How to get pooled sensitivity for sparse data?

[Yuheng Zou]

Dear all,

I tended to get pooled sensitivity by midas command for a diagnostic meta-analysis, but it failed.
The data extracted for data synthesis were as follows:

study_id	tp	fp	fn	tn
1	7	0	0	35
2	37	0	0	0
3	10	0	0	11
4	9	0	0	30
5	5	0	0	8
6	13	0	0	43
7	8	0	0	17
8	42	0	1	3
9	24	0	2	35

I tries to use midas command, the code was "midas tp fp fn tn, res(sum)". It warned that "error obtaining starting values; try fitting a marginal model in order to diagnose the problem
r(459);"
Does anyone know how to solve this problem? I highly appreciate any advice! Thank you.
All the best.

Zou

[Tiago Pereira]

There is no solution to your problem. You have an almost perfect test: specificity and sensitivity very close to 100%. There are very few false positives and false negatives.

The best you can do is to present a univariate meta-analysis of proportions, summarizing the specificity.

Based on the available data, there is no need for a meta-analysis for sensitivity: it is 100% is all studies.

Hope this helps,

Tiago
P.S. Study 2 may be inappropriate to your meta-analysis or there are data extraction errors.

[Tiago Pereira]

Sorry, I missed this: use -metaprop- and/or metaprop_one

findit metaprop