We initially retrieved 9950 RCTs and excluded 1021 duplicates. After title and abstract screening, we excluded a further 8831 studies that did not meet the inclusion criteria. We screened the full texts of the remaining 98 studies and excluded 22 that were not peer reviewed, 3 that focused on the effects of medication, 25 that focused on the effect of noninvasive brain stimulations or other modalities, 7 that were not RCTs, 4 that were study protocols, 9 that were further analyses of the same study, 6 that did not focus on poststroke aphasia, 1 that focused on acupuncture, 4 that compared different timings and intensities of an intervention, and 1 with data that could not be assessed. Finally, our network meta-analysis included 16 articles involving 775 participants (Fig. 1).21–36
<! Fig. 1 near here!>
Of the 16 included RCTs, 14 were parallel RCTs,21–29,31,32,34–36 and 2 were crossover studies.30,33 According to the Cochrane Handbook for Systematic Reviews of Interventions, including crossover studies in a network meta-analysis is acceptable.17 In our network meta-analysis, we included the final outcome data rather than only the outcome data from the first period (before crossover). The studies we reviewed employed the following speech therapies: CIAT,21–23,25,27,28,30,31,33–35 computerized speech and language therapy (CSLT),24,29 MIT,26,32,36 BOX therapy,34 M-MAT,25 PACE,22,26 and conventional thearpy,21,24,28–30,33,35,36; no intervention was prespecified as placebo. Rose et al. conducted a 3-arm study comparing CIAT, M-MAT, and no intervention.27 Eleven and 5 studies included patients who had poststroke aphasia for > 6 months22–27,30−32,34,36 and < 6 months, respectively.21,28,29,33,35 Outcomes were measured at the following time points: at the end of the intervention (n = 7),22,26,29,30,32,35,36 1 week after the intervention (n = 1),34 3 weeks after the intervention (n = 1),28 4 weeks after the intervention (n = 1),33 3 months after the intervention (n = 4),23,25,27,31 16 months after the intervention (n = 1),24 and 26 weeks after the intervention (n = 1).21 A summary of the main characteristics of the 16 included RCTs is presented in Supplementary Table 1.
The quality of the included RCTs was independently assessed using the PEDro scale. All included studies had PEDro scores between 6 and 9, with one study having excellent quality,31 and the remaining 15 studies having good quality.21–30,32−36 The detailed results of the bias risk assessment are listed in Table 1.
Table 1
| 1* | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | Total | Rating |
Kurland et al., 201622 | V | V | | V | | | V | V | V | V | V | 7 | Good |
Ciccone et al., 201521 | V | V | V | V | | | V | V | | V | V | 7 | Good |
Nenert et al., 201723 | V | V | | V | | | V | V | V | V | V | 7 | Good |
Pierce et al., 202325 | V | V | | V | | | V | V | | V | V | 6 | Good |
Rose et al., 202227 | V | V | | V | | | V | V | V | V | V | 7 | Good |
Sickert et al., 201328 | V | V | | V | | | V | V | | V | V | 6 | Good |
Stahl et al., 201730 | V | V | | V | | | V | V | V | V | V | 7 | Good |
Szaflarski et al., 201531 | V | V | V | V | | V | V | V | V | V | V | 9 | Excellent |
Vuksanović et al., 201833 | V | V | | V | | | V | V | V | V | V | 7 | Good |
Wilssens et al., 201534 | V | V | | V | | | | V | V | V | V | 6 | Good |
Woldag et al., 201635 | V | V | | V | | | V | V | V | V | V | 7 | Good |
Palmer et al., 201924 | V | V | | V | | | V | V | | V | V | 6 | Good |
Spaccavento et al., 202129 | V | V | | V | | | | V | V | V | V | 6 | Good |
Van Der Meulen et al., 201632 | V | V | V | V | | | V | V | | V | V | 7 | Good |
Raglio et al., 201626 | V | V | | V | | | V | V | V | V | V | 7 | Good |
Zhang et al., 202136 | V | V | | V | V | | V | V | V | V | V | 8 | Good |
PEDro scale criteria: 1, eligibility criteria and source of participants; 2, random allocation; 3, concealed allocation; 4, baseline comparability; 5, blinded participants; 6, blinded therapists; 7, blind assessors; 8, adequate follow-up; 9, intention-to-treat analysis; 10, between-group comparisons; 11, point estimates and variability. *Not included in the calculation of the total score |
<! Table 1 near here!>
A network diagram of the included speech therapies in terms of the speech performance of the patients is given in Fig. 2A. The pooled SMDs of speech performance revealed that CIAT was significantly more effective than no intervention (Fig. 3A). We also synthesized head-to-head studies separately to assess the differences among speech therapies. The results of the pairwise meta-analysis and network meta-analysis for the speech performance achieved with various speech therapies are shown in Table 2. We also analyzed the distribution of probabilities of effectiveness for each speech therapy, yielding the following ranking from the most to the least effective approach: MIT, CIAT, BOX therapy, M-MAT, PACE, CSLT, conventional therapy, and no intervention (Fig. 4A).
