This study included all primiparous women (first birth) who conceived without assistance a singleton live birth in Victoria, Australia between 2005 and 2015. Data on the second birth were included up until December 2017, ensuring that every participant had a minimum of 2-year follow-up after their first birth. This was the maximal sample size available because linked IVF data for Victoria were available until 2017 only. We excluded all twins and higher-order multiple pregnancies, stillbirths, women who had previously required any form of ART for conception in their first pregnancy, those with missing mode of birth, and those for whom a first and second birth were not able to be reliably linked. Stillbirth data were not reliably captured across all data sets and were thus excluded from the cohort.

Perinatal outcomes were obtained from routinely collected, audited birth outcome data through the Victorian Perinatal Data Collection (VPDC). ^, By law, every birth after 20 weeks’ gestation within the state is registered within the VPDC, with over 160 data points relating to pregnancy, labor, and birth outcomes collected. This register is regularly audited for quality and completeness. ^, IVF data were obtained from the 3 largest IVF units in Victoria at the time of the study, which provided maternal records from all cycles that resulted in a birth from January 2005 to December 2017. Previous studies using the same linked data set estimated that this includes >95% of all IVF conceptions during this time period. These databases were linked using probabilistic linkage by the third-party Centre for Victorian Data Linkage. Cases that were not linked were equally likely to be unassisted conceptions as to have been conceived through IVF. This data set has been used in separate research examining childhood outcomes following IVF conception, and further details on the linkage process can be obtained from related publications.

Our study exposure was birth by cesarean delivery, compared with vaginal birth. Cesarean delivery included all cesarean deliveries, both elective and in-labor (emergency) procedures. Vaginal births included unassisted vaginal birth and instrumental births (forceps and vacuum).

Two primary outcomes were examined: (1) the likelihood of giving birth to a live-born infant within the 12-year study period; and (2) among the cohort with a reported second birth, whether the use of IVF or other ARTs was required to achieve this pregnancy. “IVF” is used collectively to describe conception through any form of IVF, including standard IVF, IVF with intracytoplasmic sperm injection, and any other associated laboratory techniques required. “Other ART” covers other techniques used to overcome subfertility, such as intrauterine insemination or ovulation induction.

We also assessed 2 secondary outcomes. First, the interpregnancy interval between the first and second birth. Interpregnancy interval was calculated as the difference in time between the date of the first birth and the calculated date of conception of the second ongoing pregnancy (derived from the date of the second birth and the gestational age at time of birth). Next, we assessed the likelihood of having a miscarriage between the first and second birth, as well as the number of miscarriages. Miscarriage was a derived variable, calculated from the difference in documented gravidity at the time of the second birth, compared with the gravidity at the time of the first birth.

Covariates included in the statistical analysis models were decided a priori by the multidisciplinary authorship team, which included those with expertise in perinatology, reproductive endocrinology, epidemiology, and statistics. To identify covariates that may impact both mode of first birth and secondary subfertility, a directed acyclic graph was generated to illustrate the structure of the relationship between variables ( Supplemental Figure ). All covariate data were obtained from the VPDC data set. Hypertensive disorders of pregnancy and diabetes in pregnancy were identified using ICD-10 (International Classification of Diseases, Tenth Revision) codes.

Final covariates adjusted for included Socio-Economic Indexes for Areas (SEIFA) quintile, maternal age, chronic hypertension, and preexisting diabetes. Maternal body mass index (BMI) was considered as a potential confounder but was not recorded in the VPDC before 2009 and was therefore unable to be included. Age at second birth was also adjusted for all reproductive outcomes in which a second birth occurred (use of ART, interpregnancy interval, and miscarriages rates). Miscarriage was also included in the model for interpregnancy interval given its inevitable impact on the timing of a subsequent birth, but was not included in the model for the primary outcomes. Although miscarriage is on the causal pathway between first pregnancy delivery mode and interpregnancy interval length, adjusting for it provides a better estimate of the direct impact of cesarean delivery on the time taken to conceive the subsequent child.

Descriptive statistics were calculated and reported for each outcome and covariate among the overall cohort, and by mode of first birth, according to the type and distribution of data.

For our first primary outcome (the probability of having a second ongoing pregnancy within the 12-year study period), Cox proportional hazards regression was used, given the variable duration of follow-up. The proportional hazards assumption was tested statistically and using Kaplan–Meier survival curves. The start point for the model was the date of birth of the first pregnancy, and the end point was the estimated conception date of the second pregnancy that led to a live birth within the 12-year study period. As a planned sensitivity analysis, we restricted the data to those who had their first birth before December 2012, which ensured a minimum of 5-year follow-up after first birth. This reduced the potential impact of electively longer interpregnancy intervals. For our second primary outcome (the probability of requiring IVF or other ART for conception to achieve the second birth), risk ratios (RRs) were generated using Poisson regression modeling.

For all outcomes, crude and adjusted estimates were calculated. Maternal age was assessed as both a continuous and categorical variable, as well as with squared and cubic terms. In the final multivariable model, age and age squared were included after identifying the best-fitting model using likelihood ratio tests and assessment of postestimation statistics.

For both primary outcomes, we performed planned sensitivity analyses, restricting the cohort to term births only. In planned subgroup analyses, we also analyzed the exposure groups separately according to whether the cesarean delivery was elective or in-labor (emergency). Additionally, for the cohort for which BMI data were available, we stratified by BMI category to determine the potential impact of BMI on the risk estimates.

For the secondary outcomes, miscarriage rates were analyzed using RRs calculated using Poisson regression. The interpregnancy interval was not normally distributed, and the median difference in interpregnancy interval was thus calculated using quantile regression.

Those with missing exposure data (mode of delivery) were excluded. No ART outcome data were identified as missing because any missing variables were considered spontaneous conceptions by default. For the covariates used in the analysis (SEIFA and maternal age), missingness was low (<1%), and analyses were conducted on complete cases. For hypertensive disorders and diabetes in pregnancy, the presence of a documented ICD-10 code was considered evidence of a diagnosis.

Ethical approval for the project was obtained from the Mercy Health Human Research Ethics Committee (HREC) (2018– 017), the Monash Health HREC on behalf of Monash IVF (MH 16172M), and the Melbourne IVF HREC (58/18-MIVF). Each data custodian provided contractual approval for data access and linkage.