Table 1.

Research questions investigated and description of applied missingness models and prior structures for δ_ik

Analysis	MOD model	Determination of δik				What is compared?
		Structure	Structural assumption	Scenario	Prior specification
A1	pattern-mixture	identical	intervention-specific	MAR on average	${\delta }_{\mathit{ik}}={\delta }_{t_{\mathit{ik}}}$, ${\delta }_{t_{\mathit{ik}}}~N\left(0,1\right)$	MAR on average is compared with extreme scenarios (on average). Intervention 1 is the reference of the network.
				MME on average	${\delta }_{\mathit{ik}}={\delta }_{t_{\mathit{ik}}}$, ${\delta }_{t_{\mathit{ik}}}~N\left(\mathit{ln}\left(2\right),1\right)$
				MMNE on average	${\delta }_{\mathit{ik}}={\delta }_{t_{\mathit{ik}}}$, ${\delta }_{t_{\mathit{ik}}}~N\left(-\mathit{ln}\left(2\right),1\right)$
				BC on average	${\delta }_{\mathit{ik}}={\delta }_{t_{\mathit{ik}}}$, ${\delta }_{t_{\mathit{ik}}}~N\left(\mathit{ln}\left(2\right),1\right)$,  tik ≠ 1 ${\delta }_{\mathit{ik}}={\delta }_{t_{\mathit{ik}}}$, ${\delta }_{t_{\mathit{ik}}}~N\left(-\mathit{ln}\left(2\right),1\right)$,  tik = 1
				WC on average	${\delta }_{\mathit{ik}}={\delta }_{t_{\mathit{ik}}}$, ${\delta }_{t_{\mathit{ik}}}~N\left(-\mathit{ln}\left(2\right),1\right)$,  tik ≠ 1 ${\delta }_{\mathit{ik}}={\delta }_{t_{\mathit{ik}}}$, ${\delta }_{t_{\mathit{ik}}}~N\left(\mathit{ln}\left(2\right),1\right)$,  tik = 1
A2	pattern-mixture	identical	intervention-specific	MAR fixed	${\delta }_{\mathit{ik}}={\delta }_{t_{\mathit{ik}}}$, ${\delta }_{t_{\mathit{ik}}}=0$	‘On average’ scenarios are compared with corresponding ‘fixed’ scenarios. Intervention 1 is the reference of the network.
				MME fixed	${\delta }_{\mathit{ik}}={\delta }_{t_{\mathit{ik}}}$, ${\delta }_{t_{\mathit{ik}}}=\mathit{ln}\left(2\right)$
				MMNE fixed	${\delta }_{\mathit{ik}}={\delta }_{t_{\mathit{ik}}}$, ${\delta }_{t_{\mathit{ik}}}=-\mathit{ln}\left(2\right)$
				BC fixed	${\delta }_{\mathit{ik}}={\delta }_{t_{\mathit{ik}}}$, ${\delta }_{t_{\mathit{ik}}}=\mathit{ln}\left(2\right)$,  tik ≠ 1 ${\delta }_{\mathit{ik}}={\delta }_{t_{\mathit{ik}}}$, ${\delta }_{t_{\mathit{ik}}}=-\mathit{ln}\left(2\right)$,  tik = 1
				WC fixed	${\delta }_{\mathit{ik}}={\delta }_{t_{\mathit{ik}}}$, ${\delta }_{t_{\mathit{ik}}}=-\mathit{ln}\left(2\right)$,  tik ≠ 1 ${\delta }_{\mathit{ik}}={\delta }_{t_{\mathit{ik}}}$, ${\delta }_{t_{\mathit{ik}}}=\mathit{ln}\left(2\right)$,  tik = 1
B1	pattern-mixture	identical	common-within-network	MAR on average	δik = δ, δ~N(0, 1)	Identical is compared with hierarchical for each structural assumption.
			trial-specific		δik = δi, δi~N(0, 1)
			intervention-specific		${\delta }_{\mathit{ik}}={\delta }_{t_{\mathit{ik}}}$, ${\delta }_{t_{\mathit{ik}}}~N\left(0,1\right)$
		hierarchical	common-within-network	MAR on average	δik~N(Δ, σ2), Δ~N(0, 1), σ~U(0, 1)
			trial-specific		${\delta }_{\mathit{ik}}~N\left({\Delta }_i,{\sigma }_i^2\right)$, Δi~N(0, 1), σi~U(0, 1)
			intervention-specific		${\delta }_{\mathit{ik}}~N\left({\Delta }_{t_{\mathit{ik}}},{\sigma }_{t_{\mathit{ik}}}^2\right)$, ${\Delta }_{t_{\mathit{ik}}}~N\left(0,1\right)$, ${\sigma }_{t_{\mathit{ik}}}~U\left(0,1\right)$
B2a	pattern-mixture	identical	as in analysis B1	MAR on average	as in analysis B1	Structural assumptions are compared with each other.
B2b	pattern-mixture	hierarchical	as in analysis B1	MAR on average	as in analysis B1	Structural assumptions are compared with each other.
B3	pattern-mixture	identical	intervention-specific	MAR on average	${\delta }_{\mathit{ik}}={\delta }_{t_{\mathit{ik}}}$, ${\delta }_{t_{\mathit{ik}}}~N\left(0,1\right)$	Pattern-mixture model is compared with selection model for each structural assumption.
	selection
C1	pattern-mixture	identical	intervention-specific	MAR on average	${\delta }_{\mathit{ik}}={\delta }_{t_{\mathit{ik}}}$, ${\delta }_{t_{\mathit{ik}}}~N\left(0,1\right)$ – moderate ${\delta }_{\mathit{ik}}={\delta }_{t_{\mathit{ik}}}$, ${\delta }_{t_{\mathit{ik}}}~N\left(\mathrm{0,0.25}\right)$ – liberal ${\delta }_{\mathit{ik}}={\delta }_{t_{\mathit{ik}}}$, ${\delta }_{t_{\mathit{ik}}}~N\left(0,4\right)$ – conservative	Moderate prior variance for δik is compared with liberal and conservative prior variance.

Abbreviations: BC Best-case scenario, MAR Missing at random, MME More missing cases are events, MMNE More missing cases are non-events, MOD Missing outcome data, WC, worst-case scenario