Maternal cocaine use and low birth weight newborns: a meta-analysis

Many epidemiological studies published on the association between maternal cocaine/crack use and birth weight have either lacked precision or failed to control for major confounding, predominantly by tobacco smoking. Meta-analysis enables a single summary measure of effect to be calculated by combining data from any number of individual studies, thus enhancing statistical power. We undertook a number of meta-analyses using only studies that had adjusted for tobacco smoking to estimate more precisely the effect of maternal cocaine use on birth weight.

A meta-analysis of five studies presenting data for 'any' prenatal cocaine exposure, adjusted for tobacco smoking but unadjusted for gestational age, produced a pooled relative risk estimate from a fixed effects analysis of 2.15 (95% CI 1.75-2.64). However, there was substantial heterogeneity among studies (p < 0.001), and the relative risk from a random effects analysis was smaller (1.65) with a confidence interval that included unity (95% CI 0.94-2.83). Addition of a further study adjusted for gestational age had minimal effect on the pooled estimate: the fixed effects relative risk was 2.14 (1.77-2.60) and the random effects estimate 1.77 (1.15-2.71). When data on more intense prenatal exposure were analysed, the fixed and random effects analysis produced the same pooled estimate of the relative risk of 4.42 (2.24-8.71), suggesting that more frequent cocaine exposure was associated with a higher relative risk for low birth weight. Data from studies on mean reduction in birth weight produced a pooled estimate of 112 g (95% CI 62-161 g).

The current study suggests that maternal cocaine use causes low birth weight, and that the effect is greater with heavier use. However, despite the adjustment for tobacco and the adjustment by some studies for other confounders such as race, maternal age, gravidity and socio-economic status, it could be argued that other life-style factors not controlled for may account for the observed effects. While this argument is not supported by some other types of study, the issue of residual confounding can only be finally addressed by analytical studies which adequately control for important variables.