Birth Spacing and Risk of Adverse Perinatal Outcomes:  A Meta-analysis FREE

Both short and long intervals between pregnancies have been associated with increased risk of several adverse perinatal outcomes, such as preterm birth, low birth weight (LBW), small for gestational age (SGA), and perinatal death.^1- 3 However, there has been disagreement on whether the relationship is due to confounding by other risk factors. For example, some researchers have argued that short intervals between pregnancies merely designate women already at higher reproductive risk, either because of underlying disorders, socioeconomic status, or lifestyle factors.^4- 5 Furthermore, previous research in this area has several methodological limitations, such as small sample size, lack of control for potential confounding factors, dichotomization of the measure of birth spacing on the basis of an arbitrarily defined cut point, and use of birth interval (time elapsed between the woman's last delivery and the birth of the index child) instead of interpregnancy interval (time elapsed between the woman's last delivery and the conception of the next pregnancy) as the measure of birth spacing. The use of birth intervals overestimates the risk of adverse perinatal outcomes for very short intervals between pregnancies.

This issue is relevant to public health and clinical practice because if short and/or long interpregnancy intervals are found to be independently associated with increased risk of adverse perinatal outcomes, birth spacing might then be considered an intervention to prevent such adverse outcomes, mainly in the developing world. Therefore, we performed a systematic review, including meta-analysis, of the relationship between birth spacing and the risk of adverse perinatal outcomes that provided an overall summary of the effect measure and determined both the riskiest and the optimal interpregnancy intervals. In addition, we determined whether estimates of the effect measure depend on dimensions of study quality of the primary studies and whether the relationship differs in subgroups defined by the characteristics of women, and we highlight deficits that need to be addressed in future studies.

We used a prospective protocol prepared specifically for this purpose. The systematic review was conducted following this protocol and reported using the checklist proposed by the Meta-analysis of Observational Studies in Epidemiology (MOOSE) group for reporting of systematic reviews of observational studies.⁶

A search was conducted by the investigators in MEDLINE (1966 through January 2006), EMBASE (1980 through January 2006), ECLA (1980 through January 2006), POPLINE (1980 through January 2006), CINAHL (1982 through January 2006), and LILACS (1982 through January 2006), using a combination of Medical Subject Headings or key word terms for birth spacing and adverse perinatal outcomes. Terms for birth spacing included interpregnancy interval, birth interval, interbirth interval, pregnancy spacing, pregnancy interval, birth spacing, intergenesic interval, birth to birth interval, birth to conception interval, delivery to conception interval, and interdelivery interval. Terms for adverse perinatal outcomes included perinatal outcomes, infant outcomes, pregnancy outcomes, adverse outcomes, low birth weight, preterm delivery, preterm birth, small for gestational age, intrauterine growth retardation, intrauterine growth restriction, Apgar scores, neonatal depression, neonatal intensive care unit, fetal death, stillbirth, perinatal death, fetal mortality, perinatal mortality, perinatal morbidity, perinatal outcomes, neonatal death, neonatal mortality, and neonatal outcomes. Proceedings of several international meetings on birth spacing and bibliographies of the retrieved articles were also searched by hand. No language restrictions were imposed. In the case of studies discussing more than 1 outcome, each outcome was considered independently. To find unpublished studies, we contacted relevant researchers in the field. Twelve authors were contacted as well, in an attempt to obtain additional data.

Studies were included if (1) they were cohort, cross-sectional, or case-control studies that evaluated the relationship between birth or interpregnancy interval and any adverse perinatal outcome; (2) the definition of interpregnancy interval corresponded to the period between delivery of the previous infant and conception of the current pregnancy. Although the use of birth-to-birth interval overestimates the risks of adverse perinatal outcomes for very short intervals, studies using birth interval were included and analyzed separately; and (3) the authors of the studies adjusted their results for at least maternal age and socioeconomic status (measured indirectly by occupation and work status, educational level, income, housing, or other variables), because we considered these variables to be the most important confounding factors in the association between birth spacing and adverse perinatal outcomes. Studies were excluded from the systematic review if they were case series or reports, editorials, letters to the editor, or reviews without original data; if they exclusively used univariate analysis; if they did not adjust for at least maternal age and socioeconomic status; or if they did not provide data. Studies included in the systematic review were also included in the meta-analyses if they met the following additional criteria: (1) used interpregnancy interval as measure of birth spacing; (2) provided data for 4 or more interpregnancy interval strata; and (3) reported odds ratio (OR) or relative risk estimates and 95% confidence intervals (CIs) or data to calculate them. Studies of different designs and different measures of birth spacing that are included in the systematic review were analyzed separately because of different threats to their internal validity.

