancy in France. Lancet 2000; 356(9243): 1735-6
4.
McBride W. Thalidomide and congenital abnormalities. Lancet 1961 (Dec 16); 2: 1358
5.
The Clinical Practice Research Datalink. Medicines and healthcare products regulatory agency; 2013 http://www.cprd.com Accessed November 18th, 2013
6.
Charlton RA. The General Practice Research Database as an alternative to registries for studying drug safety in pregnancy: anticonvulsants as a case study. University of Bath, Bath, UK. 2012: PhD Thesis
7.
.Campbell J, Dedman J, Easton S, et al. Is the GPRD GOLD population comparable to the UK population? Pharmacoepidemiol Drug Saf 2013 (22); S1: 1-521 http://onlinelibrary.wiley.com/doi/10.1002/pds.3512/pdf Accessed November 18th, 2013 [Abstract 567]
8.
Independent Scientific Advisory Committe for MHRA database research. http://www.cprd.com/ISAC/ Accessed November 18th, 2013
Thérapie 2014 Janvier-Février; 69 (1)
Drug Safety in Pregnancy Using the CPRD
9.
Snowball JM, de Vries CS. Determination of pregnancy on the General Practice Research Database. Pharmacoepidemiol Drug Saf 2007; 16: S1–274 DOI: 10.1002/pds [Abstract 249]
10. Devine S, West S, Andrews E, et al. The identification of pregnancies within the General Practice Research Database. Pharmacoepidemiol Drug Saf 2010; 19(1): 45-50 11. Lacroix I, Hurault C, Sarramon MF, et al. Prescription of drugs during pregnancy: a study using EFEMERIS, the new French database. Eur J Clin Pharmacol 2009; 65(8): 839-46 12. Charlton RA, Hutchison A, Davis KJ, et al. Asthma management in pregnancy. PLoS ONE 2013; 8(4): e60247
89
24. Jick SS. Pregnancy outcomes after maternal exposure to fluconazole. Pharmaotherapy 1999; 19(2): 221-2 25. Ruigomez A, Garcia Rodriguez LA, Cattaruzzi C, et al. Use of cimetidine, omeprazole and ranitidine in pregnant women and pregnancy outcomes. Am J Epidemiol 1999; 150: 476-81 26. Hardy JR, Holford TR, Hall GC, et al. Strategies for identifying pregnancies in the automated medical records of the General Practice Research Database. Pharmacoepidemiol Drug Saf 2004; 13: 749-59 27. Hardy JR, Leaderer BP, Holford TR, et al. Safety of medications prescribed before and during early pregnancy in a cohort of 81 975 mothers from the UK General Practice Research Database. Pharmacoepidemiol Drug Saf 2006; 15: 555-64
13. Charlton RA, Weil JG, Cunnington M, et al. Comparing the General Practice Research Database and the UK epilepsy and pregnancy register as tools for postmarketing teratogen surveillance: anticonvulsants and the risk of major congenital malformations. Drug Saf 2011; 34(2): 157-71
28. Wurst KE, Ephross SA, Loehr J, et al. The utility of the General Practice Research Database to examine selected congenital heart defects: a validation study. Pharmacoepidemiol Drug Saf 2007; 16: 867-77
14. Sammon CJ, McGrogan A, Snowball JM, et al. Pandemic influenza vaccination during pregnancy: an investigation of vaccine uptake during the 2009/ 10 pandemic vaccination campaign in Great Britain. Hum Vaccin Immunother 2013; 9(4): 917-23
29. Wurst KE, Ephross SA, Loehr J, et al. Evaluation of the General Practice Research Database congenital heart defects prevalence: Comparison to United Kingdom national systems. Birth Defects Res A Clin Mol Teratol 2007; 79: 309-16
15. Sammon CJ, Snowball J, McGrogan A, et al. Evaluating the hazard of foetal death following H1N1 influenza vaccination: a population based cohort study in the UK GPRD. PLoS ONE 2012; 7(12): e51734
30. Devine S, West SL, Andrews E, et al. Validation of neural tube defects in the full featured - General Practice Research Database. Pharmacoepidemiol Drug Saf 2008; 17: 434-44
16. Toh S, Mitchell AA, Werler MM, et al. Sensitivity and specificity of computerized algorithms to classify gestational geriods in the absence of information on date of conception. Am J Epidemiol 2008; 167(6): 633-40
31. Charlton RA, Cunnington MC, de Vries CS, et al. Data resources for investigating drug exposure during pregnancy and associated outcomes: the General Practice Research Database (GPRD) as an alternative to pregnancy registries. Drug Saf 2008; 31(1): 39-51
17. Walley T, Mantgani A. The UK General Practice Research Database. Lancet 1997; 350(9084): 1097-9 18. Charlton RA, Weil JG, Cunnington M, et al. Identifying major congenital malformations on the General Practice Research Database: a study reporting on the sensitivity and added value of photocopied records and free text. Drug Saf 2010; 33(9): 741-50 19. Royal College of obstetricians and gynaecologists. The management of early pregnancy loss. 2006 http://www.rcog.org.uk/files/rcog-corp/uploaded-files/ GT25ManagementofEarlyPregnancyLoss2006.pdf Accessed November 18th, 2013 (18 pages) 20. Royal College of obstetricians and gynecologists. Registration of stillbirths and certification for pregnancy loss before 24 weeks of gestation. 2005 http://www.rcog.org.uk/files/rcog-corp/uploaded-files/ GoodPractice4RegistrationStillbirth2005.pdf Accessed November 18th, 2013 (4 pages) 21. Howard LM, Goss CLAU, Leese MORV, et al. Medical outcome of pregnancy in women with psychotic disorders and their infants in the first year after birth. Br J Psychiatry 2003; 182(1): 63-7 22. Chi CC, Mayon-White RT, Wojnarowska FT. Safety of topical corticosteroids in pregnancy: a population-based cohort study. J Invest Dermatol 2011; 131(4): 884-91 23. Jick S, Terris BZ. Anticonvulsants and congenital malformations. Pharmacotherapy 1997; 17(3): 561-4
© Société Française de Pharmacologie et de Thérapeutique
32. Vasilakis-Scaramozza C, Aschengrau A, Cabral HJ, et al. Antihypertensive drugs and the risk of congenital anomalies. Pharmacotherapy 2013; 33(5): 476-82 33. Vasilakis-Scaramozza C, Aschengrau A, Cabral H, et al. Antidepressant use during early pregnancy and the risk of congenital anomalies. Pharmacotherapy 2013; 33(7): 693-700 34. Vasilakis-Scaramozza C, Aschengrau A, Cabral HJ, et al. Asthma drugs and the risk of congenital anomalies. Pharmacotherapy 2013; 33(4): 363-8 35. Margulis AV, Abou-Ali A, Strazzeri MM, et al. Use of selective serotonin reuptake inhibitors in pregnancy and cardiac malformations: a propensityscore matched cohort in CPRD. Pharmacoepidemiol Drug Saf 2013; 22(9): 942-51 36. McGrogan A, Snowball JM, de Vries CS. Pregnancy losses in women with type 1 or type 2 diabetes in the UK: an investigation using primary care records: pregnancy losses in women with diabetes. Diabetic Medicine 2013; Sep 30: doi: 10.1111/dme.12332
Correspondence and offprints: Rachel Charlton, Department of Pharmacy and Pharmacology, University of Bath, Claverton Down, Bath, BA2 7AY, United Kingdom. E-mail: [email protected]
Thérapie 2014 Janvier-Février; 69 (1)