Biostatistics Home Page Biostat Connections Contents

Biostat Connections
October 1997

The History of Cancer Clinical Trials at Harvard:
Recollections of Marvin Zelen

by Paul Catalano and Dianne Finkelstein

Marvin Zelen was a leader in the early efforts to provide a mechanism for implementing cancer clinical trials. The concepts and program, which first developed in the early 1950's at NCI, blossomed into a model coordinating center at SUNY, Buffalo before moving to its current home at Harvard. It served as the model for the current Statistical and Data Analysis Center of the AIDS Clinical Trials Group, as well as for all major coordinating centers for chronic diseases throughout the world. In light of the central importance of clinical trials to the educational purposes of our department, this article focuses on the history of this effort.

In the 1940's, and the early 1950's, the research center of biostatistics was largely based at the National Institutes of Health (NIH). While many prestigious universities such as Johns Hopkins and Harvard had programs in biostatistics, these were primarily teaching departments in public health, and produced little methodologic research. Also, at the time there were few large collaborative projects involving biostatisticians; the interdisciplinary research efforts were primarily individual faculty members involved in clinical research areas of their own interest. Research in biomedical statistics was most concentrated at the NIH and most specifically at the National Cancer Institute (NCI) through the efforts of Harold Dorn, an epidemiologist who was influential in recruiting and organizing a statistical unit in the NCI. Names now familiar to all biostatisticians formed the core of researchers at the NCI: Cornfield, Greenhouse, Mantel, Schneiderman and Leiberman. The group later expanded into other NIH institutes and more statisticians joined the NIH. Dr. Zelen joined the NCI in 1963.

While the statistics group at the NCI was actively working on a variety of important problems, at the time there was no organized large-scale clinical research effort to evaluate new therapies for treating cancer, until a World War II malaria researcher named Gordon Zubrod conceived of the idea to conduct randomized clinical trials in cancer. He proposed the creation of a series of regional cancer groups to conduct studies in new therapeutics, mostly chemotherapy. Hospital members of the regional group would pool their patients into common clinical trials. Working together, academic hospitals around the country could enroll enough patients to perform serious, rigorous clinical research. The now familiar concept of the cancer cooperative group was born and the U.S. National Service Center (NSC) was created to act as their scientific and administrative center with Dr. Zubrod at the helm. Zubrod recognized the need for strong statistical support and direction, and in the late 1950's, he recruited Ed Gehan (now at Georgetown University) to be the lead statistician for the NSC.

As mentioned earlier, the original cancer cooperative groups (with one exception) were regional and were so named; i.e. East, Southeast, Southwest, Mid-west, and West. In addition, two groups were devoted to leukemia (Leukemia A and B). They were initially to operate independently, coordinated by a central office at the NCI. Today of the original nucleus only the Eastern Cooperative Oncology Group (ECOG), Southwestern Oncology Group (SWOG) and Cancer and Acute Leukemia B (formerly Acute Leukemia B) remain. Eventually they broke off from NIH, with their own central offices taking more of a scientific and leadership role in the conduct of their research efforts. As they developed more autonomy in the 1960's, statistical offices were established in each group (Ed Gehan moved to MD Anderson and headed the coordinating center of SWOG, for example). At this time, ECOG invited Dr. Paul Carbone, then a member of the NCI, to become Group Chair. Dr. Carbone needed leadership in ECOG's statistical office and in about 1970, he asked MZ, who he had known at NCI, to lead the Statistical Center. During the period of decentralization of the cooperative groups, MZ had left the NCI to join the faculty at SUNY at Buffalo, where he had begun a collaboration with lung cancer researchers in the Veterans Administration. Thus Carbone's offer to join ECOG came at a time when MZ and his group in Buffalo were starting to become interested in the scientific and statistical problems of clinical research in cancer.

At the time of MZ's first involvement in ECOG, clinical trials were not designed very well; concepts that are taken for granted today had yet to be implemented. For example, there was no computerization or quality control procedures (secretaries used to type up clinical patient information in their spare time), and no central registration or randomization (clinicians sometimes held the treatment assignment envelopes up to the light to determine the assignments; they would also wait for several patients to enter a study so they could open several envelopes at once and assign the therapies they felt were "best" for each patient!). At his first ECOG meeting, MZ met the Principal Investigators and persuaded them to close the existing (6-8) studies as they were not well designed. Quality control of the data turned out to be a very serious problem. He convened a Statistical Center committee to address data quality control issues. Their recommendation was to institute educated, well-trained individuals to coordinate, review and computerize clinical data. Since no one was trained for this job, a training program had to be initiated. Thus the genesis of the "Data Manager." The title of "Manager" was given to the position so that MZ could convince the Buffalo administration that these professionals deserved a salary commensurate with their educational level and responsibilities.

