This systematic review and meta-analysis is reported according to the Preferred Reporting Items for Systematic reviews and Meta-Analyses (PRISMA) recommendations for systematic reviews and the Cochrane guidelines for systematic reviews (19, 20). The study protocol has been registered in PROSPERO (ID CRD4202014460).

Our PICO (Population, Intervention, Comparator, Outcome) question to guide the systematic review was formulated as follows: in adult patients over 18 years of age with non-fluent aphasia due to ischemic stroke, does the MIT, as compared to no therapy or other therapy, improve functional communication, expressive language (naming and repetition), and comprehension?

The following databases were searched: Cochrane Central register of Controlled Trials (CENTRAL), PUBMED, EMBASE, Clinical trials gov. (http://clinicaltrials.gov/), and Clinical trials results (www.clinicaltrialresults.org). We also performed a manual search of reference lists in other prior systematic reviews on the same topic as well as in guidelines to identify further potentially eligible studies.

The terms used in combination for the search were: “aphasia,” “language disorder,” “stroke,” “post-stroke,” “speech or language therapy,” “melodic Intonation Therapy or MIT,” “randomized controlled trial,” “random allocation,” “controlled clinical trial,” “control group,” “double-blind,” “single-blind,” “cross-over studies,” “functional communication,” “communication evaluation,” “Boston Diagnostic Aphasia Examination,” "Western Aphasia Battery,” “Aachen Aphasia Test” “Communicative Abilities in Daily Living,” “Communicative Activity Log,” “Sabadel,” “Amsterdam Nijmegen Everyday Language Test.” The last literature search was performed on September 20, 2019 using neither language nor publication date restrictions.

We selected randomized clinical trials that included adult patients over 18 years of age with non-fluent aphasia due to ischemic stroke whose intervention was MIT vs. no therapy or other therapy. We excluded non-randomized clinical trials, those without a control group, inadequate randomization processes, mixed populations including patients with aphasia of non-stroke etiology, those with no availability of post-stroke aphasia-specific data, non-speech and language therapy studies, and incomplete studies.

The search results were merged with the reference management software (Mendeley Ltd). Duplicate records were deleted. Those separate reports from the same study were linked and evaluated as a single study. Studies analyzed in previously published systematic reviews were manually included when not obtained by the database search (5). Only published data were included in the review.

The evaluation of study eligibility was performed by two authors (AHM and CPA) with the supervision of the review coordinator who identified all potentially relevant articles. After examining the titles and abstracts, clearly irrelevant reports were discarded and the full text of potentially relevant reports was reviewed.

The following information was included in the data collection form: eligibility of the study and/or the reason for exclusion, study design, study duration, allocation and blinding process, possible sources of bias, total number of participants, study setting, diagnostic criteria, age, sex, relevant comorbidity, dates of the study, total number of intervention groups, specific interventions, outcome definitions, time-point of reported outcomes related to stroke onset, number of participants allocated to each study group, number of outcomes in each study group, missing data (lost to follow-up), summary data for each intervention group and outcome (2 × 2 table for dichotomous data; means and SD for continuous data).

For the analysis of the extracted data, we used the Review Manager 5 software (Version 5.3.5; Copenhagen: The Nordic Cochrane Centre, The Cochrane Collaboration, 2014).

The quality of the included studies and the risk of bias of each study were evaluated following Cochrane Collaboration recommendations available in the Cochrane handbook of systematic reviews of interventions: (20) sequence generation, allocation sequence concealment, blinding, incomplete outcome data, selective outcome reporting, and other potential sources of bias. Quality control and assessment of bias were performed independently by two authors. Disagreements were resolved by discussion until consensus was achieved.

Under the coordination of the principal investigator, the entire team participated in this stage. Publication bias was assessed with the help of funnel plots. The results of the data analysis were imported into the GRADEpro Guideline Development Tool (McMaster University, 2015; developed by Evidence Prime, Inc.).

Three areas of communicative ability were analyzed as outcomes: functional communication, expressive language (naming and repetition), and comprehension.

The main outcome was improvement in language skills or in functional communication constructs as measured by a formal evaluation with validated tools including the Boston Diagnostic Aphasia Examination (BDAE) (21), the Aachen Aphasia Test (AAT) (22), Sabadel (23), Amsterdam Nijmegen Everyday Language Test (ANELT) (24) and the Communicative Activity Log (CAL) (25). Naming, repetition, and comprehension were considered secondary outcomes and were measured by similar validated tools. In brief, the BDAE evaluates conversation and expository speech, auditory comprehension, oral expression, reading, and writing. The AAT evaluates six subtests including spontaneous language, repetition, auditory comprehension, and naming. The Sabadel mainly consists of a story retelling task that measures functional language. Finally, both the ANELT test and the CAL evaluate verbal communication in daily life.

Collected data for each outcome were mean and standard deviation after the treatment period as well as the number of participants in the experimental and control groups. Standardized mean difference (SMD) was the summary statistic chosen, given that it allowed for the comparison of various psychometric scales. Data were analyzed on a random-effects basis. Results were summarized as standardized mean differences (SMD) and 95% confidence intervals (CIs), and the results from tests evaluating the same outcome were pooled in forest plots for a more comprehensive analysis of the global effect across studies. Heterogeneity across the studies was evaluated considering clinical reasoning and statistical measurements such as the chi-squared and the I² test. Sensitivity analyses were performed for each diagnostic tool to better identify possible sources of heterogeneity.

