A total of 537 articles were retrieved. The majority of the papers were excluded on the basis of being noncontributory, such as clinical guidelines, antenatal health policy or prevalence studies that did not measure pregnancy outcomes. One study identified by an English abstract was written in Spanish and was translated for this review.9
Fourteen studies met the eligibility criteria for the systematic review.
Two of us (C.M. and B.S.) independently assessed all the papers. Interrater reliability was moderate (intraclass correlation coefficient = 0.58).10
The scores awarded by each author were then averaged for each study. These final scores clustered in 2 groups, fulfilling 55%–70% or 75%–90% of Bracken's criteria. We selected all studies that had a combined quality score of 75% or more for our meta-analysis.
The characteristics of the studies selected for meta-analysis are outlined in . The research was conducted in the United States, Australia and Norway.11,12,13,14,15,16,17,18
Two case–control studies were included,16,18
in which women who delivered an infant with LBW were compared with women who delivered infants of normal birth weight. In the Norwegian study,16
participants (86 cases, 92 controls) were recruited consecutively from the University Hospital of Trondheim, which serves a geographically defined population, and thus represents the only study based on a population sample. The study by Campbell and colleagues18
differed from the others in that separate analyses were conducted for infants over 38 weeks' gestation (63 cases, 189 controls) and under 38 weeks' gestation (238 cases, 88 controls).
Sample sizes ranged from 178 to 1897 women.11,12,13,14,15,16,17,18
Amaro and colleagues,11
Berenson and colleagues,12
McFarlane and colleagues,14
and Curry and colleagues17
performed cohort studies of predominantly low-income and minority women. The study by Dye and colleagues13
was unique in that it was the only selected study conducted in a rural setting. Finally, Webster and colleagues15
studied a large cohort of Australian prenatal patients.
The wide range of prevalence rates of abuse in pregnancy documented in the studies selected for meta-analysis (5.6%–16.6%)11,12,13,14,15,16,17,18
may reflect actual differences in the specific clinical populations or geographic locations. The wide variation may also be attributed to the definitions of abuse employed by researchers, the methods of ascertainment of exposure to abuse and the time period of the inquiry. For example, Webster and colleagues15
inquired about physical, sexual and emotional abuse. Dye and colleagues13
relied on a 2-question abuse screen, and Amaro and colleagues11
used one question related to physical abuse or threats. In contrast, Berenson and colleagues,12
McFarlane and colleagues,14
Grimstad and colleagues,16
Curry and colleagues17
and Campbell and colleagues18
used specific abuse-focused instruments. These instruments, which have been previously validated in the research literature on violence and pregnancy, included the Abuse Assessment Screen,19
the Index of Spouse Abuse19
and the Conflict Tactics Scale.20
The time and frequency of inquiries about abuse varied across the studies. Most studies interviewed women during the prenatal period.11,12,13,14,15,17
Four studies reported conducting only one interview during the prenatal period.12,13,15,17
McFarlane and colleagues14
interviewed women 3 times during the prenatal period. Amaro and colleagues11
interviewed once prenatally and once in the postpartum period. Campbell and colleagues18
interviewed once within 72 hours of delivery, and Grimstad and colleagues16
interviewed women during the postpartum period or 1 year after delivery.
For the meta-analysis, data were extrapolated from each study to assess the impact of abuse that occurred during the current pregnancy. This was not possible for the Norwegian study.16
For this study, the data included in the meta-analysis were based on women abused in the current pregnancy or the current relationship, or both.
The definition of the outcome of LBW is a standard measurement of morbidity accepted as birth weight of less than 2500 g.1
defined LBW as less than or equal to 2500 g. The cause of LBW may stem from prematurity or poor intrauterine growth. In most studies, however, the mechanisms for LBW were not addressed separately.
The research findings are summarized in . In studies that only provided the percentages of LBW infants,13,14
raw data were extrapolated using the totals in each exposure group. We contacted 3 authors11,15,17
to obtain the number of LBW infants in the exposure groups.
Only one study14
revealed a significant association between abuse and LBW. This study found that pregnant women who were abused were 1.8 times more likely to deliver a child with LBW (95% CI 1.1–2.9). However, after adjusting for age, ethnic origin, marital status, education, parity, poor obstetric history, inadequate weight gain, interpregnancy interval, infection, hemorrhage, anemia, smoking, and alcohol and drug use, this relation was no longer statistically significant.14
Three cohort studies13,14,15
found statistically significant associations between abuse and mean birth weight. In the studies by McFarlane and colleagues14
and Webster and colleagues,15
the unadjusted mean birth weight was significantly lower for babies of abused women (133 g lower and 132 g lower respectively). On adjustment for differences in age, smoking, alcohol, education, ethnic origin, marital status, parity, number of terminations, antenatal visits and gestational age, Webster and colleagues15
found that the relation between mean birth weight and current abuse was no longer statistically significant. The association between abuse and mean birth weight (164 g lower, p
< 0.05) uncovered by Dye and colleagues13
was not adjusted for confounders.
In a multiple regression analysis, controlling for biomedical risk, age, ethnic origin, smoking, and marijuana and other drug use, Curry and colleagues17
found a statistically significant abuse and smoking interaction and an abuse and marijuana interaction associated with birth weight. For women who reported abuse, smoking and the use of marijuana increased the risk of delivering an infant with a lower birth weight.
A standard test for heterogeneity found that the pooled data were homogeneous and appropriate to test our general hypothesis that women who reported physical, sexual or emotional abuse during pregnancy were more likely than nonabused women to give birth to a baby with LBW. When the 8 studies were entered into the meta-analysis, an odds ratio of 1.4 was found (95% CI 1.1–1.8) (). To further examine our findings, we performed a sensitivity analysis. Removing the 2 case–control studies from the meta-analysis reduced the odds ratio to 1.3 (95% CI 1.0–1.8).
Fig. 1: Odds ratios for an association between abuse during pregnancy and low birth weight (LBW). CI = confidence interval.