Association of Timing and Duration of Prenatal Analgesic Opioid Exposure With Attention-Deficit/Hyperactivity Disorder in Children

Key Points Question Is prenatal analgesic opioid exposure associated with attention-deficit/hyperactivity disorder (ADHD) in children? Findings In this cohort study of 73 480 children, with a mean follow-up of 11 years, no association between timing of analgesic opioid exposure during pregnancy and ADHD was found. The risk of ADHD diagnosis was elevated after exposure to opioids for 5 or more weeks compared with exposure for 4 weeks or less. Meaning The increased risk of ADHD observed in this study may be driven by longer duration of exposure; however, the role of residual or unmeasured confounding cannot be excluded, and this finding requires further study.


Outcome
We examined childhood ADHD using both a parent-reported ADHD symptoms scale and ADHD diagnosis and/or filled prescriptions for ADHD medications. The first may capture more subtle daily problems noticed by parents not reaching the threshold of a clinical diagnosis, whereas ADHD diagnosis is indicative of a certain level of severity. 1 ADHD symptoms: Parent-reported symptoms of ADHD in children at five years of age were measured by 12 items from the Conners Parent Rating Scale-Revised Short Form (CPRS-R(S)) included in the MoBa questionnaire at five years. 2,3 CPRS is a tool for obtaining parental reports of childhood behavior problems, and the 12 selected items are from the areas of inattention and hyperactivity/impulsivity. Parents reported how much each item has been a problem for the child during the past month. The items are scored on a four point Likert scale (1-4) ranging from "not true/seldom" to "very often". Mean scores were calculated and standardized. Higher z-scores indicated more symptoms of ADHD. The Cronbach's α for the CPRS-R was 0.9, showing good internal consistency between items of the scale. For this analysis, the ADHD diagnosis sample was further restricted to those with available outcome data in MoBa Q-5years.

Potential confounding factors
The Mothers were asked to complete the short form of the "Adult ADHD Self-Report Scale (ASRS) in Q6 (child age 3 years). ASRS is a self-report screening scale of adult ADHD and includes six questions. Four questions capture symptoms of inattention and two questions entail symptoms of hyperactivityimpulsivity. The five response options range from "never=1" to "very often=5". We dichotomized the response options (≤ 3=0, ≥ 4=1) and summed them across the six questions, as indicated in Kessler et al. 6 We then categorized the final ASRS score into three clinical categories as recommended (no ADHD symptoms=score 0-1, mild symptoms=score 2-3, moderate to severe symptoms=score 4-6). Due to a large amount of missing in the maternal ASRS (41.6% in the ADHD diagnosis sample, and 18.4% in the ADHD symptoms sample), we used this variable only in a sub-analysis.

Child characteristics
Child

Sensitivity analyses
We performed many sub-group and sensitivity analysis to test the robustness of our findings.
We conducted separate models for all exposure definitions that took into account additional paternal and child factors under alternate model specifications (eTable 3).
We descriptively examined the cumulative incidence of ADHD by prenatal analgesic opioid exposure status among boys and girls (ADHD diagnosis sample) and we performed stratified analysis with ever/never exposure to analgesic opioid exposure in pregnancy in order to better understand the role of child sex on ADHD risk.
We performed a positive control analysis with women using opioid containing cough medications during pregnancy (ATC R05D). We used propensity scores (PS) with inverse probability of treatment weights (IPTW) to account for measured confounders. First, we fit a logistic regression model to estimate the probability of exposure to R05D during pregnancy, relative to unexposed to R05D in pregnancy, conditional on confounders in model 1 (cf eTable 3) and opioid use during pregnancy. Then we performed Cox regression analysis to estimate HR and generalized linear models to estimate standardized mean differences in ADHD symptoms, as described earlier.
The following substances are marketed in Norway under R05D: ethylmorphine, hydrocodone, codeine, noscapine, and combinations (cough syrup; ethylmorphine and codeine). 7 We performed an analysis among women using opioids not in combination with paracetamol. Thus we excluded those who had used the drug with ATC-code N02AA59 (paracetamol + codeine) during pregnancy. PS with IPTW was applied as described earlier.
We calculated the E-value in order to determine how strong an unmeasured confounder had to be in order to explain away the observed exposure-outcome association. [8][9][10] We replicated the main analysis in a sub-sample of pregnancy-child dyads born in 2004 or later (ADHD diagnosis sample, n=54 740), so that all children had available outcome data since birth.
We excluded 644 children (0.8% of the ADHD diagnosis sample) who had emigrated, because the study did not have information about dates of potential emigration or death.
In the time-to-event analysis, the proportional hazard (PH) assumption was not met (timing analysis only). We therefore split the follow-up time at child age 7 years, estimating period-specific HRs. The follow-up time was split at this time-point based on the weighted Nelson Aalen curves (see eFigure 6 and eFigure 7).
In our propensity score (PS) models we adjusted for a range of variables, including anxiety/depression measured at gestational week 17, number of pain episodes throughout pregnancy and co-medications used during pregnancy. We performed additional sensitivity analyses were we 1) removed anxiety/depression and number of pain episodes, from the main PS models, and 2) adjusted for use of co-medications at baseline (MoBa Q1) in order to see if some of these variables could potentially be mediators in the exposure-outcome relationship.
Under the assumption that data were missing at random, we imputed incomplete data via multiple imputation with chained equation (ten replications). 11 The imputation procedure included exposure and outcome variables, baseline hazard, and auxiliary variables (e.g., maternal age and illnesses, parity, co-medication, risk factors for the outcomes). 11,12 Imputed data were used in all analyses.
We chose to impute 10 datasets as this number (m=10) was considered to be a good trade-off to address bias, between (a) the computational effort of having a high number of imputations, and (b) the low percentage of incomplete observations in our study (<1 to 11% for individual variables, leading to 19.4% observations having at least one covariate with missing value) Additional details on «Results»

