Behavioral Training as New Treatment for Adult Amblyopia: A Meta-Analysis and Systematic Review

Abstract

Purpose.: New behavioral treatment methods, including dichoptic training, perceptual learning, and video gaming, have been proposed to improve visual function in adult amblyopia. Here, we conducted a meta-analysis of these methods to investigate the factors involved in amblyopia recovery and their clinical significance.

Methods.: Mean and individual participant data meta-analyses were performed on 24 studies using the new behavioral methods in adults. Studies were identified using PubMed, Google Scholar, and published reviews.

Results.: The new methods yielded a mean improvement in visual acuity of 0.17 logMAR with 32% participants achieving gains ≥ 0.2 logMAR, and a mean improvement in stereo sensitivity of 0.01 arcsec⁻¹ with 42% of participants improving ≥2 octaves. The most significant predictor of treatment outcome was visual acuity at the onset of treatment. Participants with more severe amblyopia improved more on visual acuity and less on stereo sensitivity than those with milder amblyopia. Better initial stereo sensitivity was a predictor of greater gains in stereo sensitivity following treatment. Treatment type, amblyopia type, age, and training duration did not have any significant influence on visual and stereo acuity outcomes.

Conclusions.: Our analyses showed that some participants may benefit from the new treatments; however, clinical trials are required to confirm these findings. Despite the diverse nature of the new behavioral methods, the lack of significant differences in visual and stereo sensitivity outcomes among them suggests that visual attention—a common element among the varied treatment methods—may play an important role in amblyopia recovery.

Amblyopia is a unilateral, or less commonly bilateral, decrease in best-corrected visual acuity that cannot be explained by structural abnormalities in the eye. It occurs as a result of abnormal visual experience during early childhood¹ and is associated most commonly with early childhood strabismus, anisometropia, or both (i.e., mixed).² In the Western world, amblyopia is estimated to affect 2% to 4% of the population.^2–5 Because of its prevalence, amblyopia is a significant burden on society. It was estimated that untreated amblyopia can cause a yearly loss of $7.4 billion in earning power and that it is associated with substantial personal costs due to reduced quality of life and limited career choices.^6,7 

The standard treatment for amblyopia is occlusion therapy, which involves patching of the dominant eye. In young children this treatment is quite effective, with 75% of children showing improvement in visual acuity.^8,9 However, its effectiveness decreases in older children and adults.^9,10 This decrease in treatment effectiveness has traditionally been attributed to decreased plasticity in the mature brain (discussed in detail in Refs. 11, 12) at the end of the critical period of development.¹³ More recently, however, the adult brain has been shown to be much more plastic than it was once believed to be.^14,15 In support of this evidence, new behavioral treatments have been developed over the past decades for adults with amblyopia, with the goal to improve visual acuity, stereo acuity, and/or contrast sensitivity (for general reviews of each particular method see Refs. 11, 16, 17). These methods include dichoptic training, perceptual learning, and video gaming (see next section). Moreover, several studies using occlusion therapy in adults, in conjunction with refractive correction and near-vision activities (e.g., tracing contours, red-green pen and paper exercises), reported increases in visual acuity.^18–20 

Given the importance of the potential for treatment of amblyopia in adulthood, it is crucial to understand the factors involved in adult amblyopia plasticity and assess the clinical significance of the observed visual function improvements. Additionally, understanding the mechanisms of plasticity in adult amblyopia could reveal the neural substrates of the disorder. Here, we report mean and individual participant data analyses²¹ on the new behavioral methods for the first time. The goals of this article were to (1) briefly review the nature and proposed mechanisms of the new behavioral treatment methods; (2) identify the key factors affecting treatment efficacy in adults; (3) compare the effectiveness of different treatment methods; (4) assess the clinical significance of treatment effects in adults with amblyopia; (5) discuss the underlying mechanisms of amblyopia in light of these analyses; and (6) identify key research and methodological issues requiring further investigation. 

Behavioral Treatments of Adult Amblyopia

Dichoptic Training

Dichoptic training originates from the idea that amblyopia is an inherently binocular disorder rooted in interocular suppression, and that an effective treatment should engage both eyes (e.g., Hess et al.²²). In dichoptic training, participants with amblyopia are trained on tasks in which stimuli are presented dichoptically with the contrast of the image to the fellow eye attenuated in order to encourage binocular combination of the two inputs. The original task was a motion coherence task, in which the participant had to identify the direction of motion of signal dots among noise dots, with signal dots presented to one eye and noise dots to the other eye.²² With training, the difference in contrast between the two eyes could be reduced. Later, a Tetris-based game was developed in which the game pieces were presented dichoptically such that some of the blocks were seen only by the amblyopic eye, some only by the fellow eye, and some by both eyes. By attenuating the contrast of the fellow eye blocks, the signals from the two eyes could be combined in order to play the game effectively.^23–25 These dichoptic training techniques improved amblyopic eye visual acuity, and in some cases stereo acuity.¹⁷ The authors suggested that the binocular circuitry in amblyopia is not completely lost but instead is rendered functionally monocular by the active suppression of the weak amblyopic eye input by the stronger fellow eye.^22,26 Accordingly, the authors proposed that dichoptic training helps reduce this interocular suppression, leading to improved monocular and binocular vision. 

Perceptual Learning

Perceptual learning refers to the ability to improve performance on a sensory task by repeated practice.¹¹ It induces plasticity in the normal visual system and thus has been suggested as a potential method for amblyopia recovery in adults.^12,16,27 Perceptual learning for amblyopia entails extensive training (sometimes tens of thousands of trials) on a perceptual task with the fellow eye patched (e.g., detecting a Gabor patch when its contrast is varied). Participants normally improve on the task they have been trained on, and often the improvement transfers to the other eye, to other tasks, and/or to visual and stereo acuity. The tasks yielding the greatest improvement in visual function involve repeated measurements of contrast sensitivity.¹² Initially, it was suggested that perceptual learning reduces lateral inhibition at the early stages of visual processing.²⁸ Later studies suggested that perceptual learning reduces internal noise inherent in amblyopia and retunes the relative weighting of the binocular input (template retuning).^29,30 Alternatively, it has been argued that the complete learning transfer observed in perceptual learning suggests a higher-level rule-based cognitive learning process.³¹ 

Video Gaming

Video gaming has been shown to improve certain visual functions in the normal visual system (for review see Ref. 32). Li, Ngo, Nguyen, and Levi³³ proposed the use of video games to induce plasticity and improve visual function in adult amblyopia. They showed that playing either action (Medal of Honor) or nonaction (SimCity) video games with the fellow eye patched improved visual and stereo acuity in amblyopia. They suggested that video games induce essentially the same changes as perceptual learning—a reduction of noise and an increase in sampling efficiency. Since then, two additional studies have examined the effect of video games on amblyopia, with both binocular³⁴ and monocular tasks.³⁵ 

Methods

Selection Criteria

Studies were identified using PubMed and Google Scholar, with the search strings “dichoptic amblyopia,” “perceptual learning amblyopia,” and “video game amblyopia,” as well as published reviews. The searches were performed by two authors (IT and LC) during September and November of 2014. We searched all available literature without limiting it by publication year. We selected studies that have reported a statistically significant, nontransient (longer than a few hours) change in visual and/or stereo acuity in adults with strabismic, anisometropic, or mixed amblyopia. We selected studies using methods that produced a significant change in visual/stereo acuity because we concentrated on treatments that induced plasticity in adult amblyopia. We did not include contrast sensitivity in our analysis because it was not reported in many studies, and those that did report it used methods that differed too much for meaningful comparison. Table 1 shows the 26 studies we selected for inclusion in the meta-analyses (24 studies using new behavioral methods and 2 using occlusion therapy in adults as a component of the treatment). Supplementary Table S1 lists all 45 studies considered for this review initially, as well as the reasons for exclusion for those not included in the final meta-analyses. Of the 45 studies we considered (see Supplementary Table S1), only two studies, both using perceptual learning, showed no statistically significant improvement in visual acuity. As has been noted by Levi and Li,¹⁶ some perceptual learning methods are more effective than others. Since we had a very good sample (n = 16) of perceptual learning studies that produced significant results, removing the two papers with ineffective methods did not reduce the power of our analysis. 

Table 1

 

Studies Included in the Final Analysis

 

Data Collection

Individual participant data (IPD) were collected using characteristics tables from published articles, combined with individual posttraining visual acuity and stereo acuity scores. These posttraining scores were either provided in the papers in table format, provided by the authors through personal communication, or collected from plots in the papers using the software DigitizeIt (DigitizeIt, Braunschweig, Germany). The source of data is indicated in the “Raw Data” column in Table 1. The following information was collected when possible: amblyopia type, age, initial visual acuity and stereo acuity, final visual and stereo acuity, treatment type, and training duration. In total, there were 220 individual scores from 23 out of the 24 studies using new behavioral methods (the study by Polat, Ma-Naim, Belkin, and Sagi²⁸ was excluded since no IPD were available). When the two occlusion therapy studies were added, there were 243 individual scores. 

Visual acuity data were converted to logMAR and stereo acuity data to stereo sensitivity (arcsec⁻¹). We chose stereo sensitivity over stereo acuity since nil stereo acuity on any test could be easily represented as zero stereo sensitivity and thus be used in quantitative analysis. The alternative would be to select a stereo acuity that is one level above the highest level on each particular test (e.g., 1600 arcsec for Preschool Randot, in which 800 arcsec is maximum). However, this would result in inhomogeneous and arbitrary stereo acuity scores being assigned to participants who have no demonstrable stereopsis, which affects the robustness of the statistical analysis (we tried assigning the same arbitrary value of stereo acuity to all stereoblind participants; however, the magnitude of this value had a strong influence on the statistical outcomes and the estimates of true effect size). 

Test–Retest Reliability

Any improvement in visual function must be at or above test–retest reliability to ensure that any change is not merely the result of inherent measurement variability. In the case of visual acuity, test–retest reliability depends on the particular type of vision chart used, as well as the method of acuity calculation. Charts with high levels of chance performance such as the Tumbling E (25% chance) yield higher test–retest reliability than charts with lower chance performance, such as the ETDRS (e.g., Lovie-Kitchin⁵⁴). Line-by-line scoring yields higher test–retest reliability threshold than letter-by-letter scoring (e.g., Rosser, Cousens, Murdoch, Fitzke, and Laidlaw⁵⁵). Test–retest values for different charts and scoring methods measured in the literature are shown in Table 2. These values range from 0.26 to 0.07 logMAR with a mean of 0.15 logMAR (1.5 chart lines). For simplicity, we adopt this mean as the critical measure of test–retest reliability in our analysis for visual acuity. 

Table 2

 

Test–Retest Reliability for Visual Acuity Charts, Visual Acuity Values in logMAR

 

For stereo acuity, test–retest reliability ranges from 0.78 to 2.27 octaves with a mean of 1.6 octaves⁶⁴ (an octave is a halving/doubling of the score) and thus depends on the initial stereo acuity score (see Table 3). Since the vast majority of the studies analyzed here used the Randot test, which has a test–retest reliability of 2 octaves, we use this value as the test–retest reliability in our analysis for stereo acuity. 

Table 3

 

Test–Retest Reliability for Stereo Acuity Tests (in Octaves)

 

Data Analysis

We conducted two types of meta-analysis. First, mean data were analyzed using a mixed-effects model on means from the metafor package for R.⁶⁵ Second, we performed an IPD meta-analysis. Unlike conventional meta-analysis in which each study is represented by one mean score, in IPD, each study is represented by all the individual scores in that study, which are analyzed together, while taking into account the clustering of the scores into studies (using study ID as a random variable). Individual participant data are becoming increasingly popular (for review, see Ref. 21) because of their many advantages over conventional meta-analysis. For example, they allow for new subgroup analyses to be performed and the usage of new statistical methods not used in the original study.⁶⁶ Because individual data can be difficult to obtain, using a combination of mean and IPD analyses is the current recommended practice for meta-analyses.⁶⁶ In this study, IPD were analyzed with a mixed-effects model using PROC MIXED in SAS (SAS Institute, Inc., Cary, NC, USA). To assess the significance of each factor (i.e., amblyopia type, age, initial visual acuity and stereo sensitivity, treatment type, and training duration) with respect to the outcome variables (visual acuity and stereo sensitivity), we first conducted univariate analyses with only one factor and then included all the factors with P values smaller than 0.2 in the subsequently tested multivariate model. We then subtracted nonsignificant factors from the multivariate model one by one and compared the resulting models' Akaike information criterion (AIC) to find the best-fitting model. For all tests, an α level of 0.05 was used. 

Results

Visual Acuity

We first fitted a mixed-effects model to the mean change in visual acuity for 21 of 24 studies using new behavioral treatments. Three studies were excluded. The study by Fronius, Cirina, Cordey, and Ohrloff⁴¹ was excluded because there was only one participant, so variance could not be computed. The studies by Ding and Levi⁴⁶ and Ooi, Su, Natale, and He⁵² were excluded because they did not measure visual acuity. The mean and 95% confidence interval (CI) of visual acuity improvement from each of the studies are shown in a forest plot in Figure 1 (see also Table 1). This analysis assessed the true effect size (i.e., an estimate of the change in visual acuity that would result from a study that includes the entire population of adults with amblyopia). We found that the estimated true effect size was 0.17 logMAR (1.7 chart lines; 95% CI = ±0.02 logMAR) and that it was significantly different from zero (z = 16.34, P < 0.0001) (see Fig. 1). The effect size 95% CI was also just above the mean test–retest reliability of 0.15 logMAR. However, only four of 21 studies showed a mean improvement in visual acuity and 95% CI that exceeded the test–retest reliability of the particular visual acuity test used (gray vertical lines in Fig. 1, computed from Table 2). These four studies included one using dichoptic training,³⁶ two using perceptual learning,^25,27 and one using video games.³⁴ Nevertheless, on average, 56% of the participants showed improvements in visual acuity at or above mean test–retest reliability of 0.15 logMAR, and 32% improved by 0.2 logMAR (two chart lines) or more (see Fig. 1). 

Figure 1

 

Means and 95% CI for visual acuity (VA) improvement in 21 studies used in the mixed-effects analyses. The red dashed line shows the mean test–retest reliability of 0.15 logMAR (see Methods). The short gray lines show the test–retest reliability for the test used in each individual study. The size of the square symbols represents the relative weight given to each study in the model. The diamond symbol on the bottom row shows the estimated true effect size. The two middle columns show the percentage of participants who had visual acuity improved by 0.15 logMAR or higher and by 0.2 logMAR or higher.

To understand the factors affecting the improvement of visual acuity, we performed an IPD meta-analysis using individual participant data. The relations between the various factors and the improvement in visual acuity are shown in Figure 2. The results of the univariate and the multivariate analyses are shown in Supplementary Tables S2 and S3, respectively. In the model that generated the best fit (AIC = −392), only initial visual acuity [F(1,192) = 77, P < 0.0001] was a significant factor. The effect of initial visual acuity can be seen in Figure 2D. Greater improvements in visual acuity corresponded to worse initial visual acuity; that is, participants with more severe amblyopia benefited more from treatment than those with milder acuity deficits. We also performed separate mixed-model analyses with participants who underwent treatment with dichoptic training (from five articles; see Supplementary Fig. S1), perceptual learning (from 13 articles; see Supplementary Fig. S2), and video games (from three articles; see Supplementary Fig. S3). For all three methods, only initial visual acuity was a significant factor in visual acuity outcomes in the best-fitting models (dichoptic training: F(1,52) = 50, P < 0.001; perceptual learning: F(1,111) = 38.2, P < 0.001; video games: F(1,27) = 20.6, P < 0.001). 

Figure 2

 

Relations between different factors and improvement in visual acuity. Black dashed lines in bar plots (A, B) show the mean test–retest reliability for visual acuity (see Methods), and the error bars show the 95% CI. The black dashed lines in the scatterplots (C–F) show regression lines fitted by the univariate models (see text). The black solid line in (D) shows the improvement needed to attain 20/20 vision. Different symbols in the scatterplots correspond to different treatment types. DC, dichoptic treatment; PL, perceptual learning; VG, video games; VA, visual acuity; SS, stereo sensitivity.

Finally, we compared the new behavioral methods to occlusion therapy in adults. This was done by adding participant data from two studies (shown in Table 1) in which occlusion therapy was used in combination with refractive correction and near-vision activities.^19,20 We then repeated the mixed-effects analysis of the main effect of treatment type, which was significant [F(3,214) = 3.7, P = 0.012] (in the original analysis, without occlusion therapy, the main effect of treatment type was not significant [F(2,193) = 1.44, P = 0.24]). Post hoc analysis with Tukey-Kramer adjustment revealed that occlusion therapy induced a significantly greater improvement in visual acuity than perceptual learning [t(214) = −3.24, P = 0.007]. However, comparisons between all the other treatment methods were not significant. The best-fitting multivariate model now included treatment as a significant factor. 

Stereo Sensitivity

We first fitted a mixed-effects model to the mean change in stereo sensitivity for 13 studies (Fig. 3). Ten studies were excluded because they did not measure stereo acuity (see Table 1), and one was excluded because stereo acuity was measured for one participant only. The estimated true effect size was ∼0.01 arcsec⁻¹ (95% CI = ±0.004), which was significantly different from zero (z = 4, P < 0.0001) (see Fig. 3). On average, 42% of all participants (with or without initial stereopsis) showed improvements in stereo sensitivity at or above test–retest reliability of 2 octaves. In this calculation, the initially stereoblind participants who improved their stereo sensitivity were automatically counted as improving beyond 2 octaves. This finding should be interpreted with caution because participants with gross stereopsis could sometimes show nil results on one visit and be able to pass the first level of the test on the next visit without any treatment intervention whatsoever.⁶⁴ To address this limitation, we performed an additional analysis only on participants who had demonstrable initial stereopsis. We found that 34% of them improved by 2 octaves or more. 

Figure 3

 

Means and 95% CI for stereo sensitivity (SS) improvement in 12 studies used in the mixed-effects analyses. The size of the square symbols represents the relative weight given to each study in the model. The diamond symbol on the bottom row shows the estimated true effect size. The vertical dashed line shows the zero improvement point. The middle columns show the percentage of participants with initial stereopsis who improved by 2 octaves or higher and the percentage of all participants (with and without initial stereopsis) whose stereopsis improved by 2 or more octaves.

Next, we performed an IPD meta-analysis on individual stereo sensitivity data (Fig. 4; Supplementary Tables S4, S5). The multivariate model that generated the best fit (AIC = −674) included two significant factors: initial visual acuity [F(1,94) = 5.8, P = 0.018] and initial stereo sensitivity [F(1,94) = 6.9, P = 0.01]. Better initial visual acuity was associated with greater improvements in stereo sensitivity (Fig. 4D). This is opposite to the effect that initial visual acuity had on the improvement in visual acuity. Similarly, better initial stereo sensitivity correlated with greater improvements in stereo sensitivity (Fig. 4F). These results suggest that better initial visual acuity correlates with better initial stereo sensitivity. We tested this relation using a mixed-effects model with initial visual acuity as the fixed factor, study as the random factor, and initial stereo sensitivity as the dependent variable. The relation between initial visual acuity and initial stereo sensitivity was indeed statistically significant [F(1,108) = 23.2, P < 0.0001]. 

Figure 4

 

Relations between different factors and the improvement in stereo sensitivity. Error bars in bar graphs (A, B) show the 95% CI. The black dashed lines in the scatterplots show regression lines fitted by the univariate models (see text). Different symbols in the scatterplots (C–F) correspond to different treatment types. DC, dichoptic treatment; PL, perceptual learning; VG, video games; VA, visual acuity; SS, stereo sensitivity.

We also performed mixed-model analyses of stereo sensitivity with participants who underwent dichoptic training only (from five articles; see Supplementary Fig. S4) and with participants who underwent perceptual learning only separately (from six articles; see Supplementary Fig. S5). We did not test video gaming studies separately because only partial data were available on stereo sensitivity and thus the sample size was too small for meaningful analysis. For perceptual learning, only initial stereo sensitivity had a significant effect on final stereo sensitivity [F(1,29) = 10.9, P < 0.0025]; that is, better initial stereo sensitivity resulted in greater improvements in stereo sensitivity. For dichoptic treatment, only visual acuity showed a trend toward significance [F(1,52) = 3.44, P = 0.069]; that is, better initial visual acuity was associated with greater improvement in stereo sensitivity. Regardless of the differences in statistical significance of factors, all the data trends looked similar for the two treatments (compare Supplementary Figs. S4, S5). 

