Effectiveness of Sound Therapy Durations in Tinnitus Rehabilitation: A Systematic Review and Meta-Analysis

Eligibility criteria

For an article to be eligible, publication year must be between 2000 and 2023 as well as having therapy that involved using sound stimulation or masking. Participants must have had tinnitus for 3 months or more. Types of stimulation employed in therapy were not limited as long as the therapy itself did not deviate from conventional masking methods. ST paired with counseling/education were considered acceptable. There were no restrictions to age or acoustic hearing ability.

As for the exclusion category, research that carried out sound therapy simultaneously with other methods such as pharmaceutical products, acupuncture or transcranial direct current stimulation were excluded. Studies that researched cognitive behavioral therapy that combine ST are also excluded. Any literature reviews, retrospective studies, meta-analysis or systematic review were excluded. No animal model studies were included.

Information sources and search strategy

An author conducted searches in four databases of the following: ScienceDirect, Cochrane Library, PubMed, and Korean Citation Index (KCI). Data collection was conducted between February and March of 2023. Search terms were decided on by determining two necessary terms; tinnitus and RCT as well as: (“tinnitus” AND (“masking therapy” OR “tinnitus retraining therapy” OR “music therapy” OR “sound therapy”) AND (“randomized controlled trial” OR “RCT” OR “randomized”)). A different set of terms was used for KCI to broaden results by inputting only “tinnitus” AND “sound therapy” into KCI’s search engine. Search terms were formulated by both reviewers beforehand.

Selection process

Conducted searches were collected per database into google spreadsheet tabs so that all articles could be traced back to the respective database. Collected information included title, author, year of publication, abstract and keywords if available. One reviewer (JWB) initiated exclusion of articles based on the year of publication and details based on titles and abstracts. Any systematic reviews or meta-analysis were filtered out. After the removal, remaining articles were put into one compiled list for the first screening which consisted of double checking the initial removal as well as to eliminate any duplicates.

Second screening was conducted to check for RCT. The second reviewer (JB) joined the screening process at this juncture. The remaining 105 articles were screened separately by the two reviewers to ensure that studies were RCT and had one of the approved treatment methods (ST, tinnitus retraining therapy or tinnitus masking) as well as availability. Both reviewers marked each article with Xs or Os. All rejections were debated before finalizing. Rejection elements during this screening included no access, not RCT, or that therapy was combined with another treatment. Thirty-five articles passed the second screening.

Third screening involved the process of double checking the second screening as well as a full body text check. Articles that did not use sound therapy as the main form of treatment were disqualified. All remaining articles were checked for participants, intervention, comparator, outcomes, and study type as well as available data. Articles that did not include extractable data, or feasibility as well as preliminary trials were excluded. Fifteen articles passed and were included in this analysis.

Data collection process

The collection process included checking method sections of each article for participants, level of hearing loss, used stimuli, outcome measures and mean values or presented graphics with survey outcome results. Many studies collected results in multiple formats and measured outcomes through various methods ranging from interviews, subjective evaluations to objective evaluation methods. All mean values (or mean differences), standard deviations and sample size from control groups and experimental groups were taken as data of interest. Control groups were defined as either standard of care or wait list control while experimental groups were any form of therapy that involved usage of masking sounds. These included sound therapy, tinnitus retraining therapy, placebo (in which sound was used), and tinnitus masking. The collected data values were outcomes measured through tinnitus handicap inventory (THI), tinnitus frequency inventory (TFI), tinnitus questionnaires (TQ), tinnitus handicap questionnaires (THQ), hospitals anxiety and depression scales (HADS), visual analogue scale (VAS), and World Health Organization quality of life brief version. However, to extract effect size without causing unit of analysis errors, only primary outcomes were extracted for computation. If a study had multiple primary measures for different sound therapy stimuli, then those were aggregated (Harrer et al., 2021) into one effect size.

If there were no mean values provided in a study and instead charts were provided as reported results, then a third-party program called Web Plot Digitizer (Automeris, Audistin, TX, US) was used to approximate pixel close value (Rohatgi, 2022). Although it would not provide an exact value as an outright reporting, the web program provided closer approximation than visual estimation (Figure 1).

Figure 1.

A flow diagram depicting the process of article selection.

Data items

Duration groups were determined based on frequently analyzed outcome measure intervals found on the gathered studies. Baseline group contained data of pre-treatment results. From pre-treatment measurements and measurements taken at 3 months were considered short groups. This group consisted participants who had received treatments for anywhere from 1 day to 3 months. Medium duration group consisted of measures taken anywhere between 3 months and a day to 6 months. Short and medium durations’ outcome measures consisted of a wide range of time due to different intervals at which questionnaires were completed among studies. The long duration group contained measures taken at the 12 months mark with duration of therapy ranging from 6 months and a day to 12 months. Lastly, the extended groups consisted of outcome measures completed at 18 months.

