This study provides confirmation of the multiple risks related to drug use, including inadequate material resources, stress, depression, delayed or no prenatal care, poor nutritional status, as well as concurrent use of tobacco and alcohol, and is the first to our knowledge to control for multiple social, psychosocial, behavioral, and biomedical factors when evaluating specific drug use effects on birth outcomes. In unadjusted associations, marijuana, cocaine, and opiate use were all strongly related to continuous birth weight and LBW. After controlling for overlapping substance use, these effects were attenuated but still significant for cocaine and opiates on continuous birth weight. After adjustment for all associated risk factors for reduced birth weight and LBW, however, only the use of cocaine was related to continuous birth weight. No substance was significantly more likely to result in LBW—an outcome of greater clinical significance. Altogether, these results suggest that illicit drug use is a stronger risk marker than a risk factor for adverse birth outcomes.
The observed independent effect of cocaine use on birth weight (−142 g) is similar in magnitude to that found in a large, multisite cohort study that adjusted for gestational age but no social or psychosocial factors (−151 g).10
This same study, however, also reported a sizeable LBW effect.11
Of the illicit drugs examined, cocaine has the most biologically plausible effect on birth weight via vasoconstriction and reduced uteroplacental transfer.42
The greater birth weight decrement observed among infants of women with a positive labor and delivery urine screen is consistent with a dose–response effect and suggests that only heavier use later in pregnancy may be consequential.
Between 70% and 90% of birth weight decrements and increased odds of LBW related to the three drug types was explained by factors associated with drug use. Other substance use, psychosocial, and behavioral factors accounted for considerable proportions of cocaine-related effects on birth weight. In particular, smoking and stress individually accounted for 10% to 20% of unadjusted cocaine effects on continuous birth weight. For opiate use, smoking, early prenatal care, and the biomedical nutritional status indicators accounted for a substantial proportion of the estimated effects on both LBW and birth weight.
Thus, where birth outcomes are concerned, emphasis on illicit drug use per se may be misplaced. For example, the birth weight effects of moderate to severe stress (−243 g, p
0.001, not shown) and heavy smoking (−158 g, p
0.04) exceeded that of cocaine and accounted for a significant portion of unadjusted cocaine-related effects. Chronic stress is consistently related to birth weight and gestational age with hypothesized mechanisms via behavioral and biological pathways.43,44
In relation to drug use, stress may promote and reinforce substance use as a coping mechanism.45
And in contrast to any substance, moderate to severe stress was associated with twice greater odds of LBW (OR
0.03). Similarly, living in public housing and having an external locus of control were associated with both birth weight decrements (−172, p
0.02; −157, p
0.01) and LBW (OR
0.01). Cigarette smoking is also strongly related to fetal growth restriction and much more widely prevalent than cocaine use with general population estimates as high as 25%.7,46
Alcohol too has established teratogenic effects including growth restriction;7
the lack of effect observed in this sample may reflect lighter consumption patterns as alcohol is less commonly used by low-income, Black women.47
Although nutritional status may be a consequence of the anorexic effects of opiates, the use of opiates was not significantly related to either birth weight outcome before adjustment of potentially indirect biomedical factors. The absence of early prenatal care may also be a consequence of the depressive effect of opiate use.29
While observed effects of prenatal care on birth outcomes may reflect unmeasured positive selection characteristics,48,49
the linkages to ancillary social and psychosocial services provided in prenatal care are likely to be of greatest benefit to substance users and other women with multiple psychosocial risks.49
Thus, community outreach and other efforts to identify and enroll pregnant drug users into prenatal care are imperative.
In contrast to the findings of some studies, physical abuse and medical risk factors were not more common among drug users than nonusers and did not account for drug-related effects on birth weight. The experience of physical abuse was relatively high in this low-income sample and might reflect less severe forms of abuse that are common in some families. The report of stress because of abuse, an item in the hassles scale, was indeed twice as high among drug users (20% versus 10%; data not shown). Another study of this sample found that this single item related to abuse accounted for most of the effect of stress on birth weight as measured by the entire scale.50
The significantly lower prevalence of sexually transmitted infections among opiate users is likely to be the result of surveillance bias due to inadequate health care utilization. As many biomedical risk factors were less common among drug users, their control introduced a form of negative confounding and generally increased estimates of drug-related effects.
In addition to the measurement and control of multiple social and psychosocial confounders, a major strength of this study was the capacity to assess independent drug effects. Unlike some previous studies that have relied on statistical control among polydrug users, there were sufficient numbers of women in our study who used only cocaine (n
48) and only opiates (n
20) to estimate independent effects. Similarly, illicit drug use and use of tobacco and alcohol were correlated but not collinear. The combination of sources for drug use ascertainment, including self-report, the medical record, and urine toxicology screens, is another major strength that reduces misclassification and increases confidence in the estimates of drug-related effects. Several sensitivity analyses were also performed. Because neighborhood disadvantage has been associated with both maternal drug use and birth outcomes,51
the residential census tract was initially included with fixed effects to control for all observed and unobserved neighborhood attributes. Drug effects were not significantly altered, however, and only individual level models were presented to preserve statistical power and the precision of effect estimates. Additional sensitivity analyses based on the propensity score, a technique that can improve covariate balance beyond that achieved by ordinary multivariable regression,52
essentially confirmed the results reported herein.
This study also has several limitations that deserve mention. First, the effect of amphetamine use—a drug of increasing popularity—could not be examined since only one woman reported its use and toxicological screens were not routinely performed in the mid-1990s. However, the pharmacologic effects of amphetamine use are similar to cocaine and other epidemiologic studies suggest similar effects on birth weight.25,53
Second, differences in gestational age or preterm birth were not evaluated out of concern for the reliability of gestational age estimation in a sample overrepresented for drug use and lack of prenatal care. The reported date of last menstrual period is less reliable for women with social risk factors and ultrasound-based estimation is most accurate when performed in the first trimester.54
Moreover, the existing literature is more suggestive of illicit drug effects on birth weight or fetal growth than the length of gestation.7
Third, relative to the number of model parameters estimated, the sample size may have been limited to assess drug effects and interactions with great precision; there were large confidence intervals in most cases (for example, the fully adjusted odds ratio for LBW associated with opiate use of 1.74 could not be distinguished from 1). Finally, the potential for recall bias and/or error exists in any retrospective study. However, there is little empirical evidence of differential reporting of exposures according to pregnancy outcome.55,56
Moreover, the clinic-based prospective alternative to a retrospective study would have excluded the third of drug users who did not receive prenatal care—a very high risk group. Thus, concerns of selection bias exceeded those of recall bias.