Clinical trials computing was also in its infancy with lots of problems to be solved. For example, because of scarce computer resources all of the cooperative groups had to link into their local academic timesharing computer facilities for data processing. This meant that computing was done on a "catch as catch can" basis: delays of several days, unreliable service, and downtime were commonplace. Interactive computing did not exist. In addition to the early lack of access to computing hardware, there was no database management software. In ECOG, for instance, each study had its own quirky data format making it a nightmare to retrieve data for analysis. Whenever a data field changed, programs would require revisions. The concept of a clinical research database system had yet to be born. A young programmer, I.M. "Andy" Chang recognized the need and single-handedly developed the program Quire, a database management system for clinical trials. Thus ECOG also became the first clinical trials group to develop and implement a database system for clinical research. This effort, in conjunction with MZ's implementation of self-coding data collection forms and different forms for each disease site, placed great strain on the very modest university computer at SUNY Buffalo. This motivated MZ to submit a grant to NCI for a dedicated computer for clinical trials. The grant was successful and a Digital Equipment DEC-20 was purchased in 1975. Thus ECOG became the first clinical trials group to own its own data processing machine and established Buffalo as a model center for scientific computing in clinical research.

MZ built a biostatistics group at Buffalo. In the early 1970's, the department expanded. Many of the current faculty joined the department at that time: Steve Lagakos, Rich Gelber, David Schoenfeld, and Ken Stanley. Many other statisticians were members of that group, and had their careers molded by their experience at Buffalo including Jack Kalbfleisch (University of Waterloo), Ross Prentice (University of Washington), Stuart Pocock (London School of Hygiene and Tropical Medicine), and Colin Begg (Memorial Sloan Kettering Cancer Center). MZ and this group of young statisticians at Buffalo were responsible for many statistical innovations. For example, the now ubiquitous permuted block algorithm for randomizing treatment assignments was developed in ECOG as well as adaptive randomization algorithms. In addition, early research in survival analysis was focused in Buffalo. Because of the momentum of established techniques in some scientific disciplines, statisticians often find it difficult to implement innovative statistical thinking and new methods of analysis. In the formative days of clinical cancer research, however, there was little resistance to using new ideas in design and analysis from the statistical community since there were few established methods.

Because of the innovations in the science, direction and management of cancer trials pioneered by MZ and his group, Buffalo quickly became a national and international model for a clinical trials coordinating center for statistics and data management. By the early 1970's, in addition to the Eastern Cooperative Oncology Group, the V.A. Lung Group, the Working Party on Lung Cancer, the Radiotherapy Oncology Group (RTOG), the Gastrointestinal Tumor Study Group (GITSG), the Ludwig Group (European Breast cancer group) were all coordinated out of Buffalo.

While MZ's various research programs were expanding in Buffalo, considerable activity was happening in Boston. Dr. Emil (Tom) Frei, head of the (then) Sidney Farber Cancer Institute (SFCI), was interested in bringing MZ to lead the Biostatistics Division at SFCI. At the same time, the Dean at the Harvard School of Public Health (HSPH) was also interested in expanding the department. MZ discussed HSPH options with Fred Mosteller. Mosteller came out to Buffalo in the winter of 1976, and interviewed each of the members of the faculty in MZ's department. Mosteller returned to Boston and recommended that Harvard recruit MZ and his entire research staff. In all, 27 staff and 1 DEC-20 computer moved to Boston and an unprecedented 12 faculty appointments were created at HSPH. In February 1977 Fred Mosteller became chair of the Department of Biostatistics at HSPH; in 1981 MZ became chairman.

In looking forward to the future of cancer clinical research, MZ had several thoughts on the conduct of comparative trials, and the role of biostatistics. MZ felt that because of the difficulties of running randomized trials, especially issues involving recruitment of patients and obtaining informed consent, there will be considerable and growing interest in the conduct of non-randomized comparative trials. Such trials may be possible when there is an intermediate event, for example a biologic marker of disease progression. Understanding about the association of the marker with survival may make it possible to test assess the efficacy of new therapies. As more basic scientific advances are introduced in the clinical setting, use of high quality intermediate endpoints should play a larger role in the design of clinical trials.

MZ also feels that early detection of cancer offers the greatest promise for saving lives. He believes that methods for early detection of disease will demand more attention from the statistical and research communities. He also thinks the rapid development of basic science methods, especially those in molecular biology, will be a fertile ground for new statistical methodology.

When asked about where biostatistics research might flourish in the future, MZ was skeptical that it will remain in academia. He felt that because of the growing communities of statisticians in industry and government, these areas of current collaborative statistical activity might become the future funds centers of statistical research, especially in an era of dwindling for traditional statistical research grants.

Maintained by the Biostatistics Webmaster
Last Update: 9 October 1997