Caroliene Meijndert

125 The effect of implant-abutment connections on peri-implant bone level change Risk of bias ROBINS-1 was used on the prospective non-randomised trials and the domain with a high risk of bias was ‘bias due to confounding’. Low risk of bias was detected in ‘bias due to deviations from intended interventions’ and in ‘bias in selection of the reported result’. Five studies had a high risk of bias on at least one domain (20.8%), 15 studies had a moderate risk of bias on at least one domain (62.5%), and 4 studies had a low risk of bias (16.7%). See supplementary figure 1 . RoB-2.0 was applied to the RCTs and a high risk of bias was seen in the domain ‘bias due to deviations from intended interventions’. Low risk of bias was detected in the domain ‘bias due to missing outcome data’. Eighteen studies had a high risk of bias on at least 1 domain (90%), and the remaining two studies had low risk of bias (10%). See supplementary figure 2 . Publication bias The funnel plot showed no studies on the lower right part of the plot, indicating a possibility of publication bias. See supplementary figure 3 . Quantitative synthesis i. Bone level change All the study groups reported bone level changes (n=48). The pooled bone level change per year for all the groups was -0.24 mm (95% CI: -0.27 to -0.20). In the PS-conical group, it was -0.16 mm (95% CI: -0.19 to -0.13), while in the PS-parallel group it was -0.14 mm (95% CI: -0.3 to -0.06) and in the PM-parallel group it was -0.48 mm (95% CI: -0.63 to -0.32). A meta-regression analysis revealed that the differences in bone change per year were statistically significant between the two platform switched and the platform matched group; PS-conical vs. PM-parallel (p=0.00), PS-parallel vs. PM-parallel (p=0.00). The difference between the conical and parallel platform switched groups (p=0.52) was not significant. The Forest plots of the random effects meta-analysis are depicted in figures 2-4 . 7