Standardized Classification of Amblyopia Type

The criteria used to classify amblyopia type differed across studies, and in some cases they were not specified explicitly. To assess the effect of amblyopia type on visual acuity outcome more rigorously, we reclassified all participants using the refractive errors and eye deviation data provided (where available) based on specific well-accepted clinical definitions. Anisometropic amblyopia was defined as amblyopia in the presence of at least 1 diopter difference of refractive error between the eyes allowing for up to 8 prism diopters of microstrabismus (clinically it is understood that the amblyogenic component is the anisometropia and not the microstrabismus⁶⁷). Strabismic amblyopia was defined as amblyopia in the presence of a manifest eye misalignment at distance and/or near fixation without anisometropia. Mixed amblyopia was defined as amblyopia in the presence of a combination of anisometropia and strabismus of ≥8 prism diopters. When assessment was not possible due to lack of either refractive error or eye deviation information (9 out of 24 studies), we maintained the original classification provided by the studies. The classification has changed for ∼16% of participants with the use of the standardized definition. We repeated all the analyses using this standardized classification, and found again that amblyopia type (anisometropic, strabismic, or mixed) was not a significant factor influencing visual acuity or stereo sensitivity outcomes. 

Discussion

Factors Influencing Visual and Stereo Acuity Outcomes

Age.

Age has been implicated as a pivotal factor in treatment effectiveness in children.^9,10 It is thus reasonable to predict age to be a significant factor for treatment effectiveness across the life span. Our analysis showed that age does not have a significant effect on visual and stereo sensitivity outcomes in adults, irrespective of treatment methods. Levi and Li¹⁶ also found no effect of age on visual acuity improvement in their review of perceptual learning. This suggests that the adult visual cortex retains its plasticity into adulthood, which is supported by evidence of plasticity in visual attention and learning in normal vision.^68,69 

Amblyopia Type.

It has been shown that participants with different amblyopia types demonstrate distinct patterns of deficits.^70–72 However, the nature and degree of suppression of the amblyopic eye was found to be largely the same across types.⁷³ Our analysis showed that visual and stereo sensitivity improvements do not depend on amblyopia type. A similar absence of the effect of amblyopia types was found by Stewart, Fielder, Stephens, and Moseley⁷⁴ in a large study of factors influencing occlusion therapy outcomes in children. Flynn, Schiffman, Feuer, and Corona⁹ found that mixed amblyopia had a worse outcome when compared to anisometropic and strabismic amblyopia in their meta-analysis of occlusion therapy, but this effect was only marginally statistically significant. These results suggest that although the underlying causes might differ between anisometropic and strabismic amblyopia,⁷⁵ the capacity for change is similar across types. Note, however, that in our data, participants with microtropias or deviations small enough to allow for some degree of binocularity to exist (deviations < 15 PD) were more common than those with larger deviations that preclude binocularity. It remains to be investigated whether the improvements generalize to this latter group. 

Training Duration.

Training duration was a significant factor in normal adult vision as shown in perceptual learning experiments (e.g., Hussain et al.⁷⁶). Longer duration (up to a certain point⁷⁷) was also shown to result in greater visual acuity improvements with occlusion therapy in childhood (e.g., PEDIG⁸ and Keech et al.⁷⁷). However, our analyses showed that training duration was not a significant factor in visual acuity improvement or stereo sensitivity improvement regardless of treatment methods. Similar absence of the effect of training duration on visual acuity improvement was also reported by Levi and Li¹⁶ for perceptual learning. Another important question is whether there is a minimal duration required for amblyopia treatment. In children, most of the improvement with occlusion therapy occurs within the first 150 to 200 hours (which clinically corresponds to approximately 4 weeks duration if patching is done on a near full-time basis).⁷⁸ Figures 2 and 4 show that some adult participants can benefit greatly from just a few hours of behavioral training; however, to establish an average minimum training duration, improvement in visual outcomes for each method would have to be systematically monitored in a large-scale clinical trial. 

Initial Visual Acuity.

Depth of amblyopia was demonstrated to be a significant factor in the success of occlusion therapy in childhood.^9,74 We also found that initial visual acuity was a consistent statistically significant predictor of improvement in visual and stereo sensitivity. However, initial visual acuity had opposite effects on these two outcomes. Participants with more severe amblyopia showed larger improvements in visual acuity (see Fig. 2D). This finding was consistent across treatment types (see Supplementary Figs. S1–S3). It could be argued that for participants with milder visual acuity deficits, the magnitude of improvement is constrained by the upper limit on vision (a ceiling effect). Consequently, they show less improvement relative to participants with severe amblyopia, creating a reverse correlation between visual acuity and amblyopia depth. However, Figure 3D and Supplementary Figures S1 through S3 suggest that this is not the case. The solid black lines on the plots show the magnitude of improvement required for attaining 20/20 vision. For amblyopic vision of 0.17 logMAR and worse, which encompasses mild, moderate, and severe amblyopia, all data points fall on the right side of the solid black lines. The few participants with initial vision better than 0.17 logMAR cannot be driving the observed relation. These observations suggest that improvements in visual acuity are constrained by a limited capacity for change rather than the upper limit on vision, with more severe amblyopia having a greater capacity for change. 

Amblyopia depth had an opposite effect on stereo sensitivity. Participants with better initial visual acuity improved more in stereo sensitivity (see Fig. 4D; Supplementary Figs. S4, S5). Additionally, better initial stereo sensitivity resulted in larger stereo sensitivity improvements (discussed below). Thus, it is possible that better initial visual acuity is associated with better initial stereo sensitivity. Indeed, we found that participants with mild amblyopia had significantly better stereopsis. Another possibility is that stereo sensitivity is limited by the resolution of the amblyopic eye's image. Participants with severe amblyopia, even though they may have some binocularity, cannot achieve fine stereopsis because fine stereopsis requires a high-resolution image from each eye.^79,80 The image of the amblyopic eye could have low resolution either because of its falling outside the fovea of a deviated eye in strabismus, or because of blur in untreated or improperly corrected anisometropia. Thus, eye misalignment or anisometropia might have to be corrected to allow for better image resolution in more severe amblyopia before a substantial improvement in stereopsis can occur.⁸¹ In support of this hypothesis, it has been found that stereo acuity improvement correlates with visual acuity improvement in children.^82,83 

Initial Stereo Acuity.

