advertisement
Abstract #15455 Published in IGR 1-3
Collaborative normal-tension glaucoma study (letters)
Sommer A; Drance SM; Anderson DR; Schultzer MAmerican Journal of Ophthalmology 1999; 128: 776-777
Editor: A large body of evidence suggests that intraocular pressure (IOP) is a critical factor in most patients with glaucomatous optic nerve damage, whether their baseline pressure are above or below 21 mmHg. 1,2 Although the Collaborative-Normal-Tension-Glaucoma Study (CNTGS) 3,4 is an interesting and thoughtful addition to the literature on the subject, it does not provide the definitive data that many seem to believe. 5,6 The principal problem is its failure to ensure the essential condition for any definitive intervention trial: that treatment and control groups be identical in every way except for the purported intervention (in this instance, reduction of IOP). The foremost approach of a succesful trial is strict randomization of subjects. This provides the greatest likelihood that comparison groups are appropriately balanced, not only for characteristics recognized and measured, but even more importantly, for characteristics not measured. The second approach maintains equality by analyzing the two groups in identical fashion. The CNTGS falls short on several counts, including: - A potential failure in randomization. Of the 145 randomized subjects, 79 ended up in the control arm and only 66 in the treatment arm. Although this was not a statistically significant deviation from the anticipated 50:50 split, it raises concernes that potentially critical but unknown factors suffered the same maldistribution. - The authors' straightforward, intent-to-treat analysis ("once randomized, always analyzed") failed to demonstrate any difference in the rate of progression between treated and control groups. Suspecting that the differential development of cataracts masked a real benefit by artificially depressing visual field scores, the investigators "censored" patients for cataract (in essence, removing afflicted subjects from the analysis). However, this differentially removed treatment subjects who had undergone surgery, half of whom went on to develop cataracts (compared with only 14% of control subjects) 4 It is reasonable to fear that subjects treated with filtering surgery had IOPs higher or more difficult to control that the average subject and by removing them, unbalancing the groups being compared. In trying to assess the impact of the reduction of IOP instead of the attempt to reduce IOP on progression of optic nerve damage, a complete series of "adjustments" were applied to the comparison of treatment and control groups. Unfortunately, all these adjustments required significant, untestable assumptions and introduced further opportunities for additional bias. The most obvious was the exclusion of five treated subjects who failed to achieve the desired IOP endpoint. A comparable group was not removed from the control arm (how would one even identify five comparable subjects with unknown characteristics that randomization had been meant to balance?). Deterioration in the remaining 61 treatment subjects was measured from the time they achieved stable IOP reduction, over 200 days after they had been randomized to treatment. In contrast, deterioration in the 79 control subjects was measured from the time of randomization itself. Exclusion of five treatment subjects and this complex, asymmetrical, analytic approach further compromised any aspirations of balance and maintenance of "equality" between the two study groups. The authors identified several sources of real and potential bias, including more frequent examination of control subjects, which increased their likelihood of reaching a detectable "endpoint" earlier in follow-up. Attempts to compensate for those differences introduced additional opportunities for unrecognized bias. None of these concerns proves that the conclusions reached by the investigators are wrong. They only indicate there is simply no way of knowing with any (known) certainty that they are right. Alfred Sommer, M.D., MHS
References
1 Sommer A. Intraocular pressure and glaucoma. Am J Ophthamol 1989; 107: 186-188.
2. Sommer A. Glaucoma: facts and fancies. The Doyne Memorial Lecture. Eye 1996; 10: 295-301.
3. Collaborative Normal-Tension Glaucoma Study Group, Comparison of glaucomatous progression between untreated patients with normal-tension glaucoma and patients with therapeutically reduced intraocular pressures. Am J Ophthalmol 1998; 126: 487-497.
4. Collaborative Normal-Tension Glaucoma Study Group, The effectiveness of intraocular pressure reduction in the treatment of normal-tension glaucoma. Am J Ophthalmol 1998; 126: 498-505.
