Eligible studies
The search strategy yielded 395 potentially relevant articles. After removing duplicates, there were 286 studies. Then 55 potentially eligible articles were obtained by screening the titles and abstracts carefully. After careful full-text screening and assessment of the risk of bias, 20 RCTs met the inclusion criteria [4, 10, 12, 15, 28, 34-48]. This selection process and exclusion reasons were presented in the flow chart (Fig. 1). These studies, published between 2006 and 2019, included 758 TMD patients, which consisted of 358(47.23%) for LLLT, 73(9.63%) for TENS, and 327(43.14%) for placebo. 494 of 758 participants were women and 162 were unclear. The search results contained 20 direct comparisons, of which 17 were LLLT versus placebo, two were TENS versus placebo, one was LLLT versus TENS, and one was LLLT versus TENS versus placebo. Then the one-month follow-up network contained seven RCTs, including six comparisons of LLLT and placebo and one comparison of LLLT, TENS and placebo. The numbers of participants in these studies were 246. Ten (50%) of all included trials recruited patients from Brazil, four (20%) from Italy, three (15%) from Turkey, two (10%) from Iran, and one (5%) from Austria. Among 14 trials, the average participant was 14-76 years old, and the ages of the others were not mentioned.
Risk of bias assessment
Quantification of the risk of bias assessment was presented in Fig. 2. A random sequence was adequately generated with the help of a computer program or a random number list in five studies, and the risk of bias in the domain of randomization was judged to be 'low'. Allocation concealment was classified as an unclear risk in 17 studies because of no specific reporting of allocation concealment. The participants and personnel were successfully blinded in most studies (n=13, 65%), and the risk of bias was judged to be 'low'. Other studies without similar reports were rated as unclear risk of bias in this domain. As for blinding of outcome assessment, most studies were rated as a low risk of bias (n=15, 75%). Reviewers argued that the outcome assessment was unlikely to be influenced as long as the participants were blind since the pain intensity was assessed by the participants. All studies were determined to have a low risk of bias for the incomplete outcome data element because the information stemmed from studies with a low risk of bias, and there were no missing data. Besides, reviewers rated all studies as low risk of reporting bias and other bias. Because the pre-specified outcomes in the studies were reported, and no other significant bias problems were found.
Results of the network meta-analysis
Figure 3 showed two network plots for the pain reduction immediately after the final treatment and a month after therapy finalization. The size of the circle and the thickness of the edge were proportional to the number of participants and studies, respectively. Connecting lines represented direct comparisons between the two connected interventions, while the intervention pairs without connection can be compared indirectly through a network meta-analysis. Reviewers identified transitivity between studies after taking into account similarities among participants, interventions, outcomes measurements, and trial methodologies. The node-splitting method found no inconsistency because the statistical test of the difference for each group was not statistically significant (P>0.05).
According to the results of network meta-analysis, LLLT and TENS were associated with significantly lower pain intensity compared with placebo immediately after treatment[MD=1.99,95%CI:(1.07,2.92);MD=2.16,95%CI:(0.27,4.04),respectively]. However, there was no statistically significant difference between LLLT and TENS [MD =0.16, 95%CI:(-1.79,2.11)]. As for pain reduction one month after treatment, statistical significance was observed only for LLLT when compared with the placebo [MD=1.74, 95% CI: (0.02, 3.46)]. No statistically significant differences were found for placebo versus TENS or LLLT versus TENS [MD=1.54, 95%CI:(-2.26,5.35); MD=-0.20,95%CI:(-4.00,3.61), respectively].
The ranking probability of each intervention in terms of two periods of time was illustrated in Fig. 4. The possibilities of LLLT, TENS being the best intervention were 43.7%, 56.3% (immediately after treatment) and 53%, 45.5% (one month after treatment). And placebo always ranked the last.
Across the network, reviewers considered that three comparisons (A vs B, B vs C, A vs C, A, B, and C were used to describe the LLLT, placebo, and TENS respectively) were of low quality immediately after treatment. Due to the inconsistency or imprecision or both in the results, the confidence of the evidence in the studies was moderate or low. Meanwhile, reviewers found that there was a typical asymmetry in the funnel plot, which was mainly reflected in the fact that the studies on the left side of the mean effect size were much more than those on the right side, indicating potential publication bias. So, reviewers rated down all the direct and indirect comparisons. From the evidence after a month of therapy completion, one comparison (B vs A) was rated as “moderate”, and the other two comparisons (B vs C, A vs C) were rated as “low”. No publication bias was identified among these pairwise comparisons. Reviewers incorporated the GRADE judgments in Table 1. All the reasons for downgrading were labelled.