Isn't the hazard ratio different because the models themselves are different? The reference group is the same, but the variable has 3 levels in one analysis and 2 in another. If a group of participants is omitted from one of the models, there will be a new estimates of the model parameters.
... View more