Descriptive characteristics of the cohorts

The cohort with two pregnancies consisted of 612 935 women (full cohort), of which 512 515 had a spontaneous onset of the second birth (subcohort) (Table 1). The first delivery was by CS in 79 965 pregnancies (13.0%) in the full cohort and in 44 264 pregnancies (8.6%) in the subcohort of only spontaneous onsets of the second birth. The distribution of the outcome variable, the gestational duration of the subsequent pregnancy categorised, is presented in the lowermost part of Table 1. Analysing the subcohort, the proportions of deliveries within the different gestational age intervals differ between the groups: the previous-CS group has higher birth rates in earlier and later weeks (Fig 3). Other differences between the groups are higher prevalence of diabetes, hypertensive disorders, fetal malformations, growth alterations and baby boys in the previous-CS group. Moreover, the birth rate through CS in the second pregnancy after a CS is 26.4% compared to 1.9% after a vaginal delivery.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 3. Multivariable logistic regression analyses, first two pregnancies.

Adjusted odds ratio for gestational age at birth within the specified interval, in the second pregnancy after cesarean section in the first pregnancy. Only mothers with spontaneous onset of birth in the second pregnancy were included. aOR: adjusted odds ratio.

https://doi.org/10.1371/journal.pone.0317492.g003

Download:

• PNG
  larger image
• TIFF
  original image

Table 1. Cohort characteristics, first two pregnancies.

https://doi.org/10.1371/journal.pone.0317492.t001

The cohort with three pregnancies consisted of 157 581 women (Table 2). The onset of the third birth has a big impact on the observed proportion of preterm births: for women with two previous CSs, the rate of preterm birth is 5.1% for the group as a whole, but 18.1% if only women with a spontaneous onset of the third birth are included. The Swedish obstetrical practice of recommending vaginal delivery after one previous CS is illustrated by the birth rates through CS being 26.4% for the second delivery after one CS, and 98.7% for the third delivery after two CSs.

Download:

• PNG
  larger image
• TIFF
  original image

Table 2. Cohort characteristics, first three pregnancies.

https://doi.org/10.1371/journal.pone.0317492.t002

For two consecutive pregnancies, a previous CS shifted more weight to the extremes of the distribution of the gestational duration of the second pregnancy (Fig 4(A)). For three consecutive pregnancies (Fig 4(B)), the seemingly shorter gestational duration of the third pregnancy for the group with two previous CSs stands out. However, the graphs only include women with a spontaneous onset of the last birth and illustrate the bias discussed before (overinclusion of women with delivery in early gestational weeks), which happens when excluding mothers with an iatrogenic onset of the outcome pregnancy. This illustrates the difficulty in separating the true impacts of previous CSs on the duration of future pregnancies from technical artifacts.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 4. Gestational duration of the second and third pregnancies.

The distributions of gestational duration of the second and third pregnancies, stratified by previous delivery modes. Only mothers with spontaneous onset of birth in the final pregnancy of the series. CS: cesarean section.

https://doi.org/10.1371/journal.pone.0317492.g004

Logistic regression shows an increased risk of spontaneous preterm and postterm birth after previous cesarean section

For the cohort with two consecutive pregnancies (Fig 3), previous CS was associated with a higher risk of spontaneous preterm and postterm birth, with larger effect sizes for intervals further away from term (logistic regression, all p-values highly significant, <0.001). The three-pregnancy cohort (Fig 5) also shows an association between CS and subsequent spontaneous preterm birth. The risk of spontaneous preterm birth is however not higher in the third pregnancy if a mother had a CS in the first pregnancy followed by a vaginal birth in the second pregnancy, compared to mothers with two previous vaginal deliveries (aOR 1.15, 95% CI 0.96–1.36). This specific result stands out in relation to the rest of this study, where a CS consistently has an association with spontaneous preterm birth in a subsequent pregnancy.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 5. Multivariable logistic regression analyses, first three pregnancies.

Adjusted odds ratio for preterm birth in the third pregnancy for the specified comparison between different sequences of previous delivery modes. Only mothers with spontaneous onset of birth in the third pregnancy. CS: cesarean section, Vag: vaginal delivery.

https://doi.org/10.1371/journal.pone.0317492.g005

As a sensitivity analysis, regression models were created which only included pregnancies where the gestational duration was estimated by ultrasound. The results (Supporting Information S1 and S2 Figs) show no meaningful differences compared to the models relying on the best estimate of the gestational duration.

