This study is based on the Fetal Programming of Semen Quality (FEPOS) cohort [17], nested within the Danish National Birth Cohort (DNBC) [18]. In the DNBC, from 1996 to 2002, pregnant women were enrolled and interviewed by computer-assisted telephone interviews conducted around 16 and 30 weeks’ gestation, and at 6 months after childbirth.

The participants in the FEPOS cohort were recruited from March 2017 to December 2019. Recruitment to FEPOS is described in detail elsewhere [17]. In short, sons were eligible for invitation if their mothers responded to the first two telephone interviews during pregnancy in the DNBC and provided a gestational blood sample for the biobank within the DNBC. Invitation to the FEPOS cohort was further restricted to the following criteria: Son of at least 18 years and 9 months of age, living in close proximity to the FEPOS clinics in Copenhagen (the capital and largest city in Denmark) or Aarhus (the second largest city in Denmark). Sons were encouraged to decline participation if they had undergone sterilization, cancer treatment, orchidectomy, or had one or no testicles in the scrotum.

During the study period, 5697 of the 21,623 eligible young men were randomly invited to participate in FEPOS. Of these, 1174 (21%) responded to an online questionnaire on health behaviors, whereafter 1058 delivered a semen sample (participation rate 19%). After excluding sons with no semen sample and/or missing data on delivery mode, 1044 mother-son pairs (18%) were included in our main analysis (Fig. 1).

Flow chart of participants, the FEPOS Cohort, Denmark, 1998–2019

Information on delivery mode was extracted from the Danish Medical Birth Registry (DMBR) [19, 20] based on ICD-10 diagnostic codes and procedure codes according to the Nordic Medico-Statistical Committee classification of surgical procedures. Delivery mode included vaginal delivery (ICD-10: DO800, DO801, DO808, DO809, DO810, DO813, DO814, DO815, DO840, DO841), elective CS before labor (ICD-10: DO820 and operation codes KMCA10B or KMCA11), emergency CS during labor (ICD-10: DO821, DO842 and operation codes KMCA10D, KMCA10E) and unspecified CS (ICD-10: DO828, DO829 and operation codes KMCA10A). The latter three categories of CS (elective CS, emergency CS and unspecified CS) were merged into an overall CS exposure category in the main analysis due to low numbers in some of the separate categories.

Participants provided a semen sample for analysis either at the clinic or at home. If collected at home, participants were recommended to keep the sample warm during transportation to the clinic and deliver the sample within one hour of ejaculation. Recommended abstinence time was 2–3 days.

Collection and analysis of semen followed recommendations by the WHO 2010 [21]. Analysis was conducted by two trained biomedical laboratory technicians and started within a maximum of two hours from ejaculation in 1029 out of 1044 semen samples (99%). In total, 786 (75%) semen samples were analyzed within one hour. Semen volume was measured by weighing of the sample in a pre-weighed container assuming 1 g = 1 ml. After liquefaction at 37 °C, sperm concentration was determined on two aliquots of semen using an Improved Neubauer Hemocytometer. Total sperm count was calculated by multiplication of semen volume and concentration. Sperm motility was determined by assessing progressive, non-progressive and immotile spermatozoa for at least 200 spermatozoa within each of two fresh drops of semen. The outcome used in the study was progressive motility, which, to ensure optimal model fit, was assessed as non-progressive motility modeled as non-progressive + immotile spermatozoa in percentage. Morphology was analyzed at the Reproductive Medicine Centre, Skåne University Hospital, in Malmö, Sweden. The percentage of normal spermatozoa was assessed in approximately 200 spermatozoa per slide.

External quality control with the European Society of Human Reproduction and Embryology (ESHRE) External Quality Assessment scheme (Centre for Andrology, Karolinska University Hospital, Stockholm, Sweden) indicated no distinct differences between the FEPOS laboratory technologists and the expert reference examiners [17].

