Crohn’s Disease Care: Diving into the GEM Integrative Risk Score

Dr. Turck:
Hello and welcome to GI Insights on ReachMD. I'm Dr. Charles Turck, and today, we're diving into some exciting research with Dr. Sun-Ho Lee, specifically his latest study on the GEM Integrative Risk Score, or GEM IRS for short, and how it's helping us predict crohn's disease risk in healthy first-degree relatives. Dr. Sun-Ho Lee is an Assistant Professor and clinician scientist at the Inflammatory Bowel Disease Center at Mount Sinai Hospital in Toronto, Canada. Dr. Lee, it's great to have you with us here today.

Dr. Lee:
Thank you.

Dr. Turck:
So, let's jump right in, Dr. Lee. What got you interested in developing the GEM IRS, and how is it different from other risk models for Crohn's disease?

Dr. Lee:
So far, many research groups have investigated and studied the biomarkers that precede the development of IBD, especially for Crohn's disease. Usually, these biomarkers are just one or two, however, there has been no studies that really integrated all the available biomarkers along with the prediction model that predicts who develops Crohn's and who does not, especially among healthy first-degree relatives.

Dr. Turck:
Would you explain the patient cohort you followed and walk us through how you integrated so many different biomarkers and fecal microbiome data into the GEM IRS? It sounds like a complex process.

Dr. Lee:
This is part of the Crohn's and Colitis Canada GEM project, which spells out Genetic Environmental and Microbial Study. This was started in 2008 by Kenneth, who is a clinician scientist at Mount Sinai Hospital in Toronto and the lead PI of this multi-center study involving many countries in the world, including Canada, U. S., U. K., Israel, Sweden, Australia, New Zealand, etc. We enrolled healthy first-degree relatives of patients who have Crohn's disease. So, these relatives have to be either offspring or a sibling. And they have to be aged between 6 to 35 years of age at enrollment.

So far, more than 5, 000 healthy first-degree relatives were enrolled and prospectively followed for a median of about 10 years. And among them, 120 developed Crohn's disease so far, and interestingly, about 22 developed ulcerative colitis. And so, what we decided with this integrative risk score is we looked at multiple different biomarkers in the past.

The first one was actually something called the lactulose mannitol ratio, which is a urinary biomarker that reflects the gut barrier of intestinal permeability. And we were able to show for the first time that this abnormal LacMan ratio, in other words, altered intestinal permeability and predicts who develops Crohn's and who does not. And this precedes the diagnosis up to two to three years. We also looked at the antimicrobial antibodies in the past, and that was also associated with future development of Crohn's disease in healthy first-degree relatives. But we have never really integrated all the biomarkers together.

Individually, these biomarkers have a decent but not impressive performance for predicting who develops Crohn's and who does not. So, the rationale behind this study was to combine all the available biomarkers in the prior cohort and given the high dimensional dataset of the microbiome data based on 16S, we used a machine learning tool to combine and develop the model.

And in a subset of the cohort, by dividing the cohort into 50 50, especially the North American cohort, which is about 80 percent of the cohort, we developed the model in the discovery cohort. We validated the performance in an independent dataset in the validation cohort. And we also kept the Israeli cohort as a separate external validation cohort. And we were able to show in those validation cohort that this model that we developed showed an impressive predictive performance.

Dr. Turck:
Now, I'm curious about how the GEM IRS stacks up against other predictive models. You started talking about the C index, and I was wondering if you could tell us what that tells us about how effective your model is.

Dr. Lee:
Right, that's a very good question. So, there are several ways to assess how well the model predicts an outcome or event. For us, the outcome is time to event. So, among these different measures, one of the most common ways to assess the predictive performance in a time to event analysis is the concordance index.

And it's a measure which tells you how well the risk model predicts the order of events, so who develops first, who does not. And so, 0. 5 would be almost meaningless, and 1. 0 would be a perfect prediction of the order of events. So, if what we showed is an index of 0.8 in the two-validation course, that is a decent prediction performance, especially in these types of time to event analyses, using machine learning models.

Dr. Turck:
And your study noted that the GEM IRS was consistently linked to Crohn's disease across different groups and conditions. So, what made the model so reliable in this diverse range of subgroups and scenarios?

Dr. Lee:
So, we wanted it to be a very robust performance, and we wanted to first see if the model could still predict the onset of Crohn's disease after adjusting for several confounders that we thought might influence the model, such as the genetic risk score.

Using the genotyping data, we were able to generate a Crohn's disease polygenic risk score. And we adjusted the model with Crohn's disease polygenic risk score, and the effect size did not change, meaning that it was robust to those conditions and to the confounding effect from genetics.

