Are geographic atrophy algorithms taking these patients into account?
Simrat K. Sodhi, BA, MSc, MB BChir (Cantab), a graduate of the University of Cambridge, talks about a study she was part of that was presented at the 2024 AAO Annual Meeting, titled “Evaluating the Use of 3-D Volumetric OCT Data to Accurately Segment Area of GA Secondary to AMD: The GEODE Study.”
Question:
Can you discuss the design of the study and its findings?
Simrat Sodhi, BA, MSc, MB BChir (Cantab):
I think it’s kind of on trend right now to do GA analysis. Our group did a multicenter study. We looked at 715 OCT volumes, both Cirrus and Spectralis data. We used this data to create an automated algorithm that would be able to delineate geographic atrophy. In doing this, we created an algorithm that had a great Dice coefficient. It was about 0.82 across both machines and a really good coefficient determination, which was about 0.9, again, between the 2 modalities. With this, again, there’s been other areas and other groups that have also found similar Dice coefficients and coefficients of determination in their algorithm. We want to look a little bit deeper into maybe the areas or the types of scans that bring down the Dice coefficient. We found that if you have smaller atrophic islands all clumped up together, so not a lot of space in between, your Dice coefficient starts to go down.
The best atrophic region to really delineate is a large area of atrophy. But another kind of area of atrophy, well, maybe it’s a little bit more complex if you have larger islands, but that they have space between, you also have a good Dice coefficient. It’s, I think, important for us to kind of share this with other groups doing another, if they’re also developing algorithms, to really just show that there might be this subset that’s bringing down the Dice coefficient. Or even within sometimes in clinical trial data, those patients may not be included, but in a real life setting, those might be some patients that we may have to look at a little bit closer to really make sure that the automated algorithm is doing them justice. Maybe taking a little bit more of a pointed glance at those patients. That’s kind of the key elements to take away from our study.
Question:
How might these findings influence future research in geographic atrophy?
Simrat Sodhi, BA, MSc, MB BChir (Cantab):
Again, I think looking at at least the subset of patients and maybe which people are not performing as well, where there’s disagreement between a manual grader and the automated algorithm, it’s good to see that for the future. We can maybe make the algorithm a little bit more granular, really pay attention to why there’s that disagreement and how we can make it better. I think for us, in terms of our future research, we did look at GA only eyes and GA with other pathologies including neovascular AMD. But I think it’s important to look at other pathologies as well, because most of these patients that come in with GA maybe will have other pathologies. You really want to make sure that they’re able to delineate the atrophy even if there is other things going on in the background. Then for our algorithm, we want to continue to do longitudinal studies, more repeatability and reproducibility studies and really validate the algorithm as well.
Question:
What are some persistent unmet needs in GA?
Simrat Sodhi, BA, MSc, MB BChir (Cantab):
I think the therapeutics that have been launched has been great because it really has opened this wave of new research and the potential to really give your patients something that can help with the progression of the disease. But I think now with not having approval in Europe, it really, maybe we need to open the door even further for other drugs to come into that space. But I also think with the approvals not happening in certain countries, we may also have to look at the modalities that we’re using and the outcomes that we’re measuring and really see if we’re using the correct ones to track what the patients are getting out of these therapeutics.
I think there’s a lot with these algorithms that maybe we can look at biomarkers or things that are happening early on to really see if that is halting the progression. I went to Euretina and I’m assuming at AAO, similar findings were presented where microperimetry is kind of being talked about as well in that same. I think working with industry professionals, but also with researchers to really understand what are the outcomes and what are the modalities we need to be using for GA because it may be different than what we’ve been used to in the past.
Question:
What does the future hold for GA?
Simrat Sodhi, BA, MSc, MB BChir (Cantab):
I think in general with GA and all these algorithms, I think all of us are trying to come after the same goal. I think each of these programs and each of these algorithms is going to be going after a different method. But I think what’s really crucial is really looking at the subset of when these algorithms are continuously failing and then working together to really make sure that you’re finding ways that we can improve that. But in general, GA is a really hot topic right now and it’s really exciting, and I personally am really excited to see what happens in the space.
This content is independent editorial sponsored by Astellas. Astellas had no input in the development of this content.
