Designing clinical trials for amblyopia

Abstract

Randomized clinical trial (RCT) study design leads to one of the highest levels of evidence, and is a preferred study design over cohort studies, because randomization reduces bias and maximizes the chance that even unknown confounding factors will be balanced between treatment groups. Recent randomized clinical trials and observational studies in amblyopia can be taken together to formulate an evidence-based approach to amblyopia treatment, which is presented in this review. When designing future clinical studies of amblyopia treatment, issues such as regression to the mean, sample size and trial duration must be considered, since each may impact study results and conclusions.

The need for randomized clinical trials

Randomized clinical trials (RCTs) are considered one of the highest levels of evidence, above cohort studies, case series, and case reports, primarily because the process of randomization minimizes bias, increasing the probability that potentially confounding factors would be equally distributed between treatment groups. These potentially confounding factors include age, gender, race, and severity of disease, and in studies of amblyopia, specifically include baseline refractive error, amblyopia subtype (e.g. anisometropic, strabismic, and combined), and previous treatment.

One illustration of the importance of randomization is to consider the hypothetical situation where one is the treating eye care provider participating in a non-randomized study comparing treatment versus control. It is hard to remain truly dispassionate about the decision to assign each patient to either treatment or control. If the patient’s condition is on the severe end of the spectrum, the temptation may be to offer the active treatment rather than the control, whereas if the condition is mild, the temptation may be to offer the control rather than the active treatment. Randomization eliminates this potential bias, maximizing the chance that even unknown confounding factors will also be balanced between treatment groups.

Nevertheless, randomization alone is not sufficient to eliminate all forms of bias. Allocation concealment is important, preventing the investigator from knowing to which group the next patient will be assigned, otherwise the investigator might not offer participation to the next patient for the same reasons that the investigator might assign treatment to one group or the other. Masking of outcome assessment is also preferable, preventing the examiner from knowing the treatment group, so that the examiner does not consciously or subconsciously influence the result of testing. In addition, appropriate sample size is important to reduce the chance of concluding there is no effect (based on study results) when in fact there is an effect (a type II error).

Despite the preference for randomized clinical trials, there is a clear role for non-randomized studies. Preliminary data are needed to obtain an estimate of effect in both the proposed treatment group and the control group. In addition, for some treatments, randomization may not be acceptable to investigators, to patients, or to parents. When conducting a non-randomized cohort study, it is important to have clearly defined inclusion and exclusion criteria, a clear protocol with a defined follow-up schedule, a pre-established primary outcome measure, and standardization of outcome assessment. A control group should also be considered in a non-randomized study design, either concurrently or historically, recognizing that the same objections may exist for a concurrent control group in cohort studies as for RCTs, specifically investigator, patient and/or parent unwillingness to be assigned to “control.”

Recent evidence from amblyopia treatment studies

The above principles of study design have been applied to clinical amblyopia studies conducted by the Pediatric Eye Disease Investigator Group (PEDIG), a multicenter network of over 200 pediatric ophthalmologists and pediatric optometrists across North America, from private practices and academic institutions.¹ The network is funded by the National Institutes of Health to conduct large simple RCTs and observational studies. Typically 50-80 investigators participate in each study and all studies are approved by the relevant Institutional Review Boards, obtaining appropriate informed consent from participants and parents.

Regarding amblyopia treatment studies, all PEDIG amblyopia studies thus far pertain to anisometropic, strabismic, or combined amblyopia, not deprivation amblyopia. One hallmark of PEDIG amblyopia studies is the standardization of visual acuity outcome assessment, presenting single surrounded optotypes at logMAR intervals, by using the Amblyopia Treatment Study (ATS) HOTV visual acuity protocol (for 3 to <7 year olds),² or by using the eETDRS protocol (for 7 year-olds and older),³ both presented on the Electronic Visual Acuity tester.⁴ Importantly, the ATS HOTV visual acuity protocol can be performed as a matching task, which increases testability among younger children, and, typically, visual acuity assessment is performed by masked examiners who do not know the subject’s treatment assignment. The way in which PEDIG amblyopia studies have profoundly changed the treatment of amblyopia is described in the following sections.

