The association between psychostimulant use in pregnancy and adverse maternal and neonatal outcomes: results from a distributed analysis in two similar jurisdictions

Abstract

Background

Conflicting evidence suggests a possible association between use of prescribed psychostimulants during pregnancy and adverse perinatal outcomes.

Methods

We conducted population-based cohort studies including pregnancies conceived between April 2002 and March 2017 (Ontario, Canada; N = 554 272) and January 2003 to April 2011 [New South Wales (NSW), Australia; N = 139 229]. We evaluated the association between exposure to prescription amphetamine, methylphenidate, dextroamphetamine or lisdexamfetamine during pregnancy and pre-eclampsia, placental abruption, preterm birth, low birthweight, small for gestational age and neonatal intensive care unit admission. We used inverse probability of treatment weighting based on propensity scores to balance measured confounders between exposed and unexposed pregnancies. Additionally, we restricted the Ontario cohort to social security beneficiaries where supplementary confounder information was available.

Results

In Ontario and NSW respectively, 1360 (0.25%) and 146 (0.10%) pregnancies were exposed to psychostimulants. Crude analyses indicated associations between exposure and nearly all outcomes [OR range 1.15–2.16 (Ontario); 0.97–2.20 (NSW)]. Nearly all associations were attenuated after weighting. Pre-eclampsia was the exception: odds remained elevated in the weighted analysis of the Ontario cohort (OR 2.02, 95% CI 1.42–2.88), although some attenuation occurred in NSW (weighted OR 1.50, 95% CI 0.77–2.94) and upon restriction to social security beneficiaries (weighted OR 1.24, 95% CI 0.64–2.40), and confidence intervals were wide.

Conclusions

We observed higher rates of outcomes among exposed pregnancies, but the attenuation of associations after adjustment and likelihood of residual confounding suggests psychostimulant exposure is not a major causal factor for most measured outcomes. Our findings for pre-eclampsia were inconclusive; exposed pregnancies may benefit from closer monitoring.

Key Messages.

• In this population-based study of nearly 700 000 pregnancies, our crude analyses showed elevated rates of adverse outcomes (pre-eclampsia, placental abruption, preterm birth, low birthweight, small for gestational age and neonatal intensive care unit admission) with psychostimulant exposure.

• After applying inverse probability of treatment weighting based on propensity scores to adjust for measured confounders, nearly all associations were attenuated, resulting in null or small associations.

• The attenuation of associations after adjustment and likelihood of residual confounding suggests that psychostimulant exposure is not a major causal factor for most measured outcomes, although our findings for pre-eclampsia are inconclusive.

• This reassuring evidence supports informed decision making regarding use of psychostimulants in pregnancy, and indicates that closer monitoring may be warranted given the remaining uncertainty surrounding the relationship between psychostimulant exposure and the risk of pre-eclampsia.

Introduction

Stimulants (often termed psychostimulants) are the most prescribed treatment for attention deficit hyperactivity disorder (ADHD)^1–3 and include methylphenidate, amphetamine, dextroamphetamine and lisdexamfetamine. Other indications include narcolepsy and binge eating disorder. ADHD is generally first diagnosed in children, but there is increasing recognition of ADHD continuing into adulthood.⁴ Consequently, psychostimulant use among pregnant women has grown. Recent international studies^5–7 have estimated the prevalence of psychostimulant exposure during pregnancy to range between 0.5% and 1.3%, compared with <0.1% in 1998.⁷

The safety of psychostimulant exposure during pregnancy remains uncertain. Use of psychostimulants has been associated with increased blood pressure in children, adolescents and adults,^8,9 and it has been hypothesized that the vasoconstrictive effect of psychostimulants could lead to placental insufficiency when used during pregnancy, resulting in adverse perinatal outcomes including pre-eclampsia, placental abruption and preterm birth.¹⁰ Previous studies, all observational, have shown inconsistent results. Some population-based studies demonstrated increased risks of pre-eclampsia, preterm birth and admission to neonatal intensive care with exposure to amphetamine, dextroamphetamine, lisdexamfetamine and methylphenidate during pregnancy,^10–12 but these findings were not consistent with other studies reporting null or inconclusive results.^13–18

The inconsistency in the existing evidence is likely due in part to differences in the extent to which individual observational studies have addressed confounding. To better understand the impact of confounding, we compared results of sophisticated analyses of observational data from two similar jurisdictions, Ontario (Canada) and New South Wales (NSW, Australia), and applied a variety of approaches to address confounding. We analysed databases with different capture of relevant variables, including underlying diagnosis of ADHD and smoking during pregnancy, two key potential confounders that have been addressed to varying degrees in prior research.¹⁷ Our aim was to assess whether there is a relationship between exposure to prescribed psychostimulant medication during pregnancy and the risk of pre-eclampsia, placental abruption, preterm birth, low birthweight, small for gestational age or admission to neonatal intensive care unit.

