Caesarean delivery and subsequent pregnancy interval: a systematic review and meta-analysis

The main objective of this systematic review and meta-analysis is to collect and interpret the published literature on the association between a Caesarean delivery and subsequent sub-fertility (time to next pregnancy or birth) and to calculate a pooled estimate of the odds of sub-fertility following a Caesarean delivery.

The outcome of interest is subsequent sub-fertility, defined as pregnancy interval (the waiting time to next pregnancy or birth). Pregnancy interval has been classified according to the standard definitions used in the literature to date for the purpose of this systematic review and meta-analysis as follows:

The outcome will be a dichotomous measure of the time to next pregnancy within the follow-up time period reported in each study. Where a study reports estimates of two separate intervals as a measure of sub-fertility (i.e. a period of >1 year or a period of >3 years trying to conceive), the longer time interval will be used in the meta-analysis. Furthermore, where a study provides stratified estimates and sub-group analyses, only the overall estimate of sub-fertility will be included in the meta-analysis.

The protocol for this systematic review and meta-analysis was registered on PROSPERO, the international prospective register of systematic reviews (unique identification number: CRD42012003166) and is available in full on the National Institute for Health Research (NIHR) website [26]. This systematic review and meta-analysis was conducted in accordance with the preferred reporting items for systematic reviews and meta-analyses statement (PRISMA) [27], which is a detailed checklist of items specifically designed for this purpose. We performed a comprehensive search of the published literature using a combination of medical subject headings (MeSH) or key word terms for Caesarean delivery and sub-fertility (time to next pregnancy including IPI or BI). We searched the following databases for potentially eligible studies: CINAHL (1981 to October 17^th 2012), the Cochrane Library (1993 to October 17^th 2012), Embase (1974 to October 17^th 2012), Medline (1966 to October 17^th 2012), PubMed (1966 to October 17^th 2012), SCOPUS (1960 to October 17^th 2012) and Web of Knowledge (1945 to October 17^th 2012). Terms for Caesarean delivery included Caesarean section, Caesarean delivery, abdominal delivery, C-section, mode of delivery, and post Caesarean delivery. Terms for pregnancy interval included birth interval, birth spacing, pregnancy interval, first birth interval, inter-delivery interval, pregnancy gap and pregnancy spacing (Additional file 1). We enhanced our electronic search by cross-checking the reference lists of all relevant studies. We included studies which examined the association between mode of delivery and sub-fertility defined as time to next pregnancy or birth. No date or language restrictions were imposed. Eligibility criteria for inclusion in the systematic review included:

Eligibility criteria for inclusion in the meta-analysis (in addition to the above criteria) included:

Titles and abstracts of studies retrieved from the search strategy were reviewed independently applying the appropriate inclusion and exclusion criteria by the researchers. The full text article was obtained for all potentially eligible studies for further appraisal. Using a standardised data abstraction form, two assessors (SMON, ASK) individually entered the following data from each study: first author’s name, year, study design, location of the study (country), study period and follow-up, data source (i.e. hospital database, nationwide registers, patient records), total sample size, measure of pregnancy interval used (IPI or BI, and whether it was measured as a continuous or categorical outcome), statistical tests used, the indication for Caesarean delivery (breech, elective, emergency), the population sampled (i.e. primiparous, multiparous women), the exclusion criteria and the main results and conclusions of each study. Discrepancies in data abstraction between assessors were resolved through consensus.

Heterogeneity between the included studies was assessed by examining the differences in study characteristics including study setting (high-income developed country versus low-income developing country), study design (case–control, cohort or cross-sectional), sampling frame (population-based or single/select institutions), and definition of the outcome measure used. In the meta-analysis, we used the I² statistic to ascertain statistical heterogeneity, following the Cochrane Handbook for Systematic Reviews threshold recommendations [29]. An I² statistic value between 0% and 40% suggests heterogeneity may not be important; 30% to 60% represents moderate heterogeneity; 50% to 90% represents substantial heterogeneity; and 75% to 100% represents considerably significant heterogeneity. The magnitude and direction of the effects (Chi-squared test P-value, 95% confidence interval for I²) determine the importance of the I² statistic according to the handbook.

The methodological quality was assessed using a quality assessment tool based on six different types of bias common in observational studies (selection, exposure, outcome, analytic, attrition and confounding) for each study included in the review. This bias classification tool has been described in detail elsewhere [30]. Study bias was recorded as minimal, low, moderate or high according to the perceived degree of bias present for each of the six types of bias. For example, for confounding factor bias a study would be classified as minimal in bias if it adjusted for at least three key confounders including maternal age. An overall likelihood of bias was then estimated based on the total amount of bias perceived to be present.

Ethical approval was not required for this systematic review and meta-analysis as it did not include any experimental research on humans or personal information which would otherwise require informed consent.

The database searches produced 13,658 citations, of which the titles and abstracts of 10,241 were screened after removal of 3,417 duplicates. Of these, 12 were considered potentially relevant and the full text was obtained (Figure 1). From here, four were excluded with the main reasons being lack of data on the exposure (mode of delivery) or the outcome of interest (sub-fertility defined as time to next pregnancy/birth) and one study [31] used the same data source as another eligible study [16] and so in order to avoid duplication of data was removed. After hand-searching the reference lists of eligible studies, a further three studies were obtained yielding a total of 11 studies which met the inclusion criteria of the systematic review.