Observational studies of refractive correction alone

In a non-randomized prospective observational study, 84 children 3 to <7 years old with previously untreated anisometropic amblyopia were studied (visual acuity between 20/40 to 20/250).⁵ Optimal refractive correction was provided and visual acuity was measured with the new spectacle correction at baseline, confirming the presence of amblyopia and then measured at 5-week intervals until visual acuity stabilized or amblyopia resolved.

Amblyopia improved with optical correction by 2 or more lines in 77% of the subjects and resolved in 27%.⁵ Improvement took up to 30 weeks before stabilization. Mean improvement was 2.9 lines. Even after apparent stabilization, additional improvement occurred with spectacles alone in 21 (62%) of 34 subjects⁵ followed as a control group in a subsequent randomized trial,⁶ and amblyopia resolved in 6 of those subjects.^5,6

In a subsequent prospective observational study⁷ of 146 children 3 to <7 years old with previously untreated strabismic amblyopia (n=52) or combined amblyopia (n=94), optical treatment alone was provided as spectacles based on a cycloplegic refraction (allowing plus sphere to be cut by up to +0.50D). At 18 weeks, amblyopic eye visual acuity improved a mean of 2.6 lines, with 75% of children improving ≥2 lines. Resolution of amblyopia occurred in 32% of the children and visual acuity improved regardless of whether eye alignment improved.⁷

Independent of PEDIG, Moseley, Stewart, Fielder et al^8-10 and Clarke et al¹¹ have also provided evidence that marked improvement in amblyopic eye visual acuity can be obtained with spectacles alone, in both strabismic amblyopia and anisometropic amblyopia. In many cases, treatment with spectacles alone eliminated the need for patching or atropine. Taken together, the practical implication of these studies is that it is very reasonable to start with spectacles first in the management of anisometropic, strabismic, and combined amblyopia. If spectacles alone are insufficient treatment, then there are three primary subsequent options; patching, atropine, and Bangerter filters, discussed in the next sections.

Studies of patching dose

If patching is chosen as the next step, then the eye care provider must decide what dose of patching to prescribe. In a patching dose study, prescribed full-time patching (all or all but 1 waking hour a day) was compared with prescribed 6 hours of daily patching in 175 children, 3 to <7 years old, with severe amblyopia (best-corrected visual acuity 20/100 to 20/400).¹² Both groups were also prescribed at least 1 hour of near visual activities during patching. Visual acuity in the amblyopic eye improved by a similar amount in both groups at the 17-week primary outcome exam, averaging 4.8 lines in the 6-hour group and 4.7 lines in the full-time group (P= 0.45).

In a parallel patching dose study in moderate amblyopia (best-corrected visual acuity 20/40 to 20/80), 189 children were randomized to either prescribed 6 hours of daily patching or prescribed 2 hours of daily patching.¹³ Both groups were also instructed to perform at least 1 hour of near visual activities during patching. Visual acuity improvement of the amblyopic eye was similar in each group at the 17-week outcome visit, averaging 2.4 lines.¹³

These patching dose studies concluded that, in children 3 to <7 years of age, 6 hours of prescribed daily patching produces an improvement in visual acuity that is of similar magnitude to prescribed full-time patching in severe amblyopia, and 2 hours of prescribed daily patching produced improvement in visual acuity of similar magnitude to prescribed 6 hours of daily patching in moderate amblyopia. It was also noteworthy that there was no difference in the rate of improvement between different doses of patching.^{12, 13}

Independent of PEDIG, Stewart et al.¹⁴ also conducted an RCT comparing 12 hours per day with 6 hours per day of patching and used an occlusion dose monitor to measure the actual wearing time. They found that actual patching time was similar between groups (4.2 hours vs. 6.2 hours, P = 0.06), suggesting that one reason for the lack of superiority of more intense regimens might be reduced compliance with intense patching, and the fact that children and families find it very difficult to wear a patch for many hours a day. Nevertheless, the study of Stewart et al¹⁴ confirmed that many children who actually wore the patch for only 2 hours per day do in fact respond. What remains unexplained is why a few children who actually wore the patch for 12 hours a day showed very little response, and further work is needed to identify such children earlier in their treatment course and to develop new treatments that might address this type of resistant amblyopia.

