Abstracts 421

P49 RANDOMIZED CLINICAL TRIALS ON MEDICAL TREATMENT OF GLAUCOMA:

ARE THEY APPROPRIATE TO GUIDE CLINICAL PRACTICE?

Luca Rossettl, llarla Marchettl, Nlcola Orzalesl, Nlcola Scorplgllone, Valter Torrl, and Alessandro Llberatl

Clinica Oculistica de/l’lJniversita lstituto Scienze Biomediche

Ospedale S. Paolo, and Lab. Epidemiologia Clinica lnstituto di Ricerche Farmacologiche Wario Negri”

Milano, Italy

One-hundred and two randomized controlled trials (RCTs) on medical treatment of primary open-angle glaucoma (POAG) comparing various medical treatments tested on a total of 4882 patients enrolled over the last sixteen years have been reviewed. Thirty-two percent of the trials were cross-over, evaluating a total of 677 patients (mean 20 pts per RCT). Sixty-nine trials had a simple randomization design; they included more than 85% of the total patients (4215) with a mean follow-up of 10 months. There were numerous serious methodologic drawbacks and a general lack of consistency and coordination. The vast majority had a short follow-up (only 10% had follow-up longer than one year), were too small to detect any effect of treatment on clinically relevant end-points (only five trials had more than 50 patients per arm), focussed on an intermediate end-point (70% trials looked only at intraocular pressure (lop)), and were heterogeneous as regards the treatments tested (38 active combinations and 24 types of drug were tested) and the definition of standard treatment. Trials methodology was also largely unsatisfactory both in terms of design and execution: com- pliance was thoroughly measured in 23% RCT and the final analysis included all patients in 47% trials. The key message from this review is that the efficacy of medical treatment of POAG has been so far tested using an end-point (IOP reduction) whose validity and clinical relevance should be seriously challenged. The prevailing belief that glaucoma patients may benefit from a treatment able to reduce IOP is not supported by sound scientific evidence.

P49 PUBLICATION OF RANDOMIZED CLINICAL TRIALS IN VISION RESEARCH SUBMITTED AS

ABSTRACTS TO NATlONAL OPHTHALMOLOGY MEETINGS

Roberta W. Scherer and Kay Dlckersln University of Maryland

Baltimore, Maryland

Unpublished results of randomized clinical trials (RCTs) are not generally available to those who are interested in the results. In order to obtain a rough estimate of the rate of publication of RCTs in vision research, we determined the rate of subsequent publication of abstracts describing results of RCTs presented in 1988 or 1989 at either the American Academy of Ophthalmology or the Association for Research in Vision and Ophthalmology. Of the total 6014 abstracts presented, 149 were selected which appeared to describe results of randomized trials. Letters were sent to authors, requesting information regarding randomization and subsequent publication of the results presented in the abstract. We received 110 responses (74%). For non-respondents, subsequent full publication was determined via a search of MEDLINE using author(s) name.

The overall rate of full publication was 56% (841149). of the 110 replies, 79 authors confirmed that the trials had been randomized (n = 74) or quasi-randomized (n = 5) e.g. using methods such as alternation. Sixty-three percent @O/79) of these trials had been published in full, with no differences in publication rate of randomized (64%) or quasi-randomized (60%) trials. This rate may be an overestimate, since nonrespon- dents published less frequently overall (51%; 20/39) than respondents (58%; 64/l 10). The responding authors indicated that 52% of the unpublished trials were either incomplete or in the process of publication. Lack of time was most frequently stated (61%) as the reason for non-submission of results from completed trials to a journal.

Direct examination of the abstract showed that statistically significant results were reported more often in abstracts which were subsequently published in full than in those not published (58% vs 38%). Trials not published were less likely to have presented both experimental and comparison sample sizes in the abstract (28% vs 48%), possibly reflecting differences in study quality between these groups.