All published studies deemed suitable were retrieved and reviewed independently by 2 authors (A.C.-A., A.R.-B.) to determine inclusion. Disagreements were resolved through consensus. The degree of agreement was expressed as percentage agreement and κ statistics.

Study methodological quality was judged by the following 6 validated criteria believed to be important for the quality of observational studies evaluating the relationship between birth spacing and adverse perinatal outcomes^7- 8: (1) pregnancy interval used (adequate if the study used interpregnancy interval; inadequate if the study used birth interval); (2) categorization of exposure (adequate if the study examined ≥4 categories of pregnancy intervals; inadequate if the study examined <4); (3) birth spacing measurement and inquiry of outcomes (adequate if birth spacing measurement and ascertainment of outcomes were made by medical records or direct measurement; inadequate if not); (4) blinding of both birth spacing status and ascertainment of outcomes (adequate if assessment of both birth spacing status and outcomes was blinded; inadequate if not blinded or unreported); (5) loss to follow-up or exclusions (only for cohort and cross-sectional studies) (adequate if loss to follow-up or nonvalid exclusions [eg, improper elimination of records] was <10%; inadequate if ≥10% or unreported); and (6) control for confounding factors (adequate if the study additionally controlled for ≥2 of 5 confounding factors [parity, outcome of the most recent recognized pregnancy, access to prenatal care, breastfeeding, and maternal nutritional status]; inadequate if additionally controlled for <2).

Assessment of methodological quality of each study was carried out by 2 of the authors (A.C.-A., A.R.-B.) working independently. Differences of opinion were resolved through discussion.

Data were extracted independently from each article by 2 investigators (A.C.-A., A.C.K.-G.) by means of a standardized and pilot-tested data collection form. The following information was sought from each article: title, first author's name, year, geographic location of the study (country and region), study design, characteristics and source of the study population, sample size, measures of outcome, measure of birth spacing used, categorization of intervals, method of data collection, exposure measurement and ascertainment of outcome(s), blinding of birth spacing status and ascertainment of outcome(s), loss to follow-up or invalid exclusions, confounding factors controlled for by matching or adjustment, and unadjusted and adjusted relative risks or ORs and their 95% CIs for individual adverse perinatal outcomes associated with all pregnancy intervals.

The studies included in our meta-analyses differed in the units used for measurement of the interpregnancy interval (days, weeks, months, or years). Therefore, we converted these different units of interpregnancy interval to months. We used 3 different meta-analytical techniques to investigate whether a relationship exists between interpregnancy interval and the risk of adverse perinatal outcomes.

Meta-regression Analysis. We first examined the shape of the dose-response relation between interpregnancy interval and risk of adverse perinatal outcomes. For this purpose, we used the method proposed by Greenland and Longnecker⁹ and Berlin et al¹⁰ for meta-analysis of epidemiologic dose-response data. The dose-specific confounder-adjusted natural logarithms of the ORs from all studies were pooled, and a curve using weighted quadratic spline meta-regression with no intercept term was fitted. This method was chosen because several studies have reported finding a nonlinear, J-shaped relationship between interpregnancy interval and the risk of adverse perinatal outcomes such as preterm birth, LBW, and SGA. The main fields in the data set were the value x of exposures (expressed in months) assigned as the midpoints for the ranges of the reported categories of interpregnancy intervals and as 1.2 times for the lower bound of the open-ended upper categories as suggested by Berlin et al,¹⁰ and the y-axis estimates of natural logarithm of the adjusted OR for each exposure level.

