Aim 2: exploring sources of variation in rate of reporting AEs

We hypothesised that:

Poorer psycho-social status and health at baseline will be associated with higher reported rates of both Medical and Social AEs (baseline variables thought to reflect poorer psycho-social status listed below)

Poorer psycho-social status at baseline will more likely be associated with Social rather than Medical AEs

AEs reports will be more likely for those in the trial intervention arm (hypothesised to be due to surveillance bias, having received up to 64 visits from a family nurse)

Rate of AE reporting will vary by trial site (due to various system level differences between sites which could include variability in research nurse approach; e.g. actual funded time, total number of participants at site being monitored, quality of links to local Family Nurses or other local staff). Site was a predictor we sought to modify during the course of the trial, but despite our efforts differences in site were not eradicated.

Baseline variables that we considered to indicate poorer psycho-social status were younger age at recruitment, the woman’s status being classified as NEET (Not in Education, Employment, or Training), being in receipt of benefits, having ever been homeless, having lower socio-economic status (Index of Multiple Deprivation score), lower family and lower personal subjective social status, lower relationship quality, lower social support, lower family resources, lower self-efficacy, and lower adaptive functioning.

All participants were categorised as having experienced either no, or at least one Social AE. They were also categorised as having experienced either no or at least one Medical AE (regardless of severity). These formed the two dependent variables in subsequent analyses. For each dependent variable the following sets of analyses were performed. Baseline characteristics were summarised between those who experienced either no or at least one AE (Social and Medical) using number (%), mean alongside standard deviation (SD) and median alongside the 25th to 75th centiles. Baseline characteristics included socio-demographics listed above, e.g. age; health (e.g. health status, psychological distress) and group allocation. Logistic regression models were run to examine univariable associations between baseline characteristics and AEs. Baseline characteristics that were associated at the 10% significance level were retained and entered as candidate predictors for the multivariable model to detect all characteristics independently predictive based on a significance level of 0.05 of AEs. Trial site was adjusted for by its inclusion as a random effect in all models. Multi-collinearity in each model between candidate predictors was assessed by detecting the tolerance and its reciprocal, the Variance Inflation Factor (VIF). As a rule of thumb a VIF of 1 indicates no collinearity but a VIF greater than 4 (a tolerance of 0.2) might warrant further investigation and greater than 10 would indicate that multi-collinearity is a problematic.