Table 2
Network Meta-Analysis Results for Speech Performance in All Patients.
Pairwise meta-analysis |
MIT | - | - | - | 0.26 [− 0.63; 1.14] | - | 0.66 [ 0.02; 1.30] | 0.04 [− 0.93; 1.01] |
0.16 [− 0.35; 0.68] | CIAT | −0.23 [− 1.55; 1.09] | 0.16 [− 0.14; 0.46] | 0.11 [− 0.69; 0.91] | - | 0.26 [− 0.03; 0.55] | 0.36 [ 0.07; 0.65] |
−0.07 [− 1.49; 1.35] | −0.23 [− 1.55; 1.09] | BOX therapy | - | - | - | - | - |
0.28 [− 0.29; 0.86] | 0.12 [− 0.17; 0.41] | 0.35 [− 1.00; 1.70] | M-MAT | - | - | - | 0.23 [− 0.10; 0.56] |
0.26 [− 0.39; 0.92] | 0.10 [− 0.54; 0.74] | 0.33 [− 1.14; 1.80] | −0.02 [− 0.71; 0.68] | PACE | - | - | - |
0.40 [− 0.18; 0.98] | 0.24 [− 0.16; 0.64] | 0.47 [− 0.91; 1.85] | 0.12 [− 0.37; 0.61] | 0.14 [− 0.58; 0.86] | CSLT | 0.06 [− 0.22; 0.35] | - |
0.47 [− 0.04; 0.97] | 0.30 [ 0.02; 0.58] | 0.53 [− 0.82; 1.88] | 0.18 [− 0.22; 0.58] | 0.20 [− 0.46; 0.87] | 0.06 [− 0.22; 0.35] | Conventional therapy | - |
0.48 [− 0.07; 1.03] | 0.31 [ 0.03; 0.59] | 0.54 [− 0.81; 1.90] | 0.19 [− 0.11; 0.50] | 0.21 [− 0.47; 0.89] | 0.07 [− 0.40; 0.55] | 0.01 [− 0.37; 0.40] | No intervention |
Network meta-analysis |
Data are expressed as SMD [95% credible interval]. Significant results are underscored. “-” means data not applicable. CIAT, constraint-induced aphasia therapy; PACE, promoting aphasics’ communicative effectiveness; CSLT, computerized speech and language therapy; MIT, melodic intonation therapy; M-MAT, multimodality aphasia therapy. |
<! Fig. 2–4 near here!>
<! Table 2 near here!>
Network plots contain nodes, which represent the interventions in the network, and lines, which highlight the available direct comparisons between pairs of interventions.17 Our network plot (Fig. 2A) depicts one triangle loop (connecting no intervention, CIAT, and M-MAT); the loop-specific heterogeneity revealed no significant inconsistency between the results of direct trials and indirect comparisons (Supplementary Table 2).
We compared the differences between traditional pairwise meta-analyses and network meta-analyses. The results, presented as forest plots (Supplementary Fig. 1), revealed that none of the differences were significant.
We assessed motor and sensory aphasia performance as our secondary outcomes. The pooled SMD of motor aphasia performance revealed that CSLT and CIAT were significantly more effective than no intervention (Fig. 3C). We examined the probabilities of effectiveness for all therapy types, yielding the following ranking, from the most to the least effective approach: CSLT, CIAT, BOX therapy, M-MAT, PACE, MIT, conventional therapy, and no intervention (Fig. 4C). The results of the secondary outcomes are presented in Figs. 2C and E, 3C and E, and 4C and E; Supplementary Figs. 3 and 5; and Supplementary Tables 5, 6, 9, and 10.
We conducted a subgroup analysis of patients who had poststroke aphasia for > 6 months.22–27,30−32,34 A network diagram of the included speech therapies in terms of speech performance during the chronic phase of aphasia is given in Fig. 2B. The pooled SMDs of speech performance revealed that CIAT was significantly more effective than no intervention (Fig. 3B). We also synthesized head-to-head studies separately to assess the differences among speech therapies. The results of the pairwise meta-analysis and network meta-analysis of speech performance for all therapy types are presented in Supplementary Table 3. We analyzed the distribution of probabilities of effectiveness for each speech therapy, yielding the following ranking, from the most to the least effective approach: BOX therapy, CIAT, CSLT, conventional therapy, MIT, M-MAT, PACE, and no intervention (Fig. 4B).
Our network plot (Fig. 2B) depicts one triangle loop (connecting no intervention, CIAT, and M-MAT); the loop-specific heterogeneity revealed no significant inconsistency between the results of direct and indirect comparisons (Supplementary Table 4).
We compared the differences between traditional pairwise meta-analyses and network meta-analyses; the results, presented as forest plots (Supplementary Fig. 2), revealed that none of the differences were significant.
The results of the subgroup analysis of the secondary outcomes are presented in Figs. 2D and F, 3D and F, and 4D and F; Supplementary Figs. 4 and 6; and Supplementary Tables 7, 8, 11, and 12.