Trajectories of ADHD medication use before, during, and after pregnancy: A population‐based study from Norway and Sweden

To describe ADHD medication use trajectories around pregnancy in Norway and Sweden.


Key Points
• The use of ADHD medication before, during and after pregnancy follows multiple distinct patterns: continuation, discontinuation, interruption, and initiation.
• Most women discontinue ADHD medication use in pregnancy and continuation is rare but has increased over time.
• ADHD medication continuers tend to be older, less likely to live with a partner, more likely to have had multiple previous births, more likely to smoke, and also have higher use of different classes of psychotropics compared to women who discontinue, interrupt, or initiate medication use after pregnancy.

Plain Language Summary
We aimed to study ADHD medication use from one year before to one year after pregnancy in Norway and Sweden. We captured pregnancy and medication data from national registers of both countries. We identified ADHD medication use in different parts of the study period and applied a machine-learning technique called group-based trajectory modeling to cluster women based on patterns of ADHD medication use. Of 13 286 women who received ADHD medication in the study period, we identified four distinct trajectory groups: those who had a high likelihood to continue medication use throughout the study period (continuers, 5.7%), those who had a high likelihood to interrupt medication use during the study period (interrupters, 23.8%), those who had a high likelihood of discontinuing medication use during the study period (discontinuers, 49.5%), and those who had a high likelihood of starting medication use later in the study period (late initiators, 21.0%). On further describing the characteristics of each of these four groups, we observed that continuers were older, less likely to live with a partner, more likely to have had multiple previous births, more likely to smoke during pregnancy, and use other drugs for psychiatric conditions, compared to the remaining groups.

| BACKGROUND
More women are reaching reproductive age while being treated for ADHD or are diagnosed and treated for the first time as adults. [1][2][3][4][5] There is a paucity of data addressing ADHD prognosis in the perinatal and postnatal period, yet the combined burden of having an ADHD diagnosis and being pregnant can be especially challenging. Women using ADHD medication during pregnancy seem to have poorer socioeconomic status 2,6 in addition to having risky health behaviors such as smoking. 6,7 While changes in perinatal hormonal levels might affect ADHD symptom manifestation, 8 stimulant medication used in the treatment of ADHD may be associated with certain adverse pregnancy outcomes. [9][10][11][12] It is likely that the nature of medication use among this patient population is dynamic. Adequately managing ADHD with pharmacological treatment before, during, and after pregnancy requires attention by prescribers, with respect to finding a balance between risks and benefits. 1 Additionally, ADHD medication use can be impacted by maternal characteristics, pregnancy history, and use of other psychotropic medication. Therefore, novel clustering methods such as group-based trajectory modeling (GBTM) can be used to capture complex medication use patterns. 13-16 Elucidating these patterns can guide future research on ADHD medication safety and efficacy for women in their reproductive years.
The aim of this study was to describe ADHD medication use trajectories from one year before to one year after pregnancy in Norway and Sweden and to describe the characteristics of women associated with each trajectory pattern.

| MATERIALS AND METHODS
We conducted a drug utilization study on ADHD medication use among pregnant women giving birth in Norway and Sweden. We used nationwide linked data from the birth and prescribed drug registers of both countries-Medical Birth Registry of Norway (MBRN), Norwegian Prescription Database (NorPD), Swedish Medical Birth Register (SMBR), and Swedish Prescribed Drug Register (SPDR). The registers from Norway and Sweden provide population-based, prospectively collected individual-level data. Upon birth or immigration, residents are assigned a unique personal identity number, which is used to link data across registers. 17  We included women who filled one or more prescriptions for the following ADHD medications (classified as per the Anatomical Therapeutical Classification [ATC] code 19 ) from 12 months before pregnancy start defined as the first day of the last menstrual period (LMP), to 12 months after delivery: amphetamine (N06BA01), dexamphetamine (N06BA02), methylphenidate (N06BA04), atomoxetine (N06BA09), lisdexamfetamine (N06BA12, available from 2013), and guanfacine (C02AC02, available from 2013 and 2016 in Sweden and Norway, respectively). 20,21 LMP was calculated as the date of birth minus the gestational age in days.
We classified exposure to ADHD medication in seven periods: 6-12 months before pregnancy (LMP-360 days to LMP-181 days), 0-6 months before pregnancy (LMP-180 days to LMP-1 day), first trimester (LMP to LMP + 97 days), second trimester (LMP + 98 days to LMP + 202 days), third trimester (LMP + 203 days to birth), 6 months after pregnancy (birth + 1 day to birth + 180 days), 6-12 months after pregnancy (birth + 181 days to birth + 360 days). We described ADHD medication exposure during the periods as use versus no use based on one or more prescriptions filled within each period, and total amount of drug dispensed in defined daily doses (DDDs) 22 in that period. We have not considered information on the types of product formulations. Summary of prescriptions in each of the periods are presented in Table S1.
We described the characteristics of pregnant women for each of the medication use patterns. Characteristics include maternal age, year of birth, maternal education, maternal country of birth, parity, relationship status, smoking status at the beginning of pregnancy, and use of other psychotropic medications. Other psychotropics included opioid analgesics (N02A), antiepileptics (N03A), antipsychotics (N05A), anxiolytics, hypnotics, and sedatives (N05B and N05C), and antidepressants (N06A). We classified other psychotropics' exposure in four-time periods: 12 months before pregnancy, first trimester, second and third trimesters, and 12 months after pregnancy.
In both Norway and Sweden, register-based research is exempt from informed consent, with approval from ethical review boards.

