Photo by Todd McNaught
Researchers must achieve a delicate balance between drug effectiveness and toxicity when they test new medications in clinical trials.
For instance, in considering a new drug for facilitating bone-marrow transplantation, they may find that an effective dose for one patient causes unacceptable side effects for another.
When is it safe to increase the dose? What level of toxicity requires stopping the trial?
The best way to answer these questions objectively is often with a number. That's where a team of biostatisticians steps in.
"One of our most important jobs when we design a study is to translate the scientific question into something we can answer with statistics," said Dr. Katherine Guthrie, the newest member of the Clinical Research Division biostatististics group.
Guthrie joined the team a year and a half ago, responding to a need for further statistical support for clinical trials. The majority of her research involves collaborative projects with clinicians conducting clinical trials.
Though the clinician in charge of a particular study usually collaborates with one biostatistician on the project, each member of the biostatistics team works on many projects simultaneously.
Similar to counterparts in the Public Health Sciences Division, Guthrie and colleagues work with clinicians on projects spanning the three phases of clinical trials, starting with the earliest dose-finding and drug-toxicity studies, progressing to preliminary efficacy trials, and culminating in larger, more controlled studies of drug effectiveness.
Guthrie and her colleagues Drs. Barry Storer, Wendy Leisenring, and Ted Gooley usually begin collaborating with the clinician heading a study while the trial is still in its conceptual stage.
"We come up with a set of rules for how the study will be conducted," Storer said.
In testing drugs to facilitate bone-marrow transplantation, for example, the clinician may be looking for a dose in which the immune cells in the transplant do not attack the patient's own tissue (such as graft-vs.-host disease) nor does the patient's immune system thwart the transplant (graft rejection).
"The drug may prevent graft-vs.-host disease, but if the dose used is too high, there may be an increased chance of graft rejection," Storer said. "In this situation, we can design rules to go up or down in dose, depending on two different outcomes.
"We start in the middle, and if there is a decrease in graft-vs.-host disease, we can decrease the dose, but if there is too much graft rejection, we go up," he said. "So we are balancing two competing endpoints, and we need to keep in mind that the rules also should tell us when there is no dose that is acceptable with respect to both endpoints."
The right balance
Most drugs are potentially toxic if the dose is too high, but if the dose is too low, the drug could be ineffective. Biostatisticians work with clinicians to find the right balance in phase 1 (drug toxicity) and phase 2 (drug effectiveness) trials by establishing rules governing dose increase.
"Phase 1 studies are range- or dose-finding studies," Guthrie said. At this stage, the drug may never have been tested in people. Clinicians and biostatisticians work together to find "a dose not too toxic for the patient," she said.
Is there any way to predict what doses will be optimal before the trial begins?
"We can't predict how a particular study will go," Storer said, "but we can predict, on average, how the rules will behave under different scenarios."
With computer simulations, which provide useful tools for making initial predictions, the guidelines (or "rules") for a drug take into account the fact that some patients may be more sensitive than others to the drug's possible side effects.
If, for instance, the pre-determined threshold was 10 percent of the patients in the clinical trial showing symptoms of toxicity, then the rule remains firm throughout the study.
"The toxicity might seem OK," he said, "but if the rule tells you not to increase the dose, you follow the rules and don't go up."
A constant dialogue between clinicians and biostatisticians ensures patient safety. "Sometimes, we make judgment calls," Guthrie said.
"It is always right to intervene for patient safety," Storer added. "If the clinician is not satisfied with the outcome of a dose because of toxicity, we are not compelled to go forward, even if the rules might permit it."
The trial proceeds, guided by the rules. "And when the study is completed, we'll analyze the data," Guthrie said.
Aided by computer models, biostatisticians convert the numerical data back into clinical terms. The translation of phase 1 and phase 2 trial data establishes important parameters for proceeding to larger, more controlled (phase 3) clinical trials.
Sometimes past clinical trials can provide a wealth of data for addressing current questions.
"The Hutch has extensive databases on all transplants ever conducted," Guthrie said.
By starting with different questions, biostatisticians can interpret data collected in the past, finding new information coded in the statistics.
For example, Guthrie works on this type of retrospective analysis in collaboration with center gastroenterologist Drs. George McDonald and Sangeeta Hingorani, a nephrologist at Children's Hospital and Regional Medical Center. The team combs through data from a clinical trial completed in transplant patients, looking for evidence of kidney complications of a drug originally tested for effectiveness in treating graft-vs.-host disease.
"Often with a retrospective analysis," Guthrie said, "the problem is dealing with statistical variations that seem significant but are actually artifact."
Before working with the figures, then, clinicians and biostatisticians agree on criteria for describing the characteristics of a group of patients in a clinical trial.
Improper design can invalidate a study's results. Before she joined the center, for example, Guthrie's research with Dr. Norman Breslow at the University of Washington highlighted potential pitfalls in data analysis.
Breslow directs the National Wilms Tumor Study Group's data and statistical center, housed in the Public Health Sciences Division. Wilms tumor is a childhood kidney cancer.
In a study at Johns Hopkins University, clinicians scanned computerized images of tumor biopsies in Wilms tumor patients and asked whether certain tumor characteristics could predict disease-free survival. The researchers predicted high correlation, but when the Breslow group re-evaluated the data, it found the opposite: the scans appeared to be unreliable in predicting survival.
"That was an interesting study to be part of because it re-evaluated a study that was an example of what not to do," Guthrie said.
Whereas the Hopkins group tested its prediction on the same set of patients used in the analysis, "we used the test sample to come up with a set of variables, then tested our prediction on a different set of patients," she said.
Considering these potential pitfalls, Guthrie and her colleagues work with clinicians to achieve a proper balance between drug effectiveness and side effects.
If the drug in question was to facilitate bone-marrow transplantation, was it effective? If so, what dose is the most appropriate starting point?
The answer is in the numbers.
[Danielle Ippolito is a University of Washington graduate student in pharmacology.]