rate (IMR) are reported to be higher among minority and African‐American women compared with Caucasian women ( 1 - 3 ). Similar trends have been observed in other countries with a multicultural population, such as the UK and Australia ( 4 , 5 ). Differences in sociodemographic determinants ( 6 ) and stress during pregnancy have been suggested as the explanation for such variation. Increasing migration to Western countries from Africa and Asia has contributed to the interest and concern pertaining to health‐related complications such as adverse pregnancy outcomes ( 7 , 8 ). Differences in birth outcomes among biracial couples (White mother–Black father and Black mother–White father) compared with same‐race couples have been reported ( 9 , 10 ). However, results have been discrepant. Collins et al. ( 11 ) reported no difference in LBW among biracial couples when compared with White parents after adjustment for confounders, whereas Migone et al. ( 9 ) and Gold et al. ( 12 ) reported gradients in adverse pregnancy outcomes, with maternal race contributing more than paternal race. In view of the public health importance of optimal birth outcomes, our objective was to systematically review and meta‐analyze the association between biracial couples and adverse birth outcomes.

Clinical heterogeneity was assessed and reported in Table 1 , the table of included studies. This included variation in exposure covariates such as maternal age, paternal age, details of the prenatal care, gender of the infant etc. Publication bias was planned to be examined using funnel plots. Statistical heterogeneity was determined using I‐squared estimates during meta‐analysis ( 18 ).

We first included unadjusted data for this review from all studies. As is traditional with other meta‐analyses, no adjustment for multiple analyses was made. For the purpose of this review, we counted WMWF group as the reference group. When the study population overlapped between two studies for time period (i.e. one study reporting population from 1995–98 and other study reporting from 1996–2002) or for location (one study reporting a state‐wide data and another reporting entire national data), we systematically reviewed both studies, but only included study with higher sample size for the meta‐analyses. Subgroup analyses were performed to compare outcomes between groups. Weighting of the studies in the meta‐analyses was calculated based on the inverse variance of the study. The meta‐analytic software REVIEW MANAGER was used ( 16 ). The random effect model was chosen because it accounts for between‐ and within‐study variability, as we expected a degree of clinical and statistical heterogeneity among the studies. Unadjusted and adjusted odds ratios (uaOR and aOR respectively) were reported for categorical measures, and mean difference (MD) was used for continuous measures. Summary estimates with 95% confidence intervals (CI) were calculated. It is well known that there is an interplay of various factors in the causation or events leading to LBW or preterm births. Some authors have reported both adjusted and unadjusted risks in their population controlling for confounders perceived (or statistically proven) to have an effect on the summary estimate. We pooled data from these studies, which reported adjusted risk estimates, and performed random effect model meta‐analyses using a generic inverse variance method to assess the adjusted impact of biracial couples on birth outcomes ( 17 ). These adjusted estimates were expected to either confirm or refute effects observed in unadjusted estimates.

Assessment of methodological quality and risk of bias was performed using a pre‐defined checklist based on the criteria for sample selection, exposure assessment, outcome assessment, confounder, analytical and attrition biases. This tool has been used by our group previously ( 14 , 15 ) (Supporting Information Table S1). The classification in each category was: cannot tell, no bias, low, moderate or high risk of bias. Overall assessment of the study was based on the responses for each category. The studies were assigned overall bias based on the highest risk of bias in any category.

Data from each eligible study were extracted on a custom‐made data collection form. Minor modifications such as calculation of raw percentages from available data were allowed. Factors that were studied or adjusted for in the analyses in the individual studies were reported. Both unadjusted and adjusted data as reported in the primary studies were extracted and noted in the results.

The search strategy for identification of studies included electronic databases (Medline, Embase, and CINAHL), which were searched from their inception to December 2011 for all published studies in the English language. MeSH words of “race”, “racial segregation” AND “pregnancy outcome” OR “preterm birth”, “premature birth”, “low birthweight”, “growth retardation”, “small‐for‐gestational age” alone and in different combinations were used. Alternative terminologies such as neonate, infant, newborn, growth restriction, intrauterine growth retardation, were used as text words to identify additional articles. The reference lists of the identified articles were searched to identify additional eligible studies. The articles were independently scanned based on titles and abstracts by two authors (S.S. and R.S.). Discrepancies were resolved by consensus and involvement of the third author (P.S.). Selected articles were retrieved in full and were assessed for eligibility by all authors. The reviewers were not blinded to authors or the institution.