The main goal of this analysis is to compare perceived improvements achieved by participants after certain durations and compare them to see if there were any notable clinically significant differences. As such, any mean value of outcome measures, standard deviations and sample sizes were taken as data items. If mean differences were presented instead of mean value of evaluations, then mean difference values were added back to the baseline results to unify data with mean values.

Publication bias

Publication biases were checked with the small-study effect methods and presented through funnel plots. In R package called meta (Balduzzi et al. 2019) has the funnel.meta function that generates a funnel plot by using the calculated effect size. Funnel plots were visually inspected and analyzed in order to assess publication bias.

Study risk of bias

All articles were evaluated independently in accordance with the physiotherapy evidence database (PEDro) scales by both reviewers in order to assess risk of bias. The PEDro scale is a tool composed of 11 items graded with 1 or 0 to evaluate risk of bias and completeness of research papers reporting statistics which is composed of inclusion criteria and source, random allocation, concealed allocation, similarity at baseline, subject blinding, therapist blinding, assessor blinding, completeness of follow up, intention-to-treat analysis, between-group statistical comparisons, and point measures and variability (Moseley et al., 2019). By answering 1 or 0 (yes or no), an article would be given a score of up to 11 possible points with the score of 8 or higher indicating higher reliability.

Heterogeneity

Heterogeneity was assessed through Higgins and Thompson’s I² statistic, which was directly based on Cochran's Q test. With I², a general “rule of thumb” was followed where 25% or less to be considered low heterogeneity, 50% being moderate while 75% indicated substantial heterogeneity (Higgins et al., 2003). These percent ratings indicated approximately how much of the variance was caused by between-study heterogeneity.

Statistical analysis

All statistical analyses were performed by the R project version 4.3.0 (R foundation, Vienna, Austria) (Already Tomorrow) using an integrated program, R Studio (R foundation) “Cherry Blossom” along with multiple packages which were defined as part of the R source codes (R foundation) that allow it to function with dataset provisions, default statistics and graphics functions. Default packages used were: graphics, grDevices, methods, stats, and utils. Contributed packages were codes written by various authors to create specialized statistical methods that provided increased access to data or hardware and other functions to complement R. They were meant to enhance basic functions and help facilitate analysis further than possible with only basic functions. The basic packages turned on by default are of the following: base, graphics, grDevices, Matrix, methods, stats, and utils. Contributed packages listed as follow and not in any particular order: tidyverse 2.0.0, dplyr 1.1.2, forcats 1.0.0, ggplot2 3.4.2, lubridate 1.9.2, purrr 1.0.1, readr 2.1.4, stringr 1.5.0, tibble 3.2.1, tidyr 1.3.0, effectsize 0.8.3, ggvis ver. 0.4.8, pacman ver. 0.5.1, rappdirs ver. 0.3.3, rio ver. 0.5.29, FunnelPlotR ver. 0.4.1, Meta ver. 6.2-1, metadat ver. 1.2-0, metafor ver. 4.0-0, esc ver. 0.5.1, and rstatix ver. 0.7.2.

Standardized mean differences (SMD) were calculated as effect sizes with 95% confidence results by using Hedge’s G formula for paired sample tests. An example code is of the following: m.sm = metacont (n.e = sm$n.e, mean. e = sm$mean.e, sd.e = sm$sd.e, n.c = sm$n.c, mean.c = sm$mean.c, sd.c = sm$sd.c, studlab = (sm$Author...1), data = sm, sm = “SMD”, method.smd = “Hedges”, fixed = FALSE, random = TRUE, method.tau = “REML”, hakn = TRUE, title = “Short to Medium Duration”).

The above example code calculated effect size between durations named short and medium. This method was applied to all comparisons. Effect sizes were then pooled by duration groups being compared. Baseline to short duration, short to medium durations, medium to long durations and long to extended durations were paired and pooled using a random effects model (Hedges and Vevea, 1998). The p-values that resulted from this input were collected.

The null hypothesis was set as there was no clinically significant improvement observed between repeated measures. The alternative hypothesis was set as there was a clinically significant perceived improvement difference between two measured outcomes. If the calculated effect size failed to be less than or equal to the p-value of 0.05, then the group successfully rejected the null hypothesis and proved the alternative. On the other hand, if the p-value was greater than 0.05, then said group failed to reject the null hypothesis (Flechner and Tseng, 2011).

Quality assessment of studies

PEDro scores were evaluated for all articles used in this review (Table 1). Seven studies out of 15 were rated “excellent” (Argstatter et al., 2015; Bauer et al., 2017; Hall et al., 2022; Li et al., 2017; Li et al., 2019; Searchfield and Sanders, 2022; Yoo et al., 2022). Six of the processed studies were rated “good” (Henry et al., 2016; Jin et al., 2022; Mahboubi et al., 2017; Schad et al., 2018; Scherer et al., 2019; Westin et al., 2011). The remaining two studies received “fair” grades (Davis et al., 2007; Henry et al., 2006).

Table 1.

