I don't believe this is a function yet but I wondered about the possibility of having mixed aggregate and individual level data in disease spaces where it is common for studies to report results and baseline characteristics by some common subgroup like bio-experience in Psoriasis and other immunology therapeutic areas. Essentially what I am imagining is your aggregate data entry has two entries for a single study with different baseline chars/aggregate data but with all patients in one group having 0 for the subgroup variable and the others 1.
I imagine if writing a one-off custom ml-nmr you could handle this by just sampling one d/delta the the study as a whole + whatever vector of Betas makes sense given assumptions re: classes/individual/shared/three-way interactions with the subgroup and integrate over the two sets of outcome measures separately? Or am I missing something there? @dmphillippo I know you mentioned in the past trying to find a way to fit the full ml-nmr associated with a network meta-interpolation so you might have already thought of a way around this that generalizes better.
I don't believe this is a function yet but I wondered about the possibility of having mixed aggregate and individual level data in disease spaces where it is common for studies to report results and baseline characteristics by some common subgroup like bio-experience in Psoriasis and other immunology therapeutic areas. Essentially what I am imagining is your aggregate data entry has two entries for a single study with different baseline chars/aggregate data but with all patients in one group having 0 for the subgroup variable and the others 1.
I imagine if writing a one-off custom ml-nmr you could handle this by just sampling one d/delta the the study as a whole + whatever vector of Betas makes sense given assumptions re: classes/individual/shared/three-way interactions with the subgroup and integrate over the two sets of outcome measures separately? Or am I missing something there? @dmphillippo I know you mentioned in the past trying to find a way to fit the full ml-nmr associated with a network meta-interpolation so you might have already thought of a way around this that generalizes better.