Record linkage

Since the 1970s, people who live in Scotland have been allocated a unique Community Health Identification (CHI) number, which allows record linkage across clinical databases and generations. We used the CHI number to record-link between the Walker Cohort and Scottish Morbidity Records maternity admissions data (SMR02). Data were provided and by the Health Informatics Centre at the University of Dundee. The Information Services Division Scotland provides SMR02 data to the Health Informatics Centre. All data were anonymized before analyses.

The Walker cohort

The Walker cohort is a dataset of 48,404 birth records that contains meticulously recorded details of pregnancy, labor, and care before discharge for births in the hospital in Dundee, Scotland, from 1952-1966. The details of the Walker cohort have been published previously and will not be repeated here, but there is information about PPH stored as a dichotomous variable for Walker births that occurred from 1952-1958. Information on later births was recorded on different cards that did not include details of PPH. The criteria that were used to define PPH are not described for the Walker cohort, but we have assumed it to be ≥500 mL blood loss in the first 24 hours after delivery, as diagnosed by the doctor or midwife who assessed the patient.

Maternities recorded in the Walker cohort account for 75% of all births in Dundee at this time; 34,183 of these babies (73%) can be identified through their CHI number, which presents the opportunity to link this maternity or birth information with a large number of current health-outcome datasets that cover both primary and secondary care for Walker mothers and babies.

SMR02

The SMR02 dataset contains detailed information on hospital maternity admissions in Scotland that have been collected from January 1975 to present. Outcomes are coded according to the International Classification of Diseases (ICD9 and 10). Table 1 shows the codes that were used to indicate PPH as an outcome. Again, we assume these codes were assigned based on observation of blood loss ≥500 mL in the first 24 hours after delivery.

The generations

We identified data on 3 generations of women, which are defined in the following manner: generation 1: Walker mothers (women who appear in the Walker cohort as mothers); generation 2: SMR02 mothers (women who appear in the Walker cohort as babies and the SMR02 cohort as mothers); and generation 3: SMR02 daughters (women who appear in the SMR02 cohort as babies and also as mothers, if they have had children themselves).

Data analysis

In SMR02, maternity admissions that were coded as not resulting in delivery of a child were removed. Stillbirths of a baby at >500 g were included.

We used IBM SPSS Statistics software (version 19; IBM Corp, Armonk NY) to perform a univariate analysis of the pooled (Walker and SMR02) data. We calculated unadjusted odds ratios (ORs) to assess the effects of PPH in a previous pregnancy, multiple pregnancy, high birthweight (with the World Health Organization [WHO] definition of ≥4 kg), low birthweight (with the WHO definition of <2.5 kg), maternal age under 20 years (previously identified as a risk factor for PPH ), maternal age over 40 years (previously identified as a risk factor for PPH ), parity, preterm birth (with the WHO definition of ≤37 weeks’ gestation), postterm birth (with the WHO definition of ≥42 weeks’ gestation), delivery by cesarean section, instrumental delivery, and smoking status on risk of PPH in the index pregnancy. To calculate adjusted ORs, we built factors that were identified as significant in the univariate analyses into a multivariate logistic regression model with the generalized linear model function in the R package linear mixed-effects models using S4 classes (lme4) (R version 2.15.1; R Foundation for Statistical Computing, Vienna, Austria ).

To assess intergenerational transmission of PPH, we used the CHI number to link records across generations in the following manner: (1) generation 1 was linked to generation 2; (2) generation 2 was linked to generation 3; (3) generation 1 was linked to generation 3; (4) pooled mother-daughter analysis: mothers from generations 1 and 2 were linked to daughters in generations 2 and 3.

To analyze intergenerational trends in PPH for each of these comparisons, we first calculated unadjusted ORs using logistic regression model, again using the generalized linear model function in R package lme4. These models assess the relationship between PPH in the younger generation (the dependent variable) and PPH in the older generation (the independent variable) without taking into account any other potential covariates. We then used the generalized linear mixed model function in R package lme4 to build generalized linear mixed models using a binomial distribution with a logit link. These models incorporated any other covariates that were found to be associated significantly with PPH in the univariate analyses. The models also adjust for the “random effects” that can be introduced through the appearance of the same women in different mother-daughter/grandmother-granddaughter pairs (for example, a woman could be a mother in the comparison of generations 2 and 3, but a daughter in the comparison in generations 1 and 2. Additionally, 1 woman could be a mother to >1 daughter). It is important to adjust for this nonindependence because it invalidates the assumptions of many statistical tests and can introduce bias that can mask exposure effects. Where we suspected low power, we used the R package clinfun to calculate the smallest effect size that our analyses would have been able to detect at 80% power, based on our actual sample sizes.