Methods

Setting and data sources

We conducted a distributed¹⁹ cohort study using linked health administrative data from the periods 2002–18 in Ontario and 2003–12 in NSW, whereby we applied a common protocol independently in each jurisdiction. Ontario and NSW are the most populous administrative divisions within their respective countries, with a combined population of approximately 20 million insured individuals. We leveraged datasets available from ICES²⁰ (Ontario) and from the Maternal Use of Medications and Safety (MUMS) study²¹ in NSW (Supplementary Table S1, available as Supplementary data at IJE online), previously used to assess medication safety in pregnancy.^22–25

The Ontario analysis was based on births identified in the MOMBABY database, which contains linked mother-child pairs of records on live births, stillbirths and terminations of pregnancy from 20 weeks onwards occurring in inpatient settings,²⁶ and captures approximately 98% of births in Ontario.²² The NSW analysis was based on the Perinatal Data Collection which contains records for all births (live births and stillbirths) in NSW that reach at least 20 weeks of gestation or 400 g birthweight.²⁷ In Ontario, we used the Narcotics Monitoring System (all controlled substances dispensed regardless of payer, July 2012 onwards²⁸) and the Ontario Drug Benefit database (all publicly funded medications dispensed over the entire study period) to ascertain medication use. In NSW, we used a database of publicly funded dispensing claims to ascertain medication use. Other data collections are described in Supplementary Table S1.

Study population

We created two separate cohorts of singleton pregnancies among women aged 16–50 years for whom we had complete dispensing data covering the period 180 days prior to conception up until date of birth. We focused on singletons, as multiple pregnancies may carry different risks for adverse perinatal outcomes.²⁹ Our Ontario cohort comprised pregnancies conceived between 1 April 2002 and 31 March 2017. Before 2013, we only included pregnancies among social security beneficiaries (for whom complete data on dispensing of stimulants were available), but all pregnancies were included from 1 January 2013 onwards. Although only 7% of the Ontario population aged 15–44 are eligible for social security benefits,³⁰ women with mental health conditions are more likely to receive benefits.²² Our NSW cohort comprised pregnancies conceived between 1 January 2003 and 11 April 2011 among social security beneficiaries. Social security beneficiaries comprise 20% of women giving birth in NSW.²⁵ Women were defined as social security beneficiaries based on their dispensing records and were required to have continued status from 180 days prior to conception to 180 days following childbirth (Supplementary Figure S1, available as Supplementary data at IJE online).

We treated distinct pregnancies of the same woman independently. We excluded pregnancies with exposure to known or suspected teratogens from both cohorts (Supplementary Table S2, Supplementary Table S3, available as Supplementary data at IJE online). In Ontario, testosterone and phenobarbital are classed as controlled substances and are captured in the Narcotics Monitoring System. Exposure to all other teratogens was only observable among social security beneficiaries. We also excluded pregnancies where babies were born with extreme birthweight relative to gestational age (Supplementary Table S3) as these were likely data entry errors.²⁴

We identified the date of last menstrual period (LMP) by subtracting gestational age in weeks from date of birth and calculated the conception date as LMP + 14 days. In both jurisdictions, gestational age is ascertained by clinicians based on ultrasound.^31,32 We defined pregnancy as the interval between conception date and date of birth (Supplementary Figure S1).

Exposure ascertainment

Some of our outcomes of interest (e.g. pre-eclampsia) could present from the 20th week of pregnancy onwards, but exact timing of onset was not recorded. To ensure we were measuring exposure prior to onset of outcomes, we considered a pregnancy to be exposed if there was at least one dispensing record for methylphenidate, amphetamine, dextroamphetamine or lisdexamfetamine (hereafter termed ‘psychostimulants’) at any time during the first or second trimester (first 180 days from conception date). We defined unexposed pregnancies as those where a woman had no dispensing record for either psychostimulants or atomoxetine (a non-stimulant treatment for ADHD) between conception date and date of birth.

