|
Sensitivity Analysis on Tables
Sensitivity analysis works with variables in TreeAge Pro, not tables.
If the probabilities or other values in a TreeAge Pro table are
uncertain, however, relatively minor adjustments can be made to allow
for sensitivity analysis. The most simplistic approach, is to find
every reference to the table (e.g., "tProb") and multiply by a
variable (e.g., "mult_tProb") having a default definition = 1.
Sensitivity analysis can be performed on the multiplier over a
percentage range (e.g., from 50% to 150%). This implementation may
not be flexible enough for most tables.
Varying the value column
A better approach to sensitivity analysis on tables is to add
additional columns of values to the table, describing the range of
parameter values for each index/row. The sensitivity analysis is then
performed on a variable used to select a value column (and even
interpolate between columns). Note the syntax used to reference
tables with multiple value columns: myTable[ row ; col ].
How it works
In addition to two standard columns for indexes (e.g., time or
age) and basecase values, two more columns are added for low and high
parameter values, respectively. The "column" variable has a default
definition = 1, referencing the first value column during basecase
analysis. During sensitivity analysis, the "column" variable is
varied from 2 to 3, over 1 interval. This results in two
recalculations, using the low column and then the high column. If
more intervals are desired, the table can be set to interpolate
between missing rows and columns (or more columns added). Setting
intervals = 2 would also analyze values halfway between the low and
high columns (i.e., column = 2.5). Additional columns could also be
used.
A simple example is shown in the figure below:
Displaying Path Probabilities
Q: How can I display "path probability" times "payoff" (i.e., contribution to expected value) at endnodes?
A. There are two sets of changes in preferences:
1) Calculation Method:
* check "Calculate extra payoffs"
* set "Enable payoffs..." to 1 if you have only filled in 1 payoff set
2) Terminal Nodes/Columns:
* Show terminal columns
* Add a "Custom" calculation column, with the calculation:
StateProb() * Node(1)
This will display, during roll back, the product of the payoff and the path probability at each visible endnode. These columns can be copied to Excel by choosing Edit > Copy Special...
Calculating EVPI in TreeAge
Q: What is EVPI? How EVPI is calculated in a Monte Carlo simulation?
Using a Monte Carlo probabilistic sensitivity analysis performed at a decision node,
TreeAge Pro can calculate the expected value of perfect information (EVPI) or partial EVPI.
This can be used to determine the theoretical maximum value of potentially costly efforts
to remove particular uncertainties.
Check out a mini tutorial and details of
Cost-Effectiveness Uncertainty and Value of Information.
|
