How many studies does a network meta-analysis need, and how many per treatment?

For a beginner writing a first network meta-analysis, I suggest about ten papers. Five or six can also work, but with that few it is sometimes hard to build a strong network model. If your whole story and model hold together well, five or six is a fine start.

How many studies does each treatment need? In fact, one is enough. Of course two, three or more cover a wider range of patients and are more convincing. But that is the nature of evidence-based research: we do not get as many studies as we want. We analyze as many as currently exist.

From this angle, if your field has few studies and you can pool them in a meta-analysis or network meta-analysis, you push existing knowledge a little further. Even if some treatments in the network have been compared in only one study, building the network and telling a story with new results is still very good.

How do you judge whether you have enough studies? For a beginner, I suggest around ten, that is, about ten trials, which is a good balance. At minimum, it is best to have five or six.