Patterns of medication use during pregnancy in rheumatoid arthritis

Authors


Abstract

Objective

To characterize therapies prescribed during pregnancy to women with rheumatoid arthritis (RA).

Methods

We conducted a cohort study of women with RA with pregnancies using health care utilization data from 2002–2008. We examined the distribution of RA drugs by therapeutic classes, including nonsteroidal antiinflammatory drugs (NSAIDs)/coxibs, glucocorticoids, nonbiologic disease-modifying antirheumatic drugs (DMARDs), and biologic DMARDs, during 90-day pregnancy trimesters and the 180 days prior to pregnancy. Drugs were characterized according to the Food and Drug Administration risk classification system. Differences in exposure by period were determined by chi-square tests.

Results

A total of 393 pregnancies were identified among 34,169 women with RA. Seventy-two percent of pregnancies ended in a delivery. Approximately 24% of women with RA received a DMARD during preconception. At any point during pregnancy, 23% of women with deliveries were dispensed ≥1 DMARD and the proportion of use declined from the first to the third trimester (P = 0.03). Similar to DMARD therapy, use of NSAIDs/coxibs and exposure to category D/X medications were significantly lower compared to prepregnancy use (P < 0.05). In contrast, more women were prescribed glucocorticoids during pregnancy than before pregnancy. Use of biologics occurred in 12.5% of pregnancies. Compared to women with deliveries, women who experienced abortions were more frequently exposed to NSAIDs/coxibs (P < 0.05). Dispensing of category D/X medications was also higher in women with spontaneous abortions and primarily involved methotrexate (P < 0.05).

Conclusion

Approximately 24% of women with RA received a DMARD in the 180 days before conception, and the proportion dropped during pregnancy. Glucocorticoid use remained high throughout pregnancy. Our results suggest that continued efforts directed at counseling women and their physicians about the potential risks/benefits of RA therapies during pregnancy are warranted.

INTRODUCTION

Rheumatoid arthritis (RA), the most common inflammatory arthritis, disproportionately affects women of reproductive age. During pregnancy, it is estimated that between 40% and 80% of patients will experience amelioration of synovitis beginning as early as the first trimester (T1) (1, 2). In contrast, several reports demonstrate that RA onset is highest in the postpartum period, and worsening or “flare” of existing disease occurs in approximately 90% of women 4–6 weeks after childbirth (3, 4).

Although the majority of women will have a transient reduction in arthritis symptoms during pregnancy, some will require treatment with disease-modifying antirheumatic drugs (DMARDs), glucocorticoids, and nonsteroidal antiinflammatory drugs (NSAIDs) (5). However, several commonly used DMARDs are contraindicated during pregnancy (6). There are limited data on the safety of RA therapies during pregnancy because of the paucity of studies. Furthermore, how women are actually managed in terms of if and when they continue or discontinue DMARDs and other medications during pregnancy is poorly understood.

The available information on actual DMARD use during pregnancy is largely derived from pharmaceutical registries (7–9). These studies primarily focus on one therapeutic DMARD class (e.g., anti–tumor necrosis factor [anti-TNF] agents) and inconsistently report the timing of drug exposure in relation to conception and pregnancy course. The voluntary and self-reported use of DMARDs and potential for recall bias in retrospective designs may also limit the validity and generalizability of these studies. Summary information like the US Food and Drug Administration (FDA) pregnancy risk categories is often relied upon to guide therapeutic choices (10). However, the lack of specific recommendations about the timing of when DMARD treatment needs to be discontinued during reproduction adds to the complex decision making that patients and physicians must face. To date, few studies have addressed how actual prescribing habits compare to existing guidelines, or how utilization changes over time as new agents and safety data become available.

One method to answering these questions involves the analysis of health care utilization data. These data provide a rich opportunity to study the “real-world” patterns of prescription drug use among the pregnant population (11). Furthermore, record linkage analyses provide a resource for following women over time and allow for simultaneous investigation of total medication use before, during, and after pregnancy. The aim of this study was to address the gap in knowledge of how pregnant women with RA are treated in daily clinical practice. We examined the longitudinal patterns of DMARD prescribing from 2002 to 2008 using health care utilization data.

PATIENTS AND METHODS

Data source and population.