In a subsequent PEDIG RCT of prescribed 2 hours a day of patching with near activities versus 2 hours a day with distance activities,¹⁵ no difference was found in treatment effect between types of activities performed when patched. Nevertheless, some children with severe amblyopia (20/100 to 20/400) responded to 2 hours a day of prescribed patching, and so even in severe amblyopia a dose of 2 hours a day is a reasonable option. Based on these RCTs of patching dose, if visual acuity fails to improve completely with spectacles alone, it is reasonable to initiate prescription of 2 hours of daily patching for all children with anisometropic, strabismic, or combined amblyopia, regardless of severity of amblyopia.

Atropine versus Patching

In the first RCT conducted by PEDIG, ^{16, 17} atropine 1% (one drop each morning to the fellow eye) was compared with patching of the fellow eye (prescribed for at least 6 hours per day).^{16, 17} Children included in this study had moderate amblyopia (20/40-20/100) and were <7 years old at enrollment, needing to be able to complete optotype visual acuity testing (single surrounded HOTV matching using the ATS-HOTV VA protocol²), effectively limiting the study to 3- to <7-year olds. Optimum spectacle correction was required for at least 4 weeks and no more than 2 months of amblyopia therapy in the past 2 years was allowed. In addition, the protocol specified that, if the amblyopic eye had not improved by three lines or to at least 20/30 after 16 weeks of randomized treatment, the intensity of treatment was increased by either changing the spectacle lens over the fellow eye to plano in the atropine group or increasing patching to 12 or more hours per day in the patching group.¹⁶

At the 6-month primary outcome, both groups showed similar improvement in amblyopic eye visual acuity (a mean improvement of 3.16 lines in the patching group and 2.84 lines in the atropine group).¹⁶ The difference in visual acuity between treatment groups was small (1.7 letters), and not clinically meaningful (mean difference 0.034 logMAR, 95% CI, 0.005 to 0.064).¹⁶ Improvement was initially faster in the patching group, with a mean improvement from baseline to five weeks of 2.22 lines in the patching group and 1.37 lines in the atropine group.¹⁶ The relative treatment effect did not differ by age, depth of amblyopia, or cause of amblyopia.¹⁶

Between 6 months and 2 years following randomization treatment was at the discretion of the investigator,¹⁷ but only about a quarter of the children underwent treatment using the alternative treatment (atropine to patching, or vice versa). ¹⁷ At the 2-year outcome, the mean improvement in amblyopic eye visual acuity was similar in the patching and atropine groups (3.7 lines in the patching group and 3.6 lines in the atropine group).¹⁷ It is noteworthy that only about half of the subjects in each group reached 20/25 or better in the amblyopic eye.¹⁷

At age 10 years, a proportion of the originally enrolled children were brought back for re-examination.¹⁸ For logistical reasons, only children at the higher enrolling sites (>5 subjects) were included in this phase of the study, but there was no selection of children within each site. Mean amblyopic and fellow eye visual acuities at age 10 years were similar in the original treatment groups,¹⁸ and the mean amblyopic eye acuity overall (n=169) was 0.17 logMAR (approximately 20/32) and 46% of amblyopic eyes were 20/25 or better.¹⁸

At age 15 years, the mean amblyopic eye visual acuity, measured in 147 subjects, was 0.14 logMAR (approximately 20/25) and was again similar in the original treatment groups.¹⁹ Sixty percent of amblyopic eyes had visual acuity of 20/25 or better. Interestingly, better visual acuity at the 15-year examination was achieved in those who were younger than 5 years at the time of entry into the randomized clinical trial (mean logMAR, 0.09) compared with those aged 5 to 6 years (mean logMAR 0.18; P < 0.001). ¹⁹

Although the first PEDIG study used a daily dose of atropine,¹⁶ PEDIG subsequently studied less frequent administration of atropine drops. In a study comparing weekend atropine (2 days a week) with daily atropine, there was essentially identical improvement of amblyopic eye visual acuity (2.3 logMAR lines at 4 months).²⁰

Regarding augmenting weekend atropine with a plano lens (increasing the blur induced by atropine by cutting the hyperopic correction over the fellow eye), there was no substantial improvement in amblyopic eye acuity using the plano lens when compared with weekend atropine alone.²¹ Nevertheless, more subjects in the atropine plus plano lens group had reduced fellow eye acuity at 18 weeks, although there were no cases of persistent reverse amblyopia.²¹ Because some cases of reverse amblyopia required treatment, care should be taken when adding a plano lens to atropine.

