Remove clinical notes AFTER graft failure

[comorbidity]

This is challenging chicken/egg problem

LLM is extracting the Graft Failure (study endpoint) but we need to STOP reading charts after the first failure.

@Dtphelan1 documenting for future discussion, when LLM finds first confirmed instance of "Graft Failure" we should STOP processing, at least until a human has confirmed/deny the graft failure (also known as : "loss of function" or "loss of transplanted kidney")

[Dtphelan1]

Appreciate you noting this. FWIW this will probably happen not here, but in the GitLab project we use to house the projects running these experiments right now.

[comorbidity]

Related to #35 (Death Outcome Variable)

Briefly, we only want to LLM/human chart review notes up to but NOT beyond the Endpoint (Graft Failure or Death)

[Dtphelan1]

Noted!

[comorbidity]

Related to #59

@Dtphelan1 with the recent PR you should now have a way to determine when (according to structured data at least) the Kidney Graft Rejection OR Failure visit has been coded. Note that graft REJECTION can occur multiple times (though I think we only care about the time to FIRST event) but graft FAILURE can occur only once (Nephrectomy-remove the kidney or return to dialysis)

select  distinct 
        subject_ref, 
        enc_period_ordinal, 
        enc_period_start_day, 
        rejection, 
        failure 
from    irae__cohort_casedef
where   rejection or failure 
order by 
        subject_ref, 
        enc_period_start_day, 
        rejection, 
        failure ; 

I stopped short of just cutting OFF notes after the first REJECTION episode because there could be multiple EPISODES of REJECTION.