ASH Preview: Predicting POD24

I'm continuing my look at ASH abstracts, reading the summaries of the research on Follicular Lymphoma that will be presented at the conference in a few weeks. There is some really nice research happening. Nothing that is going to change things in a huge way. But lots of smaller, incremental progress, I think.

Two presentations showed up on my list next to each other. They both deal with predicting POD24. Lots of researchers have made this a priority. As you probably know, POD24 stands for Progression of Disease within 24 months. Research have found that FL patients who have successful immunochemotherapy (like R-CHOP or R-Bendamustine), but whose disease comes back or gets worse within 24 months, have a lower Overall Survival rate than other FL patients. POD24 was first proposed about 6 years ago from data in the GALLIUM study, and it affects about 20% of FL patients. It sometimes goes by other names like EFS24 (Event-Free Survival at 24 months). Slightly different, but essentially the same concept. 

Interestingly, the two ASH presentations that try to predict POD24 both try to build on the FLIPI model. 

I haven't written much about FLIPI lately, but it stands for Follicular Lymphoma International Prognostic Index.  FLIPI uses data like a patient's age, LDH levels, and disease stage to predict survival. It should NOT be used to determine an individual patient's survival, so if you are tempted to take the "quiz" that is linked above, do not panic if it gives you a bad outcome. It means nothing. As an example -- I just took the quiz and it gave me a 50/50 chance of surviving 10 years. I'll celebrate my 16th diagnosiversary in a couple of months.

FLIP is useless in predicting anything about an individual. It's far too general to do that, and it wasn't developed to do that. Instead, it was developed to help analyze clinical trial results, by giving FLIPI scores to participants so they could measure groups, not individuals. Treat FLIPI like a Buzzfeed quiz that predicts which Disney Princess you would be, because it's about as accurate in predicting your future. (That quiz said I was Tiana, when I'm obviously Belle, though Tiana is a pretty good second choice.)

There have been a lot of attempts to build on and refine FLIPI, adding more specific criteria that make more sense as predictors. A biomarker says a whole lot more about an individual's survival than their age. 

The two ASH presentations use refined FLIPI scores to try to predict POD24. They take very different approaches to developing them.

The first is presented in "1657: The FLIPI24 Prognostic Model Identifies Poor Outcomes in Non-Immunochemotherapy Treated Patients with Follicular Lymphoma." As I say above, POD24 involves FL patients who have received Imunochemotherapy.  This group of researchers is interested in FL patients who have progressed within 24 months who did not receive Immunochemotherapy, hoping to find a way to predict progression in that group of patients. They developed the FLIPI24 model to help them do that. The FLIPI24 measures some specific things -- age and LDH levels (both in the old FLIPI model), Hemoglobin (a protein that help carry oxygen in the blood), White Blood Cell counts, and B2M (a biomarker for a gene that is involved in cancer cell growth and survival). Their research looked at 1542 FL patients who had received different treatments like watching and waiting, straight Rituxan, and radiation (and a few more unspecified). 

The researchers found that their model did a good job of predicting EFS24 (Event Free Survival within 24 months). By grouping patients by risk, they were able to identify high risk patients, and this group had a median Event Free Survival of 1.8 years and a 5 year Overall Survival of just 65.1% (the whole group had a 5 year OS of about 90%). These smaller numbers held whether the patient had high or low tumor burden at diagnosis, and even if they were watching and waiting. In other words, clinical signs at diagnosis might have looked like the disease wasn't too aggressive, but the predictor model suggested otherwise. 

The second presentation took a very different approach. This one is called "3048: Development and Validation of a Machine-Learning Model to Predict POD24 Risk of Follicular Lymphoma."As the title suggests, this group of researchers used Artificial Intelligence (machine-learning) to develop their FLIPI variation, which they call FLIPI-C.  As I have written about before, I am very interested in AI and how it can help cancer patients. Still skeptical, but hopeful.

This group of researchers sued machine learning, a type of AI that can learn to develop algorithms, or ways of making decisions based on statistical patterns. One of the potential good things about AI is that it can look at things in ways that humans can't, because of our own biases, or because looking at the data would be expensive and time-consuming. That's what this group is doing with AI. (I'm being very generous here with AI's capabilities. As I said, I'm still skeptical.)

The researchers looked at 1938 patients with grade 3a Follicular Lymphoma. There are lots of statistics involved here, which is not my strength. But essentially, the researchers looked back at the FL patients and used the AI program to identify features that were common to the POD24 patients and then developed a new FLIPI model from there. They compared their model to other FLIPI variations. 

Interestingly, there is some overlap with the FLIPI24 model above, like measuring LDH, Hemoglobin, and B2M, but also includes things like comparing some white blood cell levels and several more involving lymph nodes (not much detail in a summary of just a few hundred words -- it will be interesting to see all of it laid out in a journal article some day).

What I find most interesting about the two studies is the overlap. They came at them from different directions, and came to some of the same conclusions. That gives me some hope that an improved FLIPI model could help. I'll be very interested to see if anyone does a more thorough job of comparing these two presentations in one of the many commentaries that will come out in the next few months.

Whatever the case, I'm pleased that there is continued effort to figure out early on which FL patients are likely to be POD24. That might mean more aggressive treatment early, or more careful observation, or ideally, the development of new treatments based on newly discovered biomarkers. Some kind if help for this vulnerable group would be wonderful.

I'll keep reading and sharing. Stay tuned.