Typically you need a much more exact phrasing (in more mathematical terms) of the question to ask.
Will a patient respond to medicine X? What is the likelihood? To what amount will patient repond to medicine X? Is patient in the group that is expected to respond to medicine X?
Are slightly diffent questions that may impact choice of technique.
Furthermore, your data plays an important role. Are you missing data? Have you normalized already? Do you expect 'Marital status' or 'Education level' to have an impact on medicine efficiency? (It might if certain groups take medicine at home, but it may be less likely when taking under doctor supervision)
A priori you determine how to measure or quantize the success of a model (typically prediction accuracy).
Then typically you try a few machine learning techniques, and make a model from the most successful one.