Obtain agreed approach to new HSA data

[DCEW]

The health status adjustment expert elicitation was run slightly differently this time around. Whereas before, it resulted in a single gamma distribution, this time, each of 8 participants submitted low, mean and high predictions for the parameter, leading to 8 split-Normal distributions. There seems to be no statistically valid way to combine these into a single distribution. Paul is working on understanding whether randomly sampling from the split-Normals to create a new undefined type of distribution is ok. The next thing to work out is whether this leads to a truly non-parametric curve that cannot be approximated to be parametric. It may be that selecting once from this new curve per model run would work fine - but more analysis needed.

Paul is continuing this work without a deadline. I will keep tabs on progress and try to put in more definite deadlines and deliverables as that becomes necessary. No rush for now.

This issue is to record that we need external methodology sign-off for the implementation of the elicitation output into the module (and accompanying write up similar to this one.

[DCEW]

From Paul Seamer on 29th Jan: "Hi Claire, between us we've come up with 3 possible approaches to implementing the new HSA adjustment in the NHP codebase. I've started some proof of concept work to test whether the different approaches are workable and give the same results. As soon as I've reached a view I'll share with Tom and he can make a final call on which approach works best within NHP model framework. I'm planning to finish this by end next week. You don't need to worry about me getting sidetracked by other priorities/projects I have time to work on this planned into my schedule between now and end March. So no problem with me updating the documentation of methods, I need to do this anyway as we would like to reference it for other work. Any questions just ask."