Efficacy of vision-based treatments for children and teens with amblyopia: a systematic review and meta-analysis of randomised controlled trials

Authors: Brin TA, Chow A, Carter C, Oremus M, Bobier W, Thompson B.

Geographical coverage: Not reported

Sector: Biomedical

Sub-sector: Treatment of amblyopia

Equity focus: Children aged between 4-17 years old.

Study population: Children

Review type: Other review

Quantitative synthesis method: Narrative synthesis and meta-analysis

Qualitative synthesis method: Not applicable

Background:

Unilateral amblyopia is typically defined as visual acuity (VA) worse than 20/30 in an otherwise healthy eye, alongside a two-line interocular VA difference. However, visual deficits caused by amblyopia extend beyond reduced VA and encompass broader deficits such as impaired contrast sensitivity, stereopsis, spatial localisation and global form and motion perception. These deficits may adversely impact everyday tasks such as reading or playing sports.

Objectives: To identify differences in efficacy between vision-based treatments for improving visual acuity (VA) of the amblyopic eye in persons aged 4-17 years old.

Main findings:

Authors ultimately included 36 RCTs (1%) in the narrative synthesis. From this 36, five RCTs (14%) were included in the meta-analysis, 18 in the regression analysis (50%) and 26 in the network meta-analysis (72%).

Authors note variation in adherence rates among included studies among the 17 studies that fully reported adherence rates. Authors also found the linear regression significant, meaning that treatments with high adherence where rates show larger effect sizes favouring the intervention treatment: regression coefficient 0.022; 95% CI 0.004 to 0.040 (p = 0.020). Authors note the regression line may exaggerate the relationship of adherence and effect size.

When looking at the differences between patching and binocular treatments the meta-analysis showed they were statistically significant at the 5% level (−0.03 logMAR; 95% CI 0.01 to 0.04), but not clinically significant. Authors report high heterogeneity among the study in this analysis (I² = 80% and ‍X =  19.74, p <0.001) and a low certainty of evidence due to the significant inconsistencies and low precision.

Authors found that none of the treatments showed a statistically significant difference relative to parching 2-5 hours per day when compared to any treatment for the same amount of time.

In the network meta-analysis, authors included 26 studies comparing 14 vision-based treatments to patching 2-5 hours and yielding 26 (direct and indirect) pairwise comparisons. Most treatment comparisons involved patching or combination treatments. Authors report that the only comparison of SMD between groups that reached statistical significance was found in an indirect comparison between the combined binocular and combination group and patching 2-5 hours with the combined binocular and combination group having a greater SMD (SMD = 2.63, 95% CI 1.18 to 4.09, p = 0.9988), with a high level of certainty for the efficacy of this treatment.

Authors note that the adherence analysis showed that poor adherence may have been a factor in reducing treatment efficacy and may have affected the results of this review. With high or unclear risk of bias in almost half the included RCTs, the findings of this review should be interpreted with caution. Furthermore, high heterogeneity was reported among included studies across all comparisons.

Methodology:

Authors included studies which included patients with amblyopia aged 4-17 years old caused by strabismus and/or anisometropia with no ocular pathologies, mental illnesses, learning disabilities and/or systemic diseases. Studies had to include over five participants and use a vision-based treatment conducted in a randomised clinical trial compared to other types of vision-based treatments. Outcome of interest included change in visual acuity of the amblyopic eye from baseline. No restrictions were applied in terms of time period and setting of study. Only studies written in English were included in the review.

Authors searched PubMed (Medline), Ovid EMBASE, the Cochrane Library, Scopus and VisionCite. Two authors screened the studies independently for inclusion and extracted data and assessed the risk of bias of included studies. To assess the risk of bias, authors used the Cochrane risk of bias 2 tool. To assess whether adherence affected the effect sizes (Hedges’ g) of treatment comparisons, authors regressed Hedges’ g onto the adherence rates for 26 studies that reported adherence data for all treatment and comparator groups.

Authors conducted a meta-analysis comparing the patching to binocular treatments. The inverse variance method, DerSimonian-Laird estimator for τ2, and a random-effects model to obtain a pooled mean difference and 95% CI for the study-specific mean differences were used to carry out the meta-analysis. Authors also assessed heterogeneity and used GRADEpro to evaluate the overall certainty of evidence.

Authors conducted an exploratory regression analysis at the study level to examine the relative effect of different treatments of VA. A network meta-analysis was conducted to infer relationships between a broader number of treatments beyond those that were directly investigated in health-head trials. Authors used a random-effects model measured statistical heterogeneity using the ‍X² test and I² statistic. For each direct treatment comparison, the treatment-specific mean changes in logMAR were extracted over follow-up and obtained a common effect size, namely, Hedges’ g (a type of standard mean difference (SMD).

Applicability/external validity:

Authors note similarities on the findings to a previous network meta-analysis where it was found there was no significant difference between various types of amblyopia treatments. However, authors do not generalise the findings due to the limitations of included studies.

Geographic focus:

Geographic focus of the review was not reported.

Summary of quality assessment:

This review employs a comprehensive analysis of multiple vision-based therapies drawn from five different databases. Authors avoided bias by having two reviewers screen studies, extract data and quality assess included studies. However, important limitations were identified. The searches were not comprehensive enough to ensure that all available studies were identified, therefore important studies may have been missed in the analysis which could impact on the overall conclusions of the review. The review doesn’t make it clear the risk of bias of each included studies and analysis did not take the quality of studies into account. Therefore, there is low confidence in the conclusions about the effects of this review.