Last week, the Food and Drug Administration approved a new drug for treatment of advanced breast cancer. Media reports say its approval was based on results of a clinical trial in which 991 patients with an aggressive form of the cancer received either of two pills commonly used or infusions of the new drug. Median survival time among those receiving the new treatment was 30.9 months, compared to 25.1 months for those given the pill regimen.
These reports are heartening--new drugs provide new options and new hope to patients. Such breakthroughs are often the result of decades of research by teams of scientists, including chemists, physiologists, molecular biologists, pharmacologists, clinicians, and statisticians. Statisticians? What do statisticians have to do with developing new drugs?
Just about everything. Statisticians--and statistics--are fundamental to all steps in the drug development process. Since 1962, when Congress required drug manufacturers--or "sponsors"--to demonstrate safety and effectiveness through "well-controlled clinical studies" to market a new drug, the numbers of statisticians in the pharmaceutical industry and at the FDA have mushroomed. Here's why.
Drug discovery is formidable: Thousands of compounds may be candidates for targeting a particular disease. The challenge is screening all for evidence of interesting biological activity and identifying those deserving further study. Statistical experimental design, which dictates the most efficient ways to use available resources to detect interesting activity, and statistical modeling (e.g., to predict compounds to investigate based on results for others with similar chemical structure) are essential.
Once a compound is flagged, but before it may be introduced in humans, lab and animal studies are conducted to evaluate toxic effects and the doses at which these occur. Statisticians design studies and develop data-based models describing the dose-toxicity relationship and the drug's pharmacological properties (e.g., how well it is absorbed into the body). Compounds with unacceptable results cannot move to human testing.
Because the FDA requires painstaking documentation before human testing is permitted, sound statistical study design and analysis supporting those that advance is essential.
Once permission is granted, the path to approval involves several "phases," and statisticians are indispensable in each. Phase 1 involves studying toxic effects and pharmacological properties and determining an acceptable dose. Because this is the first time the drug is given to humans, study designs are required that expose subjects to increasing doses in a cautious fashion, yet yield sufficient data to determine the "maximum tolerated dose." Statisticians are developing new approaches to this challenge.
If the drug is determined to have an acceptable profile, it moves to Phase 2, in which its efficacy (therapeutic effect) is investigated in patients (typically 100 to 300) for the first time. Here, the sponsor works to learn all it can to inform decisions about moving to large-scale testing. Usually, ultimate approval is based on a "clinical endpoint"--like survival time--of paramount interest to patients and medical providers. However, studying the clinical endpoint may take years. Thus, a "biomarker"--a characteristic associated with the state or progression of the disease, like cholesterol in heart disease--may be studied as a "surrogate endpoint" instead. If the drug has a beneficial effect on the surrogate, it should have a similar effect on the clinical endpoint. The precise definition of a surrogate endpoint and its data-based validation requires sophisticated statistical modeling.
In Phase 2, dosing requirements also are established and safety further evaluated. Design--the number of patients, doses, how often patients are examined--and analysis plans demand the use of the most innovative statistical tools to ensure that sufficient information is collected to support "go/no-go" decisions.
Sponsor and FDA statisticians are essential at the end of Phase 2, when the FDA and sponsor must agree on the clinical endpoint and the design and analysis plan for large Phase 3 clinical trials in which the new drug is to be compared to a different treatment, often a placebo. These costly studies can involve thousands of patients over several years. A fundamental statistical principle is randomization: Patients are assigned randomly to receive the new drug or the placebo so that the two groups are similar, therefore protecting against unfair comparisons resulting if, say, physicians gave the new drug preferentially to sicker patients.
At the end, the clinical endpoint is compared in the two groups using a statistical test that evaluates the strength of the evidence supporting a real difference. FDA statisticians review these analyses. Sometimes, approval can depend on a single statistical analysis and whether the strength of the evidence meets the standard demanded by the FDA.
In fact, although this final action sounds straightforward, the underlying considerations are not. Part of the design of many trials is a statistical "monitoring plan" to review the accruing data periodically --if statistical evidence is sufficiently strong that the new drug is superior, ethically, the study should be stopped and the new drug offered to all patients.
A major challenge is dropout; some patients end study participation prior to collection of their endpoints. As a result, data are missing, and failure to account for this can generate misleading results. Statistics is essential to handling this complication, leading the FDA to request that the National Research Council convene a panel of expert statisticians to develop recommendations.
FDA statisticians--more than 150 in the FDA's center for drugs alone--are involved in the development of FDA recommendations--called guidance--on appropriate statistical methods. Statistics also is essential in the FDA's other centers, including those for biologic products and medical devices.
So the next time you read about FDA approval of a new drug, know that statisticians substantially contributed to its development. Statisticians and statistics are even more fundamental in this era of personalized medicine, as sponsors seek to target treatment to patients most likely to benefit and develop "adaptive" study designs to identify these patients sooner.
Davidian is president of the American Statistical Association and William Neal Reynolds Professor of Statistics at North Carolina State University in Raleigh, N.C.