Figure 1.
Flow diagram of literature search.
Figure 2.
Combination treatment versus single DMARD.
The effect on all studies is −0.33 SMD (CI: −0.36, −0.29). Test for overall effect: Z = 17.66 (P<0.00001). Heterogeneity: Chi2 = 201.54, df = 44 (P<0.00001); I2 = 78%. One study [27] contributed to heterogeneity due an extreme effect (−3.71 SMD). The elimination of this study resulted in a little more conservative estimate (−0.31 SMD (CI:−0.35, −0.28), Z = 16.81), but eliminated the significant heterogeneity (I2 = 20, p = 0.13). Consequently, reference [27] was excluded from all comparisons. N, combination: 6725; N, single: 5446.
Table 1.
Study Characteristics and Risk of Bias Factors.
Figure 3.
Star shaped network showing the 6 different combination treatments anchored on single treatment as the common comparator.
The loops (grey lines) with corresponding numbers (1, 2, 3) show the subgroups, which were directly compared in addition to being indirectly compared. N indicates the number of patients in the groups.
Figure 4.
Double DMARD versus single DMARD: The effect of the Double DMARD treatment was highly significant (Z = 6.40).
All 18 Double studies showed heterogeneity (I2 = 89%). The exclusion of one reference [27], which had an extreme effect (−3.71 SMD), eliminated the significant heterogeneity (I2 = 17%).
Figure 5.
Triple DMARD versus single DMARD: The effect of the Triple DMARD treatment was highly significant (Z = 6.13).
The 6 Triple studies showed no heterogeneity (I2 = 0).
Figure 6.
TNF inhibitor combined with methotrexate versus single DMARD (methotrexate): The effect of TNF inhibitor was highly significant (Z = 10.84).
The 13 TNF inhibitor studies showed no significant heterogeneity (I2 = 42%, p = 0.06). The borderline heterogeneity was due to two golimumab studies (GoBefore, GoForward) [46]. The exclusion of these, did, however, not change the overall result (−0.33 SMD (CI: −0.39, −0.27)).
Figure 7.
Abatacept combined with methotrexate versus single DMARD (methotrexate): The effect of abatacept was significant (Z = 3.08).
The 2 abatacept studies showed no heterogeneity (I2 = 0).
Figure 8.
CD20 inhibitor treatment combined with single DMARD versus single DMARD: The effect of CD20 inhibitor treatment was highly significant (Z = 7.87).
The 5 CD20 inhibitor studies showed no heterogeneity (I2 = 0).
Figure 9.
Tocilizumab combined with methotrexate versus single DMARD (methotrexate): The effect of tocilizumab is significant (Z = 4.70).
Figure 10.
Indirect comparisons of different combination treatments.
There is a trend towards triple treatment being superior to abatacept and TNFi. All other differences between the combination treatments are non-significant. Abbreviations: SMD: Standardized mean difference. WMD: Weighted mean difference (SMD1-SMD2).
Table 2.
Observed Frequencies of bias factors for treatment groups.
Figure 11.
Funnel plot of all combination studies ([27] eliminated).
The left lower corner is empty compared with the right lower corner. This asymmetry may indicate that small studies with no effect was not published (publication bias). However, this asymmetry is quantitatively small, and probably does not affect the overall result. Exclusion of the three lower right studies [18], [19], [44] to eliminate the asymmetry did not change the overall result shown in Figure 2: −0.31 SMD (CI: −0.35, −0.27), test for overall effect: Z = 16.49 (P<0.00001). Heterogeneity: Chi2 = 48.41, df = 40 (P = 0.17); I2 = 17%. Abbreviations: SMD: Standardized mean difference.
Figure 12.
Analyses of bias factors and confounders, which differed significantly across treatment groups.
Only 1 bias factor (TNFi studies: Complete outcome versus incomplete outcome, line 9) had a significant influence on the outcome. Abbreviations: SMD: Standardized mean difference. WMD: Weighted mean difference (SMD1-SMD2); DM: DMARD; GC: Glucocorticoid; DN: DMARD naive; DIA: DMARD inadequate responder; D: double; T: Triple; Sp: Sponsoring; DB: double-blind; CO: Complete outcome; IO: Incomplete outcome; Dur: Disease duration at baseline; PARPR: Percentage of annual radiographic progression rate; L: low; H: High.
Table 3.
Other possible confounders across treatment groups.