A population based, retrospective, cohort study was conducted among the Kaiser Permanente Northern California member population of women who delivered a live birth from 1995 to 2008. The study was approved by the Kaiser Permanente Northern California Institutional Review Board. Kaiser Permanente Northern California has well established automated clinical and pharmacy databases that capture delivery, diagnosis of malformation, dispenses of prescription drugs, and presence of hypertension during pregnancy. In addition, through linkage to various databases including California birth certificate data and the Kaiser Permanente Northern California Alpha-Fetoprotein Prenatal Screening Program, we were able to obtain additional variables (such as maternal pre-existing diabetes, pregnancy weight, age, ethnicity, education) for adjustment as potential confounders. To examine the risk of malformation in offspring, we included all live births in the study population.
Ascertainment of maternal exposure to ACE inhibitors and other antihypertensives during pregnancy
The Pharmacy Information Management System is a computerised pharmacy prescription and dispensation database in Kaiser Permanente Northern California. This captures all prescription drugs dispensed, with information on date dispensed, dose, and days of supply. Use of ACE inhibitors and other antihypertensives during pregnancy was ascertained through linkage of the information from the database to women who delivered live births from 1 January 1995 to 31 December 2008 identified from the Kaiser Permanente Northern California clinical databases.
Last menstrual period was determined based on gestational age recorded in the Kaiser Permanente Northern California clinical databases in combination with the calendar date at delivery (last menstrual period=date at delivery−gestational age). Through linkage to information on dates of drug dispensation and days of supply from the pharmacy database, we were able to identify women who were exposed to ACE inhibitors and other antihypertensives during pregnancy. We classified the timing of exposure as: (a) use during the first trimester only, (b) use during the second or third trimester only, and (c) use at any time during pregnancy. Two cohorts of women who were exposed to antihypertensive drugs during pregnancy were established: a cohort exposed to ACE inhibitors and a cohort exposed to antihypertensives other than ACE inhibitors. Any pregnant women who used ACE inhibitors in isolation or in combination with other drugs were classified into the cohort exposed to ACE inhibitors. We excluded a few users of angiotensin receptor blockers in this cohort since they have some pharmacological similarities with ACE inhibitors. Users of all other antihypertensive drugs (see appendix on bmj.com for the list) were classified into the cohort of “other antihypertensive drug users.”
One of the most difficult methodological challenges in pharmacoepidemiological studies is to control for the underlying indication for the drugs being studied. Therefore, establishing a comparable control group is crucial. Since the underlying condition for ACE inhibitors is hypertension, we identified pregnant women who had a diagnosis of hypertension at any time from one year before their last menstrual period to the end of their pregnancy but who were not prescribed any antihypertensive drugs (based on our pharmacy data). This group served as a control for the underlying indication of antihypertensive drugs (that is, hypertension controls). The remaining pregnant women who had neither a diagnosis of hypertension nor use of any antihypertensive drug formed the normal control group.
Ascertainment of malformations
All live born infants in the identified eligible mother-infant pairs were linked to clinical databases (both inpatient and outpatient data) including electronic medical records to identify diagnoses of major malformations (ICD-9 CM (international classification of diseases, ninth revision) codes 740.0–759.9). In order to allow comparisons, we included malformations of similar types to those included in the study by Cooper et al that reported the association between ACE inhibitor use in the first trimester and malformation risk in offspring.5
All infants were followed to the end of the study period, 31 December 2008, to maximise the use of existing clinical information for a more complete ascertainment of malformations. However, most malformations (for example, 89% of congenital heart defects identified) were diagnosed in the infants before 1 year of age.
While the Kaiser Permanente Northern California automated databases have been shown to be comprehensive and accurate,8 9
we conducted a validation study of the diagnoses of malformations by comparing the diagnoses between medical charts and our clinical databases. We randomly selected 237 maternal-infant pairs for a validation study. The diagnoses of malformations were compared between the two data sources. While medical records may not necessarily be error-free, the diagnoses of malformations from our clinical databases and medical records were quite consistent. Compared with medical records, the sensitivity and specificity for diagnoses of malformations recorded in the Kaiser Permanente Northern California clinical databases were 100% and 94% respectively for any malformations, and 98% and 99% respectively for diagnoses of congenital heart defects.
Ascertainment of potential confounders and effect modifiers
In addition to the underlying indications for use of antihypertensive drugs, pre-existing diabetes before the index pregnancy and being overweight during pregnancy were probably the two most important potential confounders or effect modifiers. Women in the study cohorts with a pre-existing diagnosis of diabetes were identified through linkage to the Kaiser Permanente Northern California diabetes registry. Kaiser Permanente Northern California has a well established diabetes registry with over 200
000 members that ascertains, verifies, and follows up all Kaiser Permanente Northern California members with a diagnosis of diabetes (type 1, type 2, or previous gestational diabetes). The registry has been estimated to be 98% sensitive for diagnosed diabetes (compared with self reports derived from a survey in 2000).10
This registry allowed us to identify accurately women with a diagnosis of pre-existing diabetes and provided an opportunity to examine the potential interaction between use of ACE inhibitors during pregnancy and pre-existing diabetes.
In the current study population, about 70% of pregnant women had information on body weight during the second trimester (around 16–20 weeks of gestation) when they took part in prenatal screening. Information on height was not available for most subjects, however, so body mass index could not be calculated. We used the 90th centile of body weight in the second trimester (207 pounds (94 kg)) in our study population as the cut-off point for overweight.
From the Kaiser Permanente Northern California databases, we also had information on maternal age, ethnicity, and parity for the pregnant women included. Through linkage to the California State birth certificate data, we were able to obtain information on some additional potential confounders including maternal education levels and smoking during pregnancy. However, because the California State birth certificate data were available only up to 2006, we had missing data for those two potential confounders for the births in 2007–8.
We used logistic regression to control for confounders (maternal age, ethnicity, parity, pre-existing diabetes, and overweight) while examining the effect of maternal exposure to ACE inhibitors and other antihypertensive drugs during pregnancy on the risk of malformations in offspring. Because the rate of malformation is relatively low, odds ratios and associated 95% confidence intervals were used as approximation for relative risk to measure the strength of the association and stability of the estimates.
In addition to establishing a “hypertension only” cohort to control for the underlying indication for antihypertensive drugs, we extensively examined two important potential confounders, pre-existing diabetes and overweight. Both were adjusted for in the analyses and stratified for examination of potential modifying effect. Other potential confounders including maternal age, ethnicity, and parity were also controlled for throughout the analyses. Finally, for some variables that were available only from the California State birth certificate data (thus, available only for births up to 2006), we conducted separate analyses adjusting for those few variables among a subset of the study population including births up to 2006.