Why didn't we change number of epochs?

After I increased the number of epochs from 5 to 10, my validation accuracy increased to above 0.99. I believe that increasing the number of epochs makes comparison impossible while also changing the other hyperparameters (because one model simply as more iterations to learn) but if our final goal was to make a better performing model, I don't understand why we didn't increase the number of epochs first (and see where overfitting happens if any), before messing with other hyperparameters further.