Objective. The objectives of this study were to estimate the effect of prenatal cocaine exposure on fetal growth and gestational age after controlling for exposure to alcohol, tobacco, and marijuana and other covariates; to evaluate whether prenatal cocaine exposure has a disproportionate adverse effect on head circumference compared with overall somatic growth; and to assess whether the effect of prenatal cocaine exposure on fetal growth is mediated by cocaine’s suspected effect on gestational age.
Methods. The study population includes 476 neonates participating in the Miami Prenatal Cocaine Study, a longitudinal follow-up of in utero cocaine exposure. The sample, restricted to full-term neonates born to African-American inner-city mothers, included 253 infants exposed prenatally to cocaine (with or without alcohol, tobacco, or marijuana exposure) and 223 non–cocaine-exposed infants, of whom 147 were drug-free and 76 were exposed to varying combinations of alcohol, tobacco, or marijuana.
Results. Evidence based on structural equations and multiple regression models supports a hypothesis of cocaine-associated fetal growth deficits (0.63 standard deviation) and an independent mild effect on gestational age (0.33 standard deviation). There was no evidence of a disproportionate adverse effect on birth head circumference once the impact on overall growth was estimated. There was evidence that some but not all of the cocaine effect on fetal growth was direct and some was indirect, acting via an intermediate influence of cocaine on gestational age.
Conclusions. Cocaine-associated growth deficits, symmetrical and partially mediated by gestational age, were observed in this sample of inner-city African-American full-term infants prospectively enrolled at birth. Long-term implications will be the subject of future reports from this longitudinal investigation.
Prenatal cocaine exposure has been a topic of scientific controversy for more than a decade. Initial media coverage of the “cocaine infant” epidemic was prompted by a few sentinel reports that, in retrospect, may have led to an overestimation of risk. After admonitions about an apparent bias against the null hypothesis,1 a possible rush to judgment,2 and prejudicial labeling of cocaine-exposed offspring as “crack kids,”3 the pendulum of scientific opinion shifted in the opposite direction. Currently, the perspective is more balanced, with an appreciation that in utero cocaine exposure seems to produce a continuum of reproductive casualty in individual cases, ranging from severe complications to subtle effects of no immediately appreciable clinical significance. As noted by Lester et al,4 even subtle effects may have significant societal impact and costs when assessed from a public health perspective.
One of the central issues in the discourse regarding in utero cocaine exposure is the potential impact on fetal growth and maturation. Here, the potential public health significance is elevated because low birth weight is a known determinant of infant morbidity and mortality.5 In addition, cocaine-associated deficits in head circumference may have long-term neurodevelopmental and behavioral sequelae. A large body of evidence has been published indicating that in utero cocaine exposure is associated with modest to moderate deficits in fetal growth. Evidence is drawn from animal models,6 clinical studies of human neonates as reviewed elsewhere,7,8,9,10 and meta-analyses.11,,12
This report provides an introduction to the study design of the Miami Prenatal Cocaine Study (PCS), a longitudinal investigation of the effects of in utero cocaine exposure in a sample of full-term, inner-city African-American neonates enrolled at birth. The study addresses prenatal cocaine exposure’s effect on fetal growth and gestational age in full-term infants and illuminates 3 important ancillary issues not well represented in the literature on physical growth outcomes. First, the potential confounding influences of concurrent exposure to other substances and other relevant maternal and infant characteristics are considered within a multivariate approach to estimating the influence of prenatal cocaine exposure. Second, the hypothesis that prenatal cocaine exposure is associated with a disproportionate adverse effect on head circumference compared with somatic growth, as suggested by Little and Snell,13 is explored. Third, the mediating influence of gestational age on the relationship between prenatal cocaine exposure and fetal growth, as reported by Jacobson et al in a study of preterm and term infants,14 is investigated. The latter issue may be relevant even in a study restricted to full-term infants because of the well-known correlation between gestational age and birth growth parameters.
The Miami PCS sample is large enough to allow the use of structural equation and measurement models with the capacity to shed new light on the estimation of prenatal cocaine exposure as a determinant of fetal growth and, independently, gestational age. Within this framework it is possible to examine each of the abovementioned subsidiary issues using multivariate statistical procedures not previously applied in this context. Statistical adjustments are included for potentially confounding characteristics that covary with prenatal cocaine exposure (eg, prenatal exposure to other drugs such as alcohol, tobacco, and marijuana) and individual characteristics that might affect fetal growth (eg, maternal height, primigravida). Complementing the study protocol’s restriction in range of variability (eg, gestational age, social disadvantage, and race or ethnicity), attention to statistical adjustments to discern confounding provides a more definitive set of estimates of cocaine-associated fetal growth deficits. With respect to possible asymmetrical effects on head circumference, the use of multivariate methods is important because measures of fetal growth (birth weight, length, and head circumference) are highly interdependent, in part because these indicators share some of the same causal determinants (eg, infant sex, maternal nutritional status). In previous research, indicators of fetal growth were studied individually as if they were independent, uncorrelated responses. In the current study, a multivariate response model provides an estimate of prenatal cocaine exposure’s impact on fetal growth and also evaluates the differential influence of prenatal cocaine exposure on head circumference beyond the general impact on fetal growth.