Outcomes

As outcomes, we focused on pre-eclampsia, placental abruption, preterm birth (<37 weeks gestation), low birthweight (<2500 g), small for gestational age (SGA) and admission to neonatal intensive care unit (NICU).^10,33,34 We assessed low birthweight, SGA and NICU admission among live births only, with no stillbirths occurring among exposed pregnancies. We used validated^35–37 ICD-10-CA (Ontario) and ICD-10-AM (NSW) codes to ascertain outcomes from the hospital delivery record; in NSW we also supplemented ascertainment by using perinatal register data (see Supplementary Table S3 for complete definitions). For completeness, we repeated analyses with any major congenital anomalies as the outcome, noting that the study was not adequately powered for this outcome.

Covariates

We included as covariates several risk factors for adverse perinatal outcomes which could potentially confound the relationship between psychostimulant use and maternal and neonatal outcomes.^{10,12,14,34,38} We defined characteristics of interest using hospital diagnoses and dispensing claims. To maximize the best available information, we supplemented ascertainment in the Ontario cohort with physician billing data; in NSW we leveraged the perinatal register data (Supplementary Table S3). For each pregnancy we obtained details on women’s demographic characteristics at conception [age, country of birth (NSW only), neighbourhood income quintile/social disadvantage,³⁹ urban residence], obstetric characteristics [calendar year of conception, parity, previous caesarean delivery (NSW only)], health risks (number of previous admissions, smoking during pregnancy) and comorbidities (metabolic conditions, epilepsy, chronic renal disease, thyroid conditions, ADHD, mental and behavioural disorders).

We also identified use of opioids and related substances (Ontario cohort only) and psychotropic medications during pregnancy^{6,10,11,14,34,38} (Supplementary Table S3). In Ontario, we only had complete capture of controlled substances from July 2012 onward. Although these included most medications of interest (e.g. narcotic analgesics, benzodiazepines), we were unable to capture use of antidepressants, antipsychotics or lithium among non-social security beneficiaries.

Statistical analysis

We used descriptive statistics to describe baseline characteristics and standardized differences to assess the balance of covariates between exposed and unexposed pregnancies, with differences greater than 0.1 considered unbalanced.⁴⁰ We calculated the number of events, event rates and unadjusted odds ratios (OR) for each outcome.

To adjust for confounding, we applied inverse probability of treatment weighting (IPTW) using a propensity score based on the covariates specified above. This approach effectively creates a synthetic population in which the distribution of all measured covariates is balanced between the exposed and unexposed pregnancies.⁴¹ We trimmed the non-overlapping regions of the propensity score and defined the weights such that they allowed estimation of the average treatment effect in the treated.^41,42 We fitted logistic regression models in the weighted population (with the intention to adjust for any unbalanced covariates) to estimate the effects of exposure on each outcome.

Ontario cohort restriction

As an additional step, we restricted the Ontario cohort to social security beneficiaries so as to evaluate the impact of additional information on potential confounders available in this population. Specifically, this restricted cohort had improved exclusion of pregnancies exposed to teratogenic medications and increased ascertainment of opioid and psychotropic medication use (included in the estimation of propensity scores) relative to the full Ontario cohort.

As a secondary analysis, we restricted the main analysis to pregnancies with a recorded diagnosis of maternal ADHD, using the physician billing data available in Ontario. This was considered secondary as ascertainment of ADHD through physician billing records has not been validated.

Sensitivity analyses

We conducted a subgroup analysis of pregnancies without any exposure to opioids (Ontario only) or psychotropics to exclude possible additive effects of these medications with psychostimulants. As dispensed medications are not necessarily taken, we employed a sensitivity definition of exposure whereby we required two or more psychostimulants dispensing claims during pregnancy. We also conducted a sensitivity analysis whereby we excluded any unexposed pregnancies with psychostimulant exposure in the 180 days prior to conception.

To account for correlation between siblings, we conducted a sensitivity analysis incorporating a generalized estimating equation approach. We conducted a post-hoc calculation of the E-value⁴³ for observed effects to assess the robustness of our findings against unmeasured confounding. For each restricted and sensitivity analysis we recalculated propensity scores based on all measured covariates and fitted logistic regression models using IPTW. We conducted all analyses using SAS v9.4, Cary, NC.