Pooled ORs. Depending on data availability in the original studies, we categorized interpregnancy interval into 6 groups: shorter than 6 months, 6 to 11, 12 to 17, 18 to 23, 24 to 59, and 60 months or longer. Odds ratios were used as the measure of the relation between interpregnancy interval and adverse perinatal outcomes. The interval of 18 to 23 months was used as the referent category, because this was the interval with the lowest risk for preterm birth, LBW, and SGA. Data abstracted from each study were arranged in 2 × 2 tables. Then, ORs with their 95% CIs for each adverse perinatal outcome considered were calculated separately for 5 predefined categories of interpregnancy interval. Separate analyses of the associations in 2 × 2 tables were combined to produce pooled unadjusted ORs and corresponding 95% CIs. We also calculated pooled adjusted ORs within each category using the estimated adjusted effect and its estimated standard error (often obtained indirectly from the CI) reported in each study.

Dose-Response Regression Slopes. Under the assumption of independence of the dose-specific OR, we estimated the dose-response regression slopes of each study using the OR, 95% CIs, and the midpoint of the exposure interval.⁹ For open-ended intervals, a point 20% higher than the low end of the interval was used. Pooled dose-response slopes and estimates of risks were then obtained from random-effects models applied to the study-specific slopes. The exponentiation of the slope gave the OR for a unit increase or decrease of the interpregnancy interval (1 month). To overcome the problem of assuming independence of dose-specific ORs (which is incorrect, as they have a common reference group), we adjusted the standard error of the within-study slopes estimating the covariance.

Heterogeneity of the results between the studies was formally tested with the quantity I², which describes the percentage of total variation across studies that is due to heterogeneity rather than chance. The I² can be calculated from basic results obtained from a typical meta-analysis as

I² = 100% × (Q − df)/Q

where Q is the Cochran heterogeneity statistic.¹¹ We pooled results from individual studies using DerSimonian and Laird random-effects models¹² because moderate to high heterogeneity (I²≥50%) was present in the majority of results.

To further explore the origin of heterogeneity, we restricted the analyses to subgroups of studies defined by study characteristics such as study quality, date of publication, and sample size. Moreover, we calculated separate estimates according to race/ethnicity and study setting (developed vs developing countries). Since a number of the largest studies were multinational, we could not analyze by country. Subgroup and sensitivity analyses were performed pooling adjusted ORs provided by the studies.

To detect publication and location biases, we explored asymmetry in funnel plots. This was examined visually, and the degree of asymmetry was measured using the Egger unweighted regression asymmetry test, with P<.10 indicating significant asymmetry.¹³ All statistical analyses were performed using STATA version 8.0 (StataCorp, College Station, Tex).

One hundred thirty studies were considered relevant, and the complete manuscripts were obtained. Of the 130 studies, 122 were published in English, 4 in Spanish, 3 in French, and 1 in Portuguese. Sixty-three studies were excluded, the main reasons being lack of adjustment for confounding factors at statistical analysis (46%) and lack of data on the relationship between birth spacing and adverse outcomes considered (35%). (The list of excluded studies is available from the corresponding author on request.) A total of 67 studies (52 cohort or cross-sectional studies^14- 65 and 15 case-control studies^66- 80), including 11 091 659 pregnancies, met the inclusion criteria. The computerized search located 64 of the studies, 2 were found in proceedings of meetings on birth spacing, and the remaining 1 was found through contact with a relevant researcher in the field.

Twenty studies (30%) were conducted in the United States. The remaining 47 were conducted in 61 countries from Latin America (22 countries), Asia (20 countries), Africa (11 countries), Europe (7 countries), and Australia. Overall agreement on the inclusion of studies was 97% (κ = 0.84).

