From the Salvia Jain Lab at Massachusetts General Hospital comes the PETAL Consortium, a global team of T-cell lymphoma experts spanning 19 sites and 10 countries. Focusing on the rarity of this hematologic malignancy, this coalition integrates machine learning and genomics to predict outcomes for newly diagnosed, relapsed, and refractory mature T-cell and natural killer (NK)–cell neoplasms.
Blood Cancers Today spoke with researchers from the Salvia Jain Lab to learn more about the goals and initiatives of the PETAL Consortium. Leora Boussi, MD, Angela Koh, and Jessy Xinyi Han discuss their study of a novel scoring system using one of the largest international global cohorts of 925 patients with relapsed or refractory mature T-cell and NK-cell lymphomas.
Leora Boussi, MD: The idea behind the study is we’re dealing with a rare and heterogeneous group of patients who, in the relapsed or refractory setting, don’t have a standard of care for how to approach their clinical management. We compiled a large database that came together into the PETAL Consortium.
The questions we wanted to ask are: “How do novel single-agent therapies that have been introduced into the disease space in recent years stack up against conventional chemotherapy regimens that have been around for a long time that come with associated toxicity and challenges with administration? Can we understand which patients are going to derive benefit from these novel single agents or what type of novel single agents? Is there a way to prognosticate who is going to do well with a single agent, and can we make sure to make that drug available to that patient?”
Angela Koh: We recognize that numerous prognostic risk factors and scoring systems, like the Prognostic Index for T-cell lymphoma (PIT) and International Prognostic Index (IPI) we use in the clinic, were available and that there are histological subtype-specific prognostic scores for extranodal NK T-cell lymphoma, angioimmunoblastic T-cell lymphoma (AITL), and adult T-cell leukemia/lymphoma patients. But, there’s no scoring system that specifically addresses the variability in survival outcomes for relapsed or refractory patients. So, we aimed to design a new scoring system using one of the largest international global cohorts. Our study is a retrospective global cohort of 925 relapsed or refractory mature T-cell NK-cell lymphomas from 13 institutions representing 10 countries across the six continents. We believe this is one of the largest retrospective global cohorts. They were diagnosed with lymphoma between 2010 and 2021 and had to receive either cytotoxic or chemotherapy or novel single agent as a second-line therapy to be eligible.
For the Methods section, we first split this global cohort into 80% training and 20% testing set. We identified 21 available demographic histological laboratory and radiologic nontreatment characteristics we had in the data, and we identified 11 variables on univariable analysis to be associated with inferior overall survival from the start of second-line treatment. We did the step-by-step selection in the multivariable Cox regression, and with clinical consideration, our final variable model included six of the 11 variables based on the highest concordance index using the testing set. The six variables in the final model included age greater than 60, primary refractory disease in contrast to relapsed status, histological subtypes other than AITL, extranodal sites greater than one, Ki-67 proliferation index greater than or equal to 40, and absolute lymphocyte count around diagnosis being lower than the lower limit of normal. This led to the six-point scoring system named the PIRT score, which stands for Prognostic Index for Relapsed or Refractory Mature T-cell and Natural Killer Cell Lymphomas.
Jessy Xinyi Han: We have this new scoring system with six different scores, from score zero as the minimum to score six as the maximum. We assign a score of 1 to each unfavorable feature. All the clinician needs to do is take the patient profile and look at the six different features and see if the result matches with the unfavorable feature identified in our scoring system. We observed that as the score goes up for the patient, there is a decline trend of overall survival since second-line treatment. This matches with our hypothesis that the greater the score is, the worse the survival outcome is.
Later, the patients are stratified into three different final risk groups. The low-risk group has zero to one risk factor, the intermediate-risk group has two to three risk factors, and the high-risk group has four or more risk factors. We see similarity of survival between the predicted survival and their actual survival. Later, we also compared our performance of the PIRT score with other commonly used scores like PIT score or IPI score using our test set and an external validation set.
The way we do this validation is by bootstrapping the testing sets 1,000 times and then taking the average to see how the performance goes. We see that our PIRT score has a better prediction probability with an average C-index of 0.7 compared to the IPI score, which has an average C-index of 0.56, and the PIT score that has an average C-index of 0.59. We also conducted paired t tests between IPI and PIRT and PIT versus PIRT, and we saw that there is a statistical significance of performance difference. This shows that our PIRT score does better than existing scoring systems like PIT and IPI on the testing set and on the independent set.
We also want to make this scoring system available for clinical use. So, we developed this web-based calculator specifically for the PIRT score that can be available to investigators, clinicians, and patients who are interested in knowing more about the disease. This online calculator is hosted on our PETAL Consortium website. This is a very easy to use tool that will really help clinicians have a more comprehensive understanding of patient survival, especially for relapsed or refractory patients, when they are making clinical decisions.
Leora Boussi, MD: We had several centers throughout the United States, Australia, Brazil, South Korea, South Africa, Saudi Arabia, Japan, Italy, and India. We have representation from all over the world, and I think one question that could come up when extrapolating these conclusions is, “Is this just a United States dataset? Is this localized to a certain area?” An effort was made to include patients from all over the world so that the conclusions could be as generalizable as possible. There’s a need for consortiums like this in rare diseases to take a multinational foot first and include a lot of different centers in compiling data like this so that we have enough power in our studies to come to meaningful and statistically significant conclusions.