Results

We identified 1.8 million singleton pregnancies in Ontario and 631 187 in NSW. After applying exclusion criteria, our final cohorts comprised 554 272 pregnancies (Ontario) and 139 229 pregnancies (NSW) (Figure 1). Exposed pregnancies tended to be among younger women, those with mental health conditions and those with exposure to opioids and psychotropics (Table 1).

Figure 1.

Cohort creation, Ontario and New South Wales cohorts. NSW, New South Wales. *Population estimates available from 2013 onwards. ^†Not eligible for social security benefits (where applicable), or extreme birthweight relative to gestational age

Table 1.

Baseline characteristics, Ontario and New South Wales cohorts

	Ontario cohort*				NSW cohort
Variable	Unexposed	Exposed	Unweighted Std diff	Weighted Std diff	Unexposed	Exposed	Unweighted Std diff	Weighted Std diff
	N = 552 912	N = 1360			N = 139 083	N = 146
Demographic characteristics
Calendar year of conception
2002	2971 (0.5%)	≤5	†	†	–	–
2003	3883 (0.7%)	6-11	†	†	15 762 (11.3%)	9 (6.2%)	−0.18	0.01
2004	4086 (0.7%)	14 (1.0%)	0.03	0.01	17 216 (12.4%)	10 (6.9%)	−0.19	0.01
2005	4130 (0.8%)	16 (1.2%)	0.04	0.01	17 156 (12.3%)	23 (15.8%)	0.10	−0.01
2006	4427 (0.8%)	15 (1.1%)	0.03	0.01	17 542 (12.6%)	21 (14.4%)	0.05	−0.02
2007	4623 (0.8%)	19 (1.4%)	0.05	−0.02	17 122 (12.3%)	23 (15.8%)	0.10	0.00
2008	4752 (0.9%)	15 (1.1%)	0.02	0.00	16 654 (12.0%)	17 (11.6%)	−0.01	0.01
2009	5132 (0.9%)	36 (2.7%)	0.13	0.00	16 627 (12.0%)	15 (10.3%)	−0.05	0.00
2010a	5716 (1.0%)	38 (2.8%)	0.13	−0.01	21 004 (15.1%)	28 (19.2%)	0.11	0.00
2011	5707 (1.0%)	53 (3.9%)	0.19	−0.02	–	–
2012	5798 (1.1%)	50 (3.7%)	0.17	−0.02	–	–
2013	119 481 (21.6%)	211 (15.5%)	−0.16	0.02	–	–
2014	118 768 (21.5%)	224 (16.5%)	−0.13	0.01	–	–
2015	118 412 (21.4%)	265 (19.5%)	−0.05	0.00	–	–
2016–17b	145 026 (26.2%)	393 (28.9%)	0.06	0.00	–	–
Maternal age at conception
Mean ± SD	–	–			27.3 ± 6.1	26.6 ± 6.5	−0.11	0.01
Median (IQR)	–	–			27 (23 - 32)	27 (21 - 32)
Age group
<25	99 900 (18.1%)	441 (32.4%)	0.34	−0.05	55 213 (39.7%)	63 (43.2%)	0.07	−0.01
25–34	349 748 (63.3%)	707 (52.0%)	−0.23	0.03	66 274 (47.7%)	66 (45.2%)	−0.05	0.01
35+	103 264 (18.7%)	212 (15.6%)	−0.08	0.04	17 596 (12.7%)	17 (11.6%)	−0.03	0.01
Western-bornc	–	–			106 364 (76.5%)	141-146	0.71	−0.05
Partner	–	–			87 605 (63.0%)	68 (46.6%)	−0.33	0.00
Neighbourhood income quintile
Q1: most disadvantaged	136 907 (24.8%)	439 (32.3%)	0.17	−0.01	43 790 (31.5%)	22 (15.1%)	−0.40	0.03
Q2	111 325 (20.1%)	284 (20.9%)	0.02	−0.02	23 705 (17.0%)	40 (27.4%)	0.25	−0.02
Q3	110 264 (19.9%)	247 (18.2%)	−0.05	0.00	32 662 (23.5%)	28 (19.2%)	−0.11	0.01