This report focuses on the hypothesized impact of prenatal cocaine exposure on birth growth measurements and gestational age. The Miami PCS was conducted in 2 overlapping phases: a phase I maternal survey and a phase II follow-up study. The follow-up sample (N = 476) was recruited between November 1990 and July 1993. Of these, 436 (92%) of the mother–infant pairs were recruited during a 2-year period beginning March 1991 in which consecutive African-American deliveries from specified zip codes were intensively screened. An additional 40 were recruited in the months immediately before and after the intensive screening using abbreviated screening procedures but full enrollment interviews and bioassays. The follow-up sample was assessed at birth and followed at 1, 4, 8, 12, 18, and 24 months and 3, 5, and 7 years. Sample retention for research visits through age 7 is 87%, primarily because of retention methods previously described.15 The study was conducted in accordance with procedures approved by the institutional review board. Informed consent procedures emphasized confidentiality, with support of a federal Department of Health and Human Services Certificate of Confidentiality. Separate informed consents were used for the survey interview and toxicology assays, the follow-up study, and human immunodeficiency virus (HIV) testing.
Participants were recruited from among deliveries at the University of Miami School of Medicine Jackson Memorial Medical Center, the Miami–Dade County Public Health Trust’s main delivery hospital. To minimize the number of confounding variables and improve statistical power, a homogeneous study population was recruited with regard to full-term gestational age (≥37 completed weeks), low socioeconomic status, inner-city residence, and African-American race and ethnicity. Exclusion criteria, established to limit the potential confounding influence of developmental risk factors, included maternal HIV or acquired immunodeficiency syndrome, prenatal exposure to opiates, methadone, amphetamines, barbiturates, benzodiazepines, or phencyclidine; major congenital malformation; chromosomal aberration; or disseminated congenital infection. Mothers who could not complete the informed consent process or interview because of mental or special sensory impairment were also excluded.
During phase I, the research team identified from labor and delivery admission records 2651 full-term live births delivered to US-born mothers self-described as African-American and residing in the designated geographic area. Twins were treated as a single analytic unit by random selection of 1 twin’s data. To avoid duplicate sampling of sociodemographic and medical characteristics from the same mother, follow-up recruitment was limited to 1 term infant from each mother, resulting in elimination of 171 siblings. It was not possible to obtain informed consent and conduct the enrollment interview for 383 potentially eligible mothers for the following reasons: maternal mental retardation, psychopathology, deafness, or general anesthesia recovery (N = 16); early maternal discharge (N = 149); unpredictable delivery peaks necessitating case exclusion using random number tables (N = 95); Hurricane Andrew and its aftermath (N = 42); inadvertent errors in establishing eligibility (N = 35); holidays (N = 28); interference with patient care and family visitation (N = 13); and enrollment conflict with another research protocol (N = 5). Research staff approached 2097 (85%) of the potentially eligible mothers for consent. Of these, 1505 (72%) agreed to participate in the survey and follow-up, if deemed eligible. Mothers who participated in the survey were somewhat younger than those who did not (mean ± standard deviation [SD], 23.6 ± 5.7 vs 24.9 ± 5.9; P < .05). They also were more likely to be primigravida (25.4% vs 19.5%, P < .001) and to have received prenatal care (93.9% vs 91%, P = .007). Participating infants were more likely to have Apgar scores ≤7 in contrast to nonparticipating infants (19.7% vs 15.8%, P = .018).
Of the 1505 mothers participating in the survey, 154 were found to be ineligible for follow-up because of positive or unknown maternal HIV status, and 25 additional participants were excluded for exposure to heroin or other drugs of abuse other than the study drugs of interest. From the remaining 1326, participants were selected for the longitudinal follow-up study. The final follow-up sample consisted of 253 cocaine-exposed and 223 non–cocaine-exposed comparison infants, of whom 147 were drug-free and 76 were exposed to alcohol, tobacco, or marijuana (ATM).
Birth weight was obtained immediately after delivery by the nursing staff of the admission nursery. During enrollment, the measurements of occipital–frontal head circumference and length were performed by trained research nurses assisted by a second operator who positioned the supine infant in an optimally extended horizontal position on a flat, firm surface. The mean of 3 consecutive measurements obtained by a standard metric tape measure was used for each parameter. Gestational age was assessed using the Ballard modification of the Dubowitz examination.16 If the obstetric estimate of gestational age by maternal dates corresponded within 2 weeks of the physical assessment, the obstetric estimate was accepted. However, if maternal dates were unavailable or a 3-week or greater discrepancy was found between the 2 estimates, the Ballard assessment was used. Maternal and infant medical and demographic information, including maternal age, gravida and parity, prenatal care history, Apgar scores, and other birth outcomes, was obtained by medical record review. Gestational age was assessed and anthropometric measurements were taken before the maternal interview, clinical record review, and bioassays were conducted to ensure that the examiner was unaware of exposure status.
Mothers were interviewed within the first 36 hours postpartum by a member of the research staff using a structured standardized interview format covering maternal substance use; pregnancy, educational level, employment, and criminal history; sexual and drug use habits; and additional demographic information. Interviewers were trained to use community slang for different drugs when appropriate and to engage mothers in a nonjudgmental manner, emphasizing confidentiality. The interview included questions about a number of commonly used drugs (caffeine, cigarettes, snuff, beer, wine, liquor, marijuana or hashish, speedball, cocaine, crack, heroin, methadone, other opiates, tranquilizers, amphetamines, methamphetamine or ice, hallucinogens, and inhalants) and various routes of administration.
