Ordinal regression analyses with tumour ErbB2-IR, AR-IR and Ki67-index as the independent variables and the tumour stage as the dependent variable.

Two models are shown, one with an interaction between ErbB2-IR and AR-IR (“Interaction model”) and one without this interaction (“Main effects model”). The values returned for a variable in the main effects model represent its total effect when the other variable is included. In the interaction term model, the value for the individual variable is for a constant value of the other variable. The coefficients (± standard error) were obtained from ordinal regressions (cumulative logit model) using the R statistical package (function vglm in the VGAM bundle). The tumour stages were divided into 1a+1b, 2, 3 and 4. In both cases, the assumption of proportional odds was upheld. The models return four intercept coefficients for each level of the response with the reference level being the lowest level. These were as follows: main effects model, 0.61, −0.82, −3.14; model with interaction, 2.84, 1.40, −0.92.Ordinal regression analyses with tumour ErbB2-IR, AR-IR and Ki67-index as the independent variables and the tumour stage as the dependent variable.