Treating cancer with the power of the cloud

Cancer Commons is a nonprofit network of patients, physicians, and scientists dedicated to helping patients identify and access the best personalized treatment options. Erika Vial Monteverdi, executive director of Cancer Commons, describes how the Amazon Web Services (AWS) compute infrastructure, combined with services like Amazon Comprehend Medical, enable physicians and patients to leverage the collective knowledge of the world’s top institutions. 

At Cancer Commons, our mission is to be the center of wisdom for patients with advanced cancer. We arm each patient with the knowledge they need to help them achieve their best possible outcome. Over the past decade, we have helped nearly 10,000 people in 75 countries. We aim to reach 100,000 people by 2025.

Metastatic cancer is one of the most devastating experiences one can face. People with cancer may be overwhelmed and confused by their treatment options. Often, they’ve already tried several treatments. Cancer Commons helps patients and caregivers navigate a cancer diagnosis by helping them understand their disease and informing them about additional testing and treatment options available. The options are curated by expert scientists and oncologists and refined based on actual clinical results. When necessary, we convene virtual tumor boards (VTBs), where nationally recognized experts in a patient’s specific cancer will refine the options based on an individual’s medical history and preferences. We also facilitate access to the recommended tests and treatments through clinical trials, expanded access, or reimbursement support.

Cloud computing turns data into impact and research into life-saving treatments

Every person’s cancer is unique and requires an individualized approach to treatment. The best treatments involve intelligent combinations of therapies, and there are far too many plausible regimens to test in clinical trials. Cancer Commons and our technology partner xCures are solving this problem by integrating clinical research and care to continuously learn from all patients, on all treatments, all the time. This involves a lot of computing and data.

At first, our applications spanned multiple cloud service providers. We quickly realized the benefits of having all of our applications on one infrastructure. Now, almost every part of our process is powered by AWS in some way. For example, the patient data that flows into our system from electronic health records uses natural language processing (NLP) that relies in part on Amazon Comprehend Medical and our own proprietary systems.

Throughout the oncology community, physicians reach out to colleagues to seek their opinions about how to treat a patient. This can be done informally via phone, email, or text, or in a more organized fashion, such as our VTBs. Unlike traditional tumor boards, where physicians at one institution meet in person to discuss their challenging cases, our VTBs include experts from multiple institutions and multiple disciplines. They meet asynchronously, using a Slack-like platform to track simultaneous discussion threads covering multiple cases and treatment options. The group’s collective reasoning about treatment options and rationales is captured and used to refine our options library. In this way, it can inform the treatment of other similar patients.

We developed our VTB system on AWS, using a HIPAA-compliant architecture. We also rely on Amazon Relational Database Service (Amazon RDS) for the underlying PostgreSQL database services, so we don’t need to worry about backup and replication. We plan to move this process to Amazon Neptune as we scale to take advantage of its persistent, performant graph database.

We use adaptive Bayesian models, trained on treatment rationales and outcome data, to predict the safety and efficacy of a given treatment on a patient or cohorts. All of our cancer treatment modeling runs on AWS. The discussion that happens in the VTBs is mined, using natural language parsing and other artificial intelligence (AI) techniques, and a recommended treatment plan is created from that information. A Bayesian hierarchical model of disease progression is created, where the structure is driven by the terms from the VTBs. Our model is trained using data that comes from both the external medical literature and the treatment histories of those patients who have gone through VTBs and who have generated follow-up data. Training these models is a compute-intensive task. While we currently do this on standard AWS instances, we expect to fine-tune this using GPU clusters and tools like Amazon SageMaker.

Once a treatment plan is chosen, Cancer Commons continues working with the patient to capture which treatments they chose to pursue, their rationale for those decisions, and the clinical results. All of this information is fed back into our system to refine our models and improve future treatment recommendations. Our goal is to learn from every patient’s experience and use that knowledge to help the next patient.

Whether it’s raw computing, specialty services such as NLP, specialty computing such as GPU, or optimization services such as Amazon SageMaker, AWS has the tools and services we need. Thanks to AWS, we are helping accelerate cancer research and save lives.

Learn more about the cloud for nonprofits.

The content and opinions in this post are those of the third-party author and AWS is not responsible for the content or accuracy of this post.