How robust are findings of pairwise and network meta-analysis in the presence of missing participant outcome data?

Table 3 Characteristics of the 108 pairwise meta-analyses and 34 network meta-analyses. Values are median (range) [number of pairwise meta-analyses or network meta-analyses] unless stated otherwise

Characteristic	PMA	NMA
Number of studies	4 (3 to 25)	14 (4 to 104)
Randomised sample	60 (4 to 1996)	247 (12 to 18201)
Number of interventions	2	6 (3 to 22)
Observed comparisons (%)	1	42 (13 to 100^a)
Εvent frequency (%) in study-arms (binary outcomes only)	47 (26 to 67)^b	60 (42 to 76)^b
Studies with at least one zero-cell (binary outcomes only)	2 (1 to 6) [33]	1 (1 to 4)[9]
Intervention-comparison type:
Pharma versus placebo^c	37 (34)	23 (68)
Pharma versus pharma^c	46 (43)	9 (26)
Non-pharma^d versus pharma^c	3 (3)	1 (3)
Non-pharma versus non-pharma^c	22 (20)	1 (3)
Proportion of studies associated with:
Low risk of bias due to MOD^e	33 (8 to 100) [6]^f	50 (6 to 100) [3]^f
Moderate risk of bias due to MOD^e	33 (10 to 100) [2]^f	44 (7 to 92)
High risk of bias due to MOD^e	55 (11 to 100) [9]^f	28 (4 to 80)

MOD missing participant outcome data, NMA network meta-analysis, pharma pharmacological interventions, PMA pairwise meta-analysis
^aTwo networks on the continuous outcome are closed triangles
^bValues are median (interquartile range). The range of event frequency (%) in study-arms was 0 to 100 in pairwise meta-analyses and network meta-analyses
^cValues are numbers (percentages)
^dNon-pharmacological interventions include medical devices, surgical, complex, resources and infrastructure, behavioural, psychological, physical, complementary, educational, radiotherapy, vaccines, cellular and gene and screening [32]
^eFollowing the classification by Sackett et al. [24]: a proportion of missing participants up to 5% implies a low risk of bias due to MOD, more than 5% and up to 20% indicates a moderate risk of bias due to MOD, and more than 20% indicates a high risk of bias due to MOD
^fNumber of PMAs/NMAs that include only studies with a specific risk of bias due to MOD

ISSN: 1741-7015