Substance use questions were structured to assess drug use patterns during the mother’s lifetime, during the 3 months before the current pregnancy, and during each trimester of pregnancy. The interview instrument was modified for postpartum use from a structured interview format used in a National Institute on Drug Abuse–funded longitudinal investigation of cocaine use and acquisition of HIV infection, directed by 1 of the coauthors (D.C.C.), who led the interviewers’ training for the current study. To enhance timeline recall, mothers were presented with a calendar. Each recall period was outlined and anchored to important dates such as birthdays and holidays. Questions about drug use during each time period included number of weeks used, most days per week, fewest days per week, usual days per week, and usual dosage per day. Dosage was recorded in number of cigarettes smoked per day (regardless of brand), number of marijuana joints smoked per day, and number of standard drinks consumed per day. Standard drinks for each type of alcohol (ie, beer 12 oz, wine 5 oz, and liquor 1.5 oz) were defined according to Schneiderman.17 Cocaine dosage was recorded in number of rocks of crack cocaine or lines of powder cocaine used per day. To calculate total drug usage composites for each drug for the prepregnancy, trimester, and total pregnancy time periods, product terms were calculated by multiplying the usual dosage per day by the usual days per week by the number of weeks used in each period of interest.
Biological Markers (Urine and Meconium)
Screening of maternal and infant urine and meconium for cocaine metabolite (benzoylecgonine) was performed with a standard enzyme-mediated immunoassay technique (EMIT, Syva Corporation, San Jose, CA), at a cutoff of 150 ng/mL urine and 150 ng/g meconium, respectively. EMIT was chosen for screening because it is reasonably inexpensive, is readily available in the clinical setting, and provides a prompt identification of participants. Ostrea et al18 reported the reliability of meconium screening for cocaine. EMIT-negative meconium specimens from infants selected for follow-up were dual screened with EMIT and Coat-a-Count Cocaine Metabolite (Diagnostic Products Corporation), a solid phase radioimmunoassay. Gas chromatography/mass spectrometry (GC/MS) was used to confirm cocaine-positive results in urine and meconium.19 Meconium specimens were also assayed by EMIT for marijuana and opiates. Urine specimens were also screened by EMIT with GC/MS confirmation for marijuana (cannabinoids), opiates, amphetamines, barbiturates, benzodiazepines, and phencyclidine. Urine results usually were reported within 24 hours of specimen collection, but meconium results took several days.
Of the total follow-up cohort, 100% had at least 1 biological marker, 96% had at least 2 biological markers, and 68% had all 3 biological markers. Infant urine was available for 98% and maternal urine was available for 79% of the total sample, with no differences between the groups. Infant meconium was available for 86% of the total sample, with a slightly higher collection rate in the drug-free group (95%) than in the ATM group (86%) and the cocaine group (82%).
Follow-Up Group Assignment
Based on maternal interview and biological markers, infants were categorized as follows: cocaine (documented exposure to cocaine with or without exposure to alcohol, tobacco, or marijuana), ATM (no known exposure to cocaine but documented exposure to alcohol, tobacco, or marijuana) or drug-free (no known exposure to alcohol, tobacco, marijuana, or cocaine). Whereas exposure to alcohol and tobacco was documented by maternal self-report, exposure to marijuana and cocaine was documented by maternal self-report and 1 or more positive biological markers.
Assignment to the drug-free group depended on a negative maternal lifetime history for cocaine use, negative self-report during pregnancy and the 3 months before pregnancy, and negative toxicology results on all available specimens. A random numbers table was used to select drug-free infants for follow-up to balance the recruitment across the study period. When a positive biological marker for marijuana or cocaine contradicted preliminary assignment to the drug-free group, the infant was reassigned to the appropriate group (ie, ATM or cocaine).
To qualify for the ATM group, the mother had to have a positive self-report for alcohol, tobacco, or marijuana use during pregnancy or a positive biological marker for marijuana. In addition, she had to have a negative self-report of cocaine during pregnancy and the 3 months before pregnancy and cocaine-negative results on all available specimens. Emphasis was placed on recruiting mothers who met an arbitrary minimum ingestion criterion of 20 drinks of alcohol during pregnancy. Eighty percent of the non-cocaine-using mothers who met that criterion were enrolled in the ATM group. When a cocaine-positive biological marker contradicted preliminary assignment to the ATM group, the infant was reassigned to the cocaine group.
Assignment to the cocaine group was based on documentation of positive maternal self-report of cocaine use during pregnancy or at least 1 GC/MS-confirmed cocaine-positive biological marker. In 40 cases, positive maternal self-report was the only documentation of prenatal cocaine exposure. In 80 cases, the mother denied cocaine use during pregnancy, but 1 or more biological markers were cocaine positive, confirmed by GC/MS. Of the eligible cocaine-exposed mothers and infants, 89% were engaged in follow-up.
The first step in the analysis was examination of the distribution of each variable, followed by creation of standardized response variables for birth weight, length, head circumference, and gestational age via centering (subtracting the sample mean from each value) and then division by the standard deviation. To promote interpretation, several covariates (eg, maternal age, education, maternal height) also were centered or standardized. Centering can reduce the statistical problem of multicollinearity and promotes interpretation by bringing the zero value into the range of each covariate’s distribution.
