When the Food and Drug Administration gave the go ahead earlier this year for Donald Francis, S.D.' 79, and his company, VaxGen, to begin large-scale clinical trials of an AIDS vaccine, it may have ended the stalemate over whether AIDS vaccines are ready for large-scale testing in people. A major part of the reason for the breakthrough is a new way of analyzing the results of clinical trials, a method pioneered by Assistant Biostatistics Professor Peter Gilbert.

The new method is called sieve analysis, a felicitous name because it is easy to picture a vaccine acting as an immunologic sieve that sifts out disease-causing infections in populations. Going way beyond a two-dimensional, "yes-it-protects, no-it-doesn't-protect" answer, sieve analysis allows investigators to design and then analyze trials that juggle a complex assortment of viral immunologic information: strains of HIV, antibody levels, types of cells infected. The framework for sieve analysis remains a comparison between the infections that occur in the vaccinated group and the infections that occur in the placebo group. But with sieve analysis, the investigators can identify much more precisely why an experimental vaccine is (or is not) working--and against what. Did it protect against one strain of HIV and not another? What kind of immune response did it provoke? And was that response sufficient to protect against long-lasting infection?

"You want to see what is significant for the prevention of clinically meaningful infection," says Gilbert. "Is it genetic clade that makes a difference? Is it serotype that makes a difference? Is it the CD4 co-receptor usage that makes a difference?

A dark irony of vaccine efficacy trials and sieve analysis is that the failures--the so-called breakthrough infections--are critical. "You want to see the 'