Thank God for Marvin Zelen
Garbage in, garbage out—the Harvard biostatistician Marvin Zelen, CAS/PhD ’57, probably didn’t invent the saying, but to judge from the fruits of his long career, it represents one of his key beliefs. Zelen built his reputation in part by devising the statistical methods that underlie clinical cancer trials, in which experimental drugs are tested for toxicity, effectiveness, and proper dosage. But equally important, if not more so, he put in place dramatic measures to assure that data from the trials are untainted by errors and biases, measures that have since become standard practice in the field. “It became very clear early on,” says Zelen, who first got involved in medical research in the early 1960s, “that there was no point in analyzing worthless data.” Garbage in, garbage out, in other words.
Marvin Zelen, the son of immigrants, was born and raised in New York City, where he spent his nonschool hours playing games of chance and working in his father’s candy store. “I made the best egg cream in the Bronx,” claims Zelen, who has thick arms and mournful eyes, and speaks in a soft voice that descends to a near-whisper at the end of some of his sentences.
Hoping to escape the drudgery of the candy store, Zelen attended New York’s City College, where he took a lively interest in statistics and probability. It was the only subject he encountered that connected with his outside life, he says, alluding to his boyhood gambling. After college and a master’s degree, he worked at the mathematics lab of the National Bureau of Standards in Washington, D.C.
Sitting in his utilitarian office in Boston’s Longwood Medical Area, a workspace that features scattered papers, bookcases filled with professional journals, and a white board covered with equations in bright green ink, Zelen says the lab was part of a federal government push, after World War II, to apply mathematics to practical problems. “They recruited famous mathematicians from all over the world,” he says, adding that at age 24 he was the lab’s youngest staffer, and the only one without a doctorate.
At the lab, he crunched physical science data and also started making a name for himself, publishing in professional journals and getting an offer to join the faculty at Berkeley, an offer rescinded when the department chair learned that Zelen lacked the PhD. By then, in addition to his day job, he was teaching part time at American University, and he entered the doctoral program there, finishing in two years while still working full time. The day of his last qualifying exam, typically a three- or four-hour ordeal, Zelen had a funeral to attend. Not wanting to wait another year for the degree, he whipped through the test, which he passed, in 15 minutes.
Zelen moved from the mathematics lab to the University of Wisconsin. He stayed just two years (“I hated the cold weather”), but they were crucial ones. It was at Wisconsin that Zelen first worked with cancer researchers. The researchers approached him, worried that their studies’ integrity might be compromised because patient volunteers were coming in at many different stages of the disease. Zelen sat down and worked out a procedure for dealing with the problem. More important for his future, he learned that clinical cancer research presented lots of interesting statistical problems.
His next job was a stint at the National Cancer Institute. Then, in 1967, wanting to return to academia, he joined the biostatistics department at the State University of New York in Buffalo—admittedly, another cold-weather spot, but by then he had a wife and children, and the school had made a decent salary offer.
At Buffalo, Zelen began a 19-year connection with the Eastern Cooperative Oncology Group (ECOG), which had been set up to test experimental cancer therapies. At the time, ECOG was running seven studies, all of them “terrible,” Zelen says with a thin smile. “They were poorly thought out; the data was wrong; they had poor quality control, not enough patients—everything you can think of that was antiscientific.” One study, he recalls, was attempting to test a single drug against cancers of the breast, lung, and digestive system, a monumental waste of time and money, given that, as was widely known, each of the cancers has its own unique natural history.
At an early meeting with the group, Zelen mentioned these and other problems to the physicians in charge of the ongoing studies, suggesting that the studies be terminated and restarted from scratch. The physicians agreed without protest, he says, and soon Zelen and a growing corps of statisticians had taken charge of ECOG’s studies, inventing numerous new procedures to clean up methodology.
Take randomization. To avoid physician biases, today’s clinical trials assign patients at random to an experimental drug or to the standard treatment for their disease. Few of ECOG’s early studies used this methodology. In those that did use it, drug assignments arrived in sealed envelopes, but doctors figured out how to get around the method—saving up four or five envelopes, for instance, opening them all at once, and then giving each assignment to a patient for whom it seemed appropriate. (Usually the more promising drug went to the patients with the more promising prognoses.) The practice ended in the first days of the Zelen regime, after which drugs were assigned centrally using randomization software.
