In 2013, much of the world was introduced to the burgeoning field of precision medicine when world-famous actress, Angelina Jolie, announced she had undergone a double mastectomy to reduce her chances of getting breast cancer.
With that announcement, things like BRCA genes, mutations, and proteins were thrust onto the public stage. Suddenly, precision medicine and predictive treatments weren’t only being discussed in journals or at highly technical conferences, but on morning news shows and around water coolers. The notion that there was a way to possibly predict who’s at risk for developing certain types of cancers was a bold and novel idea for most who are unfamiliar with the field.
Simple single-gene and single-mutation models were the proverbial low hanging fruit, and were an initial onboarding to a new era of medicine. But ultimately, these have not been adequate as a predictive tool on a larger scale. We’ve made many technological advances, and have learned much through clinical studies and trials, yet we still seem focused on these remedial markers. Advancing widespread innovation in precision and predictive medicine, particularly in the fields of oncology and immuno-oncology, requires us to revisit this with an innovative approach.
The Restrictive Cost of Drug Development
Ultimately, what everyone wants is to deliver the right treatment to the right patient at the right time. This starts with drug development and ultimately with successful clinical trials. Yet, the cost of bringing a drug to market is ridiculously prohibitive, and selecting patients for clinical trials has become a major bottleneck.
According to research from the Tufts Center for the Study of Drug Development, the cost of developing an approved drug exceeds $2.5 billion. Much of this cost is incurred during clinical trials. If the sheer logistics of conducting a clinical trial doesn’t give you chills, the cost surely will. Simply finding the right patients for the trial incurs huge expense, and what happens when a patient ends up being excluded or isn’t able to participate at the last minute? Replacing that patient incurs an even higher cost as a result of these delays.
Drug development is also getting more expensive as clinical trials become more complex, require more patients. According to some studies, drug candidates experience higher failure rates in drugs tested on human subjects.
And it’s the patient who pays the price. The cost of new chimeric antigen receptor (CAR) T-cell therapies, for example, approach $500,000 for a one-time treatment. While this is a new therapy and cost will surely eventually go down, many studies peg the average cost of immunotherapy treatments for cancer patients at $100,000 per patient, per year—all for treatments that may or may not work.
The cost of developing market-approved drugs, especially in a growing field like immuno-oncology, is not only prohibitive, it’s unsustainable.
The Problem with Single-Analyte Markers in Immuno-Oncology
Make no mistake: Discovering how gene mutations affect the potential development of cancers is a major breakthrough. But when you consider the wealth of data at our fingertips, we may as well be looking at a single tree in a massive forest.
The fact is response rates to immunotherapy for cancer patients are still relatively low. The most successful rates seen with immune-checkpoint blockade therapies average around 30 percent. Put another way, even at its most successful, 7 out of 10 cancer patients see no response from their prescribed immunotherapy.
This is partly due to the fact that, while we’ve learned a lot, we’re still typically only looking at single markers when developing therapy models. Essentially, we have a “yes” or “no” proposition—either a mutation is there or not—and we’re throwing Hail Marys based on those isolated markers.
We’re far more precise than we used to be, but we’re also far from being as precise as we can be.
Our approach to immunotherapy should mimic what we understand about biology itself. Disease is not the result of one single, large change in a biological system, but rather the culmination of multiple small changes.
Recognizing this will help us reach the next level of precision in medicine. But to get there, we need to start developing therapies and studying disease with a bigger, wider lens, where we can consider every tree in the forest in a meaningful way.
What’s needed is a new predictive system that will reduce costs for precision medicine and allow newer, better standards of treatment to be widely adopted. We need to know which patients will respond sooner and faster, hone in on populations that will benefit the most from targeted therapies, be able to use routine clinical samples and deliver easy-to-interpret reports.
What’s needed is predictive immune modeling.
How predictive immune modeling works: Predictive immune modeling helps to answer a critical question: Which patients will or won’t respond to immunotherapy? Using molecular diagnostics, we can start answering this question before the patient ever receives treatment. This can bring us to a new level of targeted, precision medicine: but it requires key changes to how therapies are developed.
RNA over DNA: Complex biological systems can’t be understood solely through DNA. RNA provides a meaningful, dynamic snapshot that better represents the patient at that moment in time. Using RNA, we can build multidimensional models that take into account not just the presence or absence of a single marker, but the overall progression of the disease, patient lifestyle, and other dynamic factors that allow us to better predict a patient’s response to precision treatments.
Develop health expression models: Health expression models represent multiple facets of biology—looking at both the presence or absence of RNA, as well as the dynamic expression levels that can be influenced by the state of the disease, environmental effects, therapy, and much more. Rather than focusing on a single analyte in isolation to define a cell type, patient, or therapy response, health expression models are a multidimensional representation. In the case of immuno-oncology, immune cells known to be important to and predictive of drug responses are modeled to generate an easy-to-interpret picture of a patient’s immune response, with improved quantitation and limit of detection beyond individual markers used currently. For example, a Health Expression Model of CD4+ cells is used to define the quantitative presence of Helper T cells in each sample. With Health Expression Models of many different immune cell types, the immune response for a patient sample can be fully characterized.
The good news is that early versions of these techniques are already being put to use in clinical research, and are proving effective.
Leaning on molecular assays that seek to identify more and more subsets, researchers recently produced immunotherapy response rates of 48 percent for patients with BRAF-mutant colorectal cancer (CRC). As Scott Kopetz, an associate professor in the Department of Gastrointestinal Medical Oncology and the Division of Cancer Medicine at The University of Texas MD Anderson Cancer Center, puts it: “The theme in all of this is that we’re looking for smaller and smaller subsets.”
The more dimensions to the model, the better.
In the case of invasive breast cancer, the Oncotype DX genomic test is helping to usher in a major shift in the treatment of early-stage-estrogen-receptive breast cancer. Studies show that utilizing the test makes treatment more cost effective, and “will spare a large proportion of patients with unnecessary over-treatment, while assuring that those patients with unfavorable biology will receive the whole gamut of available therapies to try to provide most survival benefit as possible.”
Predictive immune modeling: Predictive immune modeling leverages RNA-based health expression models of immune response to more accurately predict clinical outcomes in the context of disease and treatment. In this approach, patient cohorts with clinical outcomes (e.g., responders and non-responders to therapy) are evaluated via machine learning to create a Predictive Immune Model. This model captures the salient characteristics that most accurately differentiate the two cohorts. A predictive immune model for responders and non-responders can be used to determine which cohort a new patient is most likely to belong and their clinical outcome.
Evidence of the benefits to adopting predictive immune modeling keep piling up, yet we’re still facing the same challenges we’ve been facing for the last decade. We need both to pay attention to the data and adopt a foundation built on health expression models for immuno-oncology to flourish, and we need to share in our collective successes.
When Amazon built a robust, yet plastic IT infrastructure for its e-commerce website, it didn’t tuck its findings away. Instead, it opened up its technology as Amazon Web Services (AWS), revolutionizing cloud computing. As a result, the industry and technology in many fields has improved exponentially.
As immuno-oncology stands on the cusp of a similar revolution, standardizing processes and technologies isn’t going to be easy. We can turn precision medicine into predictive medicine and vice versa, creating models and multidimensional markers which truly allow us to reach the patient with the right treatment at the right time—but we’ll only get there through shared insight and going through the hard work of building a new platform that is better, faster and more cost effective than the last one.