Curious Dr. George | Plumbing the Core and Nibbling at the Margins of Cancer

The Challenges of Using Artificial Intelligence to Improve Cancer Treatment

Razelle Kurzrock, MD, and Jeff Shrager, PhD

In a previous post, CureMatch co-founder Razelle Kurzrock, MD, told us all about her company’s artificial intelligence (AI) platform that matches patients with treatments based on their cancer’s molecular profile. Here, AI expert Jeff Shrager, PhD, responds, and Kurzrock offers a rebuttal.

Shrager is Co-Founder and Director of Research at xCures, and was formerly Director of Research at Cancer Commons. He is also an Adjunct Professor of the Symbolic Systems Program at Stanford University. Email:

Kurzrock is Director of the Center for Personalized Cancer Therapy and the Rare Tumor Clinic at U.C. San Diego, and Co-Founder and Board Member of CureMatch, Inc. Email:

Shrager: Whereas I applaud Dr. Kurzrock and CureMatch for their efforts to apply machine learning in precision oncology, I want to offer a bit of a heads-up.

Whereas it is certainly true that “we live in the ‘big data’ generation,” two senses of that term are often conflated. Google and Facebook have enormous datasets with many independent observations across relatively few features. Medical data, especially at the molecular level, is exactly the opposite, having relatively few independent observations across an enormous feature space. Moreover, the settings in which modern AI (i.e., machine learning) has seen successes are those where there are either existence proofs of a solution, which can be drawn upon as a teacher, (e.g., self-driving cars, where even 16-year-olds drive cars adequately well), in closed systems for which we have excellent simulators (e.g., astrophysics), domains in which the roles are static (e.g., games), or in which experiments are basically free (games again, or any domain with a good simulator).

Medicine is completely different: We have essentially no simulations, medical experiments are extremely costly, we lack good treatments (which is why we’re bothering with this at all), and the treatment space changes rapidly. You can’t just teach your robot doctor to cure cancer by observing good doctors curing cancer, because there are no such doctors and cures—there may be some better and some worse doctors, but as far as I know, there isn’t one that can cure cancer “adequately well” who you can use as a guide; indeed, there may be no cure for cancer at all.

Heads up! Machine-learning applications in domains like medicine, where there are small numbers of samples that range over very high dimensionality feature spaces, and with the above-enumerated limitations, are exceedingly prone to getting stuck in non-optimal minima, preferring solutions that work well enough, over exploring solutions that might work better than the ones that have been observed or tried. The way out of this problem is active learning: Rather than taking the apparently best action in all cases, one must balance the strength of belief in one’s rankings against the information gain of trying something new. Doing this requires having a global view of the whole medical (or at least oncological) space, and working out some very difficult “statistico-ethical” questions. Indeed, this is what the clinical trial system is striving to do, although it is doing so horribly inefficiently, and will basically never get there. We can solve this problem, but it requires a much broader AI approach than simply treating each patient in accord with a locally-optimal solution.

(This commentary abbreviates the argument made in much greater detail in a paper I wrote with my colleagues at xCures last year for The Journal of Law, Medicine & Ethics: Is Cancer Solvable? Towards Efficient and Ethical Biomedical Science.) 

Kurzrock: I would like to thank Dr. Shrager for highlighting two excellent points pertaining to the use of AI in routine oncology practice and the inefficiency of clinical trials—I fully agree with him. Allow me to provide some brief comments.

First, I concur that current AI-containing software platforms are certainly not sophisticated enough to be “robot doctors” that could treat cancer. Indeed, decision-support platforms like CureMatch’s BionovTM are not here to replace oncologists. They are necessary tools that help oncologists process immensely complicated data, such as that revealed by next-generation sequencing of tumors. Decision-support platforms are rule-based systems that enable evaluation of complex information by utilizing prior knowledge, akin to the dimensional origami model Dr. Shrager referenced in his earlier work.

Moreover, some of the work that lends confidence to the decision-support platforms are clinical trials. I agree with Dr. Shrager’s point regarding clinical trials’ extreme inefficiency, the fact that they are indispensable to clinical oncology research, and the concept that new clinical strategies are needed, especially to address the questions raised by today’s precision medicine that utilizes complex molecular diagnostics. For example, the prospective cross-institutional I-PREDICT study demonstrated the value of customized, matched combination therapies (rather than scripted monotherapies) and of a matching score similar to that used by BionovTM. Other efforts, such as obtaining real-world data via a Master Observational Trial are also unique approaches that enhance the clinical trial process.