Latent structure analysis showed that birth weight, length, and head circumference serve well as intercorrelated manifestations of a single underlying fetal growth dimension. The interdependency of the fetal growth and gestational age measures motivated use of a multivariate response model allowing estimation of the magnitude of difference between each group in relation to levels of the underlying fetal growth dimension and gestational age after covariate control; testing whether prenatal cocaine exposure might have a specific association with neonatal head circumference, within the context of a statistical model that holds constant the overall levels of fetal growth and gestational age and the hypothesized impact of cocaine exposure on the global dimension of fetal growth; and evaluating whether a cocaine effect on gestational age might mediate the relationship between prenatal cocaine exposure and fetal growth using method outlined by Baron and Kenny.20 The Multiple Indicators, Multiple Causes (MIMIC) model served well in this context and was used to estimate prenatal cocaine exposure’s hypothesized effects on head circumference via 2 pathways. One pathway runs indirectly through a general attenuation of fetal growth, and the other runs directly to head circumference specifically. A technical description of the MIMIC model, implementing software, and examples of previous use in epidemiology have been given by Muthén,21 Muthén and Muthén,22 and Gallo et al.23,,24 Mplus software22 was used for all MIMIC analyses. Mplus uses numerical techniques based on the expectation maximization algorithm when there is need to compensate for missing data, with an unrestricted model for the mean vector and covariance matrix when data are missing at random (Option H1 Missing). The Mplus slope estimates for prenatal cocaine exposure impact did not differ appreciably with and without the H1 option, and the overall study conclusions were unaffected. The estimates presented herein are without the EM algorithm methods for imputation of missing data.
Using the general linear model, 4 univariate response regression models also were fit to the study data, 1 model for each indicator of fetal growth and gestational age. Unlike multivariate response analyses, these univariate response analyses do not address interdependency of the indicators of fetal growth and gestational age, but they help clarify whether and by how much each individual indicator might have been influenced by prenatal cocaine exposure. Stata Version 6 software25 was used for all univariate response regressions. In both the multivariate and univariate response models, emphasis was placed on estimating the regression slopes, which indicate differences between the cocaine-exposed and the non–cocaine-exposed infants and the 95% confidence intervals for these differences, with P values used solely as an aid to interpretation. The multivariate response models served as a primary focus of the analysis, with the univariate response models used to clarify loci of prenatal cocaine effect within the multivariate profile.
A latent structure regression analysis is especially useful when there are multiple highly intercorrelated or interdependent response variables (eg, birth weight and birth length are correlated in this sample, r = 0.77; see Technical Appendix A). The degree of interdependency is an indication of a shared source of common covariation, and each response variable’s variation can be partitioned into that which is shared in common with the other response variables and that which is not shared. Then, the response variables can be expressed as a function of a linear combination of covariates, with multiple regression used to estimate the influence of each covariate on the response, with all other covariates held constant. In this context, the latent structure regression analysis also allows the estimation of whether cocaine exposure might have had an influence on birth head circumference beyond a more general influence on fetal growth. In all these regression analyses, all slopes are estimated simultaneously, with each slope estimated while all other listed covariates are held constant. As demonstrated in the study findings, the statistical power is adequate for detecting prenatal cocaine effects of modest size, even with statistical adjustment for multiple covariates.
Tables 1 and 2 present selected characteristics of the 476 mothers and full-term neonates in the study sample, respectively, sorted by group. Specifically, cocaine-using mothers were significantly older and less likely to be primigravida than mothers in the ATM and drug-free groups. In addition, ATM-using mothers were significantly older and less likely to be primigravida than were drug-free mothers. Marital status and maternal education did not differ between groups. Cocaine-using mothers were less likely to be employed outside the home than were drug-free and ATM-using mothers. Cocaine-using mothers and ATM-using mothers were less likely to have received prenatal care than were mothers in the drug-free group. Maternal height did not differ between groups, but mothers in the cocaine group and ATM group had lower prepregnancy body mass indices than drug-free mothers. Pregnancy weight gain was significantly lower in the cocaine-using mothers than in the drug-free and ATM groups. Cocaine-exposed infants were significantly smaller in all 3 measures of fetal growth, were less mature in terms of gestational age, and were more often low birth weight (<2500 g) than were the infants in the other 2 groups. There were no significant differences between the 3 groups with respect to infant sex or Apgar scores.
Maternal Self-Report of Prenatal Substance Use
Table 3 compares the proportion of mothers reporting alcohol, tobacco, or marijuana use during pregnancy between the cocaine-exposed and the ATM groups. The amounts of self-reported alcohol, tobacco, or marijuana use during each trimester and the entire pregnancy are also included. All but 18 of the cocaine-using mothers also reported alcohol, tobacco, or marijuana use during pregnancy. Although the proportion of mothers reporting alcohol consumption was greater in the ATM group because of recruiting emphasis, the median amount of alcohol ingested by alcohol-using mothers in each group was not significantly different. Compared with the ATM group, the cocaine group had a higher proportion of cigarette smokers and a higher median amount of cigarette smoking. There was also a trend toward a higher proportion of marijuana smokers in the cocaine group than in the ATM group, but the median amount of marijuana use reported by those partaking did not differ between groups.
Self-reported cocaine use is also shown in Table 3. More than two thirds (N = 173) of the 253 mothers in the cocaine group had a positive self-report of powdered cocaine or crack use during pregnancy. Of the 173 self-admitted users, 93 snorted cocaine and 116 smoked crack; 36 admitted to using both substances. None of the mothers reported intravenous administration of cocaine or any other drugs.
