We conducted a systematic literature search for potentially relevant original papers using the following electronic databases from January 1980 to first week of June 2009: MEDLINE, EMBASE, CINAHL, CABS, WHOlist, SIGLE, ETOH, and Web of Science. We used following keywords and medical subject headings to identify relevant articles in electronic databases: (‘alcohol*’ or ‘ethanol’ or ‘light drinking’ or ‘moderate drinking’) AND (‘birth weight’ or ‘low birth weight’ or ‘gestational age’ or ‘small for gestational age’ or ‘preterm*’ or ‘pregnancy outcome’ or ‘pregnancy complication’ or ‘prenatal*’) AND (‘case’ or ‘cohort’ or ‘ratio’ or ‘risk*’ or ‘prospective*’ or ‘follow*’). No language restrictions were applied. Eligible studies were original publications (we excluded letters, editorials, conference abstracts, reviews, and comments) of case-control and cohort studies reporting incidence, hazard ratios, relative risks or odds ratios of alcohol consumption in comparison to abstainers. In addition, bibliographies of key retrieved articles, relevant reviews and meta-analyses were hand searched.
The strategy resulted in 1345 hits; of which 90 appeared relevant upon initial inspection. The contents of these abstracts or full-text manuscripts identified during the literature search were reviewed independently by 2 reviewers to determine whether they met the criteria for inclusion. Articles were considered for inclusion in the systematic review if they reported data from an original study (i.e., no review articles). When there were discrepancies between investigators for inclusion or exclusion, a third reviewer (J.R.) conducted additional evaluation of the study and discrepancies were resolved in consultation. To be included in our meta-analysis, a published study had to meet the following criteria:
- Reported data were from an original study (i.e., no review articles),
- Cohort or case-control study in which medically confirmed low birth weight (defined as <2500 grams), preterm birth (<37 weeks gestation) and SGA (<10th percentile of gestational age adjusted birth weights) were the end points,
- Reporting of relative risk or odds ratios or hazard ratios (or data to calculate these risks) of low birth weight, preterm birth and SGA associated with alcohol consumption.
Thirty six studies met all of the inclusion criteria and were included in the meta-analysis. Twenty four had dose-response information with at least three or more drinking exposure groups and 12 studies had exclusive data on drinker versus no drinker. Four previous systematic reviews [10
] and three meta-analyses [14
] were identified and excluded. For details on study exclusion, please see .
Results of systematic review of the relationship between maternal alcohol consumption and low birth weight, preterm birth and small-size-for-gestational age (SGA)
All data were independently extracted by means of a standardized protocol. Study characteristics recorded were as follows: title of the study, lead author surname, publication year, source of publication, country of origin, study design (cohort or case-control), characteristics of the study population (e.g., size of the sample; method of sampling; age distribution, average age, and ethnicity), measures of outcome and exposure, duration of follow-up (for prospective cohort studies), confounding factors controlled for by matching or adjustment, and the risk estimates (relative risk or odds ratios or hazard ratios) of birth outcomes studied, compared to abstainers, associated with alcohol consumption and the corresponding confidence intervals. When a range of alcohol intake was given, the midpoint of the range was taken. In cases where open-end for the highest category was given (e.g. 40+ grams/day), three-quarters of the length of the immediate previous category range was added to the lower bound and was used as the measure. Where consumption was reported in drinks and not in grams, the gram pure alcohol equivalent (of 1 drink) explained in the article was used as a conversion factor if stated, and if not, conversion was based on geographical location: for Canada 13.6 grams, United States 12 grams, the UK 8 grams and for both New Zealand and Australia 10 grams of pure alcohol. For all other countries without any clear specifications 12 grams pure alcohol was used as an equivalent of 1 drink.
Information about the level of exposures in each study, the number of cases at each exposure level, the total population at risk at each exposure level, the adjusted estimates of relative risk (RR) compared to abstention for each exposure level, and the corresponding lower and upper 95% confidence intervals (CI) of the adjusted RR were obtained.
To ensure accuracy in data abstraction, five included and five excluded studies were randomly chosen to be abstracted independently by a co-author (H.I.) and the results were compared. Both authors agreed on 5/5 articles reviewed for inclusion/exclusion, and 611/654 data points abstracted over 10 articles. Where disagreements existed, both authors reviewed the materials together until a consensus was reached.
Drinkers versus non-drinkers meta-analysis
In the drinkers versus non-drinkers meta-analysis, the DerSimonian and Laird (1986) [17
] random-effects method was used to combine the natural logarithm of the risk estimates across studies. Where a study provided a dose-response analysis only, the risk estimates for all drinking categories were pooled using the inverse variance weighted method to derive a single estimate. These statistical analyses were completed using the METAN command in STATA version 10.1, StataCorp, Texas, USA [18
Meta-regression of dose-response relationship
Based on previously published research, the associations between maternal alcohol consumption with low birth weight, preterm birth and SGA could be either linear or nonlinear. In order to be flexible in fitting the best model, we conducted the meta-regression using linear as well as first-order and second-order fractional polynomial regression with powers –2, –1, –0.5, 0, 0.5, 1, 2, 3 to estimate a best fitting curve to the data. Best-fit curves were assessed using decreased deviance compared to the reference model. Comparisons of curves to determine the best fit were made using a Chi-square distribution [19
]. The first order and second-order fractional polynomials take the general form shown in Equation 1
where x is the alcohol exposure level in grams per day, P1
are the polynomial powers and β1
are the corresponding coefficients. No intercept term exists since all models have a starting point of Log RR = 0 (RR = 1 at zero consumption). All models were fitted in STATA version 10.1, StataCorp, Texas, USA, using the GLST function [18
Heterogeneity & Publication bias
Statistical heterogeneity between studies was assessed using both the Cochrane Q test and the I2
]. Because all statistical tests for heterogeneity are weak, we also included the 95% CI for I2
] that was calculated based on the method described by Higgins and Thompson (2002) [20
]. Publication bias was assessed by visual inspection of Begg’s funnel plot, the Begg-Mazumdar adjusted rank correlation test [23
] and the Egger regression asymmetry test for funnel plot [24
]. The RR estimates were prepooled using the inverse variance weighted method because funnel plot methodology assumes one overall RR per article. Statistically significant publication bias was defined as p <0.10.