Overall, we found some variation by socioeconomic factors: less than we expected. In addition, we found consistent, somewhat surprising differences between regions. Although Switzerland maintains a long tradition of peaceful cohabitation between people speaking German, French, or Italian as mother tongues, cultural, social, and behavioural factors, including diet, tobacco smoking and alcohol consumption, voting behaviour at referendums, social policies, and organization of public healthcare vary by region [4, 41–44]. Interpreting our findings requires nuance and further research.

Several mechanisms possibly explain lower risk of neonatal mortality in French-and Italian-speaking Switzerland. First, maternal and foetal health are possibly better from a truly lower incidence of chromosomal abnormalities, severe congenital malformations, and pregnancy complications. However, we consider it rather unlikely—we found lower rather than higher birth weight and gestational age in French and Italian language regions, endpoints often taken as proxy for general foetal and maternal health [4]. Second, even if regions showed similar underlying maternal and foetal health, regions possibly report stillbirth registrations and early neonatal deaths slightly differently. For example, neonates with anencephaly might be reported as stillbirth or as neonatal death. Third, offers or uptake of prenatal screening and decisions or methods for terminating pregnancy possibly vary. In Switzerland, one-third of neonatal deaths before 28 weeks gestation are due to late pregnancy terminations for medically-approved imminent maternal harm or maternal major psychological distress about foetal problems with poor prognoses, such as congenital malformations and chromosomal aberrations [45]. Under such circumstances, infants are stillborn if feticide is conducted prior to birth, yet if pregnancy termination is solely performed by inducing labour, infants may be born alive and die shortly thereafter [46]. Although live births are registered for all infants with signs of life—independent of gestational age or birth weight—neonatal deaths are only reported from 22 weeks gestation or for birth weights above 500 grams. Given 63% of neonatal deaths in our dataset occurred before 28 weeks gestation (Table S2), regional practice variations combined with primary resuscitation differences for preterm infants born at limits of viability perhaps strongly influenced neonatal mortality rates [47].

We thus suggest systematic differences in pregnancy-related policies and actions by parents and physicians possibly explain at least part of our findings. This hypothesis is also supported by associations with religion and lower neonatal mortality among couples reporting “no religion.” We could reasonably hypothesize such parents as more likely to terminate pregnancies with lethal malformations.

We suggest future studies using routine administrative data ideally integrate information about stillbirths and late pregnancy terminations. Larger datasets would allow for differentiating between risk factors for perinatal, late neonatal, and postneonatal death. Strong evidence shows social patterning differs between these two outcome measures, with SEP explaining mainly postneonatal deaths, particularly sudden infant death syndrome [13, 48]. Including more information about birth mechanism, such as Caesarean section or instrumental deliveries, and health behaviours, such as smoking, in routine Swiss statistics, would further improve exploration of causal pathways. Finally, we need qualitative interview studies with healthcare providers, policy makers, and parents in different language regions and within population strata to shed light on potential mechanisms and develop policies.