Unveiling the Future of Liver Disease Research: Inside AIM-MASH™ and its Groundbreaking Impact on Clinical Development, as Featured in Nature Medicine - a discussion with Lara P. Murray, PhD
Successful clinical trials in metabolic dysfunction-associated steatohepatitis (MASH) rely on consistent histologic scoring, which has historically been variable. The developed AI-based measurement tool (AIM-MASH) demonstrated reproducibility and alignment with expert pathologists, and has the potential to reduce inter-reader variability in MASH clinical trials. AIM-MASH+ additionally produces continuous scores for each of the MASH CRN (Clinical Research Network) components to enable granular measurement of histologic changes.
We sat down with Senior Translational Scientist Lead, Dr. Lara Murray to hear her thoughts on the impact of the publication, learn more about the development of AIM-MASH, and more. Read the interview below!
AIM-MASH™ and Liver Explore™ is for research use only. Not for use in diagnostic procedures.
"Can you walk us through the journey of being featured in Nature Medicine, and how this publication validates the work at PathAI?"
It’s encouraging to look back at the beginning of our journey in the MASH space at PathAI and to see just how far we’ve come. When we first started, digital pathology in MASH was still very new and not widely leveraged – now the technologies in this space have come such a long way and have become an integral part of histological assessment in many MASH clinical trials. This publication is just one step along our journey, and I think what is most exciting about having our work featured in Nature Medicine is that this story will be spread to readers beyond just the MASH community. It’s a great opportunity to raise awareness of the value and impact of AI-powered pathology.
"How do you see AIM-MASH impacting research and clinical development of therapeutics in MASH?"
There are several points within the clinical development pipeline where AIM-MASH can add value. Anywhere where histological scoring of MASH components is required – including trial enrollment, patient stratification, and endpoint assessment – AIM-MASH can be leveraged to assist pathologists (AIM-MASH AI Assist) or alongside manual pathology to reduce the variability associated with histological scoring (AIM-MASH+).
Another advantage of AI-powered scoring is that it can increase the efficiency in clinical trials. AIM-MASH can be deployed at scale with quick turn-around-times, which has the potential to reduce trial timelines and resources. Ideally, AIM-MASH can help to speed up the process of identifying effective MASH therapies and getting those therapies to market, and therefore to the patients, more quickly.
One limitation of the MASH CRN scoring system is that, sometimes, the ordinal scores may not be sensitive enough to capture changes in histology that may occur over trial timelines. This is where the AIM-MASH+ continuous scores can add value, as they can provide a more granular measurement of histological change. For example, it may be feasible that a patient does not achieve a FULL stage change of fibrosis, e.g., they are a stage 3 at baseline and a stage 3 at the endpoint.However, the continuous scores could reveal that a subordinal change occurred – for example the patient might have regressed from a 3.8 to a 3.1 for instance. So the AIM-MASH+ continuous scores are particularly valuable for MASH research, for example, in retrospective analysis of clinical trial data, to improve understanding of what types of histological changes might be occurring in the treated patients.
"Can you explain the challenges of histologic scoring variability and how AIM-MASH addresses these?"
Unfortunately, after a handful of failed clinical trials, there were a few key studies that brought to light the challenges associated with histologic scoring variability in MASH clinical trials - it became evident that MASH CRN scoring system is subject to both inter- and intra- reader variability. Meaning that we see low rates of agreement between different pathologists and even between the same pathologists when given the same biopsy to score.
Despite these challenges, the FDA/EMA currently still recommends histologic endpoints for MASH clinical trials. All together - we have a challenge and a need that was well-suited for an AI-powered solution which could reduce the subjectivity of MASH CRN scoring. This is what led us to build AIM-MASH.
To address the problem of inter-reader variability, we trained AIM-MASH using inputs from >50 pathologists with expertise in MASH, and a vast and varied dataset, consisting of whole slide images from 6 completed Ph2b and Ph3 clinical trials. The goal was to ensure as much as possible that we weren’t training a model that would recapitulate an individual pathologist’s bias.
Additionally, one major benefit of an AI-powered algorithm is that if you deploy the algorithm on the same slide over & over again, it will produce the same result each time, thus addressing the issue of intra-reader variability.