Sensitivity analyses
The point estimates under alternative model specifications were generally consisted with main findings (eFigure [3][4][5]. We only had data on maternal ADHD traits for a sub-sample of the population. In this sub-sample, the point estimates were generally similar as in the main analysis, except those comparing mid/late exposure to unexposed in the time window and pre-pregnancy users only, respectively. The cumulative incidence for ADHD was greater among boys than girls in the study (data not shown). In analyses stratified by gender the weighted HR of ADHD was 1.28 (95% CI: 0.93-1.77) among boys and 1.36 (95% CI: 0.74-2.51) among girls. The point estimates were somewhat similar, however, boys seem to be at elevated risk based on the lower bound of the CI. However, girls were likely in a small sample size. Furthermore, we found no association between opioids and ADHD symptoms in children five years of age stratified by gender (Boys: βw: -0.01, 95% CI -0.13-0.12 and girls: βw: 0.06, 95% CI: -0.10-0.21).
There were 1153 (1.6%) and 492 (1.6%) women using opioid containing cough medications (R05D) during pregnancy in the ADHD diagnosis sample and the ADHD symptoms sample, respectively. We found no association between use of R05D during pregnancy and ADHD, compared to unexposed in pregnancy (Crude HR: 0.84, 95% CI: 0.59-1.21, wHR: 0.70, 95% CI: 0.47-1.05). It is not entirely clear why we observed an apparently protective effect in this analysis. It may be random or due to a possible "health seeking behavior bias". We found no association between use of R05D during pregnancy and ADHD symptoms at child age five years, compared to unexposed during pregnancy (Crude β: -0.03, 95% CI: -0.12-0.05, βw: 0.01, 95% CI: -0.10-0.12).
There were 160 (0.22%) and 60 (0.20%) women who had used opioids not containing paracetamol in the ADHD diagnosis sample and the ADHD symptoms sample, respectively. We found no associations with ADHD ( The e-value was calculated to be 2.58, which means that confounding of strength equal to 2.58 (on both sides) could completely explain away an observed HR of 1.60 between use of opioids in ≥5 weeks vs use in ≤4 weeks and ADHD, but a weaker confounder could not.
In a subsample with data available in both MoBa and NorPD (n=50 925), the average DDDs reported among women using opioids in ≤4 weeks and ≥5 weeks was 8.6 DDD (SD 8.5) and 37.2 DDD (SD 79.0), respectively. The median was 5 and 12.5 for the two groups, respectively.
The results of analyses restricted to children born in 2004 or later are presented in eTable 7 and eTable 8. The point estimates deviated, by 20-40%, from those in the main analysis. In weighted analysis, exposure in early pregnancy was associated with a borderline increased risk of ADHD when compared to unexposed women (HR: 1.75, 95% CI: 1.01-3.03). In the analysis of duration, the point estimate was higher and showed an increased risk, however, with a wide CI (use in ≥5 weeks versus ≤4 weeks; HR: 2.45 95% CI: 1.32-4.54). Moreover, the point estimates for exposure in mid/late pregnancy indicate a reduction in risk; however, the CI crosses the null.
The complete case analysis showed similar results as the main analysis (data not shown).
Analysis excluding emigrated children did not materially change the point estimates from the main results (data not shown).
The results of the period-specific HR are presented in eTable 9. When compared to pre-pregnancy users only the results align with the main results. However, when compared to unexposed the point estimates deviates to some extent, meaning that the time point > 7 is driving the association. The PH assumption was met in the duration analysis (≥5 weeks vs ≤4 weeks).
Adjusting for a mediator could potentially bias results towards the null. However, we saw little indication of this in our study after having performed sensitivity analyses to specifically explore this issue. The point estimates changed very little (<10%) and in both directions (data not shown). If a mediator effect were in place, we would have expected these point estimates to be consistently higher than those in the main analyses were. Conclusions remain the same as for the main analyses.