Purchase this article with an account.
Fabian Lehmann, Robert Wilke, Robert Patrick Finger, Helmut Sachs; The conundrum of relations in the multivariate dataset of nvAMD treatment. Invest. Ophthalmol. Vis. Sci. 2018;59(9):2384.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
The statistical evaluation of clinical trials can be challenging, as there may appear spurious correlations if the correlation of only a single determining factor and a command variable is considered. To include all affecting factors we evaluated a multivariate data analysis in case of a retrospective study investigating the therapy of nvAMD with Aflibercept or Ranibizumab.
In our statistical evaluation the influencing variables of the best corrected visual acuity (BCVA) in 1055 subjects with neovascular AMD (nvAMD) were observed after 2 years of an anti VEGF treatment. 23.591 visits were evaluated and 8.150 Ranibizumab and 1.725 Aflibercept injections were realized. Jmp 11.2.0, SAS Institute Inc. software was used and the following influencing variables were tested: BCVA at baseline, retinal thickness at baseline, age at baseline, mean visitintervall, number of anti VEGF injections, initial BCVA gain after upload, treatment change of Ranibizumab or Aflibercept. The command variable was gain or loss of BCVA at the end of year 2.
All of the listed variables on their own have a strong effect on the BCVA after 2 years of treatment. For example an initial analysis of switching between Ranibizumab and Aflibercept seems to have a strong effect on the BCVA, which was statistical significant after 2 years (p<0.0001). But in a multivariate analysis where spurious correlations can be avoided, only the following variables have a statistically significant impact on the BCVA at the end of year 2: BCVA at baseline (p<0.001), BCVA gain after upload (p<0.001), number of intravitreal injections (p=0.05), number of visits in outpatient department (p<0.01), and retinal thickness at baseline (p=0.02). These factors had a higher influence on the result compared to the change of therapy, without a statistical significant effect on the final BCVA (p=0.2) [graph 1].
In a retrospective statistical analysis it’s essential to consider not only one variable and its effect which could be luckily statistical significant on its own, as there could be many more substantial factors having an influence on the result, which you can find with a multivariate data analysis.
This is an abstract that was submitted for the 2018 ARVO Annual Meeting, held in Honolulu, Hawaii, April 29 - May 3, 2018.
In graph 1 is drawn a Pareto-Plot to illustrate the influence of each factor on the whole model. The factors put under red have a statistic significant effect on VA result after 2 years and represent 90% of the effects that are relevant in this model.
This PDF is available to Subscribers Only