Another widespread problem with the early studies boiled down to sloppy data management by the doctors and their helpers. “They’d report somebody dead,” says Zelen, “and two months later, they’d report that the person was taking therapy.” Zelen and colleagues wrote checking procedures into their database software to flag this kind of error. They also took the job of processing data away from secretaries, who had done it as a sideline, and gave it to trained data managers. They even designed a custom data form for every study, replacing the error-prone, one-size-fits-all form that had been in use.
All the while, Zelen was devising new statistical techniques for use in ECOG’s research and in clinical practice. His inventions include, among many others, the so-called semi-Markov process, which helps clinicians determine how long to keep a patient on a treatment that has yielded no results to date, and the permuted block algorithm, a sophisticated randomization procedure.
A Civil Action
When he moved on to Harvard in 1977, Zelen brought ECOG along with him, as well as much of his department at Buffalo. Harvard hired a dozen faculty members away from Buffalo, he explains, as a condition of Zelen’s accepting the job.
In the mid-1980s, a few years into his time at Harvard, Zelen, whose renown hadn’t spread too far beyond his profession, came briefly to the notice of a broader public, when he became involved in the legal case that formed the basis of the best-selling book A Civil Action. Zelen was called in by colleagues at Harvard who had themselves been approached by family members of a number of leukemic children from Woburn, Massachusetts, a Boston suburb. Starting off with the assumption that any environmental factor that might have caused childhood leukemia would probably have also caused birth defects and other ailments, Zelen designed a study that established that the cancers were not a random cluster but the likely result of exposure to contaminated well water. Eventually the families sued the corporate owners of two factories that, according to the lawsuit, had polluted the wells with an industrial solvent.
Zelen’s study drew fire from the chairman of Harvard’s own department of epidemiology, who argued that the volunteer data collectors, some 250 Woburn residents, had a personal stake in the study’s outcome. Zelen rejects the criticism, pointing out that he and his fellow biostatisticians backed up their findings with hospital records. He also notes that his critic was a paid consultant to a chemical industry trade organization.
Zelen unveiled his study results at a meeting in a Woburn church, an event treated in detail in A Civil Action. According to the book, after Zelen’s talk, someone in the room called out, “Thank God for Marvin Zelen!” and the crowd broke into cheers.
Such dramatic moments have been rare in Marvin Zelen’s life, yet his work has dramatically affected the world. From a group that once comprised a handful of doctors who knew little of the fine points of experimental science, ECOG, thanks to Zelen’s innovations, has evolved into a national research center that’s currently running 90 trials simultaneously. Among other accomplishments, ECOG’s studies have resulted in new treatments for leukemia and lung cancer. And Zelen’s methods, pioneered at ECOG, have gotten “widespread acceptance,” not only in clinical cancer trials but in clinical research generally, says John Bennett, a hematologist and leukemia researcher at the University of Rochester. “His work has served as a model for AIDS clinical trials,” adds Elizabeth Malloy, professor of mathematics and statistics at AU.
Zelen long ago moved on from ECOG, but he still does biostatistical research. His main interest these days is in public health programs that screen for common cancers, such as those of the breast, lung, prostate, and colon. His recent papers use statistical models to derive the optimal intervals and starting ages for the screening tests, results that will help the government squeeze maximum benefit from the public health dollar.
He also continues teaching at Harvard, and he’s president of Frontier Science, a nonprofit he founded to take over support functions, such as database building and quality control, from ECOG, which was starting to run out of space at Harvard. Zelen’s wife, Thelma, serves as the nonprofit’s unpaid COO.
As to retirement, it’s not in the cards, says Zelen, who turned 81 in June. “They’ll have to carry me out. I wouldn’t know what to do,” says the one-time soda fountain jockey. “My wife and I—I think they’re going to put us in the Smithsonian as the world’s oldest full-time working couple.”