The characteristics and main findings of the cohort and cross-sectional studies included in the systematic review are presented in Table 1 (developed countries) and Table 2 (developing countries); those of the case-control studies are presented in Table 3. The sample size in the cohort or cross-sectional studies ranged from 201⁴⁴ to 4 841 418.⁴³ The number of case participants enrolled in case-control studies ranged from 36⁷¹ to 416,⁶⁹ and the corresponding number of controls ranged from 50^71,79 to 1710.⁶⁷ Thirty studies provided data on preterm birth, 26 on LBW, 24 on SGA, 10 on fetal death, 4 on early neonatal death, 6 on perinatal death, and 2 on low Apgar scores. Twenty-four studies (36%) reported more than 1 adverse perinatal outcome. Among the 52 cohort or cross-sectional studies, 37 (71%) used birth-to-conception interval, 14 (27%) used birth-to-birth interval, and the remaining 1 used both intervals. Of the 15 case-control studies, 9 (60%) used birth-to-conception interval and 6 (40%) birth-to-birth interval as measures of birth spacing. The studies varied in methodological quality, with 21 cohort or cross-sectional studies (40%) meeting 5 or more criteria and only 3 case-control studies (20%) meeting 4 or more criteria. The most common shortcomings were failure to blind investigators to both exposure status and ascertainment of outcome, the report of loss to follow-up or exclusions, and the categorization of pregnancy intervals.

Overall, among the studies that provided data on preterm birth, 21 (18 cohort or cross-sectional and 3 case-control) reported an association with short intervals, 6 (5 cohort or cross-sectional and 1 case-control) an association with long intervals, and 9 (8 cohort or cross-sectional and 1 case-control) found no association. With regard to studies that reported data on LBW, 20 (18 cohort or cross-sectional and 2 case-control) found an association with short intervals, 7 (6 cohort or cross-sectional and 1 case-control) an association with long intervals, and 6 (all cohort or cross-sectional) found no association. Among the studies that provided data on SGA, 14 (13 cohort or cross-sectional and 1 case-control) reported an association with short intervals, 6 cohort or cross-sectional studies reported an association with long intervals, and 10 (6 cohort or cross-sectional and 4 case-control) found no association. Two studies did not find an association between birth spacing and low Apgar scores. With regard to perinatal mortality (fetal death, early neonatal death, and perinatal death), 10 studies reported an association with short intervals, 8 with long intervals, and 7 found no association.

It was not possible to perform a meta-analysis of the case-control studies because only 3 met the minimal inclusion criteria. Moreover, different categories of intervals and reference categories were used in the few studies. Twenty-six cohort and cross-sectional studies provided data for meta-analyses. Sixteen studies provided data for preterm birth,^{25,27,29,33,36,39- 40,45- 47,49- 50,52,55,61,64} 10 for LBW,^{16,21,27,33,36,40,47,49,56,64} 13 for SGA,^{21- 22,24,29,33,35- 36,39,45,49,52,55,64} 7 for fetal death,^{14,27,34,55,57,63- 64} and 4 for early neonatal death.^{34,57,63- 64}

The dose-response association between interpregnancy interval and the natural logarithm of the OR of the 5 adverse perinatal outcomes in cohort and cross-sectional studies was J-shaped (Figure). For preterm birth, LBW, and SGA, the highest risk was for intervals shorter than 20 months and longer than 60 months. For both fetal and early neonatal death, the highest risk was for intervals shorter than 6 months and longer than 50 months.

Infants born to women with interpregnancy intervals shorter than 6 months had pooled unadjusted ORs (95% CIs) of 1.77 (1.54-2.04), 2.12 (1.98-2.26), and 1.39 (1.20-1.61) for preterm birth, LBW, and SGA, respectively, compared with infants born to women with intervals of 18 to 23 months (Table 4). Likewise, women with intervals of 6 to 17 months were 8% to 23% more likely to give birth to infants with these adverse outcomes. Infants conceived 60 months or more after a birth had ORs (95% CIs) of 1.27 (1.17-1.39) for preterm birth, 1.49 (1.17-1.89) for LBW, and 1.36 (1.20-1.54) for SGA. The minimal increase in the risk for adverse perinatal outcomes associated with intervals of 24 to 59 months (3%-7%) was not statistically significant. It was not possible to estimate pooled ORs for the relation between interpregnancy interval and both fetal and early neonatal death, because the categories of intervals used and the reference categories did not coincide in all studies.