Survival analyses show the link between cesarean section and prolonged pregnancy

The exclusion of women with an iatrogenic onset of birth of the outcome pregnancy in a logistic regression model leads to falsely high odds ratios for delivery before a scheduled CS date. This problem is clearly seen with the unreasonably high odds ratios in Fig 5 (birth distribution if Fig 4(B)). The results in Fig 3 have the same problem (birth distribution in Fig 4(A)). Our solution to the bias generated by the unfortunate combination of the clinical management policy together with logistic regression is to use survival models.

The Kaplan-Meier curves of Fig 6(A) show a prolonged pregnancy for the previous-CS group compared to the previous-vaginal delivery group. In Fig 6(B), the analogous survival analysis for the cohort with three consecutive pregnancies shows the scenario of two previous CSs giving the longest prolongation of the third pregnancy, while the scenario of two previous vaginal deliveries corresponds to the shortest gestational duration of the third pregnancy. The scenario of a CS in the first delivery followed by vaginal delivery in the second pregnancy yet again shows a very slight difference versus two previous vaginal deliveries, consistent with the results of the logistic regression model (Fig 5) that showed no statistically significant association with spontaneous preterm birth for that group.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 6.

Kaplan-Meier curves, series of two (A) and three (B) pregnancies. The curves illustrate the estimated probabilities of “survival” (i.e. the curve value at t days is the probability that a pregnancy lasts longer than t) depending on previous delivery mode. All mothers regardless of the onset of the final birth of the series are included, the mothers with an iatrogenic onset of the final birth are censored from the gestational age at birth.

https://doi.org/10.1371/journal.pone.0317492.g006

The Cox regression models (Table 3) calculate the adjusted hazard ratios (aHR) between the different scenarios of previous delivery modes using the same covariates as the previous logistic regression models. For two pregnancies (Fig 6(A), Table 3), the aHR for previous CS versus previous vaginal delivery is 0.72 (95% CI 0.71–0.73), i.e., a net effect of a prolongation of the pregnancy following a CS. For three pregnancies (Fig 6(B), Table 3), the largest prolongation of the third pregnancy is seen when comparing two previous CSs versus two previous vaginal deliveries (aHR 0.64, 95% CI 0.59–0.69). A very slight prolongation is seen when comparing the scenario of a CS followed by a vaginal delivery versus two previous vaginal deliveries (aHR 0.95, 95% CI 0.93–0.98).

Download:

• PNG
  larger image
• TIFF
  original image

Table 3. Cox regression models.

https://doi.org/10.1371/journal.pone.0317492.t003

Main findings

Using survival models, the main result of this study is that there is an association between a previous CS and a prolongation of the subsequent pregnancy. While logistic regression shows an increased risk for both spontaneous preterm birth (aOR 1.67, 95% CI 1.57–1.77) and spontaneous postterm birth (aOR 1.55, 95% CI 1.49–1.62) in the second pregnancy after a CS in the first pregnancy, this is not the preferred way of analysing this research question since it overestimates the risk of preterm birth in the second pregnancy. Further, a CS in the first pregnancy followed by a vaginal birth in the second pregnancy is not associated with an increased risk for spontaneous preterm birth in the third pregnancy, compared to two previous vaginal deliveries (aOR 1.15, 95% CI 0.96–1.36). Previous studies have described the association between CS and subsequent preterm birth, but the main novel finding of this study is the association between CS and subsequent prolonged pregnancy, with an extensive methodological elaboration on the importance of taking the type of onset of birth in consideration when studying this subject.

Strengths and limitations

A strength of this study is the reliability and the level of detail of the included data [22, 23]. A benefit of performing this study on Swedish patient data is the relatively high rate of vaginal birth after CS in Sweden (73.6%) compared to other countries [25]. Even if we have used a survival model to adjust for the bias due to a larger proportion of iatrogenic onsets of birth in the exposure group, the robustness of the results is helped by the very high rate of vaginal birth after one CS in Sweden, since it reduces the uncertainty of the results due to a lesser need for the model and its impacts on the observed data. The study covers a large time period and is nationwide, yet the clinical practice was similar over the entire country during the whole time frame of the study. The benefit of performing this study before the CS rate has increased even further that today is that it keeps problems generated by management differences between the exposure and reference groups as small as possible. A reasonable assumption is that Sweden will follow the global trend with an increased CS rate, and therefore it is likely that the use of advanced statistical maneuvers will become even more important, and unfortunately create even more interpretation problems than seen today.