Considering atropine for severe amblyopia, two PEDIG RCTs^21-23 included children treated with atropine for severe amblyopia (20/125 to 20/400) and, although these studies were not powered to specifically address severe amblyopia, some interesting observations can be made. Sixty children 3 to 6 years of age with severe amblyopia were randomized to weekend atropine plus a plano lens or weekend atropine with full spectacle correction.²¹ Forty children 7 to 12 years of age with severe amblyopia were randomized to weekend atropine or two hours of daily patching.²² Considering the effect of atropine on severe amblyopia across these groups, in the younger cohort,²¹ visual acuity improved by an average of 4.5 lines in the atropine alone group and 5.1 lines in the atropine plus plano lens group.²³ In the older cohort with severe amblyopia,²² visual acuity improved by an average of 1.5 lines in the atropine group versus 1.8 lines in the patching group.²³ Taken together, these studies^21-23 provide evidence that atropine can improve visual acuity in children with severe amblyopia (3 to 12 years of age), and improvement may be greater in younger children.

Regarding yet a third modality of amblyopia treatment, a further PEDIG RCT compared 2 hours of daily patching to a Bangerter filter on the spectacle lens over the sound eye,²⁴ where the Bangerter filter fogs the vision of the sound eye. The study found similar improvement in amblyopic eye visual acuity between patching and Bangerter filters (2.3 lines in the patching group and 1.9 lines in the Bangerter group at 24 weeks).²⁴

Summarizing our current knowledge on treatment after spectacle correction, 2 hours of daily patching, weekend atropine administered to the fellow eye, and a Bangerter fogging filter are excellent subsequent treatments for moderate and severe anisometropic, strabismic, and combined amblyopia. It is reasonable to involve the parents and the child in deciding which treatment to implement. If that treatment modality is unsuccessful, the child could be prescribed alternative therapy or the treatment increased as discussed below.

Treating older children for amblyopia

In a PEDIG randomized trial investigating the effectiveness of treating amblyopia in older children (7 to 17 year-olds), 507 children with amblyopic eye acuity 20/40 to 20/400, were treated with either optical correction alone or optical correction augmented with patching of the fellow eye 2 to 6 hours a day, with 1 hour of near visual activities (404 children aged 7 to 12 years and 103 aged 13 to 17 years).²⁵ Younger subjects in the augmented treatment group (age 7 to 12 years) were also prescribed one drop of 1% atropine daily, but subjects aged 13 to 17 years were not prescribed atropine due to the demands of their activities.²⁵ “Optical correction alone” was chosen as the control in this PEDIG study for older children because the study was performed prior to the PEDIG observational studies of optical treatment^5,7 which subsequently demonstrated the effectiveness of such optical treatment.^5,7

In the RCT for older children,²⁵ the primary outcome was defined as the proportion of subjects in each group classified as a responder, defined as; amblyopic eye acuity improvement 10 or more letters (two logMAR lines) better than the baseline acuity, by the 24-week visit. The patient could also be classified as a non-responder at an earlier visit if there was no improvement from the prior visit or only a pre-defined minimal improvement from baseline.

In the 7- to 12-year olds, 53% of the augmented treatment group (optical plus patching plus atropine) were responders compared with 25% in the optical only group (P <0.001).²⁵ In the 13- to 17-year olds, there was no difference in the proportion of children who met the responder criteria; 25% in the augmented treatment group (optical plus patching) compared with 23% in the optical only group. Nevertheless, among older subjects who had not been previously treated for amblyopia, those in the augmented treatment group did show a greater improvement than those in the optical only group (47% versus 20%, P= 0.03).²⁵ No patient developed constant diplopia during the randomized trial.²⁵

In this RCT for older children with amblyopia,²⁵ there was a great deal of individual variability in response. It is noteworthy that even some teenagers showed a robust response to treatment, and therefore such data support offering a trial of treatment to children up to 17 years of age, particularly if age 7 to 13 years of age or if they had received no previous treatment.