Finally, we used the GRADE approach to rate the quality of evidence, and we summarized the results in an evidence profile using the GRADE Pro tool.

Our search for articles in the databases produced a total of 226 results. After removing duplicates, 88 articles remained. The abstracts of these articles were analyzed, and 40 potentially eligible studies remained. Of those 40 studies, six studies were selected for full-text evaluation. Two were excluded: one for being a topic review (16) and the other because all the patients received MIT and they were randomized to transcranial direct current stimulation (tDCS) (26). Therefore, we included a total of 4 trials involving 94 patients in this systematic review (27–30) (Figure 1). Two studies were single, and two were multicenter and one of them used a modification of MIT (MMIT).

Table 1 summarizes participant characteristics of the included studies. Conklyn et al. (27) was a pilot study with a randomized controlled design with blinded measurement of outcomes. A total of 30 acute stroke survivors with non-fluent aphasia were randomly assigned to receive MIT treatment or no treatment. They used a modification of MIT, the modified melodic intonation therapy (MMIT), in which the therapist has the freedom to modify the protocol by using novel melodic phrases that closely match the prosody of the spoken phrases in both pitch and rhythm, as well as full phrases during initial treatment. The aim of this approach is to develop individualized treatment plans and early stimulation of right brain language structures. Outcome evaluations were based on the differences between the pre- and post-test assessments of two tasks similar to the responsive and repetition subsections of the Western Aphasia Battery. These assessments were developed for the study, and were not, therefore, validated (27).

Van der Meulen et al. (28) conducted a multicenter, waiting-list randomized controlled trial with a crossover design: patients were randomly allocated to the experimental group (MIT) or the control group (control intervention followed by delayed MIT) (28). A total of 27 participants were included: 16 in the experimental group and 11 in the control group. Outcome measures were the MIT repetition task, naming, repetition, and auditory comprehension subtests from the AAT; (22) the Amsterdam-Nijmegen Everyday Language Test; (24) and the Sabadel story retell task (23). The MIT repetition task comprised 11 trained and 11 untrained matched sentences.

Van der Meulen et al. (29) also used a multicenter waiting-list RCT design. Patients with chronic (>1 year) post-stroke aphasia were randomly allocated to the experimental group (6 weeks of MIT) or to the control group (6 weeks of no intervention followed by 6 weeks of MIT) (29). Assessments were performed at baseline (T1), after 6 weeks (T2), and 6 weeks later (T3). Efficacy was evaluated at T2 using univariable linear regression analyses. Outcome measures were chosen to examine several levels of therapy success: improvements in trained items, generalization to untrained items, and generalization to verbal communication. Of 17 included patients, 10 were allocated to the experimental group and 7 to the control group.

Haro et al. (30) was a randomized, crossover, interventional pilot trial. Participants were stroke survivors with post-stroke non-fluent aphasia. Patients randomized to group 1 received MIT first (12 sessions over 6 weeks) followed by no treatment; the patients in group 2 started active treatment between 3 and 6 months after their inclusion in the study, serving as waiting list controls for the first phase. Main measures were the CAL questionnaire and the Boston Diagnostic Aphasia Examination (BDAE), evaluated at baseline and at 6 and 12 weeks (21, 25). Twenty patients were included. Four of the patients allocated to group 2 crossed over to group 1, receiving the treatment first.

The risk of bias assessments are summarized in Supplementary Figures 1A,B. The risk of selection bias was considered to be low; all articles used a computer-generated allocation sequence or a randomization table. Allocation was correct as well; two studies used consecutively numbered sealed opaque envelopes (28, 29) and in the other two studies, the patients were consecutively allocated as long as they were included in the trial (30) or allocation was performed by a nursing manager who had no prior knowledge of the order of participants (27). Performance (participant/personnel) bias was considered to be low in one study (27) and unclear in three studies (28–30). Detection bias was considered low in three studies (blinded measurement of outcomes) (27, 29, 30) and was unclear in one given that the authors acknowledged that blinding could not be maintained because the patients spontaneously informed the researcher about their therapy allocation (28). Attrition bias was considered to be low in three studies (intention to treat analysis) (28–30) and unclear in the fourth, given that some participants had incomplete or missing data and reasons for withdrawal were not reported (27). Finally, reporting bias was low in two studies and unclear in other two, given that not all of the prespecified outcomes were reported (27, 29).

Figure 2 shows the effect of MIT on functional communication. Only the trial evaluating the CAL showed a significant effect of MIT on this outcome (SMD 1.47; 95% CI 0.39–2.56). Moderate heterogeneity was identified for this outcome (I2 36%). The main source of heterogeneity identified was the psychometric test chosen according to the sensitivity analysis performed (I2 67%). Besides, a positive effect of MIT on expressive language (repetition) was also found (SMD 0.45; 95% CI 0.01–0.90) (Figure 3). However, no significant effects on comprehension measurements were found (Figure 4). Neither of the secondary outcomes assessed showed significant statistical heterogeneity.

Supplementary Figures 2–4 show the funnel plots for each outcome evaluated in the meta-analysis. Supplementary Table 1 shows the summary of findings according to the GRADE criteria for evaluating the quality of evidence.