Starting PrEP usage

[pineapple-cat]

Addresses #205.

[mmcleod89]

@andrew-phillips-1
I think we're having a difficult time interpreting something in the SAS code and therefore making it clear in the python code.
The line in question is:
if prep_any_ever ne 1 and tested=1 and (hiv=0 or (hiv=1 and unisensprep > eff_sens_vct)) then do;
(currently 5166 in the SAS code on the main branch)

unisensprep is just a random number generated
eff_sens_vct appears to be a test sensitivity based on what test is available

What I'm confused about is that it appears to be modelling false negatives using a random variable, but don't false negatives already appear in the diagnosis part of the simulation? If we test all the HIV+ people who are selected for testing in a given time step, then the people who get diagnosed will be true positives, and the people who don't will be false negatives. Why is this code separate, is it modelling something different? Can this lead to people who are e.g. HIV+, get tested, get a (false) negative result, and still don't meet the criterion unisensprep > eff_sens_vct, and how would we interpret these people?

Is this modelling a separate round of testing i.e. testing before starting prep separately (and possibly in addition to) other forms of testing? (Although it doesn't appear to lead to diagnosis?)

[andrew-phillips-1]

Thanks @mmcleod89 @pineapple-cat I will try to have a think about this before our call tomorrow. I've also let Jenny and Loveleen know.

[andrew-phillips-1]

I think @mmcleod89 @pineapple-cat that this code is because the person has just tested in this period so we have not had a chance to get to the diagnosis part of the code. So for this test we need to consider that it could be a false negative before PrEP can be started.

(The fact that we are not using the same random draw when lower down we are seeing whether the person is diagnosed is not quite right though I think and it would be good to have that working in Python version.)

Am I misunderstanding ?

@LoveleenB-M @jennifersmith203

[pineapple-cat]

Finding this out would include testing multiple modules together; where would it make the most sense to put this test?

[mmcleod89]

That's a good question; and it's tricky to say right now because we're not sure what we would need to test yet to be able to figure this out. It might be possible to define some integration tests in a file of their own if we can say definitively what some behaviour of the simulation as a whole should look like when working together.

[mmcleod89]

Looks good! We will need to look into updating CI to fix the test issue