This is one big step forward within the field of T-cell lymphoma. This is a rare and heterogenous group of diseases. Using genomic and molecular data and having highly annotated information from a diverse group of patients allows us to ask the important questions that were sometimes a bit limited in clinical trials and other settings where we have fewer patients enrolled on the study. This study frames a new way that questions could be approached and asked within the T-cell lymphoma world outside of clinical trials.
Angela Koh: This is a new calculator specifically designed for relapsed or refractory patients who have inferior survival outcomes than the general T-cell lymphoma population. Since this is the largest global cohort, we were able to do a lot of subgroup analysis that we weren’t able to do with a smaller cohort previously. That takes into account the heterogeneous nature of T-cell lymphoma in terms of treatment regimens and the histological subtypes.
Angela Koh: One of the limitations of our study was missing data. We had a total of 925 patients and had to exclude 162 patients who were not followed up since second-line therapy. Then, 515 were again excluded because we had missing laboratory variables such as Ki-67 and absolute lymphocyte count. We had enough patients to create this calculator, but we hope to overcome this missing data issue in the prospective cohort study that we plan to launch through the PETAL Consortium.
Leora Boussi, MD: In building this consortium involving numerous different countries, there are varied treatment practices in the second-line setting among different academic centers and restricted access to novel single agents in some countries relative to others. There’s also a lack of central pathology and radiology review, and all of those things may impact study conclusions. Our limitations are to be acknowledged, but at the same time, this is a meaningful effort to draw statistically significant and informative conclusions from this patient population. This is something that we have to acknowledge is there, but also still continue with the work of compiling these datasets to get the best information we can.
Jessy Xinyi Han: One very important feature about our datasets is the heterogeneity within the patient cohort. One other challenge is that we don’t have the luxury of having millions of patients’ data. We only have 925 patients to start with. To avoid problems like overfitting, we emphasize having this training testing split and an external validation set to test our scoring system. This way, we can make sure that it not just works within the training set—which is fitted perfectly for the model—but also on the testing set and external validation so that the conclusion will be valid outside our datasets. We also have the online calculator so that the clinicians can start using them, and maybe they will also provide feedback in the future to see how they feel about the calculator and if they have input within new data and other perspectives.
Angela Koh: Mark Sorial, MD, one of our leading investigators at the PETAL Consortium, compared optimal treatment sequences in relapsed or refractory T-cell lymphoma patients, which compared the overall survival since second-line treatment between subgroups that receive different sequences of treatment of either conventional chemo or single agent, such as epigenetic modifiers or small molecule inhibitors. They have 12 different scenarios in the second- and third-line treatment, and he identified the more ideal treatment sequence scenarios in specific subgroups. There is more to come!
Leora Boussi, MD: There’s an additional project looking at time to relapse that is pending manuscript submission as well.
Jessy Xinyi Han: We hope to advance novel machine learning algorithms, like causal survival analysis methods to debias the retrospective data, which often has some data collection bias or other problems. We hope to have more advanced methods that can help uncover the underlying causal mechanism within the datasets and have more accurate estimation prediction from the retrospective data.
Read More: Salvia Jain Lab Integrates Machine Learning to Predict T-Cell Lymphoma Outcomes
Leora Boussi, MD: Ultimately, the goal is to expand this further. In addition to looking at all these questions and data prospectively within the PETAL Consortium, the goal is to expand to additional countries. There is some interest from additional sites to collaborate on this consortium. We’re looking forward to more growth there.
The other component with the prospective lens is that hopefully we can incorporate high-quality molecular annotation for all these patients. That is one thing that will be added in the next phase of this project that was not as well annotated in our retrospective cohort. Ultimately, with this highly annotated clinical and molecular data from a diverse array of countries with a robust number of patients, we’re hopeful that this PETAL Consortium can support a clinical trial network across the world for expedited approval and access to drugs. That would benefit these patients and is one of the challenges within the field.
Angela Koh: We will have exhaustive clinical data in addition to what we had for the retrospective study, including quality of life and pathology confirmed by the pathologist panel, and genomics data confirmed by the genomics experts and incorporation of machine learning. We believe that by incorporating all of these very granular data, we can move on to precision medicine with use of artificial intelligence that will allow us to understand the outcomes and treatment personalization for specific groups to support the clinical decisions in the real world—not just in the United States and sites with great resources, but also in developing countries around the world by sharing our results.
Jessy Xinyi Han: We also want to use the PETAL Consortium as a platform where researchers from different backgrounds can collaborate. People with machine learning backgrounds and clinicians can come together to work on, collaborate, and understand each other’s questions or challenges. This way, we can come together to combat this very rare but dangerous cancer.
Angela Koh: We believe that this is just the beginning of the many multinational consortiums fighting for rare diseases like T-cell lymphoma.
Read More: KLRG1 Cell Depletion Offers Novel Therapeutic Strategy for T-Cell Lymphoma