Statistical analysis
Categorical variables are described as percentages, and continuous
variables are expressed as the means or medians with dispersion
measures. The Kruskal‒Wallis, and Pearson chi-square tests were used to
perform uni and multivariate comparisons of quantitative and qualitative
variables, respectively. A longitudinal analysis with repeated measures
of SI was performed using a linear mixed-effects model, where we tested
the main association of NLA on the change in SI during labour and
postpartum, comparing the association of interventions on SI across
labour and postpartum. The analyses were fitted using an unstructured
random slope and intercept, using the R package ‘lme4’. We chose this
structure following the recommendation of using it on outcomes with a
ratio scale, such as SI.
We tested whether the changing trend was different between the NLA and
non-NLA groups. The primary outcome was the SI values for a patient over
time during labour and postpartum periods. Therefore, the primary
hypothesis of the study can be formally stated as a test of whether
there is a time × intervention interaction. We tested the interaction by
fitting a multilevel model with SI as the dependent variable at each
time and with the time period, analgesia status (NLA or non-NLA), and
the NLA-time period interaction as fixed effects. We assumed that the
time between each measure (antepartum and postpartum) was equal for
statistical analysis.
The association between NLA status and SI was adjusted in the linear
mixed-effects model by maternal age (older than 35 years), haemoglobin
level, cervical dilation, and nulliparity. We categorised continuous
variables and entered them as terciles in the multilevel model to
account for the nonlinearity of the relationship with SI. We also fitted
the models for each component of the SI (HR and SBP). We excluded
confounders with p values >0.50 in the fitted model with
all covariates and excluded these nonsignificant covariates from further
analysis. The associations for all other assessments were deemed
significant if the p-value was <0.05. We reported the
β-coefficients of each fitted model and estimated the 95% confidence
interval (95% CI). Finally, we performed sensitivity analyses to
measure changes in the β-coefficients and significance by type of NLA,
including only the most frequently used type of analgesia worldwide
(i.e., epidural analgesia).