Transforming Medicine: TMed Will Match People with Life-Threatening Illnesses to Best Treatments, in Real Time

September 2015, Vol 6, No 8
Robert Goldberg, PhD
President and Co-Founder
Center for Medicine in the Public Interest
Springfield, NJ

Transforming Medicine: The Elizabeth Kauffman Institute (TMed), a research organization dedicated to using systems biology, big data, and artificial intelligence to determine the best treatments for patients facing potentially fatal illnesses, was launched in July 2015 in New York City. The institute is named for Elizabeth (Liz) Kauffman, the wife of TMed’s Chair, Stuart Kauffman, MD, a physician and the father of the computational biology used to establish the connection between how genes interact and why people get sick. Liz was diagnosed with a pancreatic adenocarcinoma in 2011.

Dr Kauffman wanted to find something other than the standard treatments that Liz was receiving. In November 2012, as she was getting worse, Dr Kauffman found that doctors at the Moffitt Cancer Center in Tampa, FL, demonstrated that in some solid tumors, the cells in the acidic areas grew faster than those in the more alkaline areas. (Tumor cells seek out more acidic environments, so there was a biologic rationale to try alkalization to curb tumor growth.) In fact, treatment with mere sodium bicarbonate slowed the growth in the acidic areas in animal experiments.

But not all tumors are equally sensitive to alkalinization. Measuring Elizabeth’s tumor-cell metabolome would have revealed whether her tumor could be suppressed by controlled alkalinization. Dr Kauffman showed the preclinical data to his wife’s oncologist, who dismissed alkalinization therapy, because it hadn’t been tested in randomized controlled trials (RCTs). Dr Kauffman ignored the doctor, and Liz received intravenous cesium chloride, a powerful alkalizing agent, under another doctor’s care, and it might well have extended her life by several months.

Liz was given the best evidence-based medicine available, but such evidence (ie, from RCTs) measures the average response to 1 drug for a large population presumed to have the same disease in the same way.

RCTs ignore hordes of relevant data about the variability of response and the multiplicity of causes at the genetic and phenotypic levels when the average patient populations are averaged. As a result, in medicine, we know only what works for the majority of patients and not what will work for the individual patient. In turn, many drugs do not work for many people (Table), which leaves huge gaps in treatment that cost time, money, and lives. Dr Kauffman realized that this hurt Liz and countless other patients.

Table

In short, the 1-drug/1-disease model of research, regulation, reimbursement, and treatment is outdated. In non–small-cell lung cancer, genetic, environmental, behavioral, and demographic factors regulating well-being have turned it and many other conditions into multidimensional diseases for which no single drug can be successful for all patients. Cells, tissues, organs, and organ systems are networks of causal interactions that are best treated by smart combinations of drugs, nutriments, and other strategies, such as immune therapy. Yet, many researchers, drug manufacturers, and regulatory agencies are organized around the 1-drug/1-disease model.

Dr Kauffman had shown nearly 40 years ago that biologic evolution is not linear. Rather, genes, the working subunits of DNA, form and re-form networks, creating countless networks that search for the best path along “fitness landscapes.” Dr Kauffman believed that the clinical fitness landscape of illness is more like a mountain range, whereas most of medicine is based on a 1-target/1-drug/1-disease model for the average patient.

A disease landscape is rugged. According to Dr Kauffman, it has multiple peaks separated by plateaus and valleys, which is more akin to climbing the Rocky Mountains than Mount Fuji. Simply climbing uphill to the highest point will not lead a hiker to the highest peak in the shortest amount of time. Rather, it will require multiple climbers sharing information and rerouting themselves to reflect the complexity of the landscape. Disease is no different. It is the result of a dynamic interaction of genes that, although complex, can be predicted mathematically.

Dr Kauffman’s models are routinely used to predict how complex interactions of genetic mutations form networks that cause disease. He wondered if the same type of predictive model could be developed to predict the right combination of treatments. Dr Kauffman decided to test his hunch with Maggie Eppstein, PhD, MS, Founding Director of the Complex Systems Center, and Jeffrey Horbar, MD, Professor, Department of Pediatrics, both of the University of Vermont, Burlington. They carefully designed a mathematical model and computational simulations to compare RCTs with an alternative approach called “team learning.” This approach generates new treatment combinations based on the current success data, the rapid diffusion of information, and the opinions of those hospitals with the best result (and combination of treatments).1

When there were only 1 or 2 factors influencing disease, RCTs did about as well as team learning in predicting outcomes. But as more factors were added to the model, the RCT approach failed dramatically, and the team-learning approach (which approximated the network effect that Dr Kauffman had established and operates at the genetic level) was more predictive. Moreover, the team-learning approach using nearly anecdotal data required less time and money than RCTs and was more predictive in complex environments.