There is, despite our best efforts to minimize it, still a risk of residual confounding that might have affected the observed results. The most significant risk we have identified is the possibility of confounding by indication, because even if we have tried to identify a reasonably homogenous group of women for inclusion and then adjust statistically for covariates differing between the groups, there will on the individual level always be a reason why the specific delivery mode was chosen, generating a difference between the exposure and reference groups that is unlikely to be fully resolved in any observational study design. A large portion of the potential bias that could be related to this phenomenon is, in our opinion, avoided through the inclusion of relatively healthy previous pregnancies, with the most critical step being the decision to only include previous-term pregnancies. Another limitation is the possibility that the outcome of previous pregnancies may affect how likely the woman is to have another pregnancy, which would alter the results of studies such as this, a limitation that cannot be resolved within a birth cohort. Other limitations are the lack of information in the MBR about the indication for the CS and at which stage of labor it was performed.

Interpretation

The worldwide CS rate is expected to keep increasing, leading inevitably to an increase in the negative consequences of CSs. The aim of this study was to investigate the link between previous delivery modes and the incidence of deviant duration of subsequent pregnancies. While at first glance, our results seem coherent with previous studies, the survival analysis of this study changes that interpretation. Including only mothers with a spontaneous onset of the subsequent birth seems reasonable since we are mainly interested in any biological effects of a previous CS on its gestational duration. However, such selection generates a bias because the group of women with a previous CS are more likely to have an iatrogenic onset of the next pregnancy (45% vs 12%, Table 1), and therefore, spontaneous births will per definition occur before the date of an iatrogenic onset. Excluding women with an iatrogenic onset of birth, while keeping those with a spontaneous onset of birth, will exclude a relatively larger fraction of women from the previous-CS group, making that group seem smaller than it really is, and with a heavy burden of preterm birth, overestimating the risk of preterm birth for women with a previous CS. The apparent effect of a previous CS leading to increased preterm birth risk will largely be due to the study design, as the increased risk of spontaneous preterm birth after previous CS is no longer observed when applying survival models. In these, mothers with an iatrogenic onset of birth in the outcome pregnancy are kept; in effect, they are “followed-up” until the time of their CS. While we do not know their “natural” delivery date, we know that most of these mothers delivered around term, i.e. they had a low risk of spontaneous preterm birth. By including these mothers, the survival analysis shows that previous CS does not increase the risk of preterm birth. Thus, retaining the full group, i.e. not excluding the mothers with an iatrogenic onset of the next delivery, is essential. This result implies that the increased risk observed with logistic regression in this and other studies is likely attributable to a technical artefact rather than a true consequence of previous CS.

In fact, with appropriately chosen analysis models, we instead see an association in the other direction–between a previous CS and a prolonged subsequent pregnancy. To our knowledge, this association has not been reported before, as previous research has focused on the increased risk of preterm birth subsequent to CS [26, 27]. A possible explanation to this phenomenon could be that the physiological alteration of uterine function resulting from a CS may exacerbate uterine intolerance or, in the case of prolonged pregnancies, diminish the sensitivity of uterine load [28]. One speculation is that the CS might cause a stiffness in the uterus during the healing process, with an ensuing problem in failing to recognize the signals of onset of labor. This might be due to either the scar itself having poor contractile properties, or the failure of the area to allow effective signaling between the body and the cervix of the uterus, altering the maturation of the cervix and the onset of birth mechanism.

Another interesting finding seen in the cohort with three pregnancies was that a vaginal delivery in the second pregnancy after a CS in the first pregnancy proved to be no different in terms of association with subsequent spontaneous preterm birth in the third pregnancy compared to having had two previous vaginal deliveries. It seems that a vaginal birth after a CS could possibly be considered “a stress test” of uterine function in terms of achieving vaginal delivery, selecting a subgroup of women with lower risk of developing complications due to their previous CS, perhaps due to a smaller or more fortunately placed incision.

Regarding differences between the previous CS- and previous vaginal delivery groups, the overrepresentation of baby boys in the CS group is noteworthy. This discrepancy might possibly be explained by the fact that baby boys seem to be at generally higher risk of fetal distress [29, 30]. The higher prevalence of SGA, LGA, preeclampsia, and diabetes in the CS group are expected, as these factors are known to increase the likelihood of developing fetal distress [26, 31, 32]. Having taken all this into account, the differences between the cohorts should be considered to be minor and adjusted for.

Since previous studies have seen increased risk for subsequent spontaneous preterm births after late-stage CSs [16–20], a subanalysis on late-stage CSs in our cohort might have added valuable information, particularly if late-stage CS also increases the risk of subsequent delayed onset of birth. For future studies, investigating if there is any evidence of alterations in uterine function after previous surgery unrelated to the scar itself, would be of interest. That could possibly be grounds for re-evaluating the results of this study.