Multivariate Response Analyses
The 3 fetal growth indicators were used to measure the previously described underlying dimension of fetal growth. The fetal growth construct and gestational age were simultaneously regressed on prenatal cocaine exposure and other suspected influences that might be imbalanced across cocaine-exposed and nonexposed groups. Of note, with inclusion of terms for levels of prenatal alcohol, tobacco, and marijuana exposure in the model, there is no separate dummy-coded indicator for the group of neonates exposed prenatally to drugs other than cocaine. Once the values of the 4 drug exposures are known (ie, cocaine, tobacco, alcohol, and marijuana), group status can be determined, and the indicator term for prenatal exposure to drugs other than cocaine becomes redundant.
Estimates shown in Fig 1 summarize in a single diagram how fetal growth, as reflected by birth weight, length, and head circumference, depended on prenatal cocaine exposure and other suspected determinants. Each of the slope estimates reported in Fig 1 and Table 4 is a result of a simultaneous multiple regression analysis in which each estimate is adjusted for relationships involving all other estimates, including the correlation between gestational age and fetal growth, as shown on the right-hand side of Fig 1 (r = 0.347). That is, when estimating the effect of cocaine exposure, all other listed covariates are held constant. The Fig 1 and Table 4 estimates show that the cocaine-exposed group was an estimated 63% SD lower than non–cocaine-exposed neonates on the summary fetal growth dimension (difference between cocaine-exposed and nonexposed neonates estimated in units of standard deviations of the fetal growth dimension, D = −0.628; 95% confidence interval [CI]: −0.823, −0.435; P < .001). Values shown in the figure are the estimated slopes (D) for the covariates (maternal alcohol, tobacco, and marijuana use, infant sex, maternal age, maternal height, and primigravida status) and the P value associated with each slope (in parentheses). The estimated effect of cocaine on fetal growth did not change appreciably when adjusted for maternal marital status, education, current employment, prepregnancy body mass index, pregnancy weight gain, prenatal care visits, and Apgar scores.
Although not shown in figure or table, a much-simplified multivariate response model indicated that the observed attenuation of growth is connected to membership in the cocaine group but not the ATM group. In this model, the fetal growth construct was regressed on 2 dummy-coded indicator terms, 1 for each drug-exposed group, with drug-free neonates serving as a reference or comparison group. This 3-group multivariate analysis produced the following estimates for the cocaine-exposed group (D = −0.713; 95% CI: −0.903, −0.523; P < .0001) and the ATM group (D = −0.141; 95% CI: −0.400, 0.118; P = .286). Specifically, the ATM group was not different from the drug-free comparison group with respect to fetal growth, the size of the estimated difference was much smaller than that estimated for the cocaine group, and the ATM and cocaine-exposed groups were significantly different at P < .05. The same general pattern is present with respect to gestational age: Cocaine-exposed neonates had slightly lower gestational ages than did drug-free comparison neonates (D = −0.208; 95% CI: −0.410, −0.006; P = .043), whereas ATM-exposed neonates were not different from the drug-free group (D = −0.126; 95% CI: −0.400, 0.148; P = .366). However, the ATM and cocaine-exposed infants were not different with respect to gestational age (P > .10).
The possibility that cocaine exposure might have a paradoxically asymmetrical impact on head circumference was examined by elaborating the MIMIC model. A direct path from prenatal cocaine exposure to head circumference was included in addition to an indirect path leading from prenatal cocaine exposure through the fetal growth construct and onward to head circumference. The multivariate response regression model was used to hold constant the level of fetal growth and take into account the estimated effects of prenatal cocaine exposure on the latent fetal growth construct. The evidence indicates essentially no additional or specific relationship between prenatal cocaine exposure and birth head circumference (estimated head circumference difference between cocaine-exposed group and comparison group, D = 0.038 SD; 95% CI: 0.068, 0.562; P = .574).
The conceptual model represented in Fig 1 was also amended to test for a possible mediational pathway leading from prenatal cocaine exposure to fetal growth deficits, with an intermediate step involving gestational age. In keeping with the logic of Baron and Kenny,20 the observed link from prenatal cocaine exposure to fetal growth deficits meets the first of the 3 conditions that lend support to an inference of mediation. The observed link from prenatal cocaine exposure to shorter gestational age is the second of the 3 conditions. In this study, the third condition is whether the cocaine–growth relationship is attenuated with statistical adjustment for fetal maturation as measured by gestational age. This third condition was evaluated by respecifying the conceptual model shown in Fig 1 so that there is a path from the gestational age term to the fetal growth term (double-headed arrow with r = 0.347). Results from this respecification of the model support the hypothesis that a small portion of cocaine’s influence on fetal growth may be mediated via an intermediate impact on gestational age. When the intermediate path is added, the estimated direct cocaine–growth relationship is reduced from the Fig 1 value of −0.628 to a value of −0.509 (95% CI: −0.323, −0.695; P < .001). Therefore, it may appear that some but not all of the cocaine-associated fetal growth deficit is mediated by gestational age. Also, there is a statistically significant path running from gestational age to fetal growth (D = 0.361 SD; 95% CI: 0.283, 0.439; P < .001; data not shown in figure or table).
Univariate Response Regression Models
Model 1 sets up a simple unadjusted contrast between the 253 cocaine-exposed neonates and the 147 drug-free comparison neonates. Model 2 contrasts the 253 cocaine-exposed infants to the 223 non–cocaine-exposed infants (including the ATM group with the drug-free group), with statistical adjustment for prenatal alcohol, tobacco, and marijuana exposure, infant sex, maternal age, maternal height, and primigravida status. Model 3 extends Model 2 with statistical adjustment for the following additional covariates: maternal marital status, educational level, employment, prenatal care visits, prepregnancy body mass index, prenatal weight gain, and Apgar scores.
