Prospective Observational Study of Diffuse Large-cell B Lymphoma

Overview

Diffuse large B-cell lymphoma (DLBCL) represents the most common type of non-Hodgkin lymphoma and is currently a curable malignant disease for many patients with immuno-chemotherapy frontline treatment. However, around 30-40 % of patients, are unresponsive or will experience early relapse. The prognosis of primary refractory patient is poor and the management and treatment are a significant challenge due to the disease heterogeneity and the complex genetic framework. The reasons for refractoriness are various and include genetic abnormalities, alterations in tumor and tumor microenvironment. Patient related factors such as comorbidities can also influence treatment outcome. Recently the progress in Machine learning (ML) showed its usefulness in the procedures used to analyze large and complex datasets. In medicine, machine learning is used to create some predictive tools based on data-driven analytic approach and integration of various risk factors and parameters. Machine learning, as a subdomain of artificial intelligence (AI), has the capability to autonomously uncover patterns within datasets. It offers algorithms that can learn from examples to perform a task automatically.The investigators tested in a previous study five machine learning algorithms to establish a model for predicting the risk of primary refractory DLBCL using parameters obtained from a monocentric dataset. The investigators observed that NB Categorical classifier was the best alternative for building a model in order to predict primary refractory disease in DLBCL patients and the second was XGBoost.The investigators plan to extend this previous study by further exploring the two best-performing models (NBC Classifier and XGBoost), progressively incorporating a larger number of patients in a prospective way.

Primary refractory disease affects approximately 30-40% of patients diagnosed with DLBCL and is a challenge in the management of this disease due to its poor prognosis. The prediction of refractory status could be very useful in the treatment strategy allowing early intervention. Indeed, several options are now available depending on patient and disease characteristics such as salvage chemotherapy and autologous HSCT, targeted therapies or CAR T-cell therapy. Supervised machine learning techniques are able to predict outcomes in a medical context and therefore seem very suitable for this matter. An approach with machine learning seems particularly interesting because there are currently no statistical models efficient enough to provide decision-making support to clinicians. The investigators showed in a previous study that algorithms can be effective in predicting the refractory status of the disease from structured data from the patient's medical record. Due to the large number of available and effective salvage therapies, intervening quickly in the patient's therapeutic pathway seems to be the right option and the most personalized way to maximize the chances of cure while reducing those of toxicity. Based on clinical judgment of physicians and the best algorithms predictions, the physicians could choose an early treatment strategy for primary refractory DLBCL. The investigators found in a previous study two interesting models (NBC and XGBoost) for predicting refractory disease on the validation set. The application of machine learning techniques can significantly contribute to the management of DLBCL patients. These algorithms hold the potential to assist clinicians in making informed decisions regarding treatment strategies, allowing for the personalization of therapies based on each patient. This study aims to validate these findings on a broader scale in a prospective cohort and the value of this technology in the intricate management of primary refractory disease in DLBCL patients.