Statistical analysis
All outcomes were analysed according to a predefined analysis strategy that was conducted in collaboration with our clinical epidemiologist (JGB).
For our main outcome, sensitivity and subgroup analyses on the difference in severity of HDFN between two subsequent pregnancies, a Wilcoxon Signed-Rank test was used. With this test, the number of positive differences in severity (+1 to +3 disease categories), negative differences (-1 to -3) and ties are ranked.
Differences in severity of HDFN between two non-paired groups were analysed with a multinomial logistic regression. In other, non-paired analyses, the Pearson’s Chi-square test or logistic regression (or Fisher’s exact test if appropriate) was used for the comparison of proportions. Comparisons of non-parametric outcomes were analysed with the Mann-Whitney U test. A sensitivity analysis was performed among patients in whom all the information on disease outcome was available and disease severity was thus not imputed. As the mechanism and thus severity of HDFN might be different if RhD antibodies are developed after giving birth to a RhD-positive child and thus after receiving anti-D at least twice (group A), or in the first pregnancy at risk for immunization (group B), a subgroup analysis was performed in these groups.
In order to identify factors possibly predicting severe HDFN (IUT or death) in a subsequent pregnancy for counselling purposes, a prediction model was constructed including variables known or thought to be associated with HDFN severity from the literature, the potential predictors. All potential predictors with a P -value<.25 in univariate analysis were included in a multivariate logistic regression model. The prediction model was further improved by applying manual backward selection, excluding the variable with the highestP -value at every step. Eventually, all variables with aP -value<.1 remained in the final prediction model.
RESULTS