Q4	108 652 (19.7%)	211 (15.5%)	−0.11	0.01	26 066 (18.7%)	37 (25.3%)	0.16	0.01
Q5: least disadvantaged	85 764 (15.5%)	179 (13.2%)	−0.07	0.02	12 860 (9.3%)	19 (13.0%)	0.12	−0.03
Urban residence	496 890 (89.9%)	1201 (88.3%)	0.05	−0.01	94 486 (67.9%)	88 (60.3%)	−0.16	0.00
Obstetric characteristics
Parity
0	230 839 (41.8%)	617 (45.4%)	0.07	−0.01	30 795 (22.1%)	52 (35.6%)	0.30	−0.03
1	199 302 (36.1%)	373 (27.4%)	−0.19	0.01	47 707 (34.3%)	41 (28.1%)	−0.13	0.00
2	79 377 (14.4%)	192 (14.1%)	−0.01	0.00	30 646 (22.0%)	23 (15.8%)	−0.16	0.02
3+	43 394 (7.9%)	178 (13.1%)	0.17	0.01	29 935 (21.5%)	30 (20.6%)	−0.02	0.02
Previous caesarean	–	–			22 980 (16.5%)	17 (11.6%)	−0.14	0.00
Maternal comorbidities
Admissions 1 year prior to conception
0	455 800 (82.4%)	1067 (78.5%)	−0.10	0.01	93 978 (67.6%)	95 (65.1%)	−0.05	0.02
1	83 890 (15.2%)	241 (17.7%)	0.07	0.00	30 479 (21.9%)	35 (24.0%)	0.05	0.00
2+	13 222 (2.4%)	52 (3.8%)	0.08	−0.02	14 626 (10.5%)	16 (11.0%)	0.01	−0.03
Smoking during pregnancy	3210 (0.6%)	26 (1.9%)	0.12	−0.03	46 785 (33.6%)	85 (58.2%)	0.51	0.01
Metabolic conditionsd	82 683 (15.0%)	211 (15.5%)	0.02	0.01	4876 (3.5%)	6 (4.1%)	0.03	0.00
Epilepsy	1985 (0.4%)	15 (1.1%)	0.09	−0.03	105 (0.1%)	≤5	0.10	−0.01
Narcolepsy	3309 (0.6%)	38 (2.8%)	0.17	−0.05
Chronic renal disease	11 075 (2.0%)	37 (2.7%)	0.05	−0.01	1265 (0.9%)	≤5	0.09	0.00
Thyroid conditions	20 621 (3.7%)	50 (3.7%)	0.00	0.02	2033 (1.5%)	≤5	−0.08	−0.01
Psychiatric diagnoses
Any mental health conditions	144 213 (26.1%)	1046 (76.9%)	1.18	−0.03	27 932 (20.1%)	80 (54.8%)	0.77	−0.04
ADHD	2119 (0.4%)	450 (33.1%)	0.97	−0.08	15 (0.0%)	8 (5.5%)	0.34	−0.08
Psychotic disorders	2091 (0.4%)	47 (3.5%)	0.23	−0.05	–	–
Mood, anxiety or related disorders	131 273 (23.7%)	910 (66.9%)	0.96	−0.03	–	–
Substance use disorders	10 598 (1.9%)	262 (19.3%)	0.59	−0.08	–	–
Other mental health conditions	6461 (1.2%)	143 (10.5%)	0.41	−0.09	–	–
Medicine use during pregnancy
Buprenorphine/naloxone	651 (0.1%)	20 (1.5%)	0.15	0.01	–	–
Methadone	1784 (0.3%)	73 (5.4%)	0.31	−0.07	–	–
Opioids	24 414 (4.4%)	190 (14.0%)	0.34	−0.05	–	–
Psychotropics	7117 (1.3%)	124 (9.1%)	0.36	−0.04	12 414 (8.9%)	48 (32.9%)	0.62	−0.02
Psychiatric care during pregnancy
Hospitalization or ED visit	5287 (1.0%)	92 (6.8%)	0.30	−0.04	869 (0.6%)	≤5	0.17	0.01

Weighted cohort trimmed to exclude individuals with non-overlapping propensity scores.

^*Population estimates available from 2013 onwards.

^†Suppressed to avoid residual disclosure of small cells.

ADHD, attention deficit hyperactivity disorder; ED, emergency department; IQR, interquartile range; SD, standard deviation; Std diff ,standardized difference; NSW, New South Wales.

