Network meta-analysis of disconnected networks: How dangerous are random baseline treatment effects?