The estimates of pooled adjusted ORs were lower than estimates of pooled unadjusted ORs (Table 4). Nevertheless, the associations between intervals of shorter than 6, 6 to 11, 12 to 17, and longer than 59 months and preterm birth, LBW, and SGA remained statistically significant. Compared with infants of mothers with interpregnancy intervals of 18 to 23 months, those born to women with intervals shorter than 6 months had a 40% increased risk of preterm birth, a 60% increased risk of LBW, and an approximately 25% increased risk of SGA. Intervals of 6 to 17 months were associated with a significantly greater risk for the 3 adverse perinatal outcomes (adjusted ORs, 1.05-1.14). On the other hand, infants born to mothers with intervals longer than 59 months faced a 20% to 43% increase in risk of the 3 adverse perinatal outcomes. There were no differences in the risk of adverse perinatal outcomes between women with intervals of 24 to 59 months and those with intervals of 18 to 23 months. All funnel plots showed no asymmetry, either visually (funnel plots available from the corresponding author on request) or in terms of statistical significance (P>.10 for all, by Egger test).

Important statistical heterogeneity among studies was present, as confirmed by I² values greater than 50% in half of meta-analyses, and this remained in the prespecified subgroups. An examination for sources of heterogeneity among studies found that a significant portion of the heterogeneity in studies evaluating the relation between intervals shorter than 6 months and both preterm birth and LBW was explained by the study by our group⁶⁴ since the estimates of pooled adjusted ORs were significantly lowered when this study was excluded (1.30; 95% CI, 1.23-1.38; and 1.48; 95% CI, 1.40-1.57, respectively). Study quality, date of publication, sample size, and study setting provided no explanation for heterogeneity in studies evaluating the relationship between intervals of 6 to 11, 12 to 17, and 24 to 59 months and both preterm birth and LBW, because the CIs in the subgroups overlapped (data available from corresponding author on request). Compared with the overall results, the summary ORs calculated from sensitivity and subgroup analyses were almost identical. Pooled adjusted ORs calculated from subgroups evaluating intervals of 60 months or longer and preterm birth were similar to the overall adjusted OR calculated from all studies. With regard to studies evaluating the association between intervals longer than 59 months and LBW, studies from developed countries were significantly associated with higher pooled adjusted ORs. There were no significant differences in pooled adjusted ORs obtained from subgroups of studies and the overall estimates obtained from all studies assessing the association between interpregnancy interval and SGA. In general, there were no significant differences in estimates of summary adjusted ORs between white and black women in subgroups that evaluated the effects of interpregnancy interval on adverse perinatal outcomes according to race/ethnicity.^39- 40,52

For each month that interpregnancy interval was shortened from 18 months, the risk increase for preterm birth, LBW, and SGA was 1.9%, 3.3%, and 1.5%, respectively (Table 5). On the other hand, the risk for the 3 adverse perinatal outcomes increased by 0.6%, 0.9%, and 0.8%, respectively, for each month that interpregnancy interval was lengthened from 59 months.

Using 3 different meta-analytical techniques, we show that birth to conception intervals shorter than 18 months and longer than 59 months are significantly associated with increased risk of several adverse perinatal outcomes, such as preterm birth, LBW, and SGA. Infants can have LBW either because they are born early (preterm birth) or are born SGA. Thus, the association between interpregnancy interval and LBW could be due to the independent effect of interval on both preterm birth and SGA. Less clear is the association between birth spacing and the risk of fetal and early neonatal death, although results from meta-regression curves suggest that interpregnancy intervals shorter than 6 months and longer than 50 months are associated with increased risk of these adverse perinatal outcomes. The strength of our inferences is based on compliance with stringent criteria for performing a rigorous systematic review. These included the use of a prospective protocol designed to address a research question; the methods used in the identification of relevant studies; no language restrictions; the exclusion of studies that did not adjust for at least maternal age and socioeconomic status; the strict assessment of methodological quality of included studies; the use of several techniques of meta-analysis (both unadjusted and adjusted analyses); the exploration of sources of heterogeneity; the quantitative summarization of the evidence; and the inclusion of a large number of women from different populations throughout the world.

