I developed a prediction model using stepAIC after univariate analysis.
glm(formula = outcome_m6 ~ age.x + lesion_size + index_cat.x + pain.x, family = binomial, data = data)
The following is the summary info of the model.
Coefficients:
Estimate Std. Error z value Pr(>|z|)
(Intercept) 2.7372 587.8717 0.005 0.99628
lesion_size.L -14.2324 1247.0614 -0.011 0.99089
lesion_size.Q 7.3856 719.9912 0.010 0.99182
index_cat.x1 2.4286 1.2482 1.946 0.05169 .
pain.x1 2.2450 0.7021 3.197 0.00139 **
age.x.L 0.8547 0.4451 1.920 0.05483 .
Lesion size has quite high p value and only one of the coefficients are negative. Should I eliminate lesion_size from the model because of the high pvalue? when univariate analysis, this predictor has smallest p value, so I am a bit confused.
ROC and calibration shows the good model performance.