Table of PEDro scale completed for all articles

Characteristics of studies

Interventions

All studies analyzed the effects of ST or tinnitus masking in some form with different stimuli, duration and format. Majority of studies used broadband noise and white noise type stimuli as the masker (Bauer et al., 2017; Davis et al., 2007; Hall et al., 2022; Henry et al., 2006; Henry et al., 2016; Jin et al., 2022; Li et al., 2019; Mahboubi et al., 2017; Schad et al., 2018; Searchfield & Sanders, 2022; Westin et al., 2011) while the remaining studies used altered music that were either tailored according to tinnitus frequency or placebo (Argstatter et al., 2015; Li et al., 2017; Yoo et al., 2022). Some studies also included counseling or education as an additional intervention to the sound therapy or tinnitus masking (Argstatter et al., 2015; Bauer et al., 2017; Henry et al., 2006; Henry et al., 2016; Scherer et al., 2019; Searchfield & Sanders, 2022; Westin et al., 2011).

Control

Since this review’s objective was to compare effect sizes between durations, control groups did not serve any meaningful purpose unless two types of stimuli were compared. In such a case, the two variables were pooled together to be combined into one result. Otherwise, controls in each study were largely ignored for this review.

Outcome

All studies used subjective evaluations from participants as outcome measures. There were various measuring methods used among studies. The number of outcome measures differed among studies. A few studies only used one form of measure (Argstatter et al., 2015; Henry et al., 2016; Jin et al., 2022; Searchfield & Sanders, 2022). Vast majority used a combination of outcome measures that mainly consisted of one or multiple questionnaires (TFI, THI, VAS, TQ, tinnitus evaluation questionnaire, THQ, HADS, The World Health Organization of life-brief, and ISI etc.) as well as other forms of measure such as matched masking level, loudness discomfort level, loudness/pitch/bandwidth matching (Bauer et al., 2017; Davis et al., 2007; Hall et al., 2022; Henry et al., 2006; Jin et al., 2022; Li et al., 2017; Li et al., 2019; Mahboubi et al., 2017; Schad et al., 2018; Searchfield & Sanders, 2022; Westin et al., 2011; Yoo et al., 2022). However, most of these were secondary measures. For this review, each article’s declared primary measures were used as long as mean differences were calculable. Lastly, this study’s focus was on sound therapy. Any outcomes measured as results of sound therapy were pooled as long as the same measures were available.

Repeated measures

To analyze effects of therapies over time, outcomes were measured at certain intervals. Repeated measures analyze outcomes measured at different times on the same therapy or treatment (Singh et al., 2013). However, most studies did not extend their investigation to the 18 months mark. For this review, only studies with duration pairs available for effect size calculations were used for each group.

Study designs

Prior to article selection, it was agreed that all studies must be conducted as randomized controlled trials to be qualified. RCTs are designed with the objective of studying the effects of a new therapy or treatments by randomizing participants. The act of randomizing contributed to minimizing biases (Hariton & Locascio, 2018). Therefore, all studies were RCTs at varying levels of blinding.

Overall results

Table 2 presented the overall result comparisons of effect sizes and p-values calculated with Hedge’s g method after mean values, mean differences and standard deviations were collected.

Table 2.

Results of effect size (SMD) and p-values calculated through Hedge’s G

Majority of effect sizes were in the negatives, which meant that ST did prove beneficial to participants. This study, however, focused on attempting to find the diminishing return for time investment of ST. The duration group of baseline to short group displayed the largest absolute value, indicating the greatest difference between two intervals. The calculated p-value was smaller than 0.0001, meaning there was a meaningful significant difference (Figure 2). Short to medium duration’s SMD exhibited enough difference to be clinically significant with the p-value of 0.0342 (Figure 3). However, this value was a noticeable jump in p-value compared to that of the previous group. As the therapy duration entered 6 months or more (medium to long and long to extended duration groups), outcome measures failed to reject the null hypothesis (Figures 4, 5). This meant that the diminishing return for patients’ time investment occurs once therapy had been ongoing for 6 months.

Figure 2.

Forest plot of effect size for comparing baseline to short. SD: standard deviation, Std.: standard deviation, CI: confidential interval, df: degree of freedom.

Figure 3.

Forest plot of effect size for comparing short to medium duration. SD: standard deviation, Std.: standard deviation, CI: confidential interval, df: degree of freedom.

Figure 4.

Forest plot comparing medium to long durations. SD: standard deviation, Std.: standard deviation, CI: confidential interval, df: degree of freedom.

Figure 5.

Forest plot comparing long to extended durations. SD: standard deviation, Std.: standard deviation, CI: confidential interval, df: degree of freedom.

Table 3 presented the heterogeneity calculated by using Higgin’s and Thompson’s I² for each comparison. Baseline to short group had a substantial heterogeneity at 67.40%. Short to medium group had a moderate heterogeneity at 37.40%. Medium to long group had no notable heterogeneity. Long to extended group had a considerable heterogeneity at 85.10%.

Table 3.

I² value of all groups