Studies reporting data on any of the following types of outcome measures were included: ( 1 ) LBW defined as birthweight <2.5 kg, and very low birthweight defined as birthweight of 500–1499 g, ( 2 ) PTB defined as gestational age <37 weeks, ( 3 ) birthweight in grams, ( 4 ) gestational age in weeks, ( 5 ) small‐for‐gestational age (SGA) defined as birthweight below 10 th centile for gestational age, ( 6 ) IUGR defined as gestation >36 weeks and birthweight <2500 g, and ( 7 ) stillbirths. Studies reporting both crude and adjusted risks were included.

The types of participants considered in the included studies were couples composed of four combinations: White father–White mother (WMWF), White mother–Black father (WMBF), Black mother–White father (BMWF) and Black mother–Black father (BMBF) who had singleton births. For assessment of exposure we included studies that reported ascertainment of maternal and paternal racial origin from various sources such as the hospital birth certificates, discharge summaries, census tract data, surveys, vital statistics data and national databases. In all published reports, race was self‐reported.

Criteria for considering studies for this review were observational studies that explored the association between biracial couples and birth outcomes. If the study provided information on the race of mother and the father and its association with any of the outcomes of interest, then the study was eligible for inclusion in the review. We only included information available from publications and did not contact primary authors. Studies published as abstracts were excluded. Studies reporting results from groups belonging to the same race without considering biracial couples were not included in this review. Thus, we did not include studies in which outcomes were reported for one racial group vs. other racial groups only (White vs. Black). For example, we excluded studies in which only outcomes of Black couples were compared with White couples without providing results of mixed racial couples. We included articles reporting on Black and White race only due to lack of availability of sufficient information about birth outcomes for other racial mixed couples. The types of studies included were observational cohort studies, cross‐sectional studies, and case control studies. Reports of data from national or local vital statistics not published as peer‐reviewed articles were not included.

We followed the Meta‐analysis of Observational Studies in Epidemiology (MOOSE) criteria for preparing this review ( 13 ). The methods previously described by our group for systematic reviews were followed in the conduct of the review ( 14 , 15 ). The data were extracted from published manuscripts and thus no ethical approval was necessary.

Clinical heterogeneity (differences in study design, methods of assessment of exposure, and confounders that were adjusted) among studies is described in Table 1 . On meta‐analysis, significant statistical heterogeneity was identified in most of the outcomes studied, with an I‐squared (I 2 ) test result of 0–100% for different outcomes. This is not surprising knowing that the baseline population and their characteristics differed among studies. The I 2 test for PTB was 99% for all the comparison groups. The results of exploration of these heterogeneities are reported in the subgroup comparisons mentioned above. Funnel plot assessment was not carried out because of the small number of studies.

BMWF vs. BMBF: Results of meta‐analyses of LBW, PTB and BW outcomes indicated favorable results for WMBF group (Table S4). Meta‐analysis was not possible for SB as an outcome in view of overlapping data ( 12 , 27 ). For SB, Getahun et al. ( 27 ) reported an OR of 0.81; 95%CI 0.74–0.88 and Gold et al. ( 12 ) an OR 1.05; 95%CI 0.67–1.63.

WMBF vs. BMBF: Results of meta‐analyses of LBW, PTB and BW outcomes indicate favorable results for WMBF group (Table S4). Meta‐analysis was not possible for SB as an outcome in view of overlapping data ( 12 , 27 ). For the WMBF group, compared with the BMBF group, Getahun et al. ( 27 ) reported an OR of 0.67; 95%CI 0.64–0.71 and Gold et al. ( 12 ) reported an OR of 0.81; 95%CI 0.61–0.79.

WMBF vs. BMWF: Results of meta‐analyses of LBW, PTB and BW outcomes indicated favorable results for the WMBF group (Supporting Information Table S4). Meta‐analysis was not possible for SB as an outcome in view of overlapping data ( 12 , 27 ) Getahun et al. ( 27 ) reported an SB risk of 0.83; 95%CI 0.75–0.91 and Gold et al. ( 12 ) reported an OR of 0.77; 0.47–1.27 for WMBF compared with BMWF, for the same comparison.

Four studies ( n = 5 205 112 pregnancies) reported on this outcome ( 12 , 28 - 30 ). All studies observed the highest birthweight for the WMWF group. An increasing difference in mean birthweight was observed from the WMBF group to the BMWF group and to the BMBF in that order, with the BMBF group having the lowest BW. The meta‐analysis of crude birth weights indicated the mean differences (95%CI) in birthweight were −52; 95%CI –144, 39 g for the WMBF group, −147; 95%CI −183, −111 g for the BMWF group, and ‐227; 95%CI −266, −188 g for the BMBF group when compared with the WMWF group (Supporting Information Table S3).

