In petedodd/HEdtree: A Package For Decision Tree Modelling

Preliminaries

First make sure that the package is loaded:

library(HEdtree)
library(data.tree)              #needed?

Making a tree

Then we can use the syntax of data.tree to define our decision tree by first creating an empty tree object and then adding to it child nodes using the AddChild() function.

myt <- Node$new('mytree')
ma <- myt$AddChild('Option A')
mb <- myt$AddChild('Option B')
ma1 <- ma$AddChild('Option A1')
ma2 <- ma$AddChild('Option A2')
mb1 <- mb$AddChild('Option B1')
mb2 <- mb$AddChild('Option B2')

We can also add incoming probabilities (p), as well as costs (cost) and health-related quality of life (qol) parameters for these nodes

## p/q/c
myt$p <- 'myp'; myt$qol <- 'myq'; myt$cost <- 'myc'
ma$p <- 'ap'; ma$qol <- 'aq'; ma$cost <- 'ac'
mb$p <- 'bp'; mb$qol <- 'bq'; mb$cost <- 'bc'
ma1$p <- 'a1p'; ma1$qol <- 'a1q'; ma1$cost <- 'a1c'
mb1$p <- 'b1p'; mb1$qol <- 'b1q'; mb1$cost <- 'b1c'
ma2$p <- 'a2p'; ma2$qol <- 'a2q'; ma2$cost <- 'a2c'
mb2$p <- 'b2p'; mb2$qol <- 'b2q'; mb2$cost <- 'b2c'

We can check we're on the right lines using ASCI-style visualization:

print(myt,'p','cost','qol')

If other variables have been attached to nodes, they can be visualized in a similar fashion.

Calculations

HEdtree relies on recursion, closures and string manipulation to perform its calculations. For example, we can generate strings that correspond to the mean cost or HRQoL for entries to the tree.

getAQ(myt,'cost')
getAQ(myt,'qol')

We can look at an exhaustive list of parameters defined using

(varnmz <- showParmz(myt)$vars)

The function showParmz also collates variables which are defined in terms of other defined parameters (using arithmetic operations e.g.). It is permissible to define a cost using such an operation, e.g. rtest$qol <- '(b1q-.1)/2'`

Functions that evaluate these strings on data.frames are built using

H <- makeTfuns(myt) 

We can check that these are working by building a dummy PSA data.frame

## dummy PSA data frame
psadf <- matrix(runif(10*length(varnmz)), ncol=length(varnmz))
colnames(psadf) <- varnmz
psadf <- as.data.frame(psadf)

## evaluate
H$costfun(psadf)                        #return answers
H$qolfun(psadf)

Use of functions

It is also permitted to call on functions that are defined in higher-level environments, e.g.

ma1$p <- 'accpfun(age)'
accpfun <- function(x) exp(x) / (1+exp(x))
psadf$age <- 1:nrow(psadf)          #add an age in 
H <- makeTfuns(myt)                 #remake the functions
H$costfun(psadf)                    #test 

Note: this example also illustrates how one can include individual characteristics from a population and use them to influence their outcomes.

Plotting

Plotting to a web browser can be achieved by

plotter(myt)

If fewer/more labels are needed, this can be achieved using

plotter(myt, varz=c('name'), edgelabel=FALSE)

If further formatting or I/O is required, ToDiagrammeRGraph exports the graph as a DiagrammeR object (see package of the same name). Or just assign the return value to some variable, e.g. test, whereupon test$x$digram will contain the associated DOT code.

Reading in parameters

TODO

Manipulating subtrees

Looping back is not permitted (these are tree models after all). However, subtrees can be copied and added, potentiall after modification.

mbCopy <- Clone(mb)
mb2$AddChildNode(mbCopy)
print(myt,'p','cost','qol')

TODO

• PSA
• parameter tables

petedodd/HEdtree documentation built on Dec. 5, 2022, 6:22 a.m.