Cohort

The cohort was derived from the Medicaid Analytic eXtract (MAX), which contains information on Medicaid beneficiaries; Medicaid is the joint state and federal health insurance program for low-income individuals in the United States.

MAX is a health care utilization database that records Medicaid enrollment and utilization claims, including those for inpatient admissions and outpatient visits as well as outpatient pharmacy dispensing claims. Using the MAX data from 2000 to 2007, a cohort was created for the study of drug utilization and safety in pregnancy, as previously described in detail.

The use of this database for research was approved by the Partners’ Institutional Review Board (Boston, MA). Women with a claim indicating delivery were linked to infants within states using the Medicaid case number (which is generally shared by families) and the infant’s date of birth. The woman’s last menstrual period (LMP) was assigned using a validated algorithm based on diagnostic codes in the maternal and infant records.

We restricted the cohort to women who were eligible for Medicaid continuously from 3 months prior to the estimated LMP month through 1 month postpartum. To ensure complete ascertainment of relevant claims, we restricted our analysis to women with 28 days or more of enrollment each month and without restricted benefits, private insurance, or certain state-specific managed care programs (that underreport claims to MAX).

To allow for an accurate capture of congenital malformations, we also required that the linked infants met the same Medicaid eligibility criteria as the mothers for at least 3 months following the birth (unless they died, in which case a shorter eligibility period was allowed). The source cohort included 891,699 completed pregnancies with linked infants.

We excluded women who were exposed to known teratogenic medications including lithium, antineoplastic agents, retinoids, or thalidomide from the estimated LMP through the date of delivery based on claims for dispensed medications or who had an infant with an inpatient or outpatient diagnosis code indicating the presence of a chromosomal abnormality.

We also excluded women who were exposed to antihypertensive medications during the first trimester but who lacked diagnosis codes indicating chronic hypertension because of a significant risk of misclassifying the presence or absence of hypertension in these patients (because the women may have received the medications for other indications, for example, beta blockers for migraine prophylaxis, or have hypertension that was not properly coded). The final analytic cohort included 878,126 pregnancies ( Figure 1 ).

Patient flowchart

Bateman. Chronic hypertension and malformations. Am J Obstet Gynecol 2015 .

Exposure

The following 3 groups of women were considered in the analysis: (1) women without chronic hypertension who did not receive antihypertensive medications in the first trimester, (2) women with chronic hypertension who were treated with an antihypertensive medication during the first trimester, and (3) women with chronic hypertension who were not treated with an antihypertensive medication during the first trimester.

Chronic hypertension was defined by codes recorded on 2 or more distinct dates indicating chronic or preexisting hypertension recorded in the maternal inpatient or outpatient record at any time from 3 months prior to the LMP through delivery. The codes were derived from the International Classification of Disease-Clinical Modification , ninth edition, and included codes 642.0x, 642.1x, 642.2x, 642.7x, and 401.xx through 405.xx. These codes are specific to chronic hypertension and are distinct from codes that indicate other hypertensive disorders of pregnancy including gestational hypertension and preeclampsia.

Antihypertensive exposure during the first trimester was defined by a filled prescription whose days’ supply overlapped the period from the LMP to 90 days after the LMP. In defining this exposure, we used prescriptions filled from 3 months prior to the LMP until 90 days after the LMP. Duration of exposure was estimated based on the number of days’ supply. We accumulated days’ supply for consecutive prescriptions of the same medication if the medication was refilled prior to the day that the prior prescription was expected to run out. The list of antihypertensive medications considered in the analysis can be found in the Appendix ( Supplementary Table 1 ).

Outcomes

The primary outcome was defined as the presence of a major congenital malformation in the offspring. Major congenital malformations were defined based on having codes on 2 or more separate days in the infant inpatient or outpatient records during the first 3 months of life indicating an organ-specific class of malformations including central nervous system malformations; eye, ear, neck, and face malformations; cardiac malformations; respiratory malformations; cleft palate or lip; gastrointestinal malformations; genitourinary malformations; musculoskeletal malformations; or other malformations.