Two studies ( n = 22 605 786 pregnancies) reported on this outcome and contributed to all of the following results ( 12 , 27 ). Meta‐analyses were not possible, since Getahun et al. ( 27 ) used USA‐wide data and Gold et al. ( 12 ) used Californian data for the overlapping year. However, both the studies revealed that the risk of SB was higher for all the race groups when compared with WMWF group. Gold et al. ( 12 ) reported an aOR of 1.04; 95%CI 0.86–1.27 for WMBF, 1.38; 95%CI 0.76–2.50 for BMWF, and 1.35; 95%CI 0.92–1.98 for BMBF when compared with WMWF group. A significantly higher risk was reported by Getahun et al with an aOR of 1.17; 95%CI 1.10–1.26 for WMBF, 1.37; 95%CI 1.21–1.54 for BMWF and 1.67; 95%CI 1.62–1.72 for BMBF group ( 27 ).

A total of six studies ( n = 5 320 141 pregnancies) reported on LBW ( 10 - 12 , 28 - 30 ) ( Table 3 ). For the reference WMWF parent population, the rate of LBW was variable in all studies and ranged from 4.9% in Mangold et al. ( 30 ) to 6% in Collins et al. ( 11 ). The results of meta‐analyses after removing duplicate data are reported in Table 3 . Compared to the WMWF group, all three other groups had higher unadjusted and adjusted odds for LBW ( Figure 2 ).

It was identified that there was considerable duplication of data. Collins et al. ( 9 ) used data from Illinois from 1982 and Migone ( 11 ) used USA‐wide data from 1982 and 1983. Both these studies were used in the systematic review because mean BW and gestational duration were described by Migone et al. ( 9 ), and growth retardation, SGA and LBW by Collis et al. ( 11 ). However, Migone et al. was considered during meta‐analyses in view of large numbers and inclusion of an unselected population ( 9 ). Similarly, Gold et al. ( 12 ), Getahun et al. ( 27 ) and Ma et al. ( 29 ) used data overlapping year 2001. Gold et al. ( 12 ) included data from California only. Ma et al. ( 29 ) included USA‐wide data that excluded Californian data. Thus these two studies were considered separate and eligible to be combined even though data from the year 2001 was used. Getahun et al. ( 27 ) used USA‐wide data that included the year 2001. Thus, for meta‐analyses, Getahun et al. ( 27 ) was preferred over Gold et al. ( 12 ) and Ma et al. ( 29 ) because of larger sample size. This was adopted to avoid duplication of the same couples in the analyses. Various birth outcomes were assessed for different race combinations and are outlined in Table 1 .

The results of the search, the study selection log, and the number of studies are reported in Figure 1 . A total of 11 studies were excluded ( 3 , 4 , 7 , 19 - 26 ). Eight studies of 26 325 927 singleton births were eligible for inclusion ( 9 - 12 , 27 - 30 ). Characteristics of these studies are reported in Table 1 . Excluded studies and the reasons for exclusion are reported as Supporting Information in Table S2. All studies included participants and data from the USA. The results of the risk of bias assessment of the included studies are reported in Table 2 . A total of six studies had low and two studies had moderate risk of bias.

Discussion

In this systematic review of eight studies, we identified that adjusted odds of LBW, PTB, SB, and SGA births in biracial couples were higher than for both parents being White, but lower than for both parents being Black. A gradient was observed, with the lowest risk of adverse pregnancy outcomes for both parents being White followed by WMBF, BMWF and BMBF groups. Analyses of birthweight revealed a similar trend.

Study of the effects of maternal race on pregnancy outcomes indicated that the risks are higher among Black than White women (1, 2, 31-34). These differences in the maternal and infant outcomes have been attributed to poor health‐related behaviors (2, 6), poor accessibility to health facilities due to factors such as socioeconomic disadvantage (2, 35), poverty (2, 36-38), lower paternal and maternal education (39), inadequate prenatal care (32), younger maternal age (40), insufficient medical insurance (35, 41), higher rates of recreational drug use, and chronic exposure to individual and institutional racism (42). The fact that the risks of adverse outcomes have remained higher even after controlling for health behaviors (43) has led to proposed theory of the “weathering phenomenon” by Geronimus (44). The “weathering phenomenon” proposes that the health of African‐American women may begin to deteriorate in early adulthood as a physical consequence of cumulative socioeconomic disadvantage (44).