We conducted a retrospective cohort study using health care utilization data in the HealthCore Integrated Research Database from January 1, 2002, to June 30, 2008. This longitudinal claims database maintains computerized information on member enrollment, outpatient prescription drug dispensing (including quantity dispensed and number of days supplied), inpatient and outpatient diagnoses, and procedures on more than 28 million subscribers to 14 Blue Cross/Blue Shield health plans across the US. We identified a cohort of subjects ages ≥18 years who had at least two visits with a diagnosis for RA (International Classification of Diseases, Ninth Revision [ICD-9] code 714.x) and a minimum of 12 months of health plan eligibility prior to the start of followup (RA cohort). Subjects were censored at death, loss of eligibility for ≥90 days, or the end of the study period, whichever came first.

Two pregnancy subcohorts were assembled from the RA cohort. Our primary cohort of interest included women with pregnancies that ended in delivery as identified by ICD-9 (72–75.9, 630–639.9, 640–648.9, 650–669.99, 670–676.9, V22–V22.2, V23–V23.9, V24.0–V24.2, V27–V27.9, V30–V39.2) or Current Procedural Terminology (CPT) codes (01960–01969, 58605, 58610–58611, 59160, 59300, 59305, 59400, 59409–59414, 59425–59426, 59430, 59510, 59514–59515, 59525, 59612, 59614, 59620, 59622, 59812, 59820–59821, 59830, 59840–59841, 59850–59857, 59870, 59899, 99436). Our secondary cohort consisted of pregnant women who had an elective or spontaneous abortion (e.g., complete or incomplete miscarriage prompting a medical visit) identified by ICD-9 (75.0, 630, 632, 634–639.9) or CPT codes (01960–01969, 59812, 59820–59821, 59830, 59840–59841, 59850–59857, 59870). All women were required to be ages 18–45 years on their estimated date of conception (EDC), and were covered by the health plan for at least 180 days prior to the EDC and throughout the duration of the pregnancy. If more than one pregnancy occurred during followup for a given woman, a minimum of 180 days was required to separate estimated conception dates. All pregnancies meeting the eligibility criteria were considered for analysis.

Determining EDC and pregnancy periods.

The EDC was calculated for deliveries as the date of delivery minus 270 days. From the EDC, we defined T1, the second trimester (T2), and the third trimester (T3) as sequential 90-day periods. Two periods prior to pregnancy were also considered. The 90 days immediately prior to the EDC were classified as Pre-1, and the 91–180 days prior to the EDC were classified as Pre-2. Therefore, five 90-day periods were evaluated overall. Since the risk of misclassification for the EDC and subsequent trimester(s) is greater for therapeutic and spontaneous abortions, we examined the 90-day period prior to the date of pregnancy termination for general patterns of medication use instead of assuming explicit trimesters.

Assessment of exposure.

We examined the use of common RA treatments by therapeutic class, including selective and nonselective NSAIDs/coxibs, glucocorticoids (excluding inhalational and topical formulations that may be indicated for other conditions), nonbiologic DMARDs, and biologic DMARDs. All drugs were additionally categorized by the US FDA risk classification system (Table 1). We defined exposure dichotomously (yes/no) as the first pharmacy dispensing or medical claim (e.g., for infusible medications) of a prescription drug from each therapeutic class during each of the five 90-day periods. We also created categories of combination therapy (prescriptions from ≥2 classes during the same period) that are typical in clinical practice. A woman was considered exposed to a drug if it was administered or filled at a pharmacy and prescribed to be taken on at least one day during the period of interest, or if treatment was filled before the 90-day period but the duration of use was expected to extend into that period.

Table 1. Rheumatoid arthritis treatments by therapeutic class and Food and Drug Administration pregnancy risk category*
NSAIDs/coxibs, category CGlucocorticoids, category CNonbiologic DMARDs, category B or CNonbiologic DMARDs, category D or XBiologic DMARDs, category B or C
  • *

    NSAIDs = nonsteroidal antiinflammatory drugs; category C = risk cannot be ruled out; DMARDs = disease-modifying antirheumatic drugs; category B = no evidence of risk in humans; category D = positive evidence of risk; category X = contraindicated in pregnancy.