Abnormal binocular interaction and interocular suppression were suggested as fundamental mechanisms of amblyopia by Burian and von Noorden more than four decades ago.⁸⁴ This assertion has been supported by abundant clinical evidence.^73,85–87 It is therefore plausible that participants with greater initial binocularity would demonstrate greater improvements in stereo acuity from treatment since the underlying binocular cortical networks are intact. Our IPD analysis showed that when all factors are considered, binocularity, as indicated solely by initial stereo sensitivity, does have the expected significant effect on stereo sensitivity improvement (Fig. 4F) but not on visual acuity (see Fig. 2F). It is not clear whether the limited improvement in binocularity in participants with worse stereo sensitivity occurred due to a limitation imposed by visual acuity or due to greater deficits in the underlying binocular mechanisms. 

Note, however, that we could evaluate binocularity only by the presence and degree of stereopsis measured by standard clinical tests used in the studies. Intact stereopsis is at the top of the hierarchy of possible degrees of binocularity. Even in the absence of stereopsis, some participants may have fusion, which is indicative of the presence of binocularity. Moreover, clinical tests of stereopsis are limited—they test primarily crossed disparities in central vision and employ coarse step sizes. Participants might have measurable stereopsis for uncrossed but not crossed disparities as these have been found to be dissociable,⁸⁸ or they might have gross stereopsis in the periphery only. 

Treatment Type.

Despite the diverse nature of the new behavioral treatment methods, we found no differences in treatment outcomes among methods. Although it has been suggested that monocular treatment might impair binocular function,¹⁷ we found that viewing condition had no impact on stereopsis outcome, as the new methods included both monocular (e.g., perceptual learning) and binocular viewing (e.g., dichoptic training). The lack of viewing condition specificity (monocular versus binocular) in treatment outcomes has been noted before in the clinical literature.⁸⁹ In addition, the separate IPD analyses of dichoptic training, perceptual learning, and video gaming did not reveal any substantial difference in visual outcomes among these treatments. These findings suggest that as long as the amblyopic eye is given a chance to work effectively either alone or together with the fellow eye, improvement ensues. Consequently, we propose that the uniform improvement in visual outcomes that occurred irrespective of tasks and viewing conditions could be the consequence of enhanced visual attention to the input from the amblyopic eye. 

Visual Attention and Amblyopia

Several psychophysical studies have documented deficits in visual attention in amblyopia during both monocular and binocular viewing.^90–92 Participants with amblyopia undercounted features presented to the amblyopic eye, which was not attributable to low-level deficits associated with amblyopia.⁹¹ Adults and children with amblyopia also had impaired multiple objects tracking, both in the amblyopic and in the fellow eyes.^92,93 In addition, attentional blink was less finely tuned temporally in the amblyopic eyes than in the fellow eyes in adults (Popple and Levi⁹⁰). Thiel and Sireteanu⁹⁴ found that participants with amblyopia showed line bisection deficits qualitatively similar to those found in unilateral neglect, a neural disorder of visual and spatial attention resulting from stroke.⁹⁵ 

A role for attention in strabismic amblyopia has also been proposed by Singer.⁹⁶ Using electrophysiological data from visual development experiments on kittens reared with various degrees of visual deprivation, he suggested that experience-dependent modifications of receptive fields in the visual cortex fail to occur in the absence of attention to the observed stimuli. Singer⁹⁶ further suggested that strabismic suppression occurs due to visual attention being directed away from the deviated eye, which eventually results in amblyopia. Human electrophysiology also supports the view that attentional processes might be involved in amblyopic deficits. Van Balen and Henkes⁹⁷ measured electroencephalogram responses over the occipital lobe of normal controls and participants with strabismic amblyopia during either attentive or nonattentive viewing. They found that the waveforms of attentive viewing in the amblyopic eye were similar to those of nonattentive viewing in the normal eyes. Based on these findings, they suggested that “squint amblyopia may very well be compared to a state of vision ‘without attention'” (p. 20).⁹⁷ 

How could visual attention relate to the amblyopic deficit? While the first locus of dysfunction in amblyopia occurs in V1,^75,98–100 a number of physiological and psychophysical studies suggest that there are also abnormalities in extrastriate and later specialized cortical areas.^98,101–104 It is possible that the suppression of one eye that leads to amblyopia is mediated by visual attention through signals coming from higher visual areas. In normal vision, attention suppresses unwanted information in both striate and extrastriate areas.^105–107 Moreover, directing attention to the dominant stimulus reduces the suppression of this stimulus during binocular rivalry.¹⁰⁸ Furthermore, attention modulates long-term plasticity in the early visual cortex in visually normal participants.¹⁰⁹ Thus, visual attention could act to suppress the amblyopic eye both transiently and in a sustained fashion. In a transient form, at the onset of the discordant visual input in childhood, interocular circuitry could be modified by recurrent attentional signals from higher areas until the amblyopic eye suppression is established. It is also possible that the suppressing effect of visual attention is sustained throughout the lifetime in addition to the permanent physiological changes in low-level cortical areas occurring early on in life. Forcing visual attention to favor the amblyopic eye, by training with behavioral tasks that require its input, might release some of these suppressive influences and possibly lead to reorganization of the suppression circuitry. In support of this proposal, it has been shown repeatedly that playing video games modifies several aspects of selective visual attention in normal vision.^110–112 

