As a general outcome prediction problem in the medical field, whenever a patient is a basis of an observation there is also some free-form information on previous medical history. It is clear that someone can select the important things and make them into features, but is there a framework that can utilize additional structures like ICD classification or some ontologies? The processing itself is not an issue and I understand that the text in question can be a challenge, but I am more interested in the optimal utilization from the model perspective.

Optimally, I am looking for a good book on machine learning in the medical field that would answerČ

  1. How to engineer features from medical history?
  2. Can we use rare diagnoses?
  3. Should entities be included as risk scores or directly as variables? etc. If there is no book, online courses would work too.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.