When improvement in visual acuity plateaus with patching

After treatment with refractive correction and patching, some children still have residual amblyopia. PEDIG conducted an RCT to evaluate the effectiveness of increasing prescribed daily patching from 2 to 6 hours in 169 children with stable residual amblyopia (20/32-20/160) after 2 hours of daily patching for at least 12 weeks.²⁶ Assessed at 10 weeks, amblyopic eye visual acuity improved an average of 1.2 lines in the 6-hour group and 0.5 line in the 2-hour group (P = 0.002). Based on these results, when amblyopic eye visual acuity stops improving with 2 hours of daily patching, increasing the daily patching dosage to 6 hours is a reasonable next step.

PEDIG also studied the role of weaning versus not weaning treatment at the end of a therapeutic course.²⁷ Children were followed prospectively after cessation of part-time patching or atropine treatment and their treatment was classified retrospectively as to whether it was weaned or not. Recurrence of amblyopia defined as a 2–logMAR level reduction of visual acuity from enrollment (cessation of patching) confirmed by a second examination. Recurrence was also considered to have occurred if treatment was restarted with a 2–logMAR level reduction of visual acuity, even if it was not confirmed by a second examination. Weaning treatment following cessation, was associated with a reduced rate of amblyopia recurrence when the treatment was intense (6 hours per day), whereas weaning did not seem to provide an advantage when treatment was less intense (2 hours per day).²⁷

Current PEDIG RCTs in amblyopia

PEDIG is currently completing a study of patching 2 hours per day with levodopa versus patching 2 hours per day with placebo, and another study of adding a plano lens to atropine treatment in children whose amblyopia eye visual acuity stops improving when using atropine alone in the fellow eye. Results of these two studies are expected to be published in early 2015. Finally, PEDIG has just launched an RCT of binocular treatment versus 2 hours of daily patching with results expected to be published in 2016 or 2017. The development of binocular treatment is described by Robert Hess Ph.D. and Eileen Birch Ph.D in accompanying articles in this issue.

Based on the summarized PEDIG studies, an evidence-based treatment plan for anisometropic, strabismic, and combined amblyopia can be formulated (Figure 1). This suggested evidence-based approach may be helpful for eye care providers who treat children with amblyopia.

Figure 1.

Suggested evidence-based approach to treating amblyopia.

Potential problems in clinical study design: Variability

Every biological measure has intrinsic variability, and in the context of amblyopia, it is important to account for variability of visual acuity. For an individual patient, it can be challenging to determine whether the amblyopia has really improved or worsened.²⁸ For example, when using optotype visual acuity as an outcome measure, if the visual acuity is measured as 20/40 (0.3 logMAR, 70 letters ETDRS) on one day and then 20/50 (0.4 logMAR, 65 letters ETDRS) on a subsequent day, it is difficult to know whether the amblyopia has indeed worsened, or has remained the same. One approach to determining whether an observed change is a “real change” is to define a “real change” as exceeding test-retest variability. Using test-retest data at the individual patient level, repeatability coefficients or 95% limits of agreement can be calculated and those definitions can be used to represent real change for an individual patient. For HOTV isolated surrounded optotype testing using the ATS HOTV visual acuity protocol, such a real change exceeding test-retest variability is 2 logMAR levels² and for eETDRS, a real change is 6.5 letters.²⁹

Nevertheless, cohort studies often report mean change of a group, and changes in the mean value may be prone to the phenomenon of “regression to the mean,” which may give the impression that the mean value has improved when, in fact, it has not. The phenomenon of “regression to the mean” is particularly evident when inclusion criteria truncate the study population, which often occurs in amblyopia studies because only subjects whose visual acuity is worse than a certain threshold are included. As illustrated in Figure 2, a patient is only enrolled in such a study if the visual acuity is worse than the threshold for inclusion, so the patient is not enrolled if visual acuity measured better than that threshold. On a subsequent day, visual acuity may measure worse (by test-retest variability) and may then meet enrollment criteria, so the patient is enrolled with that worse measurement of visual acuity. At the next measurement, the visual acuity is likely to return toward its mean value for that patient, due to test-retest variability, the phenomenon called “regression to the mean,” which will give the impression that the visual acuity has improved when, in fact, it has not. (Figure 2)

Figure 2.