DiclofenacBetamethasoneChloroquineAzathioprineAbatacept
DiflunisalCortisoneCyclosporineLeflunomideAdalimumab
EtodolacDexamethasoneGoldMethotrexateAnakinra
FenoprofenHydrocortisoneHydroxychloroquine Etanercept
FlurbiprofenMethylprednisoloneSulfasalazine Infliximab
IndomethacinPrednisolone  Rituximab
KetoprofenPrednisone   
KetorolacTriamcinolone   
Mefenamic acid    
Meloxicam    
Nabumetone    
Naproxen    
Oxaprozin    
Piroxicam    
Sulindac    
Tolmetin    
Celecoxib    
Rofecoxib    
Valdecoxib    

Statistical analysis.

We assessed baseline covariates, including age, health care utilization variables (number of physician visits and acute hospitalizations, number of prescription drugs, including DMARDs), and maternal comorbidity (Charlson Comorbidity Index), during the 180 days preceding the EDC for deliveries or during the 180 days prior to the date of abortion.

Our primary analysis consisted of evaluating the number and percentage of deliveries exposed to: 1) each therapeutic drug class, 2) individual medications within each class, and 3) combination therapies during each of the 90-day periods and during any of the 3 pregnancy trimesters (overall pregnancy period). The differences in frequency of prescriptions by therapeutic class between Pre-1 and Pre-2 were determined by chi-square tests, and differences between pregnancy trimesters (T1, T2, and T3) were determined using chi-square tests for linear trend. Additionally, the proportion of women with deliveries exposed to each therapeutic drug class during the overall pregnancy period was compared to each prepregnancy period (Pre-1 and Pre-2) via chi-square testing. All P values less than 0.05 were considered statistically significant.

The frequency of prescriptions by therapeutic class, individual drugs, and combinations of drugs was also estimated for women with abortions and comprised our secondary analysis. Drug exposure was assessed during the entire 90 days prior to the abortion date. The statistical significance of differences in medication use among women with abortions was compared to medication use during T1 for women with deliveries by chi-square tests.

We also calculated the annual frequency of drug exposure by therapeutic class for women with deliveries. Women were stratified by their year of EDC and drug exposure during the overall pregnancy period was examined to determine if temporal changes in the pattern of medication use could be detected from 2002 up to 2008.

RESULTS

During the 7-year study period, 393 pregnancies were identified among 34,169 women of reproductive age with RA. Seventy-two percent of women (n = 281) had a pregnancy ending in delivery, 9% (n = 37) had an elective abortion, and 19% (n = 75) had a spontaneous abortion. Of the women with deliveries, 21 women had more than one pregnancy during the study period. Table 2 displays the baseline characteristics of pregnant women. The mean age for women with a delivery was 32.8 years, for elective abortions was 33.4 years, and for women with spontaneous abortions was 33.9 years. The number of physician visits, acute hospitalizations, and comorbidities was low. Among women with deliveries, 61.9% had no DMARDs dispensed in the 180 days prior to the EDC, whereas 4.3% were exposed to 3 or more different DMARDs in the 180 days prior to estimated conception. A similar pattern of baseline DMARD use was observed among women with abortions.

Table 2. Baseline characteristics of women with rheumatoid arthritis during the 180 days prior to the estimated date of conception for those who delivered and during the 180 days prior to date of abortion*
 Deliveries (n = 281)Elective abortions (n = 37)Spontaneous abortions (n = 75)
  • *

    DMARDs = disease-modifying antirheumatic drugs.

  • The Charlson Comorbidity Index, adapted for use with International Classification of Diseases, Ninth Revision–coded health care utilization data, predicts 1-year mortality for a patient with any of 22 comorbid conditions. Each condition is assigned a score of 1, 2, 3, or 6, depending on the risk of death associated with the condition, and the scores are summed to provide a total score to predict mortality (12).

