There is currently debate in the ophthalmology community about whether treatment with topical glaucoma agents should be initiated in one eye or in both eyes [7, 8, 10, 12–14]. Classically, it was felt that a more reliable estimate of drug effect could be obtained with a uniocular trial. The IOP change in the untreated eye was subtracted from the change in the treated eye, to account for correlated variation between the eyes.

This retrospective review of initiation of topical glaucoma therapy has demonstrated that: 1) The traditional teaching that for uniocular trials the IOP change in the untreated eye should be subtracted from that of the treated eye to control for diurnal and other correlated fluctuations does not universally apply. 2) Uniocular and binocular trials have similar predictive value when interpreted correctly, and either may be selected based on clinical circumstances.

The reference standard for determining drug effectiveness in an individual patient is to randomly alternate administration of the drug or a placebo in a double-masked fashion over multiple visits at which the outcome variable is assessed (an "n-of-1 trial") [17, 18]. With respect to glaucoma, drug or placebo drops would be randomly administered and IOP would be recorded. This technique is not currently standard in the treatment of glaucoma. In this study, we assumed that the average IOP on treatment, following any uniocular or binocular medication trials, minus the pre-treatment average IOP, is the best available estimate of the true drug effect [11]. Because the pre-treatment average IOP is known at the time of the trial, the only unknown which must be estimated to know the drug effect is therefore average follow-up IOP. This follow-up IOP was therefore the primary dependent variable in the main regression analyses.

Some authors have suggested that the uniocular drug trial must be abandoned because of crossover drug effect [6], because the contralateral eye IOP change during a single visit may not be used as a control due to uncorrelated fluctuations (unrelated to drug effect) [7, 9], or because the drug effect itself may be independent between the two eyes [10]. Our study addressed these questions regarding the validity of the uniocular drug trial as traditionally proposed.

The major finding at odds with standard clinical teaching is that the IOP change in the untreated eye during a uniocular trial should not be subtracted when assessing drug effectiveness. An example will help to illustrate the importance of this finding. A typical patient has a baseline IOP of 25 mmHg in both eyes, treatment is initiated in one eye, and the follow-up IOP is 20 mmHg in both eyes. The standard clinical teaching is that the drug is ineffective in this patient because the bilateral IOP change relates to diurnal or other correlated fluctuation. However, this teaching involves several assumptions: the spontaneous IOP variation is correlated between both eyes and drug crossover effect is minimal. An additional assumption is that the patient, day, and eye are selected for initiation of treatment at random, so that there is no regression to the mean. According to the classic teaching, the regression coefficient for the IOP change in the untreated eye is minus one in these circumstances. When the change in the untreated eye is subtracted from the change in the treated eye, it appears that the drug had no effect.

Alternate interpretations can be made if these assumptions do not apply. For instance, a bilateral drop in IOP from 25 to 20 mmHg with uniocular treatment may in theory also be interpreted as the result of an effective drug with a substantial crossover effect. If drug crossover is so substantial that both eyes essentially receive an equal effect even with unilateral treatment (a theoretical situation which does not apply to currently available topical agents), then the drug effect can be estimated by the average IOP change. As the average change is simply one half the change in the treated eye plus one half the change in the untreated eye, the regression coefficient for the IOP change in the untreated eye would tend to plus one half with a complete crossover effect.

Another factor which may affect the interpretation is regression to the mean, which occurs in clinical practice (and even in many randomized controlled trials) because glaucoma treatments are only initiated when patients have an IOP above the target pressure. As the IOP may be above the target pressure only due to random fluctuation, as opposed to a true change in the underlying disease process, subsequent visits may see a drop in IOP simply due to the statistical phenomenon known as regression to the mean. This phenomenon applies to both uniocular and binocular trials in clinical practice. In uniocular trials it occurs not only because the average IOP may be high on a particular visit, but also because most clinicians (and randomized trial protocols [19]) choose the eye with the higher IOP for initial treatment. Regression to the mean likely explains much of the 5% drop in IOP seen with placebo treatment in randomized trials [20]. It is possible to reduce regression to the mean in a research setting by selecting random patients, asking them to come to the clinic on random days, and initiating treatment in one eye or in both eyes at random, regardless of whether the patient appears to need additional treatment. Such a protocol may have research value but involves initiating therapy in patients already below their target pressure. Because physicians only treat patients who appear to need treatment, regression to the mean is a fact of life with which clinicians must contend when trying to understand clinical observations. The statistical coefficients derived in this study allow clinicians to make predictions which take into account regression to the mean.

In most analyses, the IOP change in the untreated eye during a uniocular trial had minimal predictive value and the regression coefficient was close to zero. For dorzolamide, the change in IOP in the untreated eye had significant predictive value, but in the direction opposite from the standard clinical teaching. Although a more positive coefficient might suggest the importance of crossover effects with dorzolamide [21], the number of dorzolamide trials was small and this finding may be due to chance.

Our data indicate that the uniocular drug trial need not be abandoned in all cases. However, the interpretation of the uniocular trial must be modified. After a uniocular trial, regression equations may be used to predict the IOP in the first eye (Tables 2) and in the second eye (Table 3). Likewise, after a binocular trial, regression equations may be used to predict the IOP in the first (Table 4) or second eye (Table 5). When interpreted correctly, both uniocular and binocular trials had similar success in predicting the future IOP in the first or second eye.

Whether a uniocular or binocular trial is performed depends on a variety of circumstances. From the standpoint of safety, the uniocular trial has advantages because the patient presumably will receive half the dose systemically, and because any local allergy or intolerance will be more easily diagnosed by its uniocular presence. Drug tolerability was an important aspect of initial calls for uniocular trials [2]. On the other hand, the binocular drug trial succeeds in providing benefit to both eyes more quickly and may spare the patient an additional visit. Moreover, a binocular trial is less confusing because it avoids subsequent regimen changes, and because the patient need not remember which drug is initially taken uniocularly and which eye is treated.

This study had a number of limitations. First, several subjects contributed more than one trial with different types of medication even though the results of two trials of different medications in one individual might not be completely independent. This objection is largely addressed by the subset of trials comparing uniocular and binocular trials because only two of the 68 subjects contributed more than one medication trial. Finally, as with previous studies [9–11], the data were retrospective and patient compliance may have been imperfect. On the other hand, the data reflect real-world observations made in an urban university glaucoma practice, and may apply to similar practices.

One problem highlighted by our data is that a single-visit uniocular or binocular drug trial still leaves a great deal of unexplained variation in follow-up IOP. Even combining the information from two visits (a uniocular and binocular trial) had minimal added benefit in terms of the ability to predict IOP (data not shown). Some authors have hypothesized that plotting the diurnal variation in individual patients might make estimation of drug effect with uniocular trials more accurate [7, 8, 12, 22]. Others have advocated multiple baseline and follow-up visits in the context of binocular drug trials [11]. Undoubtedly, additional visits will permit better predictions for both uniocular and binocular drug trials, but a quantitative understanding of the improvement is lacking. Additional studies will be needed to test specific protocols. Given the difficulty of assessing patient-specific therapeutic effect, in the future appropriate medication selection may also be based in part on genetic testing [23].