"How do you envision AIM-MASH being integrated into clinical trials?"
There are 2 options for how AIM-MASH can be integrated into clinical trials:
-
The first is through what we call an AI Assist workflow – where pathologists can use this tool in their liver biopsy reads for primary endpoint assessment in MASH clinical trials. In this version, the algorithm shows colorized overlays which aid the pathologists in identifying key histologic features and also produces an AI-derived reproducible/accurate ordinal score, which the pathologists can accept or reject if they disagree by 2 points or more. This format is what has been submitted to the FDA/EMA for consideration as a drug development tool (DDT)
-
The other option is to use the algorithm on its own, with NO pathologist intervention. In this case, one would deploy AIM-MASH+ on their study images as a secondary/exploratory endpoint to essentially have an additional, unbiased, read (using the AIM-MASH+ ordinal scores). The major benefit of this approach is that it can add confidence to the manual scoring results. It allows sponsors to take another look at their dataset with an unbiased platform. Given everything at stake in these trials, having a second (or third or fourth) opinion is immensely valuable. In this use case the AIM-MASH+ continuous scores can also be analyzed to measure more granular changes in histology.
"What potential does AIM-MASH have to improve patient outcomes in MASH?"
Firstly, AIM-MASH can help to ensure that we are getting the right patients into the trial. If it’s meant to be a non-cirrhotic trial, AIM-MASH can be used at enrollment to aid in ensuring that only F2/F3 subjects are included. Secondly, AIM-MASH can build confidence in the endpoint assessment to ensure that the results reflect, as much as possible, the true drug effect.
The end goal is to make sure we can help in getting the best treatments to MASH patients who still need them. By adding AIM-MASH analysis to one’s therapeutic development pipeline, we can ideally improve confidence in the scientific community who tests and approves these drugs and in the patients who will be the ones actually prescribed these therapeutics.
"What insights does AIM-MASH provide into those histologic features critical to evaluating MASH?"
AIM-MASH provides colorized overlays or “heatmaps” which spotlight the key histologic features of MASH on these whole slide images. On the H&E-stained images, the algorithm highlights regions of hepatocellular ballooning, lobular inflammation, and steatosis. On the trichrome-stained images, the algorithm highlights regions of fibrosis. These overlays can help bring the pathologist’s attention to these key regions of interest.
We now have another algorithm product called Liver Explore™, which not only identifies key tissue regions, but also individual cell types, lobular zones, and fibrosis subtypes. This tool can enable even further insights into the microarchitecture of liver biopsies, beyond the limitations of the scoring systems.
"Can you share any case studies or examples where AIM-MASH has demonstrated significant benefits?"
As part of our submission to the FDA and EMA for the AI-Assist workflow, we completed analytical and clinical validation of the AIM-MASH tool. We demonstrated that AIM-MASH showed non-inferiority against a panel of pathologists when comparing both to a ground truth panel of pathologists.
In a trial where both histological endpoints were met successfully - for Madrigal’s ph3 for Rezdiffra - we recapitulated the primary endpoints (MASH Resolution and Fibrosis Improvement) with AIM-MASH - corroborating Madrigal’s results with an independent, unbiased, methodology.
Similarly, in a recent publication with Novo Nordisk, we recapitulate the MASH resolution endpoint for their Ph2 trial of Semaglutide. Most interesting, we found a significant improvement in continuous fibrosis from AIM-MASH where ordinal scoring by manual pathology was not significant, which demonstrates the value of the continuous scoring metrics provided by AIM-MASH+.
"What areas of future research do you think AIM-MASH will open up in liver disease and pathology?"
Other liver diseases might require fibrosis staging for clinical trials (PSC, AATD) – while AIM-MASH is built off of the MASH CRN scoring system, we can use the same approaches to make similar scoring tools for other diseases.
Currently, we do have an algorithm product called Liver Explore, which is designed to glean insights into the microarchitecture of the liver biopsy to enable exploratory research in liver diseases beyond MASH, like alcoholic liver disease, and primary sclerosing cholangitis.
