Health Economics, Cost-effectiveness analysis, Uncertainty analysis and Oncology modelling
Please clarify your question:
Do you already have a model that includes a gompertz distribution with negative shape parameter?
What uncertainty do you want to reflect in your PSA?
What do you see as problem when trying to do that?
Dear Robert,
Sorry for not providing enough clarity. Yes, I have a health economic model that is comparing two intervention in advanced cancer in terms of cost-effectiveness. The modelling approach is Partitioned Survival model.
I want to see uncertainty on the extrapolated part of the overall survival curve implementing cholesky decomposition in the PSA. However, I am getting error due to negative shape of parameter (not able to take log of negative number). I have attached the image for your reference. Thank you so much.
Masnoon.
That for clarifying.
It looks like your PSA occasionally produces invalid parameters.
First I would advise to check why this is happening. Perhaps you made an error. Or: Does this happen because you are fitting to a relatively small data set that causes a lot of random noise in the estimate produced by the Choleski transformation? Is the survival you are fitting close to an exponential distribution so that the shape parameter for the assumed Gompertz is close to 0 and therefore can easily become negative? (Then consider whether you should choose a different distribution than Gompertz) etc.
You could also consider, if you can find a good epxlanation or an error and this happens on only a small fraction of PSA runs, to be pragmatic and say: this PSA run represents an impossible set of assumptions, therefore it does not need to be part of the uncertainty analysis. And simply skip it.
Thanks Rob. Your response was very helpful. Considering alternate survival distribution seems very logical here. Thank you again for sharing your knowledge with us and I hope more people will get benefit by reading this. Best Regards, Masnoon
- Badges
- Science topic