The reasons for the association between a short interval between pregnancies and adverse perinatal outcomes are unclear. A plausible explanation is the maternal nutritional depletion hypothesis,^27,81 which states that a close succession of pregnancies and periods of lactation worsen the mother's nutritional status because there is not adequate time for the mother to recover from the physiological stresses of the preceding pregnancy before she is subjected to the stresses of the next. This results in depletion of maternal nutrient stores, with the subsequent increased risk of adverse perinatal outcomes.⁸¹ The folate depletion hypothesis claims that maternal serum and erythrocyte concentrations of folate decrease from the fifth month of pregnancy onward and remain low for a fairly long time after delivery. Women who become pregnant before folate restoration is complete have an increased risk of folate insufficiency at the time of conception and during pregnancy. As a consequence, their offspring have higher risks of neural tube defects, intrauterine growth restriction, preterm birth, and LBW.⁸² Some investigators have attributed the higher risk of poor pregnancy outcomes to several factors associated with having short intervals, such as socioeconomic status, unstable lifestyles, failure to use health care services or inadequate use of such services, unplanned pregnancies, and other behavioral or psychological determinants.^4- 5 However, the fact that the birth spacing effects are not strongly attenuated when socioeconomic and maternal characteristics are controlled for suggests that the effects are not caused by these confounding factors.

Some hypotheses have also been proposed to explain the relationship between long intervals and adverse perinatal outcomes. Zhu et al⁴⁹ have hypothesized that, after delivery, a woman's physiologic reproductive capacities gradually decline, becoming similar to those of primigravid women (ie, “the physiological regression hypothesis”). This hypothesis is supported by the observation that perinatal outcomes for infants conceived after an excessively long interpregnancy interval are similar to outcomes of infants born to primigravid women. Another possibility is that unmeasured factors, such as sexually transmitted infections or maternal illnesses, may cause both adverse fertility and pregnancy outcomes.^5,49 These factors could differ for women in developed and developing countries. Finally, residual confounding may still be an explanation for at least part of the reported associations.

Several potential limitations of our review must also be considered. First, like any systematic review, it is limited by the quality of original data. The great majority of studies calculated the interpregnancy interval using mother's recall of her previous child's date of birth and her last menstrual period, instead of birth dates recorded on the birth records and gestational age estimated from ultrasonography. In most studies the intervals were calculated as the time elapsed between 2 consecutive live births, ignoring induced or spontaneous abortions or fetal deaths between them, which can produce even longer intervals between live births. Nevertheless, this problem would not affect the findings for short intervals. In addition, several studies did not properly address the potential confounding effects of factors other than maternal age and socioeconomic status. Second, because there was considerable statistical heterogeneity in most of the meta-analyses performed, our findings should be interpreted with caution. Nevertheless, in the great majority of comparisons the estimates showed the same direction of effect, which could suggest the absence of clinical heterogeneity among the studies. Investigation of possible sources of heterogeneity provided no plausible explanations. In addition, it is possible that the I² heterogeneity test could have excessive power when there are studies with large sample size, as was the case with some of the ones included in our meta-analyses. Third, the number of studies available for analysis on the relationship between birth spacing and some adverse perinatal outcomes is still too small to provide conclusive evidence.

The effects of birth spacing on perinatal health found in our study, as well as the effects of both short and long intervals on infant, child, and maternal health,^1- 2 should furnish a strong motivating force for health personnel to provide family planning. The health sector should supply such care not only to those wishing to limit their fertility for personal, social, or economic reasons, but should also provide the needed services to those practicing family planning for health reasons. The results of our systematic review could be used by reproductive clinicians around the world to advise women on the benefits of delaying a subsequent pregnancy for approximately 2 to 5 years to improve the health of both mother and the next infant.

Despite the advances during the last 2 decades in understanding the relationship between birth spacing and adverse pregnancy outcomes, little information is available to explain the mechanisms by which birth spacing might improve the health of mothers and their children. Also, more studies are needed on whether the effects of birth spacing on perinatal health differ in developed vs developing nations. Finally, it is imperative to understand the causes for both short and long intervals in any population to interpret the data on health risks. The consequence of this may be that family planning policies and messages may need to be tailored to different populations.