^aNSW cohort: includes conception dates between 01 January 2003 and 11 April 2011.

^bOntario cohort: includes conception dates between 01 April 2002 and 31 March 2017.

^cAustralia, New Zealand, UK, Ireland, Western Europe, Northern Europe, Southern Europe, North America.

^dCardiovascular conditions, pre-existing hypertension or pre-existing diabetes.

The proportion of pregnancies exposed to psychostimulants was higher in the Ontario cohort (1360, 0.25%) than NSW (146, 0.10%) (Table 2). Exposed pregnancies were most likely to be dispensed methylphenidate in Ontario (55.1%) and dextroamphetamine in NSW (81.5%). Most pregnancies were exposed in only the first trimester (54.0% Ontario; 43.2% NSW); nearly a third were exposed across all three trimesters (28.4% Ontario; 30.8% NSW).

Table 2.

Details of psychostimulant exposure, by cohort

	Ontario cohort*	NSW cohort
	N = 554 272	N = 139 229
Exposure
Exposed during pregnancy	1360 (0.25%)	146 (0.10%)
Timing of exposure
Trimester 1 only	734	63
Trimester 2 only	44	≤5
Trimester 3 only	–	–
Trimester 1 and 2	128	24
Trimester 1 and 3	27	≤5
Trimester 2 and 3	41	†
Trimester 1, 2 and 3	386	41
Psychostimulanta
Amphetamine only	201 (14.78%)	–
Dextroamphetamine only	122 (8.97%)	119 (81.51%)
Lisdexamfetamine only	254 (18.68%)	–
Methylphenidate only	749 (55.07%)	22-27
More than one stimulant	34 (2.50%)	≤5
Psychostimulant use 6 months prior to conception
Previous exposure	1191 (87.57%)	122 (83.56%)

^*Population estimates available from 2013 onwards.

^†Suppressed to avoid residual disclosure of small cells.

^aPercentages calculated on exposed pregnancies.

Crude rates of adverse perinatal outcomes were higher among exposed pregnancies in both the Ontario (range: 2.2–22.4% exposed vs 1.2–12.9% unexposed) and NSW cohorts (range: <5.2–24.7% exposed vs 0.7–17.2% unexposed) (Figure 2).

Figure 2.

Perinatal outcomes among exposed and unexposed pregnancies, by cohort. NSW, New South Wales; NICU, neonatal intensive care unit. *Suppressed for privacy reasons

Main analyses

Results of the crude analyses in the Ontario cohort indicated an association between psychostimulant exposure and pre-eclampsia (OR 2.01, 95% CI 1.50–2.70). This association remained after weighting (OR 2.02, 95% CI 1.42–2.88) (Figure 2). Some attenuation of the association was apparent when we restricted to social security beneficiaries in whom we were better able to account for use of additional medications, but the confidence intervals were wide (OR 1.24, 95% CI 0.64–2.40). In the NSW cohort, the association was attenuated after weighting (crude OR 2.13, 95% CI 1.18–3.85; weighted OR 1.50, 95% CI 0.77–2.94), although the confidence intervals were again wide (Figure 2).

When we examined preterm birth, effect estimates dropped substantially following IPTW. In the weighted analyses in the Ontario cohort, a small increase in the odds of preterm birth remained (crude OR 1.80, 95% CI 1.52–2.13; weighted OR 1.25, 95% CI 1.01–1.55). This association was further attenuated in social security beneficiaries (crude OR 1.41, 95% CI 1.12–1.78; weighted OR 1.10, 95% CI 0.84–1.44). The findings for the NSW cohort aligned closely, although confidence intervals were wider and included 1 (crude OR 1.82, 95% CI 1.12–2.94; weighted OR 1.27, 95% CI 0.77–2.11) (Figure 2).

In the Ontario cohort, crude estimates suggested an association between exposure and low birthweight and NICU admission; however, no associations remained after weighting. Among Ontario social security beneficiaries there was only slight attenuation in the weighted estimate for placental abruption, although the confidence intervals were wide and included 1 (crude OR 1.88, 95% CI 1.31–2.70; weighted OR 1.61, 95% CI 0.95–2.71). We observed similar patterns in NSW, although results examining placental abruption were inconclusive due to small numbers of events. We did not detect evidence of an association between exposure to psychostimulants and SGA in any of the analyses (Figure 2). As anticipated, our investigation of the relationship between psychostimulant exposure and congenital anomalies was inconclusive due to wide 95% CIs around the ORs in all analyses (Supplementary Figure S2, available as Supplementary data at IJE online).