Model 1 in Table 5 summarizes the 4 univariate response regression models before statistical adjustment for covariates and indicates statistically significant differences between the cocaine-exposed and drug-free comparison neonates with respect to all individual fetal growth and gestational age indicators. The difference (D) between cocaine-exposed and drug-free neonates on birth head circumference was almost 0.5 SD (D = −0.475; P < .001). Corresponding differences for birth weight and length were more than two thirds of a standard deviation (D = −0.699; P < .001 and D = −0.733; P < .001, respectively). In regard to gestational age, full-term infants exposed to cocaine were approximately one fifth of a standard deviation less mature than the drug-free infants (D = −0.208; P = .044).
Models 2 and 3 (Table 5) compare the cocaine-exposed infants and nonexposed infants (drug-free and ATM infants), with statistical adjustments for other suspected determinants. Point estimates for fetal growth difference between cocaine-exposed and nonexposed infants fall within a range from 0.424 to 0.631 SD on all 3 growth measures, with the cocaine-exposed smaller than the nonexposed. For gestational age of these full-term neonates, the difference is of lesser magnitude, with model 3 showing cocaine-exposed neonates delivered about one fourth of a standard deviation earlier than nonexposed neonates (D = −0.245; P = .052).
Whereas the primary focus of this research is the effects of prenatal cocaine exposure, Fig 1 and Table 4 provide an overview summary with respect to each of the covariates included in the study’s multivariate and univariate response models. It is noteworthy that fetal growth depends not only on prenatal exposure to cocaine (D = −0.628; P < .001) but also on prenatal exposure to tobacco (D = −0.006; P = .038), male infant sex (D = 0.172; P = .045), and greater maternal height (D = 0.179; P < .001). In addition, gestational age is associated with prenatal exposure to cocaine (D = −0.330; P = .003) and prenatal tobacco exposure (D = 0.008; P = .010).
Evidence from the Miami PCS, consistent with previous published research, supports an inference that prenatal cocaine exposure is associated with deficits in fetal growth in full-term infants, deficits not seen in the drug-free neonates or those exposed prenatally to alcohol, tobacco, or marijuana but not to cocaine. According to estimates from multivariate response analyses, this cocaine-associated deficit is nearly two thirds (63%) of a standard deviation for fetal growth and about one third (33%) of a standard deviation for gestational age. The study’s univariate response analyses show consistent cocaine-associated deficits of similar magnitude in relation to each of the indicators (ie, birth weight, length, and head circumference and gestational age).
The current study substantiates a moderate cocaine-specific deficit in fetal growth in a large sample of full-term infants and addresses several additional issues not well represented in the literature. Perhaps most important is the statistical adjustment for numerous covariates within a multivariate model, including the mother’s alcohol, tobacco, and marijuana use during pregnancy. Results from these analyses support an inference that the observed cocaine-associated deficits in growth and gestational age are linked to cocaine exposure and not to lifestyle or background characteristics. The hypothesis that prenatal cocaine exposure might have an additional and more selective effect on head circumference, as suggested by other investigations,13,,26,,27 was not supported once the more general impact on fetal growth as an interdependent construct was considered. Results from the present study are the first to indicate that some but not all of cocaine’s effect on fetal growth is mediated through gestational age in a population of full-term infants. In a study of healthy full-term infants, Coles et al28 detected a statistically but not clinically significant reduction in the gestational age of cocaine-exposed infants, but a possible mediational influence between gestational age and fetal growth was not assessed. With the exception of the previously cited study14 that included preterm as well as term infants, most investigations of prenatal cocaine exposure have not specifically addressed this issue.
The magnitude of the cocaine-associated deficit, estimated in this study to be approximately 63% SD on the underlying dimension of fetal growth and an estimated 33% SD with respect to gestational age, can be compared with other determinants. In this study, the estimated difference between male and female neonates in the sample was an estimated 17% SD on the fetal growth dimension. There was also an estimated 12% SD difference in fetal growth between newborns of mothers reporting smoking about 2000 cigarettes (100 packs) during pregnancy and newborns of mothers who did not smoke prenatally. From the standpoint of the clinical significance of the cocaine-associated fetal growth deficit, the estimated effect of prenatal cocaine exposure in this study is roughly equivalent to smoking 10 000 or more tobacco cigarettes (ie, approximately 500 packs, or 1¾ packs daily throughout a full-term gestation).
Potential Mechanisms of Fetal Growth Deficits in Cocaine-Exposed Infants
Studies in the Long–Evans rat have shown that prenatal cocaine exposure results in dose-dependent decreases in birth weight and postnatal growth and alterations in fetal body composition with lower levels of body fat, protein, and calcium.6 In these studies, there is also an observed dose-dependent suppression of maternal weight gain and food consumption. The authors suggest that the reduction in fetal protein and body fat content results from reduced synthesis secondary to decreased placental transport of fatty acids and amino acids. They also attribute the reduced fetal calcium to decreased placental transport secondary to decreased maternal consumption, uterine vasoconstriction, and possibly calcium ion chelation.
