Study population
Since 1967, all births after 16 weeks of gestation have been reported to the Medical Birth Registry of Norway. Our study population comprised all births from 23 weeks of gestation during the period 1967 to 2005, a total of 2,263,736 births. We excluded all mothers that immigrated to Norway (N = 144,858). We want to focus on trends in long term mortality caused by improvements in maternity care and infant care. From 1967 to 2005, the proportion of deliveries by immigrant mothers increased from 2.3% to 15.6%. Children of immigrant mothers have, on average, higher mortality than the rest of the population, and there has been considerable variation over time in the countries that immigrants to Norway come from [
15,
16]. Inclusion of births by immigrant mothers would therefore make the sample more heterogeneous and complicate interpretation of results. Information on the mother’s country of origin was obtained by linking the Medical Birth Registry with the Norwegian Central Person Registry using the mother’s unique identification number.
Pregnancies recorded as lasting more than 43 weeks were excluded (N = 27,341), since gestational length may have been miscoded in some of these pregnancies. We could not with certainty determine which were miscoded, so we excluded all these pregnancies. Also, we omitted births with unknown sex of the child (N = 260) and without information on gestational age at birth (N = 116,466) leaving 1,974,811 births for the analysis.
Study factors
Gestational age at birth was calculated from the date of the last menstrual period for births during the years 1967 to 1998. For the years 1999 to 2005, gestational age at birth was based on estimates of term date at routine fetal ultrasonographical examinations in weeks 17–19 of pregnancy. Information from ultrasonographical examinations was available for 97.6% of the women, and for the remaining women the date of the last menstrual period was used. Information on vital status at birth was obtained from the Medical Birth Registry. Information about infant death and the age of the infant at death were obtained from the Central Person Registry.
We adjusted for maternal age, parity, plurality and maternal education at delivery, since these factors are associated with offspring death [
17,
18] and the composition of mothers with regard to these factors has changed during our study period. Information on maternal age (coded: five-year intervals), parity (coded: first child, other children) and plurality (coded: single, multiple births) was obtained from the Medical Birth Registry. Information about maternal education (coded: compulsory school only, upper secondary education, university/college education, unknown education) was obtained from the Education Registry [
19]. All educational institutions in Norway report yearly individual data to the Education Registry. Hence, we were able to obtain individual information about the highest education of the mother at the time of the delivery.
Statistical methods
We carried out separate analyses for each of the following periods of pregnancy: 23–28 weeks, 29–36 weeks and 37–43 weeks. In each analysis, only fetuses that were alive in utero at the beginning of the period were included. The outcome variable distinguished between pregnancies where the fetus died in utero or during birth and pregnancies where the offspring was born alive or was still alive in utero by the end of each period of pregnancy. For infants who were born alive, we did separate analyses for the following time periods after birth: ≤ 1 week and 1 week-1 month. For these analyses the outcome variable distinguished between infants who died during the period and infants who were alive at the end of the period.
To estimate absolute yearly change in fetal and neonatal death rates, the following linear probability model [
20] was fitted for each of the three pregnancy periods and two neonatal periods, and for the whole period from the 23rd week of pregnancy to one month after birth:
where Mortalityit is a dummy variable that is one if fetus/infant i in year t died during the period and zero otherwise, Boyit (Girlit) is a dummy variable turned on if the fetus/infant was a boy (girl), ϵit is an error term, and α1 – α6 are parameters to be estimated. α1 can be interpreted as the predicted death rate for boys in 1967, that is, the death rate that would have prevailed if the error term had been zero. α2 is the corresponding predicted death rate for girls in 1967. The predicted death rates in year t are α1 + (t-1967) α3 + (t-1967)2 α5 for boys and α2 + (t-1967) α4 + (t-1967)2 α6 for girls, implying that predicted death rates fell by 38 α3 + 382 α5 for boys and 38 α4 + 382 α 6 for girls during 1967–2005. The absolute yearly change in death rates in year t is α3 + 2 (t-1967) α5 for boys and α4 + 2 (t-1967) α6 for girls. α3 and α4 can be interpreted as absolute yearly changes in death rates in 1967. In 2005, absolute yearly changes in death rates were α3 + 76 α5 for boys and α4 + 76 α6 for girls.
We also estimated a simplified linear probability model where α5 and α6 are set equal to zero. In the simplified specification, absolute yearly changes in death rates are constant and equal to α3 for boys and α4 for girls. The estimates of α3 and α4 in this model can therefore be interpreted as average yearly changes in absolute death rates during the whole period from 1967 to 2005.
Since there is considerable random variation between years in death rates, we report predicted rather than actual death rates for individual years. Adjustment for the effects of maternal age, parity, plurality and maternal education on absolute yearly changes in death rates was made by including these variables as covariates in the linear probability models.
To obtain estimates of relative yearly change in death rates, the following logistic regression model was fitted:
where Λ is the logistic cumulative distribution function [
20]. Exp(α
2)-1 can be interpreted as predicted excess death rate for boys in 1967, whereas exp(α
3)-1 and exp(α
4)-1 are the relative yearly change in death rates for boys and girls respectively. Adjusted estimates of relative yearly changes were obtained by including maternal factors as covariates in the logistic regression model.
We estimated the sex-specific number of fetal and infant deaths prevented from 1967 to 2005 by comparing two scenarios: one scenario where, for each year from 1967–2005, mortality was set equal to predicted mortality for 1967, and another scenario where mortality was set equal to predicted mortality in the actual year. In these calculations, the simplified linear probability model was used.
The study was approved by the Regional Committee for Medical and Health Research Ethics (Reference number 603-07276a 1.2007.2366).