Age, mean ± SD years32.8 ± 4.833.4 ± 5.333.9 ± 5.4
No. of physician visits, mean ± SD4.9 ± 4.92.30 ± 2.12.84 ± 2.9
Acute care hospitalizations, no. (%)22 (7.8)1 (2.7)
No. of medications dispensed, mean ± SD4.8 ± 4.53.8 ± 3.23.3 ± 3.6
Different DMARDs dispensed, no. (%)   
 0174 (61.9)22 (59.5)50 (66.7)
 166 (23.4)9 (24.3)18 (24.0)
 229 (10.3)3 (8.1)5 (6.7)
 ≥312 (4.3)3 (8.1)2 (2.7)
Charlson score, no. (%)   
 094 (33.5)19 (51.4)34 (45.3)
 1164 (58.4)18 (48.6)41 (54.7)
 217 (6.1)
 ≥36 (2.1)

In the preconception periods (Pre-1 and Pre-2), more than one-third of women with a delivery were exposed to at least one DMARD (Table 3). Biologic and nonbiologic DMARDs were used in relatively the same frequency. Etanercept was the most commonly dispensed biologic DMARD. Use of other anti-TNF agents or rituximab was minimal, and no women were exposed to anakinra or abatacept during this time. The filling of prescription drugs that should be discontinued when attempting conception, i.e., methotrexate (MTX) and leflunomide (LEF), did not reach zero, but use was low in the 90 days immediately prior to pregnancy (range 1.4–6.8%). Glucocorticoids (both oral and parenteral formulations) were prescribed to >30% of women, and NSAIDs/coxibs (range 13.2–19.2%) were used less frequently. While exposure to all therapeutic classes tended to decrease from Pre-2 (91–180 days prior to EDC) to Pre-1 (1–90 days prior to EDC), none of the reductions were statistically significant (P > 0.05).

Table 3. Therapeutic class and drug-specific exposure among women with deliveries, according to gestational time period*
 Days 91–180 prior to EDCDays 1–90 prior to EDCFirst trimesterSecond trimesterThird trimesterOverall pregnancy period
  • *

    Values are the number (percentage). Category totals may be smaller than the sum of individual medication due to use of combination therapy. EDC = estimated date of conception; NSAIDs = nonsteroidal antiinflammatory drugs; DMARD = disease-modifying antirheumatic drug; category B = no evidence of risk in humans; category C = risk cannot be ruled out; category D = positive evidence of risk; category X = contraindicated in pregnancy.

  • P < 0.05 by chi-square test comparing overall pregnancy to days 91–180 prior to EDC.

  • P < 0.05 by chi-square test comparing overall pregnancy to days 91–180 prior to EDC. P < 0.05 by chi-square test comparing overall pregnancy to days 1–90 prior to EDC.

  • §

    P < 0.05 for trend by chi-square test.

NSAIDs/coxibs54 (19.2)37 (13.2)20 (7.1)9 (3.2)16 (5.7)31 (11.0)
Glucocorticoids88 (31.3)92 (32.7)93 (33.1)74 (26.3)95 (33.8)156 (55.5)
 Oral44 (15.7)44 (15.7)37 (13.2)40 (14.2)37 (13.2)59 (21.0)
 Nonoral57 (20.3)61 (21.7)64 (22.8)46 (16.4)66 (23.5)121 (43.1)
Any DMARD (≥1)97 (34.5)90 (32.0)53 (18.9)40 (14.2)31 (11.0)§64 (22.8)
Combination of ≥2 DMARDs32 (11.4)24 (8.5)11 (3.9)7 (2.5)8 (2.9)16 (5.7)
Nonbiologic DMARD (category B or C)47 (16.7)40 (14.2)26 (9.3)14 (4.9)13 (4.6)§32 (11.4)
 Chloroquine1 (0.4)
 Cyclosporine1 (0.4)1 (0.4)
 Gold1 (0.4)1 (0.4)1 (0.4)
 Hydroxychloroquine34 (12.1)30 (10.7)19 (6.8)12 (4.3)12 (4.3)25 (8.9)
 Sulfasalazine17 (6.1)15 (5.3)11 (3.9)2 (0.7)1 (0.4)11 (3.9)
Nonbiologic DMARD (category D or X)36 (12.8)23 (8.2)7 (2.5)7 (2.5)7 (2.5)11 (3.9)
 Azathioprine1 (0.4)1 (0.4)1 (0.4)1 (0.4)1 (0.4)1 (0.4)
 Leflunomide5 (1.8)4 (1.4)2 (0.7)3 (1.1)3 (1.1)
 Methotrexate32 (11.4)19 (6.8)6 (2.1)5 (1.8)3 (1.1)8 (2.9)
Biologic DMARD (category B or C)42 (14.9)47 (16.7)26 (9.3)25 (8.9)19 (6.8)35 (12.5)
 Adalimumab9 (3.2)10 (3.6)5 (1.8)6 (2.1)3 (1.1)7 (2.5)
 Anakinra1 (0.4)1 (0.4)
 Etanercept25 (8.9)28 (9.9)19 (6.8)13 (4.6)13 (4.6)23 (8.2)
 Infliximab9 (3.2)9 (3.2)3 (1.1)5 (1.8)4 (1.4)6 (2.1)
 Rituximab1 (0.4)1 (0.4)

