ASH 2024 | Refining the 2022 ELN guidelines for AML risk stratification using machine learning

As we know that the European Leukemia Net 22 risk stratification is the cornerstone of risk stratification for AML based on which treatment decisions are made. It’s exceedingly important and we use it in routine clinical practice. So, although it’s a very useful guideline, there are certain steps or certain investigations or parameters that are still not accounted for. A few of the things that we found in our study is that it doesn’t take into account variant allele fractions. And when you risk stratify patients in one ELN risk category, what happens when they have additional mutations? So this was the question we were trying to pursue. So we used supervised machine learning to look at individual ELN risk categories and then we looked at all mutations which occurred at a frequency of more than 2% in that particular risk group. And we also looked at variant allele fractions if those patients had VAF of more than 2% which was high or low. We developed, combined all of this data and used supervised machine learning. So what we found was that even in ELN risk groups, which is the favorable as well as the intermediate risk groups, there does exist some form of heterogeneity. So what we could see that even within these risk groups, there exists a dichotomy where, for example, in the ELN favorable risk group, there seem to be at least two subsets which are defined based on mutations as well as variant allele fractions such as NPM1 variant allele fraction, presence of an NRAS mutation, a high RAD21 VAF as well as an ETV6 mutation and KIT exon 17 mutations. So these helped, you know, these were combined into a form of a score and we applied the score for ELN risk favorable risk patients and we could separate them into two different categories, both of which had poorer, both of which had different outcomes. Now interestingly, when you look at the ELN favorable two risk category, the outcomes of this are similar to that of the lower ELN intermediate risk one category. Similarly when we applied machine learning for the ELN intermediate risk we could say that based on high FLT3 variant allele fraction as well as presence of WT1 mutations you could again split these patients into two risk groups and the ELN intermediate two risk group had a poorer outcome as compared to which had a poorer outcome and this outcome was comparable to that of ELN adverse risk category so we show through this cumulative analysis that that there’s more data in the ELN and there is some form of a scope where we can look at additional mutations and VAF and further help risk stratify these patients so that you can meaningfully predict outcome.

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