| Identification of GBTM groups
GBTM is an unsupervised clustering method, which identifies groups of individuals with similar trajectories, based on maximum likelihood estimation. The number of groups is not estimated, but models with different numbers of groups can be compared using fit indexes such as Bayesian information criterion (BIC), as well as the overall plausibility of the results. GBTM estimates the probability of group membership and assigns each pregnancy to the group for which they have the highest posterior probability of belonging to. 23 We used the endpoints of the seven-time periods in the year before, during, and after pregnancy as described previously to build the trajectory models ( Figure S2). We started by estimating 3-group models and tested first-order specification for the trajectory shapes. With multiple iterations, we added more groups (3)(4)(5) with increasing order (linear, quadratic, cubic, quartic specifications) for trajectory shapes. We added extra groups until the model with the best fit was established. We used multiple indicators to select the best model, starting with BIC. The model with the highest or least negative value of BIC (closest to zero) was preferred. It is to be noted that the BIC in trajectory modeling is reversed relative of other BIC measures. As additional criteria, we calculated the average posterior probability (APP) of group membership, odds of correct classification, entropy, and rejected models which gave rise to groups with few observations, defined as <5% of the study cohort.
An APP >70% and entropy value >0.80 are generally indicative of adequate model classification. [24][25][26] Also, clinical relevance of the resulting trajectory groups was taken into consideration to select the best model, by acknowledging parsimony, comprehensibility, and usefulness of the observed groups in practice. Once the best model was identified, we compared maternal characteristics across trajectory groups to each other and with the overall cohort characteristics.
All analyses were performed using Stata SE v17.0 for Windows.
The country-specific datasets were pooled and analyzed together as a single cohort. We expect medication use patterns in both countries to be similar, even if the proportion who follow each pattern is different.
GBTM was performed using a Stata package "traj." 27 ADHD medication exposure (use vs. no use) was modeled using logistic distribution.
ADHD medication exposure intensity (estimated daily dose) was modeled using a normal distribution. To estimate the daily dose, we divided the total DDDs in each period (Table S2) by the number of days in that period. We then log-transformed total DDDs to remove skewness in distribution and limit the influence of outliers. We used GBTM to model the exposure intensity for the group that had the most consistent use across the study period.

