06/12/2013 07:06 pm ET Updated Aug 12, 2013

Sorting the Wheat from the Chaff: How Do We Know Which Cancer Therapies Really Work?

By Stephen George

News reports of "breakthroughs" in the treatment of cancer are so common that one might be forgiven for wondering why it is still a medical problem. Many of these dramatic reports are based on laboratory findings, animal studies, or tiny preliminary studies in humans. How do we know which newly touted treatments really work (i.e., are safe and effective) and which do not?

The best way, and one that has led to steady progress in the treatment of many types of cancer in recent decades, is through the randomized controlled clinical trial -- a scientific experiment using modern statistical concepts and methods to assess the relative safety and efficacy of two or more therapies. It is arguable that the wide adoption of such randomized trials is the greatest single advancement in the history of cancer therapeutics.

In the mid-20th century, there developed a confluence of interests between medical oncologists; physicians specializing in the diagnosis, treatment and prevention of cancer; and statisticians, who apply the theory and methods of statistics to medical research. The early interactions between these disciplines led to development of the randomized clinical trial and its subsequent acceptance as the proven way to obtain reliable evidence about treatments for cancer. It is not the only way (epidemiologic studies or data mining of large clinical databases also can provide information), but no other way comes close to providing the type of reliable information needed in this setting than the randomized clinical trial.

There are examples of studies before the 20th century that had some of the elements of a modern clinical trial, such as the famous study by James Lind in 1747 concerning the relative effects of various treatments for scurvy, a serious, often fatal disease afflicting sailors on long voyages. In his study, Lind, who was a ship's surgeon, divided 12 sailors afflicted with scurvy into six pairs, each receiving one of six daily supplements to their common daily diet of gruel. One of the treatments consisted of two oranges and a lemon; the other treatments were a quart of cider, two spoons of vinegar, sea water, "elixir vitriol," and a "surgeon's electuary." The two sailors who received the citrus fruits recovered quickly and fully; one returned to duty in five days, and the second was appointed nurse to the others. None of the other treatments showed any benefit. In 1867, 120 years after Lind's experiment, the British Navy first began supplying a daily lime ration to the crews of its ships (thus the origin of "limey").

Despite a few such studies with some features of clinical trials, modern clinical trials -- ones exhibiting all the important statistical principles in use today -- are products of the 20th century. The earliest such clinical trials, conducted during World War II and in the late 1940s, were studies of treatments for malaria, pulmonary tuberculosis and pneumococcal pneumonia.

The first cancer clinical trial, which compared treatments for patients with acute leukemia, began in 1955. Clinical trials in cancer arose from the U.S. National Cancer Institute (NCI), part of the National Institutes of Health, and the cooperative clinical trials groups established by the NCI in the mid-1950s. The NCI cooperative groups are organizations that bring together medical centers and clinics to carry out cancer clinical trials and related studies in a wider environment. Originally, the number of institutions and investigators was relatively small, but the cooperative group program quickly expanded and today involves more than 3,100 institutions and 14,000 individual investigators who place more than 25,000 new patients into cancer treatment clinical trials each year.

Statisticians have been active and essential collaborators in the design, conduct and analysis of these trials from the beginning. Pioneering statisticians during the early years include Marvin Schneiderman, Ed Gehan, Bernard Greenberg and Marvin Zelen. The development of modern statistical theory and methods in the 20th century, enabling the collaboration between statisticians and oncologists, was stimulated especially by the pioneering work of R.A. Fisher, the famous English statistician and geneticist who introduced many of the statistical design concepts and principles that are now standard in modern clinical trials.

The involvement of statisticians in cancer clinical trials for the past 60 years or so has led not only to improvements in the design, conduct and analysis of these trials, but also has stimulated a remarkable surge in the development of new statistical theory and methods arising from challenges encountered in the trials. These developments have improved the quality of the cancer research and the ability to learn more from trials. Today, many universities offer courses on statistical methods for clinical trials. This synergism between cancer clinical trials and the development of new statistical methodology has been a constant feature from the first trial and promises to continue in the future.

Close collaboration between statisticians and cancer researchers has been stimulated further by recent advances in understanding the genetics and molecular biology of cancer. In particular, these advances have led to a search for individualized, or "personalized," therapy based on agents that target specific molecular changes identified in a patient's tumor. There are new challenges in designing clinical trials for efficiently assessing such therapies and, as in the past, statisticians are developing novel statistical approaches to meet these challenges.

Successful clinical trials of such agents already have proved the principle. The FDA has approved more than 30 molecularly targeted agents in cancer in recent years, including prominent examples such as trastuzumab for breast cancer, sunitinib for renal cell cancer and bevacizumab for colorectal, non-small cell lung and renal cell cancers. This work promises to expand considerably in the near future as results from such efforts as The Cancer Genome Atlas project and Therapeutically Applicable Research to Generate Effective Treatments project become available.

Randomized clinical trials will remain the primary tool for evaluating potential new cancer therapies, meaning statisticians will continue to play a vital role in developing efficient designs for these trials.

George is professor of biostatistics in the department of biostatistics and bioinformatics at Duke University School of Medicine.