Choosing Endpoints in Clinical Studies and Trials

Balwantray C. Chauhan, PhD

Introduction
It is currently challenging to select endpoints in glaucoma clinical studies. Endpoints are statistical, rather than biological, events as no reference standard for progression is available, because one has to rely on the very test(s) to make an estimate of efficacy as those that are used to define glaucoma and monitor its progression. Similarly, findings from one study cannot be easily compared with those of another as strategies for detecting progression vary widely.

Event- and trend-based analyses
Some of the data available are from ‘cross-sectional’ analysis of longitudinal data, which may be counterintuitive and inappropriate. Patients are followed up longitudinally, yet cross-sectional tools are used to detect progression, i.e. when a test yields an abnormal result from a previously normal one. Compared with the event-based approach, trend-based approaches are more appropriate for providing information on the rate of progression, which is important for clinical decisions. In trend analysis, comparison is made with the patient’s own previous data. The tools that have been developed for longitudinal analysis include the glaucoma progression analysis (GPA), the Topographical Change Analysis (TCA) on the HRT and linear regression techniques. The trend based approach has certain advantages, particularly when many (> 8 or so) examinations are available as the visual field (VF) measurements over time can be used to determine the rate and magnitude of change over time. Although event analysis may identify test locations that appear progressive, with as few as three test results, it is dependent on the degree of change exceeding test–retest variability, which is high for damaged locations. To maintain reasonable specificity, most investigators have found it necessary to have glaucoma change probability (GCP) points outside normal limits to be confirmed on two or more tests. While longitudinal analysis is a more appropriate approach than cross-sectional analysis, it has only rarely been used in the recent glaucoma trials. The only two such published study are the Early Manifest Glaucoma Trial, which used the number of points that occur in a GPA-type analysis based on pattern deviation and the Canadian Glaucoma Study which used the same type of analysis but based on total deviation.

Progression rates are method-dependent
When so many different criteria have been used in studies, and in the absence of an accepted standard of glaucoma progression, it is not surprising that a review of outcomes from clinical trials shows great variation. For example, the two studies that have used an untreated arm are the Collaborative Normal-Tension Glaucoma Study (CNTG) and the EMGT study. The progression rate in the treated patients in the CNTG is 12%, while the EMGT showed an almost four times higher rate. Thus the criterion used to define progression must always be borne in mind when considering study data.

Selection of criteria for visual filed progression in the ‘real world’
It is not meaningful to compare sensitivities between different tests unless their respective specificities are equal. A test that has lower specificity (i.e. falsely flags progression where no true change has taken place) is likely to be more sensitive to change compared to a test with high specificity. The ideal approach to compare progression rates, therefore, would be to equalize the specificities with each test. However, owing to the lack of an external standard, this is difficult to achieve with reasonable confidence. Owing to sampling variation, the limits of test–retest variability derived in one sample of patients may not be identical to those that would have been derived from another sample, yet even small differences in specificity are likely to have a large effect on a test’s sensitivity to change. A meaningful comparison of progression rates, based on empirical data, is therefore very difficult to make. There are three ways in which false-positive rate can be assessed. Test-retest data are often used to derive limits for variability but are probably not entirely independent and may not be the best choice to use. Parallel cohorts of control subjects can be tested to see how frequently they progress and if this provides an indication of false-positives. This gives an independent measure of validation, but it assumes the eyes of the control subjects do not change, which is not likely in reality. The third approach is to look for parallel improvement rates in patients; this assumes that all improvement is ‘noise’. The presentation will review these three methods of estimating false-positive rates.