The oncologist, who runs a community practice in an affluent part of Florida, felt frustrated.
“I’m trying to get my patients in clinical trials. I believe in it. I do all the work to get patients signed up. I send trial samples to a big academic center in, say, New York City. But then it takes three to six months to get them enrolled.”
Two years ago, as he listened to his friend vent, Steven Stein, the senior vice president for United States clinical development at the Swiss pharmaceutical giant Novartis, knew that his friend’s patients don’t usually have that much time to wait. Most of them have cancers that have spread from the primary site to other organs. They need new treatments fast.
“I remember thinking, ‘There has to be a better way,’ ” Stein says. Researchers often bemoan the fact that only 2 to 3 percent of cancer patients participate in research. Stein wanted to figure out how to make it easier for patients to sign up for clinical trials.
Within a year, Stein’s team had designed a clinical trial protocol that turned standard research practices around 180 degrees, launching what it now calls the Signature Clinical Trial Program. Instead of a patient traveling to one of several research sites, Novartis would send the investigational drugs to his or her local oncologist’s office. Instead of testing hundreds or thousands of genetically unscreened patients, the company would accept only patients who had the genetic markers the drugs were supposed to target. Instead of waiting months, patients could access the treatments in two or three weeks. Instead of running a large-scale trial to investigate one or two questions, clinicians could conduct smaller, rapid proof-of-concept studies to quickly rule out the tumor types that don’t respond to a study agent and identify other tumor types that are potentially treatable with the drug and worthy of further study. This way, the study could “learn” without starting all over from scratch, saving time and money.
“People look at Big Pharma and they think we’re not going to do disruptive, innovative things,” Stein says. “I’m proud that Novartis was willing to do it. It’s patient-driven.”
Across the medical establishment, researchers, clinicians, regulators, and patients are beginning to recognize that the era of personalized medicine must mean not only changes in treatment, but also changes in how medical research moves forward.
Signature is just one of dozens, if not hundreds, of experiments aimed at adapting research methods to the new world created by next-generation genomic sequencing. With the growing number of molecular markers for various diseases and the realization that there are many subsets of diseases and many subsets of patients, experts are scrambling to adapt how we answer increasingly specific medical questions. Many of the new trial designs are even forcing researchers to change the kind of statistical logic they use to test their theories.
To understand how revolutionary this is, it helps to remember that actually trying to prove that medical treatments work is a relatively new innovation. For most of human history, doctors relied mostly on observation. They tried something; if it worked, great. If not, the patient probably died. Progress crawled.
In the 18th century, James Lind, a British naval surgeon investigating scurvy, came up with the idea of giving different treatments to different groups of patients and comparing how the patients fared. Yet a hundred years later, medicine continued to exist in a Wild West environment, with thousands of untested “patent medicines” marketed without any proof that they worked. Throughout the first half of the 20th century, the American government tried to bring some sort of order to drug development, with varying success. Clinical trials were developed, but their form was neither mandatory nor regulated. Drug disasters and scandals occurred regularly.
Finally, in 1961, after the worldwide thalidomide scandal, in which a drug given to pregnant women to control nausea caused grievous birth defects in their babies, governments decided to standardize how drug investigations proceeded.
In short, for the last half-century, regulators like the U.S. Food and Drug Administration have basically set this gold standard for randomized clinical trials (RCT):
- The trials need to be randomized, meaning patients are assigned to a particular treatment randomly. The hope is that this will help to eliminate bias from the results.
- The trials need to be double blind, meaning that neither the doctor nor the patient knows which drug a patient receives. Again, the goal is to avoid bias.
- The trials need to have a control group that does not receive whatever is being investigated. That way, researchers can compare whether the outcomes for patients who receive the drug are better than those for patients who don’t.
This system has required huge numbers of test subjects, usually hundreds and often thousands. It demands that patients travel to wherever the research is being done. It can take decades, and it often costs more than $1 billion to bring a single new drug to market. The system emphasizes drug approval rather than discovery. Progress is incremental.
“They test drug combination X against combo Y. They find a 5 percent difference and everybody cheers. But that’s just modest progress. Thousands of patients’ lives are lost in the meantime,” says Matthew Ellis, a professor of medicine at Washington University in St. Louis and author of several papers on innovative clinical trial designs. “My call to action is that we need to sit down and design clinical trials where genomics is the objective, where we get the right kinds of clinical specimens and right kinds of clinical data so you can do proper discovery science.”
The standard RCT, as powerful a tool as it has been, doesn’t work as well in the era of personalized medicine, many experts say. When treatments are designed to target specific molecular markers, it doesn’t make sense to test them on people who don’t have those markers. Test groups may be smaller. The diagnostic for the disease may often be developed at the same time as the treatment.