We also included known environmental risk factors that might be a confounder for onset of Crohn's disease, such as smoking history and Jewish ethnicity, and despite adjusting for these confounding effects, the model was able to predict who develops Crohn's disease really well.

We also looked at in other subgroups, such as first-year relatives with normal LacMan ratio and with normal gut barrier function. And it still was significantly associated with onset of the disease. And lastly, we wanted to see how far back can we go and, if among the subset, one year before diagnosis, two years before diagnosis, and up to seven years before diagnosis, can we still predict who develops Crohn's? And the model was robust up until seven years before diagnosis.

Dr. Turck:
What else can you tell us about fecal calprotectin and the lactulose mannitol ratio, and how they fit into the GEM IRS model, and how they're important for predicting Crohn's disease risk?

Dr. Lee:
We can do some analysis to see which variables contribute the most in predicting Crohn's disease and based on this— and this is called a feature importance which is used in these machine learning models— when we ran this analysis, fecal calprotectin And LacMan ratio were actually the factors that contributed the most for predicting Crohn's disease.

And then after those came the microbial composition, the taxa and genus level taxonomy, and also the estimated microbial functions that contributed to the model. However, in the subset where the fecal calpro and LacMan ratio were basically normal, the model was still able to predict who develops Crohn's.

And in this case, what we speculate is that it's not just fecal calpro or LacMan that are contributing to disease, but also the microbial composition and functional pathways that are contributing to predict who develops Crohn's, even in the subset with normal fecal calpro or normal LacMan ratio.

Dr. Turck:
For those just tuning in, you're listening to GI Insights on ReachMD. I'm Dr. Charles Turck, and I'm speaking with Dr. Sun Ho Lee about his study on the GEM Integrative Risk Score, or GEM IRS.

Since your study involved multiple countries, what does that geographic diversity say about how well the GEM IRS performs, and are there any tweaks needed for different patient populations?

Dr. Lee:
That's a great question. For this particular study, we used the North American cohort, and we split the North American cohort into discovery and validation cohort, but we also used a separate Israeli cohort as a validation set. Now I cannot say that this is generalizable for the other countries, and this is something that we should externally validate.

However, I am not aware of any other prospective studies that are following healthy first-year relatives and who have developed Crohn's disease. So, it would be really exciting to validate this externally.

Dr. Turck:
Dr. Lee, do you have anything to say about future directions?

Dr. Lee:
Yes, we need more studies to better define the preclinical stages with integrating not just the biomarkers that I discussed today, but other omics, including proteomics and metabolomics, which might provide more insight, and even improve the predictive performance of the model.

And I just wanted to shout out the PROMIS Consortium, which is an international collaboration that we are a part of to investigate and validate the pre-disease signatures that precede the diagnosis of IBD. And as part of this consortium, collaborative efforts are underway to apply novelomics to these pre-clinical, pre-disease cohorts to better understand the pre disease phase of Crohn's.

Dr. Turck:
And before we wrap up, Dr. Lee, do you have any final thoughts or takeaways you'd like to share with our listeners?

Dr. Lee:
As of now, we do not have an intervention that can prevent or delay the risk of developing Crohn's. However, the GEM IRS, at least in my opinion, shows the potential that by integrating various risk markers, such as fecal calprotectin, LacMan ratio, and the microbiome data, we can start to risk stratify healthy first-degree relatives who are at risk of developing Crohn's with a decent predictive performance.

I think we can also provide an estimation of the probability of developing Crohn's disease. So based on your baseline microbiome, fecal calprotectin, LacMan ratio, and the demographic data, we can start to provide the absolute risk, such as your probability of developing Crohn's disease at one year, three-year, five year, seven years, up to nine years.

And we can also provide the relative risk compared to the median. Although we don't have the intervention yet, our data also supports that it's not just the fecal calprotectin or LacMan ratio, but the microbial functions and composition, which also contribute to onset of disease, providing some information on the potential targets to modify with an intervention.

So, the hope is that these data will be used to design a primary prevention set, recruiting those at high risk of developing Crohn's with an intervention that modifies the contributors that I just spoke about and hopefully show that we can reduce the risk of developing Crohn's disease.

Dr. Turck:
Thanks so much, Dr. Lee, for giving us a peek into your truly fascinating study. It's been such a pleasure chatting with you today.

Dr. Lee:
Thank you so much.

Dr. Turck:
For ReachMD, I'm Dr. Charles Turck. To access this and other episodes in our series, visit GI Insights on ReachMD.com, where you can Be Part of the Knowledge. Thanks for listening.