| RESULTS
There were 829 793 births in Norway and 1 287 783 births in Sweden during the study period. After applying the exclusion criteria, we included 4740 (0.6%) of 813 107 pregnancies from Norway and 8546 (0.7%) of 1 269 146 pregnancies from Sweden, which were exposed to ADHD medication before, during or after pregnancy ( Figure S1).
In the total cohort of 13 286 pregnancies, we identified a model with four distinct trajectory groups as the best-fit model (Tables S3 and S4). Pregnancies were classified based on the highest probability of belonging to each group: 5.7% followed a linear trajectory of continuous ADHD medication use (denoted as "continuers"), 23.8% followed a cubic trajectory categorized by an increase, then decrease, followed by an increase in the probability of ADHD medication use (interrupters), 49.5% followed a cubic trajectory of an increase followed by a decreasing probability of ADHD medication use (discontinuers), and 21.0% followed a linear trajectory of increase in probability of ADHD medication use after pregnancy (late initiators) (Figure 1).
A high proportion of continuers (80%-90%) used ADHD medication in each of the seven time intervals. More interrupters used ADHD medication before pregnancy and in the first trimester than discontinuers, and all interrupters had re-initiated use by one year after pregnancy. Among discontinuers, 70%-80% used ADHD medication before pregnancy, which declined to no use by the end of pregnancy and remained low after pregnancy. With no previous use in the year before or during pregnancy, over 90% used medication in the year after pregnancy.
Maternal characteristics of the study population and each trajectory group are presented in Table 1 Continuers were most likely and discontinuers least likely to use other psychotropics before, during, and after pregnancy (Figure 2). Antidepressant use followed a similar pattern among all subgroups of ADHD medication users, with higher use (31%-43%) during the year before pregnancy, reduction in use during pregnancy (11%-29%), and a rise in use (19%-39%) after pregnancy. Use of antidepressants after pregnancy was comparable with use before pregnancy among continuers (39% vs. 43%), interrupters (34% vs. 37%), and initiators (36%) of ADHD medication. However, discontinuers of ADHD medication showed a relatively lower reuptake of antidepressants after pregnancy (19% vs. 31%). Similarly, discontinuers of ADHD medication showed a low reuptake of anxiolytics, hypnotics, and sedatives (14% vs. 36%) after pregnancy compared to before pregnancy. Discontinuers had the lowest use of antidepressants, antiepileptics, and opioid analgesics across all intervals surrounding pregnancy. Initiators tended to have lowest use of antipsychotics (9%), and anxiolytics, hypnotics, and sedatives (33%) in the year before pregnancy. Use of opioid analgesics among all groups of ADHD medication users declined at the beginning of pregnancy (4%-12%) but increased sharply from second trimester (7%-17%) to year after pregnancy (13%-26%) almost matching levels before pregnancy (16%-28%).
On further modeling continuers to describe the trajectories of average daily dose, we arrived at a three-group model: low-dose (20.1%), medium-dose (49.4%), and high-dose (30.5%) users. This analysis suggested that most had reduced ADHD medication dose during pregnancy. The average DDDs of ADHD medication use per day among high-dose users continued to be high and more stable throughout the study period compared to medium-dose and low-dose users. Medication use among low-dose users declined slightly after delivery; however, a reuptake was seen in the interval 6-12 months after delivery (Figure 3). Model selection criteria are described in Tables S5 and S6.
F I G U R E 1 Trajectories of ADHD medication use before, during and after pregnancy in Norway and Sweden (N = 13 286). [Correction added on 28 June 2023, after first online publication: label for 5.7% pregnancies following a linear trajectory of continuous ADHD medication has been corrected from "Initiators" to "Continuers".] T A B L E 1 Maternal characteristics according to ADHD medication use trajectory based on drug exposure from 12 months before to 12 months after pregnancy in 2006-2019 (n, %).

