The EVRS RD study is a statistical study based on 7678 cases concerning only primary procedures and PVR stages 0, A, B and C1. The goal of this study was to highlight the variables that affect the final surgical outcome.

180 surgeons from 48 countries

These cases were sent by 180 surgeons from 48 countries. Here is the graph of the cases origin.

This is a photography over a year of thousands of retinal detachments operated all over the world. This is not a randomized study; however, this bias is highly reduced by the huge number of analyzed cases. For each case, 30 different variables were specified. Some of them, mainly in the clinical findings group, were not entirely filled in as expected. For further study, the organizer should be more specific about the data definition ; there was for example no definition of PVR stage 0. Other data, such as the lens status or the vitrectomy machine parameters, were very reliable since they do not depend on any interpretation.

Some missing data were guessed when possible. Otherwise, they were indicated as missing value, and when there were too many missing values for the same case, this case was simply deleted. Every surgeon was sent his cases back after this cleaning, so that everyone was able to check what had been done.

Finally 18 variables were studied : 7 parameters about clinical findings, 3 parameters about the vitrectomy machine if a vitrectomy was performed.

3 variables about maneuvers during vitrectomy and 4 other surgery details The surgeon and country variables were not studied for the following reason :

This is the PVR distribution for all the data base. When looking at the PVR distribution for Algeria or Malaysia, it appears that the photography of the operated RDs does not the reflect the truth. It is for instance highly unlikely that in Algeria only PVR stages B and C are operated.

Moreover, in a same country, huge differences appear between surgeons. Here are the PVR distribution of cases operated by two Italian surgeons. Once again, there is obviously a selection bias which can explain these two different distributions in the same country.

One of the main goal of the study was to highlight the variables that have an influence on the result. In order to do so, it was necessary to define the notions of failure or success. Due to the percentage of success and for a scale matter, it was more logical to focus on failure rate instead of success rate. Graphically, the difference between 1% and 3% is indeed far more obvious than between 99% and 97%. Failure level 1 is the true final failure declared by the surgeon. On the graphics it appears in red. Failure level 2 designates the cases not declared as failure, but where silicone oil had not been removed in June, 2011. These cases can turn into a failure or a recurrence, after silicone oil removal and cannot be considered therefore as final success. The remaining silicone rate is in orange. Failure level 3 designates the rates of cases needing re operation for complication or recurrence. This percentage will be written in yellow on the graphics but will not be represented by any bar.

Now the methodology. First, univariate and bivariate analysis were performed for the whole data base in order to have a graphical representation of the first results. This is also a first step necessary to identify the variables that are linked to the failure rate. Second, a multivariate analysis was carried out. A step by step logistic regression with all cases on clinical findings and surgery parameters was performed. Moreover, with only vitrectomy cases, the same kind of regression was performed on vitrectomy machine parameters. This was done in order to highlight the variables that independently affect the true failure rate and in order to suggest a strategy. This was a heavy work that leaded to more than 3000 lines of calculations.