Harvard Public Health Review Home

Harvard Public Health Review/Summer 2002

677 huntington avenue

"Everybody With Me?"

Not a seat is empty in Kresge 502, the cozy auditorium tucked into the midsection of the Harvard School of Public Health's main building. Students, staff, junior faculty, several tenured professors, and at least one dean are in the audience for the noontime talk. The speaker is dishabille: his medium-length gray hair in a state of frenzy, his brown plaid shirt half tucked into jeans that puddle on the tops of his running shoes. His topic: a new algorithm for finding the optimum treatment strategy. Dry, dense stuff, but he manages to make it seem urgent. The overheads spew mathematical notation as his explanation darts between flights of jargon and attempts to bring it all down to earth ("I call this the blip-down step.").

"Everybody with me?" asks Professor James Robins several times. "Everybody in the game?"

For the first half of his 20 years at the School, the answer to that question was--perhaps a little too emphatically--no. But times have changed for Robins. After years of rejection by elite journals like the Journal of the American Statistical Association, he is now recognized as one of the leading mathematical statisticians in the country, if not the world. A gift from Mitchell L. Dong and his wife Robin LaFoley Dong last year endowed his professorship in the epidemiology department. His advanced epidemiologic methods class demands more of students than any other course at the School, but it's widely appreciated as an important rite of passage into deeper thinking about the fundamentals of epidemiology and statistics.

"Students, they don't realize the great effort he makes to explain something that for him is so absolutely trivial and for most of us is really hard," says Miguel Hernán, SM'99, an instructor in epidemiology who co-teaches the methods course and has worked with Robins to make the material more accessible. Sander Greenland, a professor of statistics at UCLA's School of Public Health, says Robins stands out because he comes up with the nitty-gritty statistical methods that put his bold concepts on solid ground. "On an intellectual level, he is one of the best I've ever worked with," says Greenland.

Robins admits that the rejection letters put a chip on his shoulder, but says he never doubted that he is right. Now he talks enthusiastically about his work being part of a "convergence": "There used to be a completely different culture and language in artificial intelligence, robotics, economics, statistics, biostatistics--we now all talk exactly the same way."

Robins grew up in suburban St. Louis. His father, Eli Robins, was head of psychiatry at Washington University School of Medicine. His mother, Lee Robins, is still a psychiatric epidemiologist there. One of four boys, Robins was competitive and rebellious: "My parents were intellectuals so, of course, I didn't like intellectual things and was totally into sports." Clearly, the anti-intellectualism only went so far. Robins hit Cambridge in 1967 as a Harvard freshmen along with some other fairly bright kids.

But the rebelliousness runs deeper. Robins, who majored in math and philosophy, never got his undergraduate degree because he refused to take the courses necessary to fulfill the language requirement (tests proved he had special problems processing spoken language, but college officials held him to the requirement). As a young doctor, he was fired by a community health clinic in Boston for trying to organize a "vertical" union that would include everyone from custodial workers to the doctors. After joining the faculty at the School in 1982, Robins wasn't an outright rebel, but he was certainly something of a misfit. Robins says he has Professor Richard Monson to thank for "protecting" him during those lean times and Dimitrios Trichopoulos, Vincent L. Gregory Professor of Cancer Prevention and former epidemiology chair, for giving him an institutional home in the department.

Robins was helping run an occupational health clinic at Yale that he had co-founded when he started to dabble in statistics. Part of his interest came from occupational health, a field full of problems relating exposure to disease causation. The real attraction, however, may have been that statistics gave free rein to his tremendous gifts for math and abstract thinking. Robins started just by reading books from the medical library, which he says, smiling, may have been the beginning of his "weird self-taught views." When he started to take statistics courses he was so confused by all the "bizarre conventions" that he got Cs. His probing questions were answered with "you really couldn't understand."

The irony: Robins's work itself is hard to understand, although he makes a valiant effort to explain it without visible condescension. At the most general level, he has come up with novel statistical methods for sorting out spurious associations from true cause-and-effects--a problem facing anyone trying to interpret lots of data. In epidemiology, the usual approach is to control for confounding factors--the variables that are associated both with the exposure and the outcome of interest. One problem with that approach is that it requires the statistician to make some educated guesses about what those confounding factors might be. Another is that the confounders can also be so-called intermediate variables--events on the "causal pathway" between the exposure and the outcome. And it is a long-standing rule in epidemiology that you can't control for an intermediate variable, so researchers were stuck.

Robins cut the Gordian knot by inventing a statistic called the "G estimator" that makes analysis of data that are simultaneously confounders and intermediate variables possible. He's branched out from there. Robins has come up with novel methods for adjusting for treatments that haven't been randomized (for example, use of aerosolized pentamidine in a clinical trial of AZT). He has devised techniques for using surrogate markers to stop clinical trials early. His Kresge 502 talk on optimizing treatment strategies was the trial run of a statistical model that could greatly reduce the amount of information needed to make treatment choices.

Practical application remains a major hurdle, however. Robins says attachment to tried--if not entirely true--methods and thinking get in the way. He is optimistic about that changing as students with some training in his methods move on to jobs and teaching positions. As for his pathway from lone wolf to top dog, Robins says, sure, he's made some adjustments but nothing major. "Now I'm a big shot, but I haven't changed."

Peter Wehrwein

Harvard Public Health Review Home

next story

previous story

HSPH home

top of page

This page is maintained by Development Communications in the Office of Resource Development.
To contact us with suggestions, comments, and questions, please e-mail: editor@hsph.harvard.edu

Copyright, 2005, President and Fellows of Harvard College