Help me, about network sidesplit in network meta-analysis

[Toshiharu Fujii]

good evening to all,

I am currently working on a network meta-analysis to assess the effects of A,B,C, and D drugs, utilizing the network command using STATA 17 SE. Specifically, I am attempting to investigate local inconsistency between A and B by running "network sidesplit A B." However, I encountered a problem as the following message appeared. Unfortunately, despite trying the "nosymmetric" option, I encountered the following persistent issue:
"Warning: there is no evidence about the A-B contrast outside trials containing A and B
mvmeta_bscov_proportional: mvmeta_mufrom sigma failed
r(459);
"
Even when I attempted to run "network sidesplit all," only the AB portion seems to be missing in both direct and indirect as follow.
Side Direct Indirect Difference
Coef. Std. Err. Coef. Std. Err. Coef. Std. Err. P>z
A B . . . . . . .
A C .2333009 .1118497 .0447364 .1813231 .1885645 .220767 0.393
A D .0834128 .1364211 .271976 .162957 -.1885632 .2207025 0.393
C D .038676 .1183973 -.1498883 .1856149 .1885643 .2207366 0.393

I have included the analysis data at the bottom of this post.
As a beginner in this field, I am seeking guidance on how to resolve this issue. Your assistance in providing clear instructions would be greatly appreciated.

study_number	drug	event	total_number
1	4	29	400
1	1	16	400
2	4	95	1989
2	3	80	1773
3	3	13	713
3	1	12	730
4	3	69	634
4	4	66	596
5	4	0	200
5	1	2	200
6	4	40	387
6	1	26	380
7	2	0	370
7	1	8	372
8	2	13	685
8	1	15	678
9	3	58	4623
9	1	48	4617
10	4	961	9235
10	1	929	9186
11	1	6	100
11	4	8	100
12	3	146	6741
12	1	111	6716
13	1	26	327
13	4	26	334

[Toshiharu Fujii]

I received following suggestion:

　The first line of the error message gives you the answer: there is no evidence about the A-B contrast outside trials containing A and B. That is, there is direct evidence but no indirect evidence. Now, the sidesplit command asks for the direct and indirect evidence to be compared. Clearly this is impossible and not interesting.

I recommend that

1. You use the nocodes option: -network setup event total_number, studyvar(study) trtvar(drug) rr ref(1) nocodes-. This keeps your existing copdes 1/2/3/4.
2. You run -network map-. This shows that direct & indirect evidence are only between treatments 2/3/4.
3. You run -network sidesplit all-. This picks out all the comparisons where there are both direct and indirect evidence.