5. Danias J, Podos SM. Comparison of glaucomatous progression between untreated patients with normal-tension glaucoma and patients with therapeutically reduced intraocular pressures [letter]. Am J. Ophthalmol 1999; 127: 623-624.
6. Anderson DR, Drance SM, Schulzer M. Comparison of glaucomatous progression between untreated patients with normal-tension glaucoma and patients with therapeutically reduced intraocular pressures [letter]. Am J Ophthalmol 1999; 127: 625-626.
Author reply
We are pleased that Dr. Sommer insists on scrupulous interpretation. We carefully studied his comments and remain confident that our study design and the conclusions were correct and statistically significant. 1. Randomization was meticulously carried out within the individual centers, based on block randomization with randomly variable block sizes. The small difference in the numbers of subjects assigned to the two arms of this magnitude is common enough in clinical trials and is certainly not statistically significant, as Dr. Sommer points out. The mean intraocular pressure (IOP) was very slightly higher in those randomized to treatment shown in Table 1 1, but otherwise the two groups differed only by the induced reduction in IOP. Repeat of each analysis with statistical adjustments for a number of covariates did not affect the result. A third paper (submitted) shows that the prerandomization natural course of patients ultimately randomized to treatment was identical to the prerandomization course of those ultimately assigned to control. We find no reason to fear that randomization failed to make the two groups match, thus meeting the criteria of an interventional trial outlined by Dr. Sommer. 2. With the two papers published back to back, 1,2 it is easy for the reader to forget that our protocol was designed primarily to determine whether the level of IOP has an impact on the clinical course. 1 For ethical reasons, judgements about any deterioration were made in the untreated group starting immediately after randomization. In the treated group baselines were established after IOP was lowered. Deterioration was unequivocally slower in the group with lowered IOP. To ensure that differing start times and other variables were not responsible for this result, we perfomed additional analyses after various data adjustments, only some of which were detailed in the publication. No matter the manner of analysis, the difference between the two groups persisted. The five patients who never achieved a 30% lowering of IOP were not excluded from analysis in the first paper, 1 but simply did not produce date after succesful IOP lowering. These patients were included in the analysis of the intent to treat in the second paper 2; therefore, no "balancing"was necessary. 3. The data monitoring committee was puzzled when the outcomes were compared with the randomization baselines and the two arms did not diverge. Cataract formation was eventually suspected to be confounding the visual field data in the surgically treated subgroup. This hypothesis was tested by identifying cataracts from acuity data and the unbiased attribution to cataract by the clinicians masked to the subsequent use of the information. These subjects were not removed from the intent-to-treat analysis, but their data were included only up to the date of verified cataract formation. The two arms now diverged significantly. Comparison of all relevant variables of surgically and medically treated subjects found no features (like a higher IOP) to distinguish them. Analysis was repeated after data "thinning" with restricted randomization methods, to make the two arms maximally comparable, and also with covariate adjustments. None of these steps altered the divergence of the two arms if data were censored after cataract formation. In both papers, the conclusions were drawn from the primary analysis. Susequent analyses with adjustments showed that the conclusions held up after correction for potential biases like those that worry Dr. Sommer.
Stephen M. Drance, M.D. OC.
Douglas R. Anderson, M.D..
Michael Schulzer.
References.
1. Collaborative Normal-Tension Glaucoma Study Group, Comparison of glaucomatous progression between untreated patients with normal-tension glaucoma and patients with therapeutically reduced intraocular pressures. Am J Ophthalmol 1998; 126: 487-497..
2. Collaborative Normal-Tension Glaucoma Study Group, The effectiveness of intraocular pressure reduction in the treatment of normal-tension glaucoma. Am J Ophthalmol 1998; 126: 498-505.
Classification:
9.2.4 Normal pressure glaucoma (Part of: 9 Clinical forms of glaucomas > 9.2 Primary open angle glaucomas)