Potential confounders were identified a priori using existing literature and Directed Acyclic Graphs (DAGs) [22] (Additional file 1: Appendix S1). Maternal age at delivery was retrieved from DMBR. Information on maternal smoking in the first trimester, pre-pregnancy body mass index (BMI) and highest socioeconomic status of the parents was provided by the mothers in the first DNBC interview. Socioeconomic status was defined according to the International Standard Class of Occupation and Education codes (ISCO-88 and ISCED). Categorization of the covariates was presented in Table 1.

Information on precision variables was recorded at the clinical visit, including place of semen sample collection (at home or in the clinic), abstinence time (in days), spillage of semen sample (no; yes), and interval from ejaculation to analysis (in minutes).

Information on pregnancy complications, labor induction, gestational age at birth (in weeks) and birth weight (in grams) were used in sensitivity analyses. Pregnancy complications were self-reported in the women’s interview around 16 and 30 weeks of gestation and 6 months after childbirth and included diabetes mellitus (type 1, type 2 or gestational diabetes), hypertension and preeclampsia. Women, who reported one or several of these conditions, were defined as having pregnancy complications (none; one or more). Information on gestational age at birth, birth weight, and labor induction (pharmaceutical or non-pharmaceutical based on ICD diagnose codes DO837 and DO847 and procedure codes KMAC00, KMAC96A, BKHD20, BKHD20A and BKHD 21) were obtained from DMBR.

Data management and statistical analyses were conducted with STATA MP, version 15.1 (StataCorp, College Station, TX). According to the requirements of regulations [GDPR, Regulation (EU), 2016/679 of 25 May 2018)], a calculated percentile must be based on at least five observations. Thus, pseudo percentile 5, 50 (median), and 95 for the semen characteristics were calculated by delivery mode using STATA’s -sumat- commands.

As the distributions of the semen characteristics were over-dispersed, we applied negative binomial regression models to examine the association between CS and semen characteristics using STATA’s -nbreg- package. We estimated crude and adjusted relative differences in percentages with 95% confidence intervals (CIs) for each semen characteristics by comparing sons born by CS to those born by vaginal delivery.

Azoospermic men were excluded from analyses of motility and morphology (n = 17), and participants reporting spillage of the semen sample were excluded from analyses of semen volume and total sperm count (n = 180). All models were adjusted for abovementioned selected confounders, place of semen sample collection and abstinence time. The analysis of sperm concentration and morphology was further adjusted for spillage, and the analysis of motility was further adjusted for interval from ejaculation as sperm motility decreases with time [23].

We applied selection weights to all analyses to account for non-participation [24]. These were estimated as the inverse probability of participation derived from a multivariable logistic regression with participation status (yes; no) as the dependent variable and the primary exposure variable (delivery mode) and the potential confounding factors, in addition to region (Aarhus; Copenhagen), parental time to pregnancy (TTP) including use of medically assisted reproduction (MAR), included as explanatory variables. TTP and MAR were considered indicators of parental fertility, and these data were obtained in the first interview in the DNBC.

Assumptions were checked for each model. We compared the observed distribution of each semen characteristic against the model-based distributions from the fitted model using QQ-plots. Furthermore, the standardized deviance residuals were plotted against the model-based predictions. The model check was compatible with the assumptions (data not shown).

We performed three sensitivity analyses to assess confounding by indication. First, we further adjusted for birth weight z-scores [25] as being small or large for gestational age might be an important indicator for CS, and birth weight for gestational age has been reported to be associated with semen quality [26, 27]. Birth weight z-score was not included in the main analyses due to the risk of collider-stratification bias [28, 29]. Second, we restricted our analyses to women without induction of labor and pregnancy complications. Labor induction has been suggested by some researchers as an independent risk factor for CS [30, 31], and the indications for induction may be the same as for CS. In addition, severe pregnancy complications might be a potential medical indicator for CS. Third, CS was further classified into elective CS, emergency CS and unspecified CS, with vaginally-born sons as reference, to examine whether the type of CS was important for semen quality.