Bivariate analysis

The connections between each explanatory variable and the response variable were analysed with bivariate statistics. The Wald test from logistic regression used a P cut-off point of 0.25 because statistical significance may not capture importance and the more traditional levels, such as P of 0.05, could fail to select variables known to be essential [13]. A cut-off value of 0.25 is supported by literature [14]. Decisions to keep a variable in the “best” model were based on clinical or statistical significance, or on the significance level of a confounder between 0.1 and 0.15 as it might, in combination with other variables, make an important contribution [13]. In the present study, variables could potentially be entered into the multivariable model if the results of the bivariate test had a value of P < 0.25.