Between 2002 and 2004, IDEAL subjects were recruited postpartum at 4 data-collection sites: Los Angeles, California; Honolulu, Hawaii; Des Moines, Iowa; and Tulsa, Oklahoma. Detailed recruitment methods for the IDEAL study have been reported previously.4,27
This study was approved by the institutional review boards at each site, and written informed consent was obtained from all subjects. A National Institute on Drug Abuse Certificate of Confidentiality was obtained that assured confidentiality of information regarding the mothers' drug use.
At recruitment, mothers were interviewed for sociodemographic and prenatal substance use information. Meconium was collected from each infant and analyzed at a central laboratory (US Drug Testing Laboratories, Inc., Des Plaines, IA) for drug metabolites.28
MA exposure was determined by self-reported MA use during this pregnancy and/or a positive meconium screen and gas chromatography/mass spectroscopy confirmation. A matched case-control study design was used. The exposed (n
= 204) and comparison (n
= 208) groups were recruited consecutively from the same sites and matched on maternal race, birth weight category (<1500, 1500–2500, and >2500 g), insurance (private/public), and education (high school completed/not completed). Inclusion in the comparison group required denial of MA use during this pregnancy and a negative meconium screen for MA. Comparison dyads with characteristics that were difficult to match (eg, Asian race, >2500 g birth weight, public insurance, high school not completed) were enrolled before a matching exposed dyad, leading to slightly different sample sizes in the 2 groups. Prenatal exposure to alcohol, tobacco, and marijuana was included in both groups as background variables.
The sample for this study included all children who were evaluated for behavioral problems at ages 3 or 5 years. The follow-up rate was 70% at 3 years and 76% at 5 years (2 cases at 3 years and 7 cases at 5 years had missing CBCL information). There were 330 subjects (262 at both ages, 26 at 3 years only, and 42 at 5 years only) or 80% of the cohort (166 exposed and 164 comparison). Comparison of the characteristics of the 330 subjects in this study with the 82 not included () revealed no significant differences on all characteristics, except mothers who were not included used more marijuana during pregnancy.
Comparison of Dyads Included and Not Included in the Study
was read to the caregiver by a certified interviewer then computer-scored to yield measures of internalizing, externalizing, and total problems and syndrome scores that aggregate co-occurring problems and are the basis for internalizing (emotionally reactive, anxious/depressed, somatic complaints, or withdrawn) and externalizing (attention problems and aggressive behavior) scores. Higher scores indicate more problems. Some items on the CBCL are consistent with Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition
diagnostic categories. We include the DSM-oriented score of attention deficit based on evidence of its clinical significance.29
Most caregivers who completed the CBCL at 3 and 5 years were biological parents (78% and 74%, respectively). Other caregivers were foster or adoptive parents (12% and 17%), relatives (9% and 5%), or nonrelatives (2% and 4%). The caregiver was the same at both ages in 84% of the cases. No significant differences in CBCL scores by caregiver were found (P values from .139 to .962).
At recruitment, demographic and neonatal characteristics were obtained from the lifestyle interview including race, gender, insurance (public/private), maternal age, single status/no partner involvement (yes or no), socioeconomic status (SES), birth weight, and gestational age. Birth length and head circumference were obtained from the infant’s medical chart. SES was calculated by using the 4-factor Hollingshead Index adapted for single parent and nonnuclear families,30,31
with Hollingshead level V indicating low SES.
Postnatal caregiver and environmental characteristics were measured on multiple visits. Measures from the lifestyle interview at 1 month and 1, 2, 3, and 5 years included physical and sexual abuse based on report to Child Protective Services (yes/no) and any change in primary caregiver (none, 207; one, 62; two, 38; three, 1; and four, 4). The Brief Symptom Inventory, administered at 1 month and 1 and 3 years, provided an overall score of caregiver psychological symptoms.32
The quality of the home environment, computed as an overall summary score, was measured at 2.5 years of age by using the Home Observation for Measurement of the Environment Early Childhood inventory.33
The personal safety section of the Substance Use Inventory at 3 years assessed domestic violence experienced by the caregiver (yes/no).
Prenatal use of MA and other drugs of abuse including quantity and frequency of use was obtained from the Substance Use Inventory,34
from which level of use was determined. Consistent with other IDEAL studies,7,8
heavy MA exposure was defined as maternal use ≥3 days per week across pregnancy. Some use was any MA use <3 days per week across pregnancy. The pattern of use according to trimester () indicated overall decline and quitting MA use over the course of the pregnancy. However, declining and quitting MA use occurred sooner in the some use versus heavy use group (first to second trimester, P
= .001 in both cases). Eighteen heavy users (62%) used in all 3 trimesters versus 21 (16%) some users (P
Frequency of Self-Reported MA Use According to Trimester of Pregnancy
Level of exposure to other drugs of abuse was calculated as cigarettes per day, ounces of absolute alcohol per day, and joints per day for marijuana across pregnancy.35,36
Postnatal use of MA, alcohol, tobacco, and marijuana (yes/no for each drug) was similarly measured at 1, 2, 3, and 5 years.
General linear models were used to compare means for continuous variables and χ2
tests for categorical variables (–). General linear mixed models (SAS PROC MIXED, version 9.1.3) tested the effects of MA exposure (any exposure; level of MA exposure) and longitudinal CBCL measures at ages 3 and 5 years and the interaction of MA exposure and age, adjusted for covariates ( and ). Level-of-use models included separate tests for heavy and some MA exposure versus the comparison group. Interactions of covariates and MA exposure or level of MA exposure were examined but did not meet criteria for inclusion (P
> .05). There were 16% and 13% missing values for the Home Observation for Measurement of the Environment scale and domestic violence, respectively. We therefore applied multiple imputation37,38
by using SAS PROC MI and MIAnalyze. The results were very similar to analysis without imputation. The final model from the imputed dataset was used to retain the largest sample size.
Sample Characteristics According to MA Exposure
Behavior Problems Scores According to Prenatal MA Exposure
CBCL According to Level of MA Exposure
A priori covariates included prenatal exposure to alcohol, tobacco, and marijuana; gender; SES; and birth weight. Any exposure to alcohol, tobacco, or marijuana exposure was included in analyses of any MA exposure, with level of exposure included in analyses of heavy MA exposure. Other variables were examined for inclusion as covariates on the basis of published literature and characteristics that differed between exposure groups (P
< .05) if not highly correlated with other covariates (r
= 0.70). Covariates measured at multiple time points were averaged (eg, caregiver psychological symptoms) or aggregated over time (eg, any postnatal tobacco use at 1–5 years) to provide the best estimate of the childrearing environment to date. Covariates were included if associated with any of the outcomes (P
< .15). All models were adjusted for prenatal tobacco, alcohol, and marijuana exposures; birth weight; gender; low SES; maternal age; no partner; primary caregiver change; domestic violence; postnatal caregiver use of MA; alcohol, tobacco, and marijuana exposure; caregiver psychological symptoms; quality of the home; and reported child abuse. Continuous covariates (eg, maternal age) were grand mean centered to increase precision and interpretation of the intercept. Subjects were nested in site to account for the correlations among the subjects from the same site and to provide more accurate estimates.39