I was interested and surprised by the fact that many predictive models, like models predicting factors associated with the highest likelihoods of different types of cancer, often have limited practical use in terms of actually fighting cancer. For example, a model might show that smoking increases the likelihood of cancer, but people will often not heed the advice of such models because of anecdotal evidence based on certain individuals who smoked and lived long, healthy lives. On the other hand, there might be genetic factors that are predicted to have a higher chance of association with cancer, but this does not necessarily provide the patient with relevant information as to what they can do to increase their chances of survival.
Going off of this, I would ask, how can such predictive models be of use to a doctor who is actually trying to cure cancer, or another disease, in a patient? If a patient has been diagnosed with cancer, is it of any use to know that they possess certain characteristics that could have helped to predict that they would ultimately get cancer, or is this information useless given that the patient has already been diagnosed? As in, is there any way that the data about correlation between certain risk factors and cancer can actually be used to treat the cancer itself.
I share your surprise that models that are extremely intricate with extremely well funded developmental processes are not as applicable as one may think. Models and data are only as relevant as the public responses to them are and I think you're touching on a very important potential area for improvement in the process of disseminating data to the public in an attempt to change their behavior.