Copyright: This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.

Curious Dr. George | Plumbing the Core and Nibbling at the Margins of Cancer

Using Artificial Intelligence to Match Combination Targeted Therapies in Oncology

A Q&A with Razelle Kurzrock, MD, Director of the Center for Personalized Cancer Therapy and the Rare Tumor Clinic at U.C. San Diego, and Co-Founder and Board Member of CureMatch, Inc. Email:

Q: The new understanding of many cancers brought about by molecular testing has led to a whole new field: precision oncology, which emphasizes targeted and immunotherapy. While promising, and sometimes spectacularly successful, targeted monotherapy has limitations. The evolution of targeted and immunotherapy by combinations of drugs offers new scientific options for cancer patients. But there are so many new molecular findings, so many new investigational drugs or drugs newly approved by the U.S. Food and Drug Administration (FDA), and so few appropriate patients, that matching patients to best drug combinations can be a mathematical nightmare. What have you and your company CureMatch to offer to help with this dilemma?

A: Thank you for this excellent question. As you correctly noted, tumors, even those that share the same histologic origin, are highly heterogenous and unique at the molecular level. Therefore, the existing paradigm of treating all cancer patients based on their tumor’s tissue of origin, even by adding minimal biomarker stratification criteria, has proven largely inadequate. The advent of molecular diagnostics allows for improved patient stratification during therapy selection; however, most patients are still treated with monotherapies, which ultimately perform poorly.

Early data show that individualized matched combination therapies targeting most of a patient’s druggable aberrations are associated with improved outcome. However, selecting the “right” combination in routine oncology practice could be challenging. The average oncologist is pressed for time, seeing approximately 350 new patients annually and up to 100 patients per week. To complicate matters, even if an oncologist wanted to rationally combine only the approximately 300 FDA-approved “oncology-specific” drugs, there would be estimated 45,000 possible two-drug and approximately 4.5M three-drug combinations. Even molecular tumor boards found in some academic centers rely largely upon expert knowledge and experience to tailor personalized combination treatment strategies for hundreds of patients with unique molecular profiles. Clearly, the drug selection process is rapidly outpacing human capabilities, and software tools are needed to help with data analytics.

Bionov™, a rule-based artificial intelligence platform developed by CureMatch, utilizes the latest data available on targeted, immuno-oncology, hormone therapy, and cytotoxic agents. Bionov™ employs an algorithm that matches patients with monotherapy and multidrug regimens based on their available tumor “omic” profile. Drug regimens provided in the Bionov™ report are ranked using a predictive “Bionov™ score” that reflects the degree to which a given regimen matches the patient’s molecular profile.

To generate our database, we curated all FDA-approved drugs relevant to oncology for their biological impact on their targets. Recently, we added oncology drugs that have been approved by the European Medicines Agency (EMA) to our database, and FDA/EMA-approved drugs are kept up-to-date based on their respective labeling changes. Further, we researched and curated preclinical and clinical literature pertaining to the efficacy of these drugs, including drug toxicities and contraindications. The CureMatch scientific team conducts literature reviews on a routine basis to ensure drug efficacy is kept current.

Our methodology has been validated in several studies, and I will highlight two of them here. First, in a retrospective meta-analysis of 70 exceptional responders for whom molecular profiling data was available, Bionov™ correctly ranked the response to all treatment regimens (including failed regimens) with 84% sensitivity and 77% specificity. This analysis demonstrates how the Bionov™ algorithm is able to discriminate, solely on the basis of the molecular fingerprints of a patient’s cancer, treatment regimens that favor a positive outcome from those that are more likely to be associated with an unsuccessful response. The second study I want to highlight is a prospective clinical trial: our group found that a higher matching score (similar to the Bionov™ score) was an independent predictor of increased disease control rate, prolonged progression-free (PFS), and overall survival rates. Furthermore, PFS was significantly improved in 75% of patients treated with combination therapies based on high matching scores.

We live in the “big data” generation. As the patient progresses through their treatment journey, massive amounts of actionable molecular information are generated, and clinical oncologists may not be entirely prepared to effectively utilize it. We believe that predictive analytics models—such as Bionov™—can provide an alternative framework for modern clinical practice, collaborating with and empowering oncologists in their decision-making process.