In our secondary analysis including only women with an underlying diagnosis of ADHD in Ontario, even after weighting we observed potential associations between psychostimulants and pre-eclampsia, preterm birth and low birthweight, although confidence intervals for pre-eclampsia included 1 (Supplementary Figure S3, available as Supplementary data at IJE online).

Sensitivity analyses

When we restricted to pregnancies without exposure to opioids or psychotropics, we observed similar patterns as the main analyses although effect estimates tended to be smaller (Supplementary Figure S4, available as Supplementary data at IJE online). In the Ontario cohort, the weighted analyses suggested a possible association between exposure and pre-eclampsia, preterm birth and NICU admission, but these associations were attenuated in the analyses based on social security beneficiaries (Supplementary Figure S4). We did not detect any associations in the NSW cohort. In both the Ontario and NSW cohorts, we found similar or slightly stronger associations as in the main analyses when we employed a stricter definition of exposure (two or more psychostimulant dispensings), and when we employed a 6-month washout period in defining unexposed pregnancies (Supplementary Figure S5, available as Supplementary data at IJE online). Neither analysis altered the conclusions drawn from the main analyses. Effect estimates and confidence intervals resulting from the generalized estimating equation models were only slightly different compared with our main analyses (Supplementary Figure S6, available as Supplementary data at IJE online) and our original conclusions were not affected.

The E-score for pre-eclampsia indicated that an unmeasured confounder would need to have OR ≥3.46 with both psychostimulant exposure and pre-eclampsia, after adjustment for all measured confounders, to explain away the observed weighted OR (2.01). An unmeasured confounder with adjusted OR ≥2.19 would cause the confidence interval to include 1.

Discussion

In our population-based study of nearly 700 000 pregnancies, our crude analyses showed higher rates of adverse perinatal outcomes among pregnancies exposed to psychostimulants. The attenuation of associations after accounting for measured confounders and the likelihood of residual confounding suggests that psychostimulant exposure is not a major causal factor for most measured outcomes. However, uncertainty around the potential association between psychostimulants and pre-eclampsia remains. Our analyses of the Ontario population indicated a positive association between psychostimulant exposure and pre-eclampsia (absolute risk 3.4% among exposed pregnancies vs 1.7% unexposed) that persisted after inverse probability of treatment weighting based on propensity scores. Our post-hoc sensitivity analysis indicated that this is unlikely to be wholly attributable to confounding. However, this signal did not persist in our analysis based on social security beneficiaries (in whom we were better able to account for other medication use) or in our NSW cohort.

Although our findings related to pre-eclampsia were inconclusive, other population-based analyses have found a potential association. Cohen et al. detected an increased risk of pre-eclampsia associated with psychostimulant use in pregnancy among Medicaid-eligible women (adjusted risk ratio 1.29, 95% CI 1.11–1.49).¹⁰ Poulton et al. also found associations between psychostimulant exposure both before and during pregnancy and pre-eclampsia (adjusted OR 1.5, 95% CI 0.8–2.6) in a population of NSW women.¹² Such an effect is biologically plausible based on the hypothesis that vasoconstrictive effects of psychostimulants could lead to placental insufficiency.⁴⁴ Women treated with psychostimulants during pregnancy may therefore benefit from closer monitoring.