In addition to explaining the similar visual outcomes from diverse behavioral treatment methods, the attentional hypothesis could explain why the V1/V2 physiological deficits found in animal models of amblyopia are not large enough to account for the behavioral deficits.^75,113,114 Sustained attentional suppression could be acting in addition to physiological deficits in V1 to exacerbate the behavioral deficits. In addition, attention could account for the limits on improvement in visual acuity that we found. The new behavioral methods might address only one aspect of the attentional modulatory effect in amblyopia because the more permanent physiological changes might be less malleable in adults. 

Clinical Implications

Clinical Significance and Test–Retest Reliability.

From a clinical perspective, a statistically significant difference does not necessarily mean that a treatment effect is clinically significant. Clinical significance can be defined as treatment efficacy that leads to a change in patient management. Although many approaches have been proposed (e.g., expert opinion, patient-noticeable difference), no consensus has been reached regarding what constitutes clinical significance.¹¹⁵ In addition, clinical significance is often contextual. For example, while a one-line improvement in visual acuity may not be deemed “clinically significant” by most clinicians, a one-line improvement from 20/60 to 20/50 could allow amblyopic patients to drive a motor vehicle in some jurisdictions. 

A key factor to consider in assessing clinical significance is test–retest reliability. Because of test measurement variability, at a minimum, a treatment needs to exceed its test–retest reliability of the outcome measure to be considered effective. We found that the mean visual acuity improvement across studies was 0.17 logMAR with its 95% CI just above the mean test–retest reliability (0.15 logMAR) of the charts commonly used to assess visual acuity. However, only 4 of 21 studies exhibited a mean visual acuity improvement with 95% CI that exceeded test–retest reliability for the particular testing method used (see Fig. 1). Nevertheless, on average, 56% of participants exhibited visual acuity improvement by at least 0.15 logMAR and 32% by at least 0.2 logMAR across studies, suggesting that some patients may benefit from these new behavioral treatments. Interestingly, extensive occlusion therapy combined with refractive correction and near-vision activities^19,20 resulted in significantly greater improvements in visual acuity than perceptual learning, but not dichoptic training or video gaming. The similarity between new behavioral methods and gold-standard occlusion therapy once again implicates the potential role of an attentional component in amblyopia recovery. 

Assessing stereo sensitivity improvement is somewhat more complicated because test–retest reliability of clinical stereo tests is expressed in octaves. Consequently, improvement depends on the initial stereo sensitivity, and the final stereo sensitivity has to be four times the initial value to exceed the test–retest reliability. We found that, on average, 42% of all participants showed improvements equal to or greater than the test–retest reliability of 2 octaves. When we considered only those participants who had measurable stereopsis at the onset, we found that 34% of them improved by 2 octaves or more. 

Application in Clinical Practice.

Our analysis suggests that these new behavioral methods could be used in the rehabilitation of amblyopia in adults. These treatments appear to be especially beneficial for individuals with more severe amblyopia, as these participants showed the greatest improvement in visual acuity. Individuals with mild amblyopia could also benefit as they showed the greatest improvement in stereopsis. The specific treatment method does not appear to matter, but from considerations of time efficiency and compliance, treatments that require shorter durations (e.g., dichoptic training) would seem to be a better choice. It should be noted that none of the discussed treatments have been standardized, and large clinical control trials have not been conducted yet. Until such trials are completed, the treatments should be used with caution. One potential clinical concern is the risk of intractable diplopia, which could be induced by reducing interocular suppression in the absence of fusion.¹¹⁶ While reducing interocular suppression may result in improved visual acuity, without intact binocular circuitry, certain patients (especially those with early-onset large-angle strabismus) would be at risk of developing diplopia that cannot be ameliorated by traditional means such as prisms and/or strabismus surgery. Dichoptic treatment was said not to induce diplopia,²⁴ while diplopia has been reported to occur after occlusion therapy in adults.¹⁹ This concern was not addressed or acknowledged in the majority of papers we reviewed here. 

Future Directions

Areas of Future Research.

This review revealed several promising avenues for future research. First, we showed that participants with more severe amblyopia exhibit greater responsiveness to treatment. We demonstrated that this effect is not an artifact of the upper limit on visual acuity. Instead, it is possible that it is associated with the suppression mechanisms in amblyopia. More severe amblyopia is associated with greater inhibition of the amblyopic input⁷³; thus, there may be more room for reduction of suppression in the more severe cases. This hypothesis can be tested by measuring amblyopic suppression before and after treatment and correlating it with the observed improvement in visual acuity. Hess et al.²³ provided some evidence for the correlation between the degree of suppression and improvement in visual acuity. They measured the contrast required to equally engage the fellow and the amblyopic eyes perceptually. The degree of reduction in contrast difference between the two eyes (believed to be evidence of reduced suppression) was positively correlated with improvement in visual acuity. However, Hess et al.²³ used a nonclinical measure of suppression; it would be interesting to see if similar results could be observed with conventional clinical tests of suppression (e.g., Bagolini red filter bar). 