Illustration of "regression to mean." Example patients with stable visual acuity, whose only changes in measured visual acuity over time are due to test-retest variability. Patient A was enrolled at Visit 1 because visual acuity met enrollment criteria, but improved by the next visit due to test-retest variability (and "regression to the mean") and still appeared improved at the outcome visit. In contrast, patient B was not enrolled at Visit 1 or Visit 2 because visual acuity did not meet enrollment criteria, but at Visit 3, visual acuity measured worse than the enrollment criteria (due to test-retest varability) and so the patient was enrolled. The visual acuity then measured better at Visit 4, and at the outcome visit, due to test-retest variability and regression to the mean. Both subjects appeared to have improved visual acuity at the outcome visit, compared with enrollment, but these “improvements” were only due to regression to the mean.

Some study design characteristics can increase or decrease the confounding effect of regression to mean. If possible, truncating the population by the parameter used as the outcome measure should be avoided, but this is difficult in amblyopia studies when defining amblyopia as a visual acuity deficit as worse than a particular threshold. One clear advantage of randomized clinical trials is that any “regression to mean” effect should occur equally in both groups, and so any observed difference between treatment groups cannot be due to regression to the mean.

As the PEDIG observational studies are reviewed, for example the study of spectacles alone for anisometropic amblyopia,⁵ it needs to be acknowledged that there was some risk of “regression to the mean,” because subjects were only enrolled if they met a threshold visual acuity deficit (20/40 or worse) and there was no control group. Nevertheless, such a large magnitude of improvement, mean 2.9 ± 1.8 logMAR lines and 27% resolution (within one line of interocular difference)⁵ cannot be entirely attributed to regression to the mean. In addition, in some studies it may be possible to statistically model the “regression to mean” effect from test-retest data, and use those data to determine how likely an observed effect might be due to such regression to the mean.

Potential problems in clinical study design: Inadequate sample size

An additional way that variability of the outcome measure influences study design is its implication for sample size. If sample size in inadequate, there is a risk of a Type II error, the risk of finding no statistically significant difference between experimental groups when there was in fact a difference between the treatments. In this context, it is important to consider the minimally clinically important difference, the difference that, if found, would influence clinical practice. This value is not intuitively obvious and much future work is needed in the amblyopia field to further define and justify a minimally clinically important difference when expressed as a mean. It is easier to consider what proportion of subjects would need to move from one visual acuity bin (category or group) to another, and then calculate what that effect would have on the mean. These calculations should be performed before embarking on a study. A mean difference of 0.75 logMAR lines or 1.0 logMAR lines between treatment groups would be considered a minimally clinically important difference by many clinicians but smaller mean differences may be important to detect. In the context of treatment versus control, if most of the treated cohort only improves a small amount, for example less than a logMAR line, then the population mean will not change markedly, and some clinicians might argue that they would be unimpressed with that treatment. Whereas if only a small proportion of the treated population improve a large amount and the remainder do not improve at all, the population mean would also not change markedly and some clinicians might argue that it is not worthwhile to treat so many for a few who might improve. Such issues need to be considered and discussed during the design of every study.

It is also important to evaluate the 95% confidence intervals around a mean difference and how the ends of those intervals relate to the minimally clinically important difference. The 95% confidence intervals can be pragmatically considered a range of values consistent with your data, i.e., a range of plausible values. If the upper end of the 95% confidence interval includes values that would be important to detect (e.g., a non-significant difference of 0.18 logMAR with 95% CI −0.01 to 0.36) the study may have missed a very large mean effect of 3.6 logMAR lines, and this example study would be seriously underpowered, such that reasonable conclusions cannot be made.

