Strategies in Retinal Detachment Surgery: The EVRS RD Study
Can nonrandomized studies answer the question?
DIDIER H. DUCOURNAU, MD
Without a statistically significant, clear-cut answer regarding the choice between available strategies, a surgeon often makes treatment decisions that are based on opinion, intuition, or assumptions. Objective answers, however, must be out there, and there must be a way to find them.
Let’s take an example: Which is the best management for a retinal detachment with stage A PVR induced by a superior tear in a pseudophakic eye without hypotony: pneumatic retinopexy, scleral buckling, or vitrectomy?
The question is difficult to answer because there have been no large-scale clinical trials that showed statistically significant differences among these treatment options.
The SPR study1 was a serious analysis of 681 eyes with RDs, contributed by 45 surgeons from five different countries. The study attempted to answer the “buckling vs vitrectomy” question, but no statistical difference between the two methods was found.
The problem with such a study, in addition to what might be viewed as a great deal of wasted effort in light of its inconclusiveness, is that some surgeons interpret “no statistical difference in this study” as meaning “no statistical difference at all” and thus choose their strategies according to their personal preferences.
|Didier H. Ducournau, MD, is the CEO of European VitreoRetinal Services Company and a retinal physician in private practice.
Dr. Ducournau can be reached via e-mail at firstname.lastname@example.org.
IS STATISTICAL DIFFERENCE SIGNIFICANT?
What we need to remember is that a lack of statistical difference is germane only when the number of patients studied is large enough to allow the difference to be expressed. When dealing with low failure rates, a very high number of cases is necessary.
In the example of a series of 681 RDs, a significant difference between buckling and vitrectomy would only be found if the failure rate of one technique was very much lower than the other, for example, 1.2% vs 3.6% or 1.8% vs 4.7%. To achieve a statistical difference (with a P value less than 0.05) between two techniques presenting, respectively, 1% and 2% failure rates, a minimum of 2,400 cases would be required.
Furthermore, if we want to compare these two techniques while taking into account the presence or the absence of one clinical detail, twice as many cases would be required.
Because a randomized study on such a scale is unimaginable (What company would sponsor the study?), there are only two practical options: Either accept that no statistically significant answer is offered to the question, or accept the results of a nonrandomized study as the only way to collect more than 5,000 cases.
PROBLEMS WITH NONRANDOMIZED STUDIES
In a nonrandomized study, surgeons can make mistakes regarding the imposed selection criteria. There is no easy way to avoid this problem completely. Additionally, surgeons can select the cases that he or she wants to contribute, even if the input is anonymous. This fact can affect the quality of the results, and nobody is supposed to trust these results.
High-level statisticians are used to dealing with this problem and can guarantee that it is avoided by mandating two criteria. First, one should never present the result of a technique as an individual result but always in comparison with at least one of the other techniques. Thus, the possibility of errors in selection will have a comparative effect in both groups.
Second, a large number of surgeons should be required for each technique so that the effects of selection errors will be negligible. Of course, randomization can also solve these problems, but who can launch a randomized study with 5,000 cases?
THE EVRS EXPERIENCE
The European VitreoRetinal Society (EVRS) asked its members to complete a prospective form for all successive RDs (Stage 0, A, B, and C1) as primary procedures, operated on between April 2010 and April 2011, with a requested minimum follow-up of three months. For each case, 25 items had to be described:
• Six items on the RD clinical findings (lens status, PVR stage, number of detached quadrants, type and number of retinal breaks, aggravating factors).
• Nine on vitrectomy details (completeness, gauge, pump type, cutting speed, posterior hyaloid and ILM removal, retinotomy, combination with phaco, other techniques).
• Two on retinopexy and tamponade.
• Three on external procedures (buckling, drainage, other).
• One on perioperative complications.
• Four on the results, including any complications, the need for reoperations, the final results, and the follow-up.
By August 2011, the reports numbered 7,678 RDs from 180 contributors in 48 countries. The subsequent EVRS RD Study2,3 was presented in Malta during the 12th EVRS meeting. The results were evaluated independently of the investigators by the French National Institute of Statistics and Economic Studies.
After cleaning the database, a global working sheet was sent to each contributor, masking the name of the other contributors, so that they could agree on the cleaning accuracy.
First, univariate and bivariate analyses were performed for the whole database, the first step necessary to identify the variables that were linked to failure rate. Multivariate analysis was then performed. Step-by-step logistic regression was performed for the whole database on clinical findings and surgery parameters for identification of the major variables. This allowed us to identify that four clinical findings were independently linked to the definite failure rate:
• Presence of PVR stage C1 (P value = 10−5; odds ratio [OR] 4.4).
• Detachment of all four quadrants (P value ≤ 10−5; OR 2.4).
• Presence of choroidal detachment/hypotony (P value = .0005; OR 2.0).
• Presence of a large/giant tear (P value = .023; OR 1.6).
Logistic regression was performed in cases in which vitrectomy was performed, taking into account the vitrectomy parameters, intraocular maneuvers, and initial clinical findings. Two vitrectomy machine parameters were independently and significantly linked to the irreversible failure rate:
• Use of a vacuum-control (Venturi pump) machine (P value = .006; OR = 2.9).
• Use of low cutting speed (P value = .014; OR = 2).
These data were reported independently and were then combined to build a strategic decision tree based on the results. For example, we determined:
• In the absence of PVR or with stage A PVR without vitreous hemorrhage, choroidal detachment or hypotony, and without posterior, large or giant retinal breaks, cases submitted to vitrectomy with a vacuum-control machine had a much greater risk of developing a definite failure than cases operated on without vitrectomy (P = .045).
• When the RD was induced by a tear and not by a hole, pneumatic retinopexy resulted in a higher irreversible failure rate, compared with a buckling procedure (P = .034).
• In cases of PVR stage B, without vitreous hemorrhage, choroidal detachment, or hypotony, not performing a vitrectomy resulted in a higher definite failure rate (P = .0017).
• In cases of PVR 0, A, B, or C1, not performing a vitrectomy resulted in a higher irreversible failure rate when a choroidal detachment/hypotony (P = .0015) or large/giant tear was present (P = 7 × 10−8).
LESSONS WE CAN LEARN
We cannot emphasize enough the value in building such a decision tree, especially for less experienced surgeons. Even the most experienced of us should be interested to know that when a patient was operated on by a surgeon using a vacuum-control vitrectomy machine, the odds of developing an irreversible failure are 2.9 times greater than if the vitrectomy machine is equipped with an aspiration flow-control pump.
Nonrandomized studies remain important tools for the modern surgeon and may bring us closer to making well-informed treatment decisions, even if they do not satisfy the commonly cited requirement that “additional randomized studies are required to confirm the obtained results.” RP
1. Heimann H, Bartz-Schmidt KU, Bornfeld N, et al. Scleral buckling versus primary vitrectomy in rhegmatogenous retinal detachment: a prospective randomized multicenter clinical study.
2. Strategy for the management of uncomplicated retinal detachments: The European Vitreo-Retinal Society Retinal Detachment Study Report 1. Ophthalmology. 2013, in press.
3. Strategy for the management of uncomplicated retinal detachments: The European Vitreo-Retinal Society Retinal Detachment Study Report 2. Ophthalmology. 2013, in press.