Copyright: This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.

Curious Dr. George | Plumbing the Core and Nibbling at the Margins of Cancer

Comprehensive Molecular Testing Needed for Stage IV Lung Cancer

A Q&A with David Spigel, MD, Chief Scientific Officer, Director of the Lung Cancer Research Program, and Principal Investigator at Sarah Cannon Research Institute. Email:

Q: You are an expert medical oncologist with particular interest in lung cancer. The various forms of lung cancer are serious diagnoses, all potentially lethal malignancies. There are theoretical, investigational, and clinical justifications to perform molecular testing of these tumors. In your opinion, should such testing target specific mutations, panels of genes, or use next-generation sequencing (NGS) for whole-exome or genome analysis? At what point in a patient’s disease should molecular testing be performed?

A: Caring for patients with lung cancer today requires broad NGS at diagnosis for stage IV disease. There are multiple potential targets, and spot testing for individual mutations is simply inefficient in my view. We need to test for mutations in EGFR, ROS, ALK, MET, and BRAF—and also TRK and PD-L1. HER2 is nearing similar importance, and others are not far behind. I need to know about mutations in these genes as soon as possible to make treatment decisions, not the least of which is deciding whether a patient will qualify for a clinical trial.

Q: What challenges do you and other oncologists face in getting the molecular tests you need?

A: It’s a bit ridiculous that we have local and “send out” testing, and each can be imperfect if the labs are not using NGS. Currently, blood has become the easiest for me to get the day I meet a patient, and it takes five to eight days for results. It gets the ball rolling so to speak. I have tissue tested by commercial vendors, but those results can take three to four weeks—and that’s simply too long to make treatment decisions for a lot of folks (and me).

In the future, we will need a one-stop shop that offers the best-in-class technology in the shortest amount of time with the least amount of material. I bet that will be blood. And lung cancer is just the first malignancy that makes the strongest case for broad upfront testing; there’s no reason this won’t be also true for every cancer we treat one day.


Copyright: This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.

Curious Dr. George | Plumbing the Core and Nibbling at the Margins of Cancer

Using Molecular Testing to Guide Treatment for Advanced Colorectal Cancer

A Q&A with Kalpana Kannan, PhD, former Scientist at Cancer Commons

Q: Colorectal cancer is common, and although many cases in earlier stages are cured by surgery alone or with adjuvant chemotherapy, it is still a lethal threat for many patients. Nonetheless, several new targeted and immunotherapeutic agents are now available. When should patients receive molecular testing for their colorectal cancer, what information should especially be sought, and which therapeutic agents are likely to be effective?

A: For every patient who is diagnosed with stage IV colorectal cancer (mCRC), complete molecular profiling of their tumor is highly recommended. It is important for patients to know their microsatellite status, RAS (KRAS and NRAS) and BRAF mutations, and HER2 amplification status at a minimum. Depending on these molecular profiling results, targeted therapy or immunotherapy may be applicable.

Microsatellite status refers to the status of short tandem repeats of DNA that are present throughout the human genome. Since microsatellites have a repetitive sequence, they are prone to mutations. These mutations are usually corrected by the DNA mismatch repair (dMMR) system. Tumors with a defective dMMR system that cannot adequately repair the mutations have microsatellites of different lengths than would be found in the germline DNA. This molecular phenotype is called microsatellite instability (MSI). Tumors are generally categorized as MSI-high (MSI-H), MSI-low, or microsatellite stable (MSS).

MSI-H or dMMR tumors are very sensitive to immune checkpoint blockade. The U.S. Food and Drug Administration has approved the immune checkpoint inhibitors pembrolizumab and nivolumab for treatment of patients with MSI-H/dMMR mCRC following progression with chemotherapy. Recently, the immunotherapy combination of nivolumab and ipilimumab showed an overall response rate of 60% in the front-line setting (before any chemotherapy) in a clinical trial with 45 patients. So, immunotherapy has an important role in the treatment of this population of patients who seem to not benefit as much with conventional chemotherapy.

For most other patients, in fact 95% of mCRC patients whose tumors are MSS or MMR-proficient, single-agent immune checkpoint inhibitors are not advisable. Combination treatments of checkpoint inhibitors with kinase inhibitors (such as regorafenib) and VEGF-targeting agents (bevacizumab) are currently being evaluated in clinical trials and are showing some promising results. In a phase 1b trial of the combination of regorafenib and nivolumab, an overall response rate of 33% was observed in patients with MSS tumors. While this is promising, it is important to focus on targeting the other alterations that may be present in these patients’ tumors.