Our analysis showed that training duration was not a significant factor in amblyopia recovery. In most studies except a few using perceptual learning, for example,²⁹ the duration of training was determined by considerations other than the time needed to reach optimal improvement (e.g., participant availability). To evaluate further the effect of training duration, as well as the minimum required duration, it would be necessary to monitor treatment progress over longer periods to allow participants to reach their asymptotic level. 

We found that initial stereo sensitivity was not a significant factor in improvement of visual acuity. We selected stereo sensitivity as a measure of binocularity because in the majority of the papers we discussed, binocularity was assessed only with clinical stereo acuity tests. It is possible that when assessed with a battery of tests for binocular vision (as suggested in the following section) and with each participant given a score based on the cumulative test results, binocularity might be revealed as a significant predictor of plasticity in adult amblyopia. A study could be conducted in which binocularity is measured prior to and following treatment with a large battery of tests, and then improvements in acuity and binocularity could be correlated with initial binocularity level. Such a study could also determine the importance of intact binocular circuitry on the ability to regain balanced vision by comparing participants lacking clinically measureable binocularity (e.g., fusion) to those with various degrees of binocularity. 

We found that treatment type had only a small differential effect on visual function improvement. This is somewhat surprising, given the large differences among treatment methods, such as monocular versus dichoptic viewing conditions, as well as the different nature of the tasks. It would be useful to confirm this result with a study directly comparing treatments administered monocularly, binocularly, and dichoptically, using a video game, for example, for stimulus delivery. 

Finally, given the small influence of treatment type, we suggested that directing visual attention to the input of the amblyopic eye might play an important role in visual acuity and stereopsis improvements. Future investigations should assess the role of visual attention in plasticity in amblyopia. Comparing performance on attention-related tasks before and after treatment could reveal a correlation between improvements in task performance with improvements in visual acuity. One rehabilitation study that used such high-level assessment showed that visual counting in participants with amblyopia improved after perceptual learning.³³ The attentional component of amblyopia deficits can also be tested by examining the correlation between the depth of amblyopia and performance on visual attention tasks (e.g., attentional blink, counting, visual search). A positive relation would provide evidence that attention modulates suppression of the amblyopic input. 

Methodological Recommendations.

Based on the analysis presented in this paper, we have formulated several methodological recommendations for future studies, which will make it easier to assess and compare various interventions. 

First, the clinical characteristics that are important to report in a study include visual acuity and stereo acuity before and after treatment, the criteria used for the classification of amblyopia types (preferably employing standardized criteria), refractive errors, eye deviation before and after treatment (to monitor for any eye alignment changes), and individual training durations. 

Second, we recommend using several tests for assessment of binocularity before and after treatment (e.g., Hess, Babu, Clavagnier, Black, Bobier, and Thompson³⁷). Fusion can be measured with Bagolini lenses and the Worth-4-dot test. If fusion is not found in participants with strabismus, prismatic lens correction can be administered to compensate for their deviation to verify whether fusion can be achieved after optical alignment of the eyes. A synoptophore can be used to assess the potential for fusion and stereopsis with proper eye alignment in participants with strabismus (thus demonstrating that these individuals indeed have an intact binocular circuitry, and that they simply cannot access it in the presence of their strabismic deviations). Stereopsis should be assessed by clinical and nonclinical tests, which could be implemented on a stesreoscopic display using random-dot stimuli and contour stereograms. These tests could include assessment of both crossed and uncrossed disparities and could employ a staircase procedure to measure stereo acuity with greater precision. 

Third, we suggest performing refractive correction adaptation as the first step in any treatment protocol. In several studies using occlusion therapy,^19,20 full refractive correction was administered to some of the participants immediately prior to patching treatment. Refractive correction is known to induce improvement in amblyopia on its own.^10,74 Thus, it is imperative to let the participants adapt to refractive correction first for 3 to 4 months, and then measure their visual acuity again before beginning any treatment to exclude the effect of refractive adaptation on amblyopia. 

Finally, in addition to reporting statistically significant differences in treatment outcomes before and after treatment, it is important to report the proportion of participants showing a mean improvement in the outcome measures (and 95% CI) that exceed the test–retest reliability of the particular tests used. This will provide clinicians the essential information to assess whether a particular treatment method is clinically relevant, and also what participant characteristics are pertinent in predicting clinical effectiveness. It is also important to include information about compliance with these new methods (as was done by Hess, Babu, Clavagnier, Black, Bobier, and Thompson³⁷), especially for studies done in children, and the numbers of participants who have been excluded from the studies due to their inability to perform the necessary experimental tasks even when they are presented at the easiest levels. 

Acknowledgments

Supported by a Natural Sciences and Engineering Research Council fellowship (IT) and Grant MOP 106663 from the Canadian Institutes of Health Research (CIHR), Leaders Opportunity Fund from the Canada Foundation for Innovation (CFI), Brandan's Eye Research Foundation, the John and Melinda Thompson Endowment Fund in Vision Neurosciences, and the Department of Ophthalmology and Vision Sciences at The Hospital for Sick Children (AMFW). The authors alone are responsible for the content and writing of the paper. 

Disclosure: I. Tsirlin, None; L. Colpa, None; H.C. Goltz, None; A.M.F. Wong, None