Melodic Intonation Therapy for aphasia: A multi-level meta-analysis of randomised controlled trials and individual participant data

Abstract

IMPORTANCE: Melodic Intonation Therapy (MIT) is a prominent rehabilitation programme for individuals with post-stroke aphasia. Despite substantial progress in recent years, the efficacy of MIT remains not fully understood. OBJECTIVE: Based on a-priori hypotheses, the present meta-analysis investigated the efficacy of MIT while considering quality of outcomes (psychometrically validated versus unvalidated measures), experimental design (presence versus absence of randomisation and control group), influence of spontaneous recovery (quantified as number of months post-stroke), MIT version applied (original versus modified protocol), and level of generalisation (performance on trained versus untrained items). DATA SOURCES: An extensive literature search in all major online databases, trials registers and the grey literature identified 606 studies (years searched: 1973-2021). STUDY SELECTION: Inclusion criteria: randomised controlled trial (RCT) data or case reports on adults with aphasia; pre-post assessment of language performance. Exclusion criteria: substantial variation from original MIT protocol; unvalidated outcomes, unless both trained and untrained items were compared; essential information not indicated/retrievable. Final sample: 22 studies. DATA EXTRACTION AND SYNTHESIS: Following PRISMA guidelines, studies were double-coded. Multi-level mixed- and random-effects models were used to separately meta-analyse RCT and non-RCT data. MAIN OUTCOMES AND MEASURES: Measures of language performance focused on aphasia severity, everyday communication ability, domain-general function, language comprehension, non-communicative language expression, and speech-motor planning. RESULTS: Unvalidated outcomes appeared to attenuate MIT's effect size by a factor of 0.29-0.43 across study designs when compared to validated outcomes. Moreover, MIT's effect size was 5.7 times larger for non-RCT data compared to RCT data. Effect size also decreased with number of months post-stroke, suggesting confound through spontaneous recovery primarily within the first year post-stroke. In contrast, variation of the original MIT protocol did not systematically alter benefit from treatment. Crucially, analyses demonstrated significantly improved performance on trained and untrained items. The latter finding arose mainly from gains in repetition tasks, rather than other domains of verbal expression including everyday communication ability. CONCLUSIONS AND RELEVANCE: Accounting for various methodological aspects, the current results confirm the promising role of MIT in improving language performance on trained items and in repetition tasks, while highlighting possible limitations in promoting everyday communication ability.