Most often, these are KRAS mutations, which are present in about 30–50% of CRCs. The most frequent mutations in the KRAS gene are in codons 12 and 13, namely, KRAS G12V, G12C, G13D, and others. Mutations at these positions result in the activation of RAS. Considered to be undruggable for a long time, only recently have efforts to target KRAS paid off. AMG 510 is an oral inhibitor of KRAS G12C. Clinical trial results with this agent in CRC indicate a disease control rate of 92% (1 partial response and 10 stable disease among 29 patients).

Various other strategies to target RAS mutations are currently in trials. These include targeting of EGFR; targeting of the downstream effectors RAF, MEK, and ERK; and targeting of synthetic lethal interactions with CDK4, SHP2, and PLK1.

Copyright: This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.

Curious Dr. George | Plumbing the Core and Nibbling at the Margins of Cancer

Navigating Pancreatic Cancer—The Basics

A Q&A with Lola Rahib, PhD, Lead Scientist, Pancreas Cancer, at Cancer Commons, Los Altos, CA

Q: Navigating a pancreatic cancer diagnosis can be overwhelming and confusing for patients and their loved ones. How can patients and their caregivers ensure having the knowledge, support, and plan they need to be able to navigate treatment options and other aspects of the disease?

A: Patients and caregivers can regain control of a chaotic and anxiety-inducing process by making sure they maintain and organize detailed medical records and information about diagnosis, treatments, and options. As a patient it is critical to advocate for yourself and your needs. If this is not possible, ensure a designated family member or caretaker can advocate on your behalf.

Q: What are the specific aspects of the disease that are most important to navigate?

A: A little over ten years ago, Dr. Brown began a series of preclinical studies to test the possibility that an important contributor to the recurrence of malignant brain tumors after radiation therapy was reconstitution of the tumor vasculature. Specifically, he hypothesized that this reconstitution stemmed at least in part from circulating pro-angiogenic cells not in the tumor at the time of radiation—a phenomenon known as “vasculogenesis.” In agreement with this concept, a finding common to all of the tumor models he tested was a major influx into the irradiated tumors of bone marrow-derived cells, most of which were macrophages, that correlated with when tumors began to grow two to three weeks after completion of radiation. Further, he demonstrated that the mechanism for this influx was a radiation-induced hypoxia that triggered a cascade that led to the secretion of stromal cell-derived factor-1 (SDF-1), which was instrumental in attracting these cells. The apparent importance of excluding these cells’ entry into tumors post-irradiation suggests a new treatment strategy, which we call macrophage exclusion radiation therapy (MERT).

In August of 2014, based on these strong preclinical data, we launched a phase I/II clinical trial of MERT. This study examined the effects of administering a four-week continuous infusion of plerixafor (Mozibil)—the only commercially available agent that blocked the SDF-1 binding receptor CXCR4—at the end of irradiation to newly diagnosed GBM patients (NCT01977677). We enrolled 29 patients and established in phase 1 that the treatment was well tolerated at a dose that resulted in plerixafor serum values being maintained above the threshold level for CXCR4 blockade.

Two findings in phase II of this trial were particularly noteworthy: (i) a persistently lower relative cerebral blood volume within the irradiated field, and (ii) a much-improved control of the cancer in the treated field.

The noted overall median survival of nearly 22 months compared favorably with the best results obtained in other studies of GBM. However, it fell short of the dramatic improvements in survival noted in our preclinical studies, which utilized whole-brain irradiation (WBRT). WBRT was abandoned by clinicians in the early 1990s as a treatment for GBM because the high rate of local recurrence did not seem to justify the associated potential treatment-related issues of irradiating the entire brain (i.e., cognitive decline). However, we have shown that MERT is actually radioprotective for cognitive decline in rats given WBRT, consistent with the fact that tissue inflammation after radiation is related in large part to macrophage entry. Therefore, we have opened a new trial (currently open to accrual)using the same basic strategy in which a modest dose of WBRT has been added. Our expectation is that the widened radiation fields will further patient survival without excessive toxicity.

It is also important to note that the MERT strategy can be applied to any solid tumor in which local control using radiation is challenging. Further study of this strategy can therefore be of benefit to a wide spectrum of cancer patients.