The heterogeneity of the 11 included studies is presented in Table 1, where the different characteristics including sample size, region, population sampling and data sources are outlined. Ten of the 11 studies were cohort studies. Only one study [10] was conducted in a low-income developing country. In addition, the studies were conducted in varying time periods, including one study from the 1950’s to early 1980’s [36]. Seven studies collected information from single institutions or interviews and questionnaire data. The I² statistic was used to quantify statistical heterogeneity and ranged from 0% to 88%, a high amount of heterogeneity between the studies. Definitions of pregnancy interval used varied between studies, with 44% heterogeneity in studies using BI compared to 45% heterogeneity in studies using IPI.

The studies varied in methodological quality (Table 4). According to the bias classification system used (Additional file 2), six were considered high in overall bias [17, 32, 33, 35‐37] due mainly to lack of adjustment for confounding or small sample size, [34] and five were considered low in bias [7, 10, 16, 18, 34]. The five studies considered low in bias and which adjusted for a minimum of three potential confounders including maternal age were included in the meta-analysis.

Strengths of the review include the thoroughness of the literature search which included seven databases, two reviewers, a detailed search-strategy compiled with a team of obstetricians and epidemiologists and rigorous cross-checking of reference lists. Furthermore, no date or language restrictions were enforced. Unfortunately, limitations to this review include firstly that any systematic review is limited by the quality of the original data collected, especially with observational studies. For example, only five out of the 11 included studies adjusted for potential confounders [7, 10, 16, 18, 34], and the other six studies used matching to reduce the likelihood of bias [17, 32, 33, 35‐37]. It was not possible to meta-analyse the six studies which only presented crude estimates as each reported pregnancy interval using very different measures making it impossible to combine the estimates. Sample sizes were quite small in three of the studies that reported an association [32, 35, 36] and thus the findings may be due to chance. Furthermore, the indication for mode of delivery (instrumental vaginal, vaginal breech, elective caesarean section, emergency caesarean section, etc.) was available in only two of the meta-analysed studies [16, 18] and so confounding by indication may also exist in some of the included studies. Short periods of follow-up may also be an issue, as some women were followed for short periods including: one year [36], three years [35], four years [34] and five years [17, 33, 37], which may not be a sufficient amount of time to allow for a subsequent pregnancy in some cases. The loss of a child and the desire to replace the loss is another important factor that is not acknowledged in all studies, with only one study stratifying analyses based on infant survival or death within the first year [16].

Finally, to know whether there is any association between Caesarean delivery and subsequent sub-fertility, it would be necessary to know whether women were deliberately delaying pregnancy (difficult deliveries in the past, desired family size obtained, oral contraceptive use, varying recommendations by doctors on the optimal time to wait following a Caesarean delivery before trying for another baby again, or change of partner). It is also not possible to establish whether any fertility treatment was used by the study populations. This in itself would be a major confounder as a woman who was already sub-fertile may have an increased risk of sub-fertility in the future. Ideally, subsequent research using registry data for instance, should aim to access the fertility registers (available in the Scandinavian countries for example) in order to exclude or account for certain groups such as women with a history of poor reproductive outcomes or women who accessed fertility treatments such as in vitro fertilisation (IVF). Sub-group analyses by other important risk factors such as advanced maternal age should also be prioritised. The main focus should be on the likelihood and timing of a subsequent pregnancy in women with a previous Caesarean section compared to vaginal delivery where no adverse outcomes occur (i.e. in live births only) in order to minimise the bias of pre-existing sub-fertility. In this review, only three studies used a cohort of women actively trying to conceive again [7, 10, 36]. It was possible to include two out these three studies in a sub-group analysis of women actively trying to conceive, generating a pooled OR of 0.82 (95% CI 0.72, 0.94) as the rate of subsequent sub-fertility. Secondly, only published literature was included in this systematic review and so publication bias may be cited as a reason for insufficient validity. It is however unlikely that any large study was missed due to the thoroughness of the literature search.

Much of the research to date investigating a woman’s fertility has been based on time to pregnancy studies, with retrospective designs using a well-defined target population, which are cheaper to conduct with the added benefit of higher external validity compared to prospective studies [40]. However, methodological issues exist in time to pregnancy studies including the inability to obtain detailed time-specific information about behaviour and risk factors [41]. Appropriate study design can minimise the biases associated with time to pregnancy studies according to Joffe et al. [40], who describe three main sampling frames: 1) pregnancy-based studies recruiting pregnant or recently delivered women who are easy to define, gain access to and recall bias is minimal, but sterile couples are excluded and sub-fertile couples under-represented; 2) cross-sectional population-based or occupationally-based studies where couples are randomly selected from the general population or an occupational group and a detailed obstetric history can be retrieved on one or more pregnancies, and failed attempts. However, long recall may lead to possible bias, for exposure variables and covariates; 3) population-based birth cohort studies use a previously defined birth cohort, who are questioned about their reproductive history, allowing for a longitudinal aspect of factors from earlier life to be studied. The sampling frames of each study included in the systematic review can be found in the supplementary material (Additional file 3).

Different biases including selection bias (including successful pregnancy attempts only), ‘wantedness’ bias (where couples say that a pregnancy was planned when it was not for fear of appearing not to want the child), planning bias (excluding couples with accidental or unplanned pregnancies), and pregnancy recognition bias (where a pregnancy is identified very early into the gestation may result in miscarriage in some cases) can be controlled if studies are well-designed and conducted, and through the use of appropriate sensitivity analyses in properly selected populations [42]. Time to pregnancy data has shown to be valid even where there is a recall of 20 years, although non-birth outcomes such as miscarriage are more commonly excluded [40, 43‐45].