“God help you if you get a rare disease,” the quip used to be. Researchers are now realizing that all patients have rare diseases because there’s really no such thing as an “average” patient. We used to think that diabetes was one thing or that cancer was one thing or that multiple sclerosis was one thing, but every month, new studies are published showing that diseases come in many subtle variations. Two patients may both have diabetes, for example, but the biological error that is causing the disease in each patient may be quite different.
“From randomized clinical trials, the most data we have is for a 155-pound white male. We don’t do as many studies on women. We don’t do as many studies on ethnic minorities. Yet we’re realizing that people have huge differences in overall metabolism, in how they metabolize drugs, in target sites,” says Andreas Kogelnik, founder and director of the Open Medicine Institute in Mountain View, California. “That’s kind of a big deal. It’s bending us away from cookie-cutter, one-size-fits-all medicine. If a breast tumor has 57 mutations, do you do a trial with 57 arms? Do you do 57 separate trials? I don’t think so.”
So far, most of the innovations in clinical trial design have involved variations on two major “rethinks.”
1. Randomized large samples. It makes sense to test responses on a large group when you’re testing for an average. But it doesn’t make sense to test the drug on everyone if the drug isn’t meant for everyone.
Researchers are now tinkering with the idea of “enriched models.” Rather than testing a large group with the disease being treated, they first screen patients to identify those who have the mutation they’re targeting with the treatment. Then they “enrich” the trial sample with those patients.
For example, there have been big successes in testing treatments with melanoma patients who have a mutation in a BRAF gene that codes for a kinase, an enzyme that acts as a kind of molecular switch. About half of melanoma patients have this mutation, and drugs that target this mutation seem to work better than other chemotherapy in patients whose disease has spread or metastasized.
“We need to start with biology first, not drug first,” says Ellis, of Washington University.
2. Testing one thing at a time. “The problem with the older clinical trial model is that you set yourself up to answer one big question and maybe a couple of secondary questions,” says Joe Miletich, senior vice president for research and development at Amgen, a California-based pharmaceutical company. “But if you don’t reach the end point you’ve set up for yourself, you don’t know what you’ve learned.”
In response, researchers are starting to design “adaptive trials.” Using Bayesian logic, they can design research systems that can learn as they go, adapting according to set standards, depending on results. For instance, a team can test drugs that are in the very early Phase 1 stage of study, and if they don’t work, they can replace them with other agents without having to start over with a whole new trial. In addition, researchers can test several things, or ask several scientific questions, at once by designing the trials to have different sections, or arms. In the United States, there are more than 20 adaptive trials in the pipeline. (The trials in the Novartis Signature protocol are adaptive trials.)
Two of the earliest successes are the I-SPY 2 trial and the BATTLE trial. I-SPY 2, a joint effort of the National Cancer Institute, the FDA, and the Biomarkers Consortium, tests up to 12 cancer drugs in women with locally advanced breast cancer. Participants are assigned to a particular arm based on how patients with similar molecular profiles have responded to various drugs. So far, researchers have logged promising results in two breast cancer subtypes. In the BATTLE trial, based at MD Anderson Cancer Center in Houston, researchers tested multiple drugs on multiple mutations related to lung cancer and found at least one agent that increased survival time for patients with non-small cell lung cancer.
“We need to get more drugs, hundreds of them, in the pipeline,” says Laura Esserman, director of the Carol Franc Buck Breast Care Center at the University of California San Francisco and a principal investigator on the I-SPY 2 trial. “We need to test them in a more organized, fast fashion, figure out the winners, and move on to the next.”
The research innovations that have been tested so far are just the beginning, experts say.
As researchers design the next generation of clinical trials, there are several questions they will try to answer. They need to figure out more ways that doctors in clinics can work easily together. They also need to figure out how researchers, clinicians, and patients can collaborate. Many are experimenting with different models for this. OpenMedNet is an online community for doctors, academics, and patients interested in sharing information and accelerating research. Cancer Commons is trying to pool the day-to-day experiences of doctors and patients in clinics. And CollabRx has recruited experts to curate the latest research and help both doctors and patients access the most recent data.
All this data sharing — much of it at a genomic level, no less — will require new ways to crunch vast amounts of data. And researchers will need to address privacy issues — how to get more patients involved while also protecting their data. They will also need to adapt to smaller numbers of patients but larger amounts of data in each trial.
To really understand the course of disease, many experts say, we need to not only understand the genes, but also the RNA that translates those genes into instructions and the proteins that result from those directions. “We need to understand the protein chemistry and the cell chemistry of cancer. We need to join the dots biologically,” Ellis says. Recognizing this need, the National Cancer Institute has started a Clinical Proteomic Tumor Analysis Consortium to try to collect some of this needed data.
But this question, researchers say, is the biggest of all: How will we answer questions that we might not have even dreamed of a generation ago?
“We’re starting to see not only adaptive trial designs, but adaptive products. How do you do a study on that? That’s adaptive trial design on steroids,” says Kogelnik, of the Open Medicine Institute. “I don’t have all the answers on better trial design, but the key is to be open to other trial designs.”