content top

Material & Methods

Left Right

We collected 2603 ME cases with a minimum follow-up of 6 months, sent by 85 retina specialists from 29 countries.

Four principal etiologies: ERM, CRVO, BRVO and DME.

The statistical significance was much more difficult to reach than the last year RD study due to the number of etiologies.

But most of all due to the treatment complexity. The number of different treatments forced us to:

analyze only the 4 major etiologies, this is to say to concentrate only on 2159 cases.

If we add the treatments previously performed elsewhere, the repetition of injections with the same or with other treatments, and the combination of treatments, you can imagine the complexity of the analysis.

The first line is easy to analyze, the second more difficult and the third quite impossible.

So that at the end we did not get sufficient data for each treatment of each etiology. So the statistician arbitrarily decided two things:First we will present trend lines. What is a trend line?

A trend line is, in statistics, a linear regression this is to say an approach to modeling the relationship between two variables X&Y. In fact we selected to use a second order polynomial regression because it fits more to illustrate the effects of a treatment according to the pre treatment vision or the evolution of the improvement according to the follow up than a strict linear regression or a order 3 polynomial regression (we cannot imagine a vision going up and then and then up again after a single treatment.I will try to explain you why a polynomial regression presentation is more exact in our study than a histogram presentation.

If you consider this population A of 80 cases, with the pre treatment visual acuity here and the post treatment here, this is to say if you consider the improvement or loss according to the pre treatment vision, we can see that there is an homogenous distribution of results which are mostly positive in the upper zone of this iso-acuity line.

The population B is quite similar, a little bit denser for the higher visual acuity levels.

If we draw the trend line, the curves does efficiently illustrate the fact that the post treatment vision is globally superior to the pretreatment vision therefore that the treatment is efficient. The trend curve illustrates the trend global effect of the treatment effect.

In the case of population B the trend line curve illustrates the same positive trend effect, confirming the general impression that both populations are quite similar.

For population A, the presentation using the classical histograms showing the average improvements according to the pretreatment vision fits with the trend line representation.

You can see here that this is not the case for population B : the histogram illustration does not fit with the trend line. Why?

It is because the population is too small, particularly for some less frequently tested visual acuity. Who test 0.04 or 0.06?. Lets take the example of pretreatment VA at 1.2: there are only two cases. If these only two cases drop their post treatment VA.

The representative histogram will be negative and will not represent the general trend. The characteristic of the trend line is that, even if it takes into consideration all the cases including the exotic ones, it also considers the entire number of cases, so that the aberrant results effect will be diluted. But, of course, you can tell me «this is absurd to analyze 80 cases with 15 columns, you should regroup the columns!» But then how to regroup them. If we decide to regroup 3 by 3, the second column regrouping 1.4 , 1.3 and 1.2 will not fit as well. So we can decide to regroup 4 by 4 and then we begin to trick according to what we would like to appear. Regression line is more exact and impossible to forge.

To have even more reliable results, the statistician decided to consider only the averages which contain at least three cases, so that the aberrant cases will not have any impact.

The counter part of this decision is that you will get some interrupted curves because there is not at least 3 cases after the dedicated date.

The last difficulty was this fact that everybody knows: the visual improvement should be considered only according to pre treatment vision level.

Here, the interrupted black line represents the gain obtained whatever the treatment for the 2159 studied cases according to the pre treatment vision. I do not understand how people can still present the percentage of improvement superior to 3 lines without considering the initial vision. You easily understand that it is easier to obtain 4 lines improvement when the pretreatment vision is count finger than when it is 20/40; this would lead the patient to see more than 20/20. The true evaluation of results should rely on classification of results according to pre treatment VA.

So when you will see this kind of presentation, you will understand that the improvement of 3 lines obtained thanks to treatment «rose», being just on the average level is better than the 3 lines obtained by the treatment «white» which is half a line below the average level.

content top