Limitations

Our study had several limitations. First, although we applied a propensity score approach to adjust for confounding, we may not have been able to fully account for all relevant factors. Although we maximized ascertainment of ADHD in the Ontario cohort by leveraging physician billing codes, this method has not been validated and some under-ascertainment is apparent; 0.4% of our cohort identified as having an ADHD diagnosis whereas the reported prevalence of ADHD among pregnant women ranges from 0.5% to 1.3%.^5–7 Furthermore, we did not have information on the severity of ADHD, which is likely related to adverse outcomes.^11–13 It is therefore possible that our main analyses and our sensitivity analysis restricted to women with a diagnosis of ADHD are subject to unmeasured confounding by ADHD severity or other unmeasured factors. Similarly, although we were better able to adjust for important confounders such as smoking in the NSW cohort, the available data had poor capture of ADHD diagnoses, mental illness and associated lifestyle factors. It is therefore possible that the observed effect in NSW may reflect confounding by unmeasured factors. Indeed, it has been hypothesized that previously observed relationships between psychostimulant exposure and adverse perinatal outcomes may be due to underlying ADHD or closely related conditions.^12,34 In addition to suboptimal measurement of some confounding factors in NSW, we lacked data on dispensed opioids that we hypothesized might have an additive effect. However, we continued to see consistent signals in the Ontario analysis where this information was available, demonstrating that our main findings are unlikely to be a reflection of additive effects. Second, we only had data on dispensed medications, which does not necessarily reflect actual consumption. However, when we restricted our definition of exposure to two or more filled prescriptions (more likely to reflect medication use during gestation), we observed elevated ORs for pre-eclampsia among pregnancies exposed to psychostimulants in both cohorts, supporting our main findings. Third, our NSW cohort had small numbers of exposed pregnancies, yielding reduced precision to detect effects. We were also unable to robustly assess associations between exposure and congenital anomalies or neurological development which may not become apparent until early childhood. Fourth, we cannot discount the possibility of selection bias arising from the absence of data on pregnancy losses and terminations before the 20th week of pregnancy. Studies have shown a possible association between psychostimulant exposure and miscarriage.^6,13,34 Last, the majority of our study population comprised social security beneficiaries. Key confounders such as smoking, use of other medications or other risky behaviours during pregnancy may differ compared with the general population, and it is therefore possible that confounding may be a greater issue in this group.

Conclusions

Our study compares the results of assessments of the association between psychostimulant exposure during pregnancy and adverse perinatal outcomes across two similar jurisdictions, datasets and cohorts, with varying capture of key confounding factors. Our findings for the outcome of placental abruption, preterm birth, low birthweight, SGA and NICU admission suggest that psychostimulant exposure during pregnancy is not a major causal factor for most adverse perinatal outcomes. This reassuring evidence helps support informed decision making by clinicians and women of reproductive age regarding use of psychostimulants in pregnancy. Uncertainty remains, however, regarding the relationship between psychostimulant exposure and the risk of pre-eclampsia. Women treated with psychostimulants during pregnancy may therefore benefit from closer monitoring in relation to this outcome.

Ethics approval

The University of Toronto Health Sciences REB (#37204), the NSW Population & Health Services Research Ethics Committee (ref HREC/12/CIPHS/44), the Australian Institute of Health and Welfare Ethics Committee (EC 2012/2/22) and the Aboriginal Health & Medical Research Council of NSW Ethics Committee (871/12) approved the study.

Data availability

The Ontario dataset for this study is held securely in coded form at ICES. Legal data sharing agreements between ICES and data providers (e.g. health care organizations and government) prohibit ICES from making the dataset publicly available, but access may be granted to those who meet pre-specified criteria for confidential access, available at [www.ices.on.ca/DAS]. The full dataset creation plan and underlying analytical code are available from the authors upon request, understanding that the computer programs may rely upon coding templates or macros that are unique to ICES and are therefore either inaccessible or may require modification. The NSW data were linked with the permission of each of the source data custodians and with specific ethical approvals. The authors do not have permission to share patient-level data because of the highly confidential nature of the data. Permission to access the data is restricted to researchers named and approved by relevant Human Research Ethics Committees.

Supplementary data

Supplementary data are available at IJE online.

Author contributions

X.C., H.Z., T.G., D.H., S.P., S.V., A.H. conceived of and designed the study. A.H., D.H., X.C. acquired the data: X.C. created the analytical datasets and conducted the statistical analysis. All authors interpreted the results. X.C. and A.H. drafted the manuscript. All authors critically reviewed the manuscript. All authors reviewed and approved the final manuscript. X.C. had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Funding

This work was supported by the National Health and Medical Research Council (NHMRC) (grant numbers 1196900, 1028543, 2005259 to X.C., 1158763 to A.S., 2007048 to H.Z.), UNSW Scientia Award to H.Z., NSW Health Early-Mid Career Fellowship to A.H. T.G. is supported by a Canadian Institutes of Health Research (CIHR) Canada Research Chair in Drug Policy Research & Evaluation. The funders had no role in the design and conduct of the study; collection, management, analysis and interpretation of the data; preparation, review or approval of the manuscript; and decision to submit the manuscript for publication.