TMed will improve the selection of treatments by adding back and analyzing all the relevant data about the variability of response and the multiplicity of causes at the genetic and phenotypic levels from thousands, if not millions, of patients (Figure 1). It will create personalized multidimensional data clouds of each individual and then use artificial intelligence to identify the patients with similar features. It will use the analytic results to develop highly predictive models and software to match the treatments, including combinations of treatments, to the right patients (Figure 2).

Figure 1
Figure 2

Matching Patients to Treatments

To set up TMed’s analytics platform, Dr Kauffman has turned to 2 friends who are world leaders in systems biology and the development of predictive disease models using artificial intelligence and machine learning—Lee Hood, MD, PhD, President and Co-Founder, Institute for Systems Biology, Seattle, WA, who helped usher in the era of genomics and biotechnology; and Colin Hill, MSc, a former student of Dr Kauffman’s and Co-Founder, Chairman, and Chief Executive Officer of GNS Healthcare, the leading big healthcare data analytics company.

Dr Kauffman, Mr Hill, and Dr Hood are joined by the author of this article, and by Kathleen O’Connell, an Emmy award–winning producer at CBS News. Ms O’Connell lost her 14-year-old son, Lucas, to diffuse intrinsic pontine glioma, a type of brain cancer. She, too, struggled to get Lucas more than only the RCT-based standard of care. Ms O’Connell promised Lucas before he died that she would do whatever she could to save the lives of other children with brain tumors.

As Mr Hill noted, “With enough data from enough people—we are talking hundreds of thousands, and sometimes, even millions of patients—we can apply analytics to build predictive models to discover which interventions will work.”2 That may seem to be an insurmountable task, but Mr Hill, Dr Hood, and Dr Kauffman have been developing the tools to do just that. Moreover, the rapid decline in the costs of data, the soaring increase in computing power, and the advances in machine learning and big data analytics now make it possible to use the experiences of a multitude of patients regarding treatments and unique and common patient characteristics to determine the best treatment for others.

In addition, consumers are increasingly deciphering biological complexity to match people to combinations of treatment via the use of social networks, multidimensional data clouds, and digital tools. They are bypassing large clinical trial networks to get better answers more quickly with individualized approaches to their disease.

Finally, there are sophisticated methods to match patients to treatments and to measure outcomes instead of requiring large randomized trials. “N = 1” experiments (in which individual patients try different treatments) now have a lot of statistical power, because they are based on the predictive and multidimensional analysis of patients.

TMed plans to focus on 4 to 5 diseases, which may include pancreatic cancer, brain cancer, sickle-cell anemia, sarcoma, and multiple myeloma. It will form partnerships with patient groups, physicians, health systems, and other organizations that are interested in accelerating the use of personalized medicine to improve treatment outcomes to acquire data and/or biosample acquisition. TMed will provide data science and computer modeling of the data, and will work with partners to create cloud-based patient–drug matching software.

The software and algorithms behind them will be shared and refined with partners who, in turn, will be able to use them to achieve that goal. Dr Hood emphasizes that the integration of technologies and missions is critical to letting partners “learn one another’s languages and work together effectively in teams.”3

TMed believes that by providing doctors and patients with information that can be used to find the best treatment quickly, it can also give the personalized medicine movement a focus and a platform for measuring impact. There are many efforts focusing on a specific aspect of personalized medicine. TMed is unique in integrating all these tools in its analytical platform and is focusing exclusively on accelerating the adoption of personalized medicine for treating patients.

As Mr Hill noted, “When big data analysis ‘turns the lights on’ to reveal which treatments will work for specific patients, babies are saved from premature birth, cancer patients live longer, chronic conditions are managed better, and billions of dollars are saved by avoiding treatments that do not work for given patients.”4


References

  1. Eppstein MJ, Horbar JD, Buzas JS, Kauffman SA. Searching the clinical fitness landscape. PLoS One. 2012;7:e49901.
  2. Hill C. Can big data save my dad from cancer? December 18, 2012. Forbes. www.forbes.com/sites/colinhill2012/12/18/can-big-data-save-my-dad-from-cancer/.
  3. Hood L. Systems biology and P4 medicine: past, present, and future. Rambam Maimonides Med J. 2013;4:e0012.
  4. GNS Healthcare. GNS Healthcare applauds precision medicine initiative. Press release; February 2, 2015. www.gnshealthcare.com/gns-healthcare-applauds-precisionmedicine-initiative/.

Related Articles