During pregnancy, glucocorticoids were the most commonly dispensed therapeutic class (Table 3). Women were more frequently exposed to glucocorticoids during pregnancy than during the prepregnancy period (P < 0.001). Furthermore, the use of glucocorticoids remained stable across T1, T2, and T3 (P > 0.05). In contrast, use of NSAIDs/coxibs, any DMARD, or the combined use of ≥2 DMARDs was lower than before pregnancy (P < 0.05). Sixty-four women (22.8%) with a delivery were dispensed at least one DMARD at any point during pregnancy, but the proportion of use significantly declined as pregnancy advanced from T1 to T3 (P = 0.03). Hydroxychloroquine (8.9%), etanercept (8.2%), and MTX (2.9%) were the most commonly used DMARDs during pregnancy. Nonbiologic DMARDs from category B or C declined in use from T1 to T3 (P = 0.04), but there were no discernable trends among other therapeutic classes. Overall, women were rarely exposed to category D or X medications, and this was significantly lower compared to use during Pre-1 or Pre-2 (P < 0.01).

Between 16.0% and 45.9% of pregnancies ending in abortion were exposed to NSAIDs/coxibs or glucocorticoids (Table 4). Use of at least one DMARD in the 90 days preceding an abortion was not significantly different (range 24.0–32.4%) than during the gestational period for women with deliveries (22.8%; P > 0.05). In contrast, women who experienced an elective or spontaneous abortion were more frequently exposed to NSAIDs/coxibs during the 90 days before the abortion than women with deliveries during the first trimester (P < 0.05). Dispensing of category D/X medications (6.7%) was approximately 3-fold higher in women with spontaneous abortions compared to deliveries (2.5%; P = 0.02) and primarily involved exposure to MTX.

Table 4. Therapeutic class and drug-specific exposure in the 90 days prior to date of abortion*
 Elective abortions (n = 37)Spontaneous abortions (n = 75)
  • *

    Values are the number (percentage). Category totals may be smaller than the sum of individual medications due to use of combination therapy. NSAIDs = nonsteroidal antiinflammatory drugs; DMARD = disease-modifying antirheumatic drug; category B = no evidence of risk in humans; category C = risk cannot be ruled out; category D = positive evidence of risk; category X = contraindicated in pregnancy.

  • P < 0.05 by chi-square test comparing exposure in days 1–90 prior to abortion to days 1–90 after estimated date of conception (first trimester) for deliveries.

NSAIDs/coxibs8 (21.6)12 (16.0)
Glucocorticoids17 (45.9)21 (28.0)
 Oral4 (10.8)9 (12.0)
 Nonoral15 (40.5)13 (17.3)
Any DMARD (>1)12 (32.4)18 (24.0)
Combination of ≥2 DMARDs4 (10.8)6 (8.0)
Nonbiologic DMARDs (category B or C)6 (16.2)11 (14.7)
 Hydroxychloroquine6 (16.2)6 (8.0)
 Sulfasalazine1 (2.7)5 (6.7)
Nonbiologic DMARDs (category D or X)2 (5.4)6 (8.0)
 Azathioprine1 (2.7)
 Leflunomide1 (2.7)1 (1.3)
 Methotrexate5 (6.7)
Biologic DMARDs (category B or C)7 (18.9)8 (10.7)
 Adalimumab3 (4.0)
 Anakinra1 (2.7)
 Etanercept2 (5.4)4 (5.3)
 Infliximab5 (13.5)2 (2.7)

The temporal exposure by therapeutic drug class from 2002 to 2007 is shown in Figure 1. During pregnancy, glucocorticoids were clearly the most common exposure. Other than a slight increase in 2003, the use of glucocorticoids at any point during pregnancy ranged from 51.9% to 63.6% for women with deliveries. Coincident with this increase, a decrease in dispensing of nonbiologic DMARDs (category B or C) was observed. The pattern of exposure to other classes was qualitatively consistent over time.

