ASH 2025 | The PRISM score: improving risk stratification of newly diagnosed AML treated with ven-based therapy

So our abstract that was selected for an oral presentation, we call it the PRISM AML score, so Prognostic Risk Integration for Survival Modeling and Acute Myeloid Leukemia, is a really lovely collaboration with collaborators from around the globe where we’ve put together this large data set, inclusive of both the internal data set, which is a training cohort and an internal validation cohort, as well as our external validation cohort. It’s comprised of around 2,300 patients that were treated with lower-intensity venetoclax-based combination therapies. So most of these patients received a hypomethylating agent, either azacitidine or decitabine with venetoclax, though a very minute portion received low-dose cytarabine in combination with venetoclax. 

And the reason that this was an important project was… or the aims of the project I should say, or the objective, was to understand if we could come up with a more nuanced prognostication model for overall survival compared to our current models, which work quite well and are good foundational models, but leave room for improvement. And so, you know, that was really the impetus of this project, as we know that traditional risk stratification models used for younger patients receiving intensive chemotherapy don’t stratify this lower-intensity venetoclax-based population well. 

We were able to identify that the PRISM-3 score… we were able to identify 17 features actually that were prognostic in the setting of this therapy, and we were able to derive a score that was a linear risk score that we then broke into tertiles to make three risk groups. We called these the PRISM-3 groups, and these PRISM-3 groups outperformed the current standard of care, which is a four gene classifier that uses mutations in FLT3-ITD, NRAS, KRAS, and TP53 to prognosticate patients receiving venetoclax-based therapy, patients with AML. The PRISM-3 score outperformed the four gene classifier and actually, importantly, re-stratified a lot of patients – about 50% of patients that were classified as favorable risk using the four gene score were re-stratified as either intermediate- or adverse risk using the PRISM-3 score. And this resulted in better separation of the survival curves and improvement in survival discrimination and modeling with an increase in the C index. This was validated in both the internal validation set as well as the external validation set with very stable C indices ranging from 0.63 to 0.64 across all the sets. 

We then look to see whether or not the PRISM-3 score in those patients that were previously classified as favorable risk and were re-stratified, whether or not it improved survival, or rather I should say stratified survival within that patient subgroup, that risk subgroup specifically, since that was a large portion of patients that were re-stratified. And indeed we saw across all three cohorts that the PRISM-3 score re-stratified patients. 

So this is a more nuanced, a more individualized risk calculator that also incorporates some clinical features in addition to genetic and cytogenetic features that appears to better prognosticate survival following venetoclax-based therapy in AML. It’s somewhat of a complicated score to put together. So in conjunction with the presentation, we also have an interactive web app that individuals can use. It’s prism-aml.com. This allows someone to calculate not only the PRISM-3 score, but also the 4-gene classifier is present there, and they can evaluate survival based on both of the scores, because we do still think there’s value in that 4-gene classifier. And then we also have what we call a discovery survival analysis on this website. So we know there’s other genetic mutations that were not selected for the model, but investigators and clinicians will be interested in outcomes in patients with those specific mutations or mutational combinations. And so using this discovery cohort, the survival discovery tab, interested individuals will be able to select and look at specific outcomes within specific genetic subgroups. So it’s really exciting work.

This transcript is AI-generated. While we strive for accuracy, please verify this copy with the video.