Pattern-mixture model in network meta-analysis of binary missing outcome data: one-stage or two-stage approach?

Table 2 Scenarios for the simulation set-up

Number of trials per comparison
typical loop	NO = 1, NP = 3, OP = 4
Trial size (\( {n}_{i,k}^E={n}_{i,k}^C={n}_i \) in trial i)
< 50 (small)	n_i~U(12, 39) placebo-controlled trials
	n_i~U(15, 49) old-controlled trials
> 100 (moderate)	n_i~U(102, 187) placebo-controlled trials n_i~U(128, 241) old-controlled trials
Initial event rates of the control arm in trial i
low events	\( {p}_{i,P}^{C,0}\sim U\left(0.05,0.09\right) \) placebo-controlled trials
	\( {p}_{i,O}^{C,0}\sim U\left(0.10,0.15\right) \) old-controlled trials
frequent events	\( {p}_{i,P}^{C,0}\sim U\left(0.27,0.40\right) \) placebo-controlled trials
	\( {p}_{i,O}^{C,0}\sim U\left(0.63,0.76\right) \) old-controlled trials
Unbalanced risk of missing outcome data (\( {q}_{i,k}^E<{q}_{i,k}^C \) in trial i)
moderate	\( {q}_{i,k}^E\sim U\left(0.05,0.10\right) \), \( {q}_{i,k}^C\sim U\left(0.11,0.20\right) \)
large	\( {q}_{i,k}^E\sim U\left(0.21,0.30\right) \), \( {q}_{i,k}^C\sim U\left(0.31,0.40\right) \)
Missingness mechanisms via log IMOR
informative	φ_{i, P}~TN(μ = − log (2), σ² = 1, a = log (1))
	φ_{i, k}~TN(μ = log (2), σ² = 1, a = log (1)) k = N, O
Treatment effects
basic parameters	LOR_NP = ln (2), LOR_OP = ln (1.5)
functional parameter	LOR_NO = LOR_NP − LOR_OP (consistency equation)
Common between-trial variance
predictive distribution	τ²~LΝ(−3.95, 1.34²) (small) τ²~LΝ(−2.56, 1.74²) (substantial)

Note: C Control; E Experimental arm; IMOR Informative missingness odds ratio; LN Log-normal distribution; LOR Log odds ratio; N New intervention; O Old intervention; P Placebo; T Truncated normal distribution; U Uniform distribution
Typical loop as defined by Veroniki et al. [32]
Using predictive log-normal distributions that correspond to all-cause mortality and generic health setting for small and substantial between-trial variance, respectively [33]

ISSN: 1471-2288