Rules of Thumb for Decision Trees

Rules of Thumb for Decision Trees

by Lekha Tummalapalli -
Number of replies: 0

Hi all, 

Dr. Marseille also reviewed your first assignment on mammography decision trees and thought they were excellent! If he had any additional comments for you, I e-mailed them to you. 

He wanted to pass along some rules of thumb for decision trees: 

  • The decision node at the beginning does not have associated probabilities - participants are "randomized" to the decision strategies. No decisions after the initial decision node. It is followed by chance nodes and terminal nodes. 
  • Probabilities stemming from chance nodes must add to 100%. In Excel, create a formula =1 - cell to accomplish this.  In TreeAge, use # symbols.
  • All chance nodes need to have probabilities assigned to them.
  • All terminal nodes need to have utilities assigned to them. 
  • Probably better to treat the screening arm as “annual screening over X years” rather than portraying the results as successive screenings. Explicitly portraying of successive screenings is good, but may be overly-cumbersome for our purposes in this class. In the Markov modeling lecture, we will cover how to do repeated cycles.

Note: I also was mistaken in several of my comments to you - The prevalence of cancer in screening group and no-screening group are the same. The true disease prevalence is the same, although screening may detect more. 

Feel free to e-mail with any questions! 

Best,

Sri Lekha Tummalapalli