Similar to some of the studies in the review, we identified a gradient effect with higher relative influence of maternal race than paternal race on pregnancy outcomes. Intergroup comparison revealed that the WMBF group had a lower risk of adverse outcome compared with the BMWF group. Proposed sociodemographic factors may explain the discrepancies in outcomes between the WMWF and the BMBF groups; however, it is uncertain whether the same factors explain the differences in outcome in the WMBF and the BMWF groups (12). Polednak et al. (10) reported on detailed characteristics of mothers in the BMWF group. These mothers were younger, had lower education, were more likely to be unmarried, had a higher rate of inadequate prenatal care and a lower level of paternal education compared with the WMBF group (10). This may partially explain the higher rates of adverse outcomes in the BMWF group. In addition, observations from the 1990 USA census indicate that White wives of Black husbands had fewer educational achievements than White wives of White husbands. Fu described this as either “Status Exchange” or “In‐group preference” hypotheses (45), which suggest that minority men marry White women in “exchange” for higher social status (45). Fu observed that White wives of Black husbands had less schooling than White wives of White husbands (45). Alternatively, it is suggested that individual and institutional racism (42), negative life events, social stress and effects of weathering among pregnant Black women with a White spouse or partner could be contributing factors for adverse outcomes in the BMWF group (40, 42, 44). In addition, genetic factors have been suggested to play some role. This may include possible role of fetal, paternal and maternal genes in the causation of LBW (46-48). Animal studies have indicated an influence of paternal genes in hypertensive disorders of pregnancy and the development of the placenta (49, 50). However, the maternal and the intrauterine environment contributes significantly, as evidenced by the association of stress and maternal general health with adverse pregnancy outcomes (51, 52). Overall, it appears that neither genetics nor environmental factors adequately explain the etiology for adverse pregnancy outcomes in interracial Black–White groups (9, 53).

To our knowledge this is the first systematic review and meta‐analysis of pregnancy outcomes of interracial couples. The strength of our review exists in the large sample size of more than 26 million pregnancies, which enables between‐group differences to be investigated, the analyses of unadjusted and adjusted estimates and the provision of robust estimates of risks. Limitations of this review include those associated with retrospective database studies. A number of pregnancies were excluded due to coding errors, and non‐recording of race or outcomes. The majority of the studies obtained data from birth certificates, which are reliable for recording birthweight but may not provide accurate details of social attributes, gestational age and medical complications (54). Obtaining data about the infant's race can be challenging as controversy exists as to which race child should be assigned to at birth (54). Another limitation could be in the assignment of Black race. Black race includes USA‐born African‐Americans, and Black immigrants from sub‐Saharan Africa, South America and the Caribbean (2). It is possible that there are differences in the risks among these Black subgroups. Even after adjusting the risks, residual confounding persists due to unaccounted factors in individual studies (35).

The included studies are heterogeneous in their conduct and reporting. The exposure covariates and the confounding factors adjusted for during the analysis varied between studies. Moreover, same data may have been used by more than one study, making some studies not mutually independent. The statistical heterogeneity is evident from I‐squared estimates during the meta‐analyses. It can be argued that systematic review and not meta‐analysis is the preferred approach, as is traditional in such a situation. We addressed this by including all the studies in the descriptive review and including those studies without the possibility of data duplication for meta‐analyses. We used a random effects model to address heterogeneity during meta‐analyses. Hence, we believe that the summary estimates are a true reflection of the available data, despite the heterogeneity of individual studies.

The researchers are highly aware of the constellation of risk factors including psychosocial, environmental, health determinants, behaviors, medical and other factors that contribute to pregnancy outcomes. The complexity of perinatal risk factors must be kept in mind when interpreting our results. The purpose of this review was to report on the association between birth outcomes and biracial couples and not to infer causality.

Notwithstanding the limitations described above, the information from this review is important from the perspectives of public health policy and human rights. The increased stillbirth, LBW, PTB and SGA rates among infants born to biracial couples compared with White parents is of concern and should be monitored continuously. However, it is evident from this review that this information should be considered during antenatal and prenatal counselling and perinatal management. The risk factors should be kept in mind and discussed with the parents. Evidence‐based strategies to reduce the risks of adverse outcomes should be offered to all expectant couples, with special attention to those at increased risk.

Future research in this area needs to examine further the reasons for adverse outcomes rather than merely repeating similar studies. Future studies should explore the interactions between social, demographic and environmental factors as they pertain to biracial couples and outcomes of their pregnancies. In addition, similar studies of pregnancy outcomes are needed for other mixed race groups. We suggest that practitioners and researchers seek a better understanding of racial differences in health and disease, and continue to implement strategies to compensate for the effects of racial vulnerabilities in birth outcomes.