Earlier this week, CancerLinQ and Owkin announced a new research collaboration to work together to use artificial intelligence to understand the treatment of non-small cell lung carcinoma, the most common form of lung cancer, with the goal of improving patient outcomes.
CancerLinQ, Owkin, and one or more research sites in Europe will use federated learning, an AI framework that allows data to be analysed without it leaving its source, to analyse real-world patient data from the European research sites and CancerLinQ Discovery, a vast repository of de-identified data from more than 100 hospitals, cancer centers, and oncology practices across the United States.
The research project aims to understand the characteristics of patients who are resistant to certain common treatments, with the ultimate goal of enabling physicians to prevent patients from undergoing unnecessary procedures and direct them towards more suitable options, including potentially enrolling in clinical trials.
Sean Khozin, CEO of CancerLinQ LLC and Executive Vice President of ASCO, and Thomas Clozel MD, Co-founder and CEO of Owkin, share their hopes for their partnership.
What do you hope to achieve together?
Sean Khozin, CancerLinQ: This is an exciting research collaboration that can unlock new insights from CancerLinQ’s de-identified real-world data set, CancerLinQ Discovery, using advanced analytical methods developed by Owkin. The ability to objectively examine the real-world experience of patients with cancer, including those who are not represented in traditional clinical trials, using next-generation analytics is a foundational theme in the evolution of CancerLinQ’s commitment to advancing cancer care quality and research.
Thomas Clozel: We are hugely excited to be working with CancerLinQ to help to make breakthroughs in research methods that we hope will contribute to advances in the way that patients are treated. A key part of Owkin’s mission is to connect the medical world together, and our federated learning project with CancerLinQ and researchers in Europe is a testament to the power of federated learning to enable collaboration at scale in the medical research community. As a former oncologist, I saw how important collective intelligence and collaboration are in allowing physicians to make the best decisions with their patients, and I am excited to help further these together with CancerLinQ.
This will be the first time that federated learning will be used to analyse data from both the United States and Europe. It is an exciting moment in the use of cutting-edge machine learning methods that protect privacy and security and respect organizational autonomy. Together, we hope to make impactful discoveries from real-world patient experiences, that could ultimately improve the treatment of a disease that affects millions of patients across the world.
How can the treatment of non-small cell lung carcinoma be improved?
Sean: Lung cancer is the leading cause of cancer deaths, claiming nearly 2 million lives each year globally. The majority of patients with lung cancer present with advanced non-small cell lung cancer (aNSCLC). Historically, first-line treatment of patients with aNSCLC involved administration of platinum-based doublet chemotherapy with an overall response rate of about 30%. More recently, with the advent of targeted therapies and immune checkpoint inhibitors, patient outcomes have significantly improved.
However, aNSCLC remains an aggressive malignancy with the majority of patients developing resistance to available therapies within the first two years of treatment. Optimizing the treatment of patients with aNSCLC requires a better understanding of clinical outcomes in the real-world as means of developing new strategies for tailoring treatment decisions to the individual needs of patients and identifying opportunities for new therapeutic modalities to address the existing unmet needs.
How can AI help to make these medical breakthroughs?
Sean: AI can accommodate objective and reproducible ways of quantifying the intrinsic and extrinsic variables influencing the outcomes of patients with aNSCLC. For example, when applied to the growing volume of genomic datasets as part of routine molecular profiling, AI can enable the identification of new prognostic and predictive variables to not only improve the efficacy-safety profile of existing therapies but also to surface new mechanistic insights for powering study of a new generation of anticancer therapies.
Furthermore, AI can augment our disease classification schemas in lung cancer that are currently largely based on human visual inspection of biospecimens. AI-derived histopathological features can supplement traditional classificational standards to identify new categories of patients with potentially unique mechanisms of disease enabling the discovery and validation of new therapeutic targets.
What is federated learning and what benefits does it offer to medical research?
Thomas: Federated learning is an AI framework that was developed to solve the problem of access to high-quality data, because AI is nothing without access to rich, multimodal data. Federated learning allows researchers to train machine learning models on data held within hospitals and other institutions, without the data leaving its secure source. Instead of gathering data on a single server, the data remains locked on their servers and the algorithms and models travel between them. Our federated data network covers hospitals and other multimodal datasets across the world – allowing us to train AI models on enough data to make long-term, transformative contributions to medical research and drug development.
What will be the long-term impact of AI on medical research?
Sean: AI is a powerful tool that can be applied to guiding both more personalized treatment decisions at the point of care and the discovery and development of new therapies. AI can advance precision to clinical decision making by allowing for a more nuanced understanding of variables that influence patient outcomes, especially outside of traditional clinical trials where patients are typically more diverse and heterogeneous.
From next-generation sequencing and gene editing to computer vision applications designed to derive new insights from medical imaging (e.g., DICOM and digital histopathology), AI can be considered necessary and foundational to the ability to extract maximal value from our modern tools of research in biomedicine.
Thomas: At Owkin, by working in collaboration with academic researchers and expert clinicians, we have already shown the possibilities for AI to improve biomarker discovery, clinical trials and patient diagnosis. A wide range of oncology stakeholders are seeing first-hand the benefits of AI, especially now that we are able to solve issues that previously held back the sharing of data. Privacy-protecting AI frameworks like federated learning are currently being implemented across the healthcare sector, allowing diverse datasets from many sources to be rapidly, safely and securely analysed. Many of yesterday’s data difficulties are being solved by technology and greater collaboration, allowing AI’s potential to be realised.
We must remember that AI’s power must be combined with the unparalleled insights of our human doctors. AI for medical purposes must be interpretable – defined as whether someone could understand the cause of a decision made – in order to ensure its design and outputs are clinically relevant. This can be achieved by working with medical experts to focus on using AI to improve our fundamental scientific understanding of disease mechanics.
But the fundamental, long-term impact of AI will be to unlock a new era of precision medicine, in which every patient receives the specific treatment that is right for their unique needs, at an early stage. We are excited to be working with CancerLinQ to make an important step in this journey.