Cocaine-associated maternal undernutrition and weight gain during pregnancy also appear to affect fetal growth in humans.14,,29 Frank et al30 also found that in utero cocaine exposure was associated with indicators of diminished lean body mass and depressed neonatal fat stores. Because this finding was present after statistically controlling for maternal weight for height at conception and pregnancy weight gain, the authors speculate that cocaine may also adversely affect intrauterine growth by vasoconstrictive impairment of nutrient transfer to the fetus or depletion of fetal nutrient stores.
Although the causes of growth retardation often are complex and multifactorial,10 the literature gives reason to surmise that prenatal cocaine exposure is a significant contributor to fetal growth impairment in humans.28,29,30,31,32,33,34,35,36,37 These reports, some with small sample sizes, have relied on limited univariate and multivariate analyses for control of confounding factors. The current report substantiates many of these previous findings, using a multivariate structural equation modeling approach that takes into consideration the interdependencies between covariates.
In an early meta-analytic result of 20 methodologically defined articles on prenatal cocaine exposure and pregnancy outcome, Lutiger et al11 showed that the results of analysis of confounding variables for growth (ie, birth weight, length, and head circumference) depended on the nature of the comparison. Comparison of cocaine-using and drug-free mothers provided a medium effect size, whereas the comparison between polydrug cocaine-using mothers and polydrug non–cocaine-using mothers yielded small to nonexistent effect sizes. Hurt et al38 reported that gestational age–adjusted birth weight and head circumference in cocaine-exposed infants were significantly smaller than in the control infants, as determined by standard growth percentiles. However, after controlling for cigarette smoking, prenatal care, and gravidity, cocaine exposure had no significant relationship to weight or head circumference. Jacobson et al14 noted that alcohol, smoking, opiate use, and cocaine use each correlated with smaller birth weight, birth length, and head circumference, but in a multivariate model, birth weight correlated only to alcohol and smoking, length only to alcohol, and head circumference only to opiates. They concluded that cocaine’s effect on birth weight is predominantly an indirect consequence of shorter gestation and poorer maternal nutrition. Eyler et al39 reported a significant negative association between prenatal cocaine use and birth weight and length, but it was a significant interaction effect such that head and chest circumferences were significantly smaller in the offspring of cocaine users who also smoked tobacco. A meta-analysis of 11 identified studies controlling for the effect of tobacco smoking led to the conclusion that maternal cocaine use during pregnancy has a moderately strong association with low birth weight.12
Growth Symmetry Versus Asymmetry
Previous investigations, with varying study designs and approaches to covariate controls, have found that prenatal cocaine exposure is associated with diminished head circumference. Microcephaly has been specifically identified among offspring of cocaine-abusing mothers.30,,33,,35,,40,41,42,43 When the pattern of intrauterine growth has been specified, the term most often associated with prenatal cocaine exposure is “symmetric growth retardation,”29,,30,,37 in contrast to so-called asymmetric growth retardation, reflecting deficits in somatic growth with relative sparing of head circumference. However, Little and Snell13 reported a more ominous paradoxical or reverse asymmetry involving disproportionate diminution of head circumference in relation to birth weight among cocaine-exposed infants. Sallee et al26 noted a disproportionate decrease in head circumference percentile compared with birth weight percentile in infants exposed to cocaine, detected by positive maternal urine and positive infant hair levels of benzoylecgonine. Bateman and Chiriboga27 also described a brain wasting pattern of asymmetrical growth retardation in newborns exposed to the highest level of cocaine exposure as documented by maternal hair benzoylecgonine.
In the current study, the MIMIC analyses, used here for the first time in research on prenatal cocaine use, represent novel applications of psychometric methods first introduced more than 15 years ago.21 These methods produced evidence that prenatal cocaine exposure might affect head circumference, birth weight, and length via a more generalized indirect path involving the latent dimension of fetal growth. If prenatal cocaine exposure were to exert a disproportionately pronounced toxic effect on brain growth, the evidence should show differentially reduced head circumference relative to other growth parameters. Using this approach, the evidence in this study does not support the hypothesis of a direct effect on head circumference beyond the more general effect on overall fetal growth. This result indicates no asymmetry in cocaine effects but does not contradict the study evidence on cocaine-associated deficit in relation to each of the individual birth growth parameters, including head circumference, shown in the univariate response models in Table 5.
Several limitations and features of this study merit attention. First, because the enrolled neonates were full-term, this study’s findings do not speak to issues related to prematurity. Second, the restriction of the study sample to African-American infants born to mothers residing in socially disadvantaged inner-city neighborhoods leads to an associated restriction of inference. The estimated between-group differences associated with cocaine exposure might differ elsewhere (eg, in less disadvantaged areas with greater access and use of prenatal care that might counterbalance some adverse effects associated with cocaine use). Third, the maternal self-report measure and bioassays were taken during the delivery hospitalization. Although interviewed by skilled interviewers, some mothers may be reluctant to divulge sensitive information or may have difficulty recalling specifics of drug-taking behaviors during pregnancy. Biological markers (eg, urine and meconium) provide complementary data but cannot supplant maternal self-report because urine reflects only very recent drug use and meconium reflects an unknown number of weeks to months before delivery. The resulting assessment of cocaine exposure may yield unknown misclassification errors of cocaine-exposed infants to the non–cocaine-exposed groups. Of course, misclassification errors tend to make groups more similar to one another, with resulting bias of the cocaine estimates toward the null hypothesis.
