Transcript Introduction to TreeAge June 1, 2005 Mendel E. Singer, Ph.D.

Introduction to TreeAge
June 1, 2005
Mendel E. Singer, Ph.D.
Case School of Medicine
[email protected]
Assumptions
(always dangerous)
You’re already familiar with the essential
components of a decision analysis
You’ve been introduced to the basic idea
of cost-effectiveness analysis
You may not be comfortable with a
decision tree, but you’ve seen a few
You haven’t programmed any decision
trees, or did once upon a time, but……
What’s TreeAge?
TreeAge (pronounced “triage”) is software
specifically for doing decision analyses, and
cost-effectiveness analyses that are based on
the decision analytic model.
Until recently it was called DATA, but the
company’s catchy name, TreeAge, is what stuck,
and everyone called it that anyway.
Originally developed for business applications,
they have been developing for health care
applications for close to 10 years now.
Why Use TreeAge?
Some people use TreeAge, others use
alternative decision analysis software like
DecisionMaker, some use Excel, and some use
computer simulation software.
TreeAge is a mature software product that is
well supported. It is regularly updated and
improved, doing a good job of keeping up with
advancements in the methods.
It is very intuitive to use because it is based
around the decision tree graphic, and not
programming code.
What Can TreeAge do?
Decision Analysis
Cost-Effectiveness Analysis
Baseline and Sensitivity Analyses
Markov Models
Monte Carlo Simulation
Influence Diagrams
Note: It’s OK not to fully appreciate what all of the
above are! You can still follow this lecture!
Getting the Software
Purchase it!
Perhaps your group already has a site
license
Download a trial copy from:
www.treeage.com
– Limited in size
– Works for 21 days from first time it is opened
WARNING!!
You won’t learn to program by watching a
lecture. You have to get hands on!
It can help to do some small programs for
practice. I will see if HERC can put this on
the web site along with this presentation.
Abdominal Aortic Aneurysm (AAA)
We need a simple, and therefore simplified problem to
use as an example.
AAAs: 5% - 7% of people over age 60
Most commonly affect Men, Age 40-70.
Usually asymptomatic, often until rupture
Options are surgery and watchful waiting, with some
decision rule as to when to operate based on the size of
the AAA.
Debate as to how large the aneurysm should be to
recommend surgery.
Our example will be based largely on the work of Katz
and Cronenwett (1994).
Problem Definition
Reference Case
– 60-year old male
– 4 cm abdominal aortic aneurysm
– Otherwise, patient is in good health
Surgery vs Watchful Waiting
Time horizon: 1 year
Outcome Measure: Survival
– Alive = 1
– Dead = 0
More Details of the Example
Simplifications and assumptions:
– If expansion occurs, it reaches 5.5 cm.
– Either the AAA ruptures, or surgery is
performed.
– If the AAA ruptures, the patient may die before
emergency surgery can be performed.
– No death from other causes during the 1 year
time horizon used for the analysis
Types of Nodes
Start with a choice node, with each possible
strategy coming out of this node.
Three kinds of basic node types:
– Choice, or Decision Node
– Chance Node
– Terminal Node
Choice or Decision
Chance
Terminal
1. Always start with a choice, or decision node
2. Names of nodes go above the branch
3. Note in the properties box that the name is cut off. We can
select the vertical bar to the right of the properties box and
drag it, to make the box wider.
To add branches there are several techniques.
1.
Double-click on the node
2.
Right-click on the node and select the choice to add branches.
3.
Use the menus to select “Options / Add Branches”
4.
Enter Control-A.
By double-clicking on the decision node you get…..
Parameter Values
Name
pDieBeforeSurgery
pDieDuringSurgeryAfterRupture
pDieElectiveSurgery
pDieSurgeryNoRupture
pExpansion
pRupture
Value
0.55
0.54
0.046
0.23
0.033
0.92
Low
0.27
0.27
0.023
0.11
0.016
0.88
High
0.82
0.81
0.069
0.34
0.049
0.96
1. Probabilities go below the branch
2. Don’t put in the actual numbers for the probabilities.
3. Instead, use variables. This makes it easy to change its value, even
if it appears in many places in the tree. More importantly, it allows us
to do sensitivity analysis on that parameter.
Naming Variables
Use descriptive names. Better to write out long variables,
than to use abbreviations you may not be sure of when
you open the model many months later to revise it.
Use a consistent naming convention. Many people begin
the name of probability variables a “p”, utilities with a “u”,
costs with a “c”, etc….
– e.g. mortality rate = pDie
– E.g. utility of diabetes = uDiabetes
Make the names easy to read. When using multiple
words in a variable name, capitalize the first letter of
each new word.
– e.g. pDieBeforeSurgery
Some use the underscore character between words.
– e.g. pDie_before_surgery
The # Sign
In TreeAge, the # sign can be used for a probability.
It indicates the leftover probability, after accounting for
the other branches emanating from the same node.
You should use this whenever possible.
Aids sensitivity analysis. When a probability changes
value, the sum of the probabilities for the nodes leaving
that chance node will no longer be 1.
Try to never use more than 2 branches out of 1 chance
node. This way, whenever one probability changes, we
know the other possibility must change by the same
amount in the opposite direction. E.g. if the baseline
estimate of operative mortality is 0.05, then if that value
were to increase to 0.10, then the probability of surviving
the surgery must decrease by the same 0.05 (0.10-0.05).
1. At each node there is a number in a box indicating the expected value of
the outcome from that point forward
2. At the choice node, it also indicates which strategy is optimal. The nodes
in the optimal path are all highlighted.
3. At all terminal nodes, it first shows the outcome score associated with
that result, and then shows the probability of the path ending in that
terminal node.
4. All of the variable names are temporarily replaced by their values.
1. Graph shows how the expected value (mean) changes based on the
mortality rate from elective surgery over the range of values specified.
2. Watchful waiting is unaffected because elective surgery isn’t in its path.
3. The point of intersection, known as the threshold value, is shown on the
graph and the point and its associated mean is shown in the right margin.