The distribution of each variable was assessed and
correlations were explored for both continuous and categorical
variables to look for unusual distribution of variable
and to determine potential correlations. Kaplan–Meier
survival plots were constructed for each of the categorical
variables. A univariate Cox proportional hazards model was
used to check for assumed proportion hazards using
univariate analysis looking at log plots and testing survival
time vs. scaled Schoenfeld residuals using both chi squared
tests and visual examination of plots. DIM at ESD was
included in all models to correct for DIM at entry into the
study. Significance of each variable in the univariate model
was analyzed using the likelihood ratio test. Variables with
a P < 0.2 were retained for inclusion in a multivariable Cox
proportional hazards model. Farm was included in the
analysis as strata effect. Explanatory variables were
removed in a backwards elimination based on Likelihood
ratio test comparison of reduced model with full model
using P > 0.05 as reason for removal. Interaction between
variables was examined and each of the removed variables
individually entered at the end to ensure correctness of the
model.