Regan et al. Reply to “Sibling Comparison Design in Birth-Spacing Studies”

Interpregnancy interval, if causally linked with adverse perinatal health outcomes, is a potentially modifiable risk factor that can be targeted through public health intervention to improve maternal and child health. However, it currently remains unclear whether interpregnancy interval has a true causal impact on perinatal health. Recent published maternally matched studies have cast some doubt on this effect, raising the possibility that uncontrolled confounding has driven many of the previous findings supporting these associations. Even among matched studies investigating the impact of interpregnancy interval, results have been highly heterogeneous (1), with conflicting evidence for harm from short or long intervals (2–5). The uncertainty provoked by these recent matched studies necessitates further investigation, and different approaches have much to offer in this regard.

While useful for assessing exposure-outcome relationships, matched designs bear their own challenges. As highlighted by Hutcheon and Harper (6) in their commentary on our study (7) in this issue of the Journal, matched models might be susceptible to residual within-mother temporal confounding. These studies also have greater setup requirements, including longer follow-up to observe 3 or more births (to enable matching of 2 or more intervals) and linkage of pregnancies to the same woman. Furthermore, results of matched studies might not apply to recurring events, which are excluded from analysis as concordant strata. Despite these limitations, matched designs provide the opportunity to explore the role of unmeasured time-invariant confounding and how this could influence our understanding of interpregnancy interval. As a result, matched designs are useful methods for establishing evidence in relation to pregnancy spacing recommendations.

In our study, we applied a matched design to the investigation of the perinatal health impacts of a single interval per mother (7). Unlike a classical matched study on interpregnancy interval, we included an observation for a woman’s first birth, which allowed adjustment for risk factors from the first pregnancy (which does not, by definition, have an interpregnancy interval). Because maternal risk factors are shared across both first and second births, and both births are used to adjust for these factors, adjustment is made using within-mother information. This also allowed for the inclusion of women with fewer than 3 children, representing 63% of multiparous pregnancies in our sample. Our model further adjusted for factors that vary at an individual level between pregnancies. When adjusting for confounders that vary between pregnancies (i.e., maternal age), which the simulation study did not (6), the effect of interpregnancy interval is not conditioned out. As is the case with case-crossover designs, only discordant pairs contributed data to the model. While this reduces the analytic sample size and has implications for power and precision, exclusion of concordant pairs is not necessarily a flaw of the design. This exclusion is analogous to case-crossover designs, which have proven to be valuable tools in further evaluating the potential impacts of other perinatal exposures (8, 9).

In the absence of randomized trials, we agree with Hutcheon and Harper (6) that further investigation employing complementary methods is still needed in order to provide appropriate guidance for family planning. Should future studies confirm that there is little impact from short or long interpregnancy interval, other factors might take precedence in a family’s decision-making, such as desired family size, maternal age, and other family circumstances. Matched designs are not intended to replace unmatched designs. Rather, they offer us the opportunity to reevaluate the influence of interpregnancy interval from an alternative perspective to previous unmatched cohort studies. Given the uncertainty around the effects of interpregnancy interval and the need to provide evidence-based recommendations to families, we believe further investigation is warranted, and novel approaches using complementary study designs (which have historically been underutilized in this area) will be important for providing such evidence.