We required 2 codes to define the presence of malformations to exclude cases in which a single mention may be recorded to justify a diagnostic test to rule out a condition. Our group has previously validated specific cardiac malformations using this approach with medical records; the positive predicted value was greater than 75%. Secondary outcomes included each of the organ-specific malformations.

Covariates

We identified 5 groups of potential confounders in the cohort: maternal demographic characteristics, comorbid medical conditions, obstetric characteristics/conditions, maternal medications, and measures of health care utilization. Demographic characteristics assessed included age at delivery divided into 6 categories (≤19, 20-24, 25-29, 30-34, 35-39, ≥40 years); race/ethnicity (grouped as white [non-Hispanic], black [non-Hispanic], Hispanic, Asian/Pacific Islander, other, or missing); region of delivery (Northeast, South, West, or Midwest); and year of delivery.

Comorbid medical and obstetric conditions were identified by the presence of 1 or more diagnosis codes in the maternal inpatient or outpatient record from 3 months prior to the estimated LMP through delivery. These included preexisting diabetes mellitus, chronic renal disease, obesity, tobacco use, alcohol abuse, and illicit drug use/abuse. Obstetric conditions included multiple gestations and gestational diabetes. Multiparity was defined based on the woman’s Medicaid eligibility type; an eligibility classification of adult with dependent children indicated multiparity.

Comorbidities and conditions were assessed until delivery because, although the relevant etiological window for exposures leading to the development of congenital malformations is the first trimester, conditions present during the first trimester (chronic hypertension, diabetes, obesity, etc) may not always be coded with fidelity during this relatively brief window. Additionally, conditions that develop later in pregnancy (eg, gestational diabetes) may be markers for risk factors that are present in the first trimester (maternal obesity or hyperglycemia).

The validity of this approach rests on the assumption that the development of a congenital malformation in the infant does not lead to the development or preferential recording of these maternal conditions. Although this is likely a safe assumption for the conditions assessed, we also conducted a sensitivity analysis to ensure that our findings were robust when the window during which exposure and covariates were ascertained was confined to the prepregnancy period and first trimester (see the following text).

Maternal medication exposures assessed included exposure to either insulin or oral hypoglycemic medications from 3 months prior to the LMP through delivery (which may be markers for the presence or severity of maternal preexisting or gestational diabetes). We also identified and considered as a covariate in our analyses exposure to any potentially teratogenic medications during the first trimester including fluconazole, aminoglycosides, folic acid antagonists, methimazole, potassium iodide, tetracycline, danazol, misoprostol, statins, coumadin, and propylthiouracil, which may act as confounders if women with chronic hypertension are exposed to these medications with a frequency that is different from normotensive women.

Finally, we considered measures of health care utilization during the 3 months prior to the LMP through the end of the first trimester, which may be markers for overall health status and thus have a relationship to the risk for birth defects. These included the number of distinct non–antihypertensive prescription and number of physician visits for any reason (both grouped as 0, 1-3, >3).

Statistical analyses

We determined the baseline characteristics of women without chronic hypertension who did not receive antihypertensive medications in the first trimester (controls), women with chronic hypertension who were exposed to an antihypertensive medication during the first trimester (treated chronic hypertensives), and women with chronic hypertension who were not treated with an antihypertensive during the first trimester (untreated chronic hypertensives) and summarized them as counts and proportions.

We conducted 2 separate comparisons in our analyses: (1) controls vs subjects treated for chronic hypertension and (2) controls vs untreated chronic hypertensives. In each analysis, we first determined the frequency of and unadjusted odds ratio (OR) and 95% confidence intervals (CIs) for the primary and secondary outcomes.

To account for the differences in the baseline characteristics in the groups that are being compared, we performed propensity score analyses. In the first analysis (comparing controls and subjects treated for hypertension), we used a logistic regression model to estimate the probability of being a subject treated for hypertension (as opposed to a control) based on the maternal demographic characteristics, comorbid medical conditions, obstetric characteristics/conditions, maternal medication exposures, and measures of health care utilization, as defined in previous text, without further selection (this probability is the propensity score).