Designing an underpowered study can be avoided by calculating the needed sample size based on variability of response in the outcome measure. That variability of response can be estimated from pilot data, and such estimation is one important rationale for performing pilot studies. A reasonable estimate of the difference between groups is also needed and can be based on pilot data. There also needs to be a pre-study agreement that finding such an effect would in fact influence practice.

One related problem with variability and sample size is underpowering a study by including a broader cohort that does not respond the same way to the treatment as your original pilot cohort. If one bases sample size calculations on pilot data from a cohort with specific characteristic (e.g., untreated amblyopia) and then decide to include subjects who have different characteristics (e.g., previously treated amblyopia) and who respond differently to treatment, the variability of the outcome measure may be higher in the planned study than observed in the pilot cohort. Higher variability in the outcome measure would require a larger sample size in the planned study to detect the same treatment effect. If attention is not paid to this issue of variance of the outcome measure, which depends on the specific cohort to be enrolled, then the study may be underpowered.

One option in reducing variability of an outcome measure is to choose an outcome measure with less variability, but this is challenging in amblyopia studies because optotype visual acuity is the accepted gold standard. One way to reduce variability is to use repeat measures and take an average, for example, a mean of two measures is better than one, and a mean of three measures is better than two. Nevertheless, there are practical barriers to obtaining multiple measures of visual acuity in young children, specifically inattention, boredom, and fatigue.

Potential problems in clinical study design: Effect of trial duration

In interpreting the results of any RCT and in designing new RCTs, there needs to be an awareness of the potential bias of choosing a particular trial duration. In PEDIG’s first amblyopia treatment trial,¹⁶ children with moderate amblyopia were randomized to occlusion (at least 6 hours/day) versus atropine (1 drop every morning). Masked assessment of visual acuity at 6 months revealed very similar improvements in visual acuity. But if the trial had been designed as a 5-week study, very different conclusions would have been made, because the mean improvement with patching at 5 weeks was 2.22 lines, and only 1.37 lines with atropine, a difference of 0.87 lines (95% CI 0.60 – 1.13).¹⁶ If the overall trial had been a short 5-week study, the conclusions would have been that patching was superior to atropine. Therefore, in the trial design process, it was important that atropine advocates insisted on a longer trial duration to adequately represent the treatment effect of atropine.

In the recently launched PEDIG RCT to compare binocular treatment of amblyopia with conventional monocular patching treatment, the decision regarding study duration was a source of controversy. Current pilot data are mostly limited to only 4 weeks of game play,³⁰ but it may not be reasonable to compare binocular treatment to patching at 4 weeks in the full randomized clinical trial, because patching treatment may take longer to be effective and assessing the primary outcome at 4 weeks may bias against patching. Conversely, it may not be reasonable to compare outcomes at 6 months, because other pilot studies of binocular treatment suggested children will not play the same binocular game every day for many months, and therefore assessing the primary outcome at 6 months may bias the study results against binocular treatment. An additional consideration is the possible regression of treatment effect following cessation of therapy, which may be different between treatments. The timing of outcome assessment may have a profound effect on study results, conclusions, and interpretation.

Regarding trial duration and the timing of outcome measure assessment, in the design phase of a clinical trial, it is important to obtain sufficient pilot data on time-course of treatment effect. It is also important to ensure advocates for each treatment provide input during the study planning process, for example, both binocular treatment advocates and patching advocates when designing a binocular treatment versus patching study.

Conclusions

Our current treatment of amblyopia should be influenced by recent RCTs and cohort studies, which lead to a suggestion for an evidence-based algorithm for treating amblyopia (Figure 1). RCTs are the preferred design because they minimize bias, but many factors in clinical study design influence outcome and conclusions. It is important to recognize the risk of regression to mean in cohort studies and account for variability of the chosen outcome measure. It is also important to recognize that specific features of study design, for example, timing of outcome assessment, may bias the study results toward favoring one treatment or the other, and these issues should be considered as future studies are designed.

Highlights.

• Randomized clinical trials are preferred in clinical amblyopia research

• Nevertheless, observational cohort studies have an important role

• Studies in amblyopia can be synthesized into an evidence-based treatment algorithm

• Future study design should consider issues that may affect study outcome

• Issues include regression to the mean, sample size, variability, and trial duration