Copyright: This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.

Curious Dr. George | Plumbing the Core and Nibbling at the Margins of Cancer

Serious Caveats in Screening for Pancreatic Cancer

A Q&A with Rama Gullapalli, MD, PhD; a a physician-scientist in the departments of Pathology, Chemical and Biological Engineering at the University of New Mexico. His research lab focuses on the role of the environment in hepatobiliary cancers. He is also a practicing molecular pathologist with an interest in emerging molecular diagnostics, next generation sequencing and bioinformatics. Email: 
Q: A recent New York Times op-ed piece from an NYU Langone Health professor urged an aggressive approach to screening for early-stage pancreatic cancer. Despite optimism, the history of cancer screening is rife with trouble, the harms often exceeding the benefits. What do you think is the best way to proceed?
A: Imagine a scenario.
A new cancer test hits the market with some impressive characteristics: a detection sensitivity of 95% and a specificity of an equally impressive 95%. If you were asked the question, “Given a positive test result, what are your chances of actually having cancer?” and you guessed a number of 80 or 90%, you would not be alone. But you’d be wrong.
The key missing information necessary to answer this question is the disease probability among the general population. The number of new cases of cancer detected every year in the U.S. is about 462 cases per 100,000 people. This means that the probability of a new cancer being detected in a member of the U.S. population annually is roughly 0.00462%. Incorporating this information leads to only an 8.1% chance of having cancer for a test that is positive! This is what is called an inverse probability problem.
Puzzled? Let me explain it in a different way. Statistics show that, in the U.S., about 462 people are newly diagnosed with cancer for every 100,000 people among the general population each year. The new test will correctly pick up 95% of these new cancer patients (i.e., about 439 patients). Of the remaining 99,538 people who do not have cancer, the test will incorrectly diagnose cancer in about 4,977 individuals! This is what pathologists would refer to as a “false-positive” diagnosis. The key point to remember is that cancer is a relatively rare disease. This basic fact enormously influences the value of any given cancer-screening test available in the market.
There has been much optimism and hype associated with cancer screening. Some cancer screening tests, such as tests for colorectal cancer or cervical cancer, have indeed made a dent in our ability to detect and treat the disease at an earlier stage. But in other cancers, such as breast cancer and prostate cancer, the results have been a mixed bag. For instance, screening for cancer in hard-to-access organs, such as ovarian cancer, led to an increase in complications due to surgery with no difference in the cancer outcomes.
A screening test with an increased false-positive rate (think of the 4,977 people in our imagined scenario who had a false-positive test result, but no real cancer), results in unnecessary and invasive testing that is ultimately of no clinical value. However, the societal costs of following up false-positive test results are enormous and include increased downstream testing and increased patient interventions. For patients, an enormous amount of anxiety and stress is expended in resolving false-positive screening test outcomes.
recent New York Times op-ed piece discussed the issue of cancer screening in one such hard-to-treat disease: pancreatic cancer. In response to beloved TV host Alex Trebek’s diagnosis of stage 4 pancreatic cancer, author Diane Simeone, MD, suggests DNA testing as a first step to identify high-risk BRCA gene mutations in potential pancreatic cancer patients. BRCA gene mutations are associated with a higher risk of some types of cancer, including breast, ovarian, and pancreatic cancers. In her op-ed, Dr. Simeone reports that her clinic identified BRCA gene mutations in roughly 15% of the pancreatic cancer patients treated there. The key point is that these mutations were detected in patients who already had pancreatic cancer.
The op-ed piece correctly states the importance of identifying individuals at a higher risk for pancreatic cancer. While it is indeed optimal to screen for these high-risk pancreatic cancer patients, the means by which we can identify these patients beforehand is unresolved and very much a work in progress. One must be especially careful in the context of hard-to-diagnose and hard-to-treat diseases, such as pancreatic, liver, and ovarian cancers.
With the dramatically falling costs of DNA testing, one may be tempted to view it as the silver bullet for early cancer detection. However, the utility of DNA testing for screening purposes in different cancers is unproven currently and needs further research. Patients and physicians must be fully aware of the potential harms of unnecessary downstream testing due to the false positive outcomes of DNA testing. DNA testing may be cheap, but the consequences of DNA testing may prove to be very costly.
Caveat emptor!
Copyright: This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.