Controls and treated subjects treated for chronic hypertension were then matched on their propensity score in a fixed 3:1 ratio using a nearest neighbor algorithm with a maximum matching distance of 0.05. The same approach was used for the comparison of controls with subjects untreated for chronic hypertension. In these matched cohorts, we again estimated the odds ratio and 95% CI for the primary and secondary outcomes.

Sensitivity and exploratory analyses

Because diabetes is such a strong risk factor for congenital malformations and the possibility exists that there may be residual confounding, even adjusting for the presence of diabetes (ie, diabetes may be more severe in the hypertensive patients), we repeated our analysis excluding patients with any codes or medications indicating the presence of diabetes.

Specifically, we excluded women who had any diagnosis codes indicating preexisting or gestational diabetes or who had prescriptions for insulin or oral diabetes medications from 3 months prior to the LMP through delivery. The total number of the excluded patients was 88,937.

We then repeated the propensity score–matched analysis, focusing on the outcomes of overall congenital malformations and cardiac malformations (which were significantly associated with both treated and untreated hypertension in the primary analysis).

We performed a second sensitivity analysis excluding patients with preterm delivery. Preterm delivery has a complex relationship to congenital malformations because of the following: (1) infants with congenital malformations are more likely to deliver preterm, and (2) infants born preterm will sometimes have conditions that might be coded as malformations (eg, patent ductus arterious, patent foramen ovale, undescended testes) that would have spontaneously resolved had the infant been carried to term.

In the first scenario, controlling for prematurity could induce an association between hypertension or antihypertensives and malformations. In the second scenario, controlling for prematurity would be justified if we were interested in the direct effect of hypertension or antihypertensives not mediated through prematurity, assuming our analyses properly adjusted for shared risk factors of malformations and prematurity.

In the primary analysis, we assumed the first scenario and did not adjust for preterm delivery. In this sensitivity analysis, we assumed scenario two and condition on prematurity through restriction. The total number of pregnancies excluded in this analysis was 98,049. We again repeated the main analysis for the outcomes of overall congenital malformations and cardiac malformations.

In the primary analysis, we collected information on covariates from prior to the LMP through delivery to improve our ascertainment of chronic hypertension and conditions that might confound the association of chronic hypertension and malformations. However, if a woman is diagnosed as carrying an infant with a malformation on prenatal ultrasound, it is possible that her medical conditions would be scrutinized more carefully and/or recorded more accurately in the medical record (surveillance bias). To ensure that this potential bias did not affect our results, we repeated our analysis defining chronic hypertension and the covariates used in the analysis based solely on codes recorded in the maternal record from 3 months prior to the LMP through the end of the first trimester.

We required codes in the infant record recorded on separate days to define the presence of an organ-specific malformation in the primary analysis. This was done to minimize the risk of identifying malformations based on codes used to justify rule-out diagnostic tests. However, to test that our results were robust to this approach, we repeated the primary analysis with the outcome defined based on 1 or more codes in infant inpatient or outpatient record from day of birth to day 90 of life.

To determine the potential effects of residual confounding in our analyses, we defined the strength (confounder-outcome relative risk) of a hypothetical residual confounder that, if present, would explain the effect of treated and untreated chronic hypertension on the risk of developing a congenital malformation. Because the strength of confounders depends on the relative prevalence, we assumed a prevalence of 5% among the normotensive controls and then a range of prevalences in the treated and untreated chronic hypertension patients (which were analyzed separately).

As an exploratory analysis, we repeated the primary analysis, examining the association of treated and untreated hypertension and specific cardiac malformations including ventricular septal defect, right ventricular outflow obstruction, single ventricle, secundum atrial septal defect, conotruncal defect, and left ventricular outflow obstruction.

Analyses were performed in SAS version 9.3 (SAS Institute, Cary, NC).