Figure 1.

Annual frequency of drug exposure. The frequency of drug exposure to each therapeutic class was evaluated by year of estimated date of conception among women with deliveries (n = 281). NSAID = nonsteroidal antiinflammatory drug; DMARD = disease-modifying antirheumatic drug.

DISCUSSION

Our results show that among a population of women with RA, more than one-third received DMARD therapy in the preconception period, but there was a reduction in DMARD and NSAID/coxib exposure during pregnancy. In contrast, the use of glucocorticoids at any point during pregnancy was high and contributed the largest number of RA therapies dispensed during the gestational period. Despite temporal changes in the availability of RA treatments, the patterns of medication use remained relatively constant over time.

There is a paucity of data on the prevalence of DMARD, NSAID/coxib, and glucocorticoid use during pregnancy among women with RA. As a result, there are few studies to which we can directly compare our findings. One recent study evaluated antirheumatic drug use in women before and during pregnancy by linking a prescription database to a national birth registry (13). The authors showed that NSAIDs and prednisolone were the most commonly dispensed drugs before, during, and after pregnancy. Similar to our findings, overall use at any point during pregnancy was particularly high for prednisolone (32%) and NSAIDs (16%), but exposure tended to decline with successive pregnancy trimesters. Only 5 of 1,411 women filled prescriptions for MTX or LEF in the 3 months before pregnancy, and 4 of 1,411 women were exposed to MTX or LEF during pregnancy (13). This percentage, lower than our reported pregnancy exposure to MTX (2.9%) and LEF (1.1%), may be the result of increased vigilance in certain populations about avoiding category D/X medications when planning or becoming pregnant. In another study, the frequency of antirheumatic drug use during pregnancy closely resembled the patterns of use identified in our cohort. De Man and colleagues found that approximately one-third of pregnant women with RA were exposed to glucocorticoids (14). Their reported use of prednisone was 35% among pregnant women and remained constant across pregnancy trimesters. In the same study, no women were exposed to MTX, LEF, or biologics, but sulfasalazine (SSZ) use occurred in approximately 33% of women during each of the trimesters (14). Another study found that MTX was discontinued in all women, but SSZ was maintained during pregnancy by 38% of women (15). This number may reflect the popularity of SSZ as a first-line agent for RA in Europe compared to our findings, indicating a very low proportion of SSZ use during pregnancy among a US-based cohort. It may also in part be due to the perceived safety of SSZ (labeled FDA category B), while MTX, designated category X, is a known teratogen and is typically avoided or discontinued in the preconception period.

Surveys of physician practice have examined DMARD use during pregnancy. Chakravarty et al found 65 DMARD-exposed pregnancies: 39 to MTX, 10 to LEF, 15 to etanercept, and 2 to infliximab (16). Three congenital malformations, all in the MTX group, were reported among the 52 pregnancies with known outcomes. A larger survey estimated that 454 women became pregnant while receiving biologic therapy and that 31.3% continued TNF inhibitor therapy throughout pregnancy without any reported fetal or maternal adverse effects (17). In our study, 12.5% of women with deliveries and 17.9% of women with abortions were prescribed a biologic DMARD at some point during pregnancy. This is higher than the findings by Viktil et al, who found a minority of women (∼1%) with at least one prescription of a biologic DMARD during pregnancy compared to ∼3% who were dispensed a biologic in the 3 months prior to conception (13). The discontinuation of biologics during pregnancy likely reflects the uncertainty that patients and providers face, since the safety of these agents during pregnancy remains largely unknown.

We used health claims data to define a cohort of women with RA who became pregnant during their course of followup. The strength of this approach is that we were able to ascertain total RA drug dispensing before and during pregnancy, without recall bias, and evaluate the spectrum of therapeutic classes instead of focusing on a single drug category. Another strength was our ability to evaluate longitudinal trends in DMARD prescriptions. This is relevant because the availability and prescribing of newer DMARDs, specifically biologic agents, surged after 2002. Our results indicate that biologic use occurs in a minority of women in the 180 days prior to pregnancy and remains low during pregnancy.

