A reduction in intraocular pressure in clinical trials can be determined through the mean intraocular pressure, through the proportion of patients who have the intraocular pressure reduced to a specific target intraocular pressure, or both. Since both these possible endpoints measure the shift of 2 intraocular pressure distributions, we recommend that only one of them be tested. In general, testing the difference between mean-values is much more efficient than testing the difference between proportions. However, proportions of successful patients are valuable in showing the clinical implication of a reduction in mean intraocular pressure, particularly when evaluating a moderate pressure reduction. The effect of a small mean intraocular pressure reduction on the probability to reach the target intraocular pressure is pointed out, particularly the fact that it can be substantial even if the mean reduction is smaller than the measurement error.
In individual patients the effect of an intraocular pressure (IOP)- reducing treatment is commonly judged by its ability to reduce the IOP below a specific level (target pressure), with the ambition to prevent further progression of the glaucoma disease by keeping the IOP below this level. 1-5 In clinical practice it is recommended that the target is individually determined and continually adjusted. 3-5 It is generally assumed that a reduction of at least 30% from the IOP at which damage occurred is a reasonable initial target IOP in eyes with a moderate damage of the optic nerve. 2 Higher or lower target pressures may be chosen depending on the degree of optic nerve damage.
In clinical trials the IOP-reducing effect is evaluated for a whole group of patients and the shift of the whole distribution of IOP values is of interest and must be summarized in an appropriate way. This can be done by several different methods and this article demonstrates how some of the most important statistics perform and the interrelations of those important statistics. This has previously been done theoretically, 6,7 but our aim is to illustrate this with realistic data from a selection of large clinical trials. 8-14 In the majority of clinical trials of IOP-reducing treatment, the mean IOP is used as an endpoint to determine the difference in treatment effect between groups of patients. However, the target pressure concept suggests that the comparison of the IOP between groups of patients should be determined by the difference in proportion of successful patients instead of the difference in mean IOP. Irrespective of whether the target IOP is an arbitrarily selected IOP level valid for all patients in a trial population or an individually predetermined IOP level for each single patient, it means dichotomizing a continuous variable. The statistical implication of this has previously been shown in detail; both theoretically based on the Gaussian (Normal) distribution and in an applied form based on actual distributions of blood pressure in patients with systemic hypertension. 6-7 Particularly, it is well known that the association measures of dichotomized variables (eg, proportion differences, proportion ratios, and odds-ratios) are dependent not only on the variable distribution but also on the selected target value (cutoff level). It is also well known that the statistical test-power of mean-value based tests of continuous variables is in general much higher than tests based on dichotomized variables.
The present work compares the strategy of analyzing mean-values of the continuous IOP variable with the strategy of analyzing percent patients who have their IOPs reduced to or below a specific target pressure (target success), using realistic numbers for a typical open-angle glaucoma or ocular hypertension trial population. The difference in mean IOP was related to the difference in target success and the difference in odds-ratio of target success. The required number of patients for each strategy was compared.
A related issue that was discussed is a common misunderstanding that a mean-value difference in IOP that is smaller than the measurement error of the IOP cannot have a sizable effect on the probability to reach a specific IOP target or be of clinical value.