search

Health-Economics-in-R / CEdecisiontree

:deciduous_tree::deciduous_tree: Lightweight cost-effectiveness analysis using decision trees.
https://health-economics-in-r.github.io/CEdecisiontree/
Other
12 stars 2 forks source link

Cdectree_expected_values and dectree_expected_values give different results #6

Open jogwalker opened 3 years ago

jogwalker commented 3 years ago

Hi, I have been playing around with this package and trying to get it to work for quite a big decision tree which would be a huge hassle to manually calculate expected probabilities. It looks like in dectree_expected_values all endpoint nodes are given their inputted cost value directly, while in Cdectree_expected_values, endpoint nodes have a value of 0.

n8thangreen commented 3 years ago

Hi Josephine (I think!), there are certainly inconsistencies with this. I've basically just kept adding bits to it over quite a bit of time whenever I needed something for a project so theres very little overall design decision. Its great that you're trying it out tho! I should sort it out so this is a good excuse. Can you send me the example you've tried which gives the problem please and I'll fix it... n.green[at]ucl.ac.uk

jogwalker commented 3 years ago

Hi Nathan Thanks! The difference can be seen using your examples ("cost" and "probs")

mt <- define_model(transmat=list(vals=cost,prob=probs)) dectree_expected_values(model=mt) 1 2 3 4 5 6 7 5.6 12.8 3.8 10.0 1.0 10.0 1.0 Cdectree_expected_values(vals=as.matrix(cost),p=as.matrix(probs)) [1] 5.6 2.8 2.8 0.0 0.0 0.0 0.0

Are you able to clarify whether this is a different output being calculated or does one of them have an error?

Also, I thought that the value for node 1 should be equivalent to the below, but perhaps I'm calculating something different or incorrectly?

probs[1,2]cost[1,2]probs[2,4]cost[2,4] + probs[1,2]cost[1,2]probs[2,5]cost[2,5] +

  • probs[1,3]cost[1,3]probs[3,6]cost[3,6] + probs[1,3]cost[1,3]probs[3,7]cost[3,7] 2 1 7.84

I can get 2.8 like this:

probs[1,2]cost[2,4]probs[2,4] + probs[1,2]cost[2,5]probs[2,5] + probs[1,3]cost[3,6]probs[3,6] + probs[1,3]cost[3,7]probs[3,7]

Or 12.8 like this:

cost[1,2] + probs[1,2]cost[2,4]probs[2,4] + probs[1,2]cost[2,5]probs[2,5] + probs[1,3]cost[3,6]probs[3,6] + probs[1,3]cost[3,7]probs[3,7]

Also, just to note that in my transition matrix I had some probabilities that were repeating decimals. This meant that some of the functions weren't working because is_prob_matrix(test[["prob"]]) was returning FALSE even though when I looked at the value it looked like a 1! So I rounded all my probabilities (to 6 decimal places) and it works fine now.

Thanks again Josephine

On Mon, Nov 16, 2020 at 3:44 PM Dr Nathan Green notifications@github.com wrote:

Hi Josephine (I think!), there are certainly inconsistencies with this. I've basically just kept adding bits to it over quite a bit of time whenever I needed something for a project so theres very little overall design decision. Its great that you're trying it out tho! I should sort it out so this is a good excuse. Can you send me the example you've tried which gives the problem please and I'll fix it... n.green[at]ucl.ac.uk

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/Health-Economics-in-R/CEdecisiontree/issues/6#issuecomment-728144089, or unsubscribe https://github.com/notifications/unsubscribe-auth/AA7OI6L2T2YHT6SJXA753P3SQFCFZANCNFSM4TXH7EEQ .