The limitations of our study should be noted. Since we had no information on the actual gestational length, the EDC was estimated as 270 days before the date of delivery. Our method for dating available drug days strictly depended on the EDC. The use of this standard gestational length is common in studies using administrative data but poses the risk of exposure misclassification if the assumptions used to construct the pregnancy period are invalid (15, 18). This may have led us to overestimate medication use in the early stages of pregnancy among those with a gestational length of <270 days (e.g., preterm or multiple births). In this circumstance, DMARD, glucocorticoid, and NSAID/coxib exposure that actually occurred in the preconception period would be falsely attributed to the first trimester. Despite these drawbacks, a study comparing administrative data with data from a birth registry showed that gestational age assumptions only result in a small proportion of exposure misclassification (19). However, the extent of misclassification was larger for FDA category X drugs in the first trimester. If we overestimated the already low prevalence of category X drug use in our study, then even fewer women may have been exposed to harmful drugs at the beginning of pregnancy than our findings suggest.

The possibility of misclassification is particularly problematic for abortions. To our knowledge, no validation studies about the sensitivity, specificity, and positive predictive value of diagnostic or procedural codes for elective and spontaneous abortions have been reported. Due to this uncertainty of gestational length and pregnancy timing, we chose to present total medication use in the 90 days prior to the date of abortion. Additionally, there is the risk of ascertainment bias, as women with abortions requiring medical care covered by the health plan may have been different than the women who did not seek medical care or who sought it privately. Despite this, our estimate of abortions is in keeping with estimates in the literature of ∼25 per 1,000 women with RA (20). It is impossible to determine if the higher exposure to NSAIDs/coxibs and category D/X medications among women with abortions was the cause of abortion, or if the increased use represented greater disease activity, which may have independently impacted pregnancy outcome. Because health care utilization data do not contain measures of disease severity and presents this problem of “confounding by indication,” we acknowledge that conclusions about the true patterns or causal relationship between DMARD exposure and risk of abortion cannot be drawn.

Misclassification may have also arisen independent of the manner in which we defined exposure to RA treatments. Our methods identified which prescription drugs were dispensed but not necessarily which were taken by women with RA. In particular, women often change their health behavior as a consequence of becoming pregnant (21). In this setting, we may have overestimated medication use for women who filled a prescription during the very early stages of pregnancy that was later stopped upon confirmation of pregnancy due to safety concerns. Furthermore, as pregnancy advances, some women with RA enter a state of low disease activity and may have discontinued DMARDs because of improvement in arthritis symptoms. Alternatively, medication use would be underestimated if a prescription was procured and taken intermittently throughout pregnancy. Use of over-the-counter analgesics was not captured by our methods and can contribute to an underestimation of total RA therapy. Although these are all plausible scenarios, previous work has shown that compliance to prescription drugs for chronic illness is high during pregnancy (22). Furthermore, prescription drug use for chronic conditions varies less from preconception to pregnancy, while occasional or over-the-counter drug use tends to decline as pregnancy ensues (23).

In conclusion, this health care utilization study shows that women with RA are exposed to DMARDs and NSAIDs/coxibs during the immediate preconception period, but that use during pregnancy is low and declines during the gestational period. In contrast, exposure to glucocorticoids is substantial before and throughout pregnancy. While pregnancy outcomes or adverse events were not reported in our study, prednisone dosages of >20 mg/day during pregnancy are associated with an increased risk of fetal cleft palate formation and maternal complications such as pregnancy-induced hypertension, gestational diabetes mellitus, and infection (24). Our results suggest that continued efforts directed at counseling women of childbearing age about the potential risks of common DMARD, NSAID/coxib, and glucocorticoid use are needed. The use of administrative data is an attractive method of assessing the understudied population of pregnant women, but methods to validate gestational length are needed to make accurate assessments of drug exposure and subsequent pregnancy outcomes. Future studies that link pregnant women to their offspring will allow examination of pregnancy outcomes and the safety of these medicines during pregnancy.

AUTHOR CONTRIBUTIONS

All authors were involved in drafting the article or revising it critically for important intellectual content, and all authors approved the final version to be published. Dr. Kuriya had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Study conception and design. Kuriya, Hernández-Díaz, Liu, Bermas, Daniel, Solomon.

Acquisition of data. Kuriya, Hernández-Díaz, Liu, Bermas, Daniel, Solomon.

Analysis and interpretation of data. Kuriya, Hernández-Díaz, Liu, Bermas, Daniel, Solomon.

Ancillary