| DISCUSSION
To our knowledge, our study is the first to have used trajectory modeling to explore longitudinal patterns of ADHD medication use around pregnancy. The best model we identified with GBTM had four trajectory groups: continuation, interruption, discontinuation, and initiation of medication use. Of the trajectory groups identified, ADHD medication continuers were characterized by older age, having multiple children and less cohabitation with a partner, more smoking, and higher use of other psychotropics. These characteristics suggest that continuers have additional comorbidities warranting use of other psychotropics and may have more severe ADHD. Some of these characteristics could be in part driven by older age. Life circumstances such as being a single mother may also contribute to treatment decisions F I G U R E 2 Other psychotropic use before, during, and after pregnancy, according to trajectory of ADHD medication use classification. We classified other psychotropics' exposure in four-time periods: 12 months before pregnancy, first trimester, second and third trimesters, and 12 months after pregnancy.
F I G U R E 3 Trajectories of ADHD medication dose among continuers (N = 751). y-axis represents antilog of log-transformed DDDs per day.
since continuers were most likely to have multiple previous births and least likely to live with a partner. To further understand how continuers used medication during pregnancy, we explored dosetrajectories. High-dose users were characterized by stable and high dose of medication use throughout pregnancy. However, the majority were in the medium-dose and low-dose groups which showed a reduction in dose during pregnancy. This reduction may have led to use of subtherapeutic levels of medication during pregnancy, which could in turn outweigh the risks over benefits.
There is limited research on patterns of ADHD medication use in pregnancy. A Canadian study exploring prevalence and trends of ADHD medication use before and during pregnancy (N = 1130), found higher medication use before pregnancy compared to during pregnancy, and a trimester-specific reduction in medication use from 0.26 to 0.14 per 100 pregnancies from the first to later trimesters. 28 A clinical study from the U.S. characterized the course of ADHD during pregnancy among women aged 18-45 years. They were observed to have discontinued (n = 5), adjusted (n = 8), or maintained their medication through pregnancy (n = 12). Significant differences were found between medication discontinuers versus adjusters, and discontinuers versus maintainers for mood and family functioning. 29 Another study using the U.S. MarketScan ® commercial claims data (2013-2018) described patterns and predictors of perinatal prescription stimulant use. Of the 15 413 pregnancies which were exposed to stimulant use prior to or during pregnancy, three cohorts emerged: discontinuation prior to conception (42%), discontinuation during first trimester (39%), and continuation into second and third trimesters (19%). 30 Similar to our study findings, they observed that those who continued medication into pregnancy were older in age and also had more severe ADHD.
In our study, most pregnant women discontinued or interrupted their ADHD medication. The will to minimize fetal exposure to medication is likely to have motivated discontinuation. Furthermore, use during pregnancy could be influenced by the relative ease of access to specialist care. While it is possible to contact the specialist care without a primary care referral in Sweden, it is not the case in Norway. 31,32 Also, breastfeeding could influence medication use after delivery. However, specific data on breastfeeding is unavailable in the registers. We have included the year after pregnancy in our study period, which is the typical period of breastfeeding in Norway and Sweden. 33,34 ADHD medication safety in pregnancy has been explored before, 9,35,36 with some safety signals arising. Although the smallest proportion of women continued ADHD medication use in pregnancy during the study period overall, we observed that the proportion of continuers has increased in recent years. Given the hazards and challenges of untreated ADHD, more women might be choosing to continue medication use. 1,29 Nonpharmacological strategies such as cognitive behavioral therapy, and attention training techniques could be recommended to manage ADHD. However, given the lack of evidence supporting these as standalone approaches, pharmacotherapy remains the mainstay strategy. 1,37 The use of nationwide registers increases the generalizability of the study findings, ensures completeness of data, and avoids poor recall. Nevertheless, we have a number of limitations. We could only study pregnancies which result in births. Therefore, these patterns may not represent the use for pregnancies ending in miscarriages or induced abortions. There is previous research suggesting an association between psychostimulants and risk of adverse perinatal outcomes including preeclampsia and preterm birth. 11,38 Therefore, ADHD medication use in pregnancy may affect gestational length, resulting in shorter pregnancies with less opportunity to fill a prescription in the third trimester. This could cause misclassification of group assignment. We cannot be certain if prescription fills translate to medication use. It is possible that prescriptions filled at the end of each period might have been used in the following period, which was not accounted for. Some ADHD medications are indicated for other diagnoses such as hypersomnia which is quite rare.
As we have studied ADHD medication exposure before, during, and after pregnancy, the resulting trajectory groups are driven by the use at the end of the period of interest. These trajectory patterns, identified for descriptive purposes, might not be relevant to study associations and perform outcome studies. However, trajectories of medication use during a narrower time period could be useful to study later pregnancy outcomes. Typically, medications for most chronic conditions may be filled for up to 3 months at a time, though prescription durations are likely shorter for psychostimulants. Therefore, we captured exposure in 3-month and 6-month intervals. Restricting to a shorter exposure interval for example, 1 month, might have led to more misclassification since prescriptions may be filled less frequently when the medication is used regularly. These trajectories might have followed a different trend if we had limited our medication exposure assessment to during pregnancy only or if we had more detailed information on use during each month, week, or day of pregnancy. A more granular definition of the time intervals would have been possible if we had self-reported data.
In conclusion, most women discontinue or interrupt their medication during pregnancy. However, more women are choosing to continue use in recent years. Those who continued ADHD medication used more psychotropic drugs which suggests the presence of other comorbidities and possibility of more severe ADHD. Also, continuers were more likely to have had multiple previous births and were less likely to be cohabiting with a partner; both these characteristics might have an impact on the decision to continue medication use. Gaining insights on medication use patterns is of clinical significance to guide rational medication use around pregnancy. Further studies are needed to acknowledge varying patterns of medication use in multiple intervals surrounding pregnancy to better understand the pregnant population with ADHD and their medication use behavior.

ACKNOWLEDGMENTS
The work was funded by grants from the Research Council of Norway, project numbers 301977 and 273366, and through its Centres of Excellence funding scheme, project number 262700.
several entities (pharmaceutical companies, regulatory authorities,