Help with MIDAS - meta analysis of diagnostic accuracy- warning r(504)

[Ana Pimentel]

Hi everyone,

I’m new in the forum and I am not used to talking about statistical programs in English, so I hope you can understand my question and help me…
I’m performing a meta-analysis of diagnostic accuracy with 4 studies in Stata 11.2 . I ran the metandi command and everything was ok (summary estimates, HSROC, deeks funnel), but I couldn’t build the SROC with AUC, and the forest plots, then I installed midas command. After that, I can’t re-run those same analysis again, or build SROC, forest plot, Fagan nomogram, etc.

The following warning arose “Hessian has become unstable or asymmetric r(504)”- matrix has missing values; You have issued a matrix command attempting a matrix operation that, were it carried out, would result in a matrix with missing values; dividing by zero is a good example.

I am not sure if I did understand the warning… One of the included studies presented no false negative case. Is that the reason?! How can I solve this?
Before installing midas I had the summary estimates, but now not even that Stata runs anymore…
I really appreciate your help, thanks in advance.

Some commands I'm trying to run:
metandi tp fp fn tn
midas tp fp fn tn, texts(0.60) bfor(dss) id(study) ford fors
metandi tp fp fn tn ,plot
midas tp fp fn tn, sroc(both)
midas tp fp fn tn, fagan(0.2)

Our data:
author tp fp fn tn
P 19 6 13 84
F 94 134 71 1315
R 12 10 2 59
S 10 7 0 54

[Marcos Almeida]

Hello Ana,

Welcome to the Stata Forum!

Midas is a user-written program. According to the instructions we read in the FAQ. we are supposed to state this whenever we are making reference to ado files.

That said, midas demands at least 4 studies and ALL of them must have true and false negatives, and true and false positives as well.

Therefore, I believe that's the reason you're getting the error message.

Hopefully that helps.

Best,

Marcos.

[Ana Pimentel]

Hello Marcos, thank you for the quick response!
Do you know if there is some correction factor available to deal with that problem? I was reading about zcf(#) command, but I am not sure if I can (and how to) apply in this situation (B.A. Dwamena, R. Sylvester and R.C. Carlos. Midas: meta-analysis of diagnostic accuracy studies, available at http://fmwww.bc.edu/repec/bocode/m/midas.pdf)
Even with one study presenting no false negative case I was able to run metandi, and this problem only started after installing Midas…
Thank you for helping me.

Ana Laura

[Marcos Almeida]

Hello Ana,

Unfortunately, this is the furthest I can go. Sorry, but I have no experience with the option zcf(#), even though it seems to related to the logtransformations, not the (absolute) lack of values related to, say, tn, fn, tp or fp.