Some observers might consider the use of latent structure regression models a limitation of this study. However, in a study of possible adverse responses to prenatal cocaine exposure, because a randomized experiment with systematic replication is not possible, one must rely on study design rules (eg, exclusions, matching) and statistical adjustments to compensate for potentially imbalanced determinants of the adverse consequences. Therefore, multiple regression analysis is essential to hold constant maternal age, primigravidity, and other possibly imbalanced covariates.
In addition, because the fetal growth indicators (weight, length, head circumference) are interdependent manifestations of a complex growth process, latent structure analysis allows partitioning of the shared covariation of these growth measurements from the nonshared variation. Use of this multivariate response procedure avoids some of the scientific complications that must be faced when interdependent responses are studied individually (eg, Bonferroni corrections). In addition, the latent structure regression analysis allows examination of the possibility of asymmetrical cocaine influence on head circumference (the nonshared part of the variation) beyond a hypothesized cocaine influence on the shared part of the variation. These are advantages not gained in the simultaneous comparison of mean growth measurements or in separate multiple regression analyses, 1 for each interdependent growth measurement.
The current investigation, drawn from the Miami PCS longitudinal investigation of full-term African-American infants, supports previous research indicating a specific cocaine-related deficit in fetal growth and gestational age after consideration of other prenatal substance exposures and potential confounding influences. The study also highlights the potential mediating influence of gestational age as a determinant of fetal growth. Results from the current study, based on analyses that consider the interdependency between fetal growth indicators, call into question the notion of cocaine-related asymmetrical deficits in head circumference relative to other growth parameters. Although results suggesting a cocaine-associated deficit in fetal growth are not novel, the improved level of covariate control achieved through the use of multivariate modeling substantiates the modest but extremely stable influence of prenatal cocaine exposure on intrauterine growth, even in full-term infants. In a related report from this project, the MIMIC model will also be used to address issues of gestational timing (using intercorrelated trimester-specific self-report) and level of cocaine exposure (using intercorrelated bioassays) with respect to fetal growth and gestational age deficits.
This current report on cocaine-associated fetal growth deficits sets the stage for the longitudinal investigation of hypothesized long-term effects of prenatal cocaine in multiple domains including cognition, language functioning, attentional processing, emotional adjustment, educational attainment, and social adaptation in this exceptionally large, well-retained cohort of socioeconomically disadvantaged inner-city minority children.
This research was supported by a grant from the National Institutes of Health National Institute on Drug Abuse (R01 DA 06556) and General Clinical Research Center (M01 RR 05280).
We are indebted to the participating families and staff of the UM Perinatal Chemical Addiction Research and Education Program and the physicians and nurses of the University of Miami–Jackson Memorial Medical Center Neonatal Services for their contributions to this research. We also gratefully acknowledge the assistance of Dr Niou-Ching Wu in toxicology assays, Dr Guoyan Zhang of the Miami–Dade County Health Department in vital statistics, Dr Fangchao Ma in editorial assistance, Richard Henderson in data management, Luz Ajuria-Londono in data entry, and Patricia J. Anthony for providing the graphics.
- Received January 9, 2001.
- Accepted May 29, 2001.
Reprint requests to (E.S.B.) Department of Pediatrics, Division of Neonatology, Perinatal Chemical Addiction Research and Education Program, Box 016960 (M-808), Miami, FL 33101. E-mail:
- ↵Zuckerman BS, Frank DA. “Crack kids”: not broken. Pediatrics.1992 ;89:337–339
- ↵Lester BM, LaGasse LL, Seifer R. Cocaine exposure and children: the meaning of subtle effects. Science.1998 ;282:633–634
- ↵Robins LN, Mills J. Effects of in utero exposure to street drugs. Am J Public Health.1993 ;83(suppl):1–32
- ↵Bandstra ES. Discussion: subject recruitment and retention in longitudinal studies involving perinatal substance abuse and human immunodeficiency virus infection. In: Kilbey MM, Asghar K, eds. Methodological Issues in Epidemiological, Prevention, and Treatment Research on the Effects of Prenatal Drug Exposure on Women and Children. Washington, DC: US Department of Health and Human Services; 1992:183–193
- ↵Schneiderman JF. Nonmedical drug and chemical use in pregnancy. In: Koren G, ed. Maternal-Fetal Toxicology: A Clinician’s Guide. New York, NY: Marcel Dekker; 1990:147
- ↵Muthén BO, Muthén L. MPlus software. The comprehensive modeling program for applied researchers. User’s Guide. Los Angeles, CA: Muthén & Muthén; 1998
- ↵Statacorp. Stata Statistical Software. Release 6.0. College Station, TX: Stata Corporation; 1999
- ↵Bateman DA, Chiriboga CA. Dose-response effect of cocaine on newborn head circumference. Pediatrics. 2000;106(3). Available at: www.pediatrics.org/cgi/content/full/106/3/e33
- ↵Richardson GA, Hamel SC, Goldschmidt L, Day NL. Growth of infants prenatally exposed to cocaine/crack: comparison of a prenatal care and a no prenatal care sample. Pediatrics.1999 ;104(2):. Available at: www.pediatrics.org/cgi/content/full/104/2/e18
- ↵Eyler FD, Behnke M, Conlon M, Woods NS, Wobie K. Birth outcome from a prospective, matched study of prenatal crack/cocaine use: II. Interactive and dose effects on neurobehavioral assessment. Pediatrics.1998 ;101:237–241
- ↵Hadeed AJ, Siegel SR. Maternal cocaine use during pregnancy: effect on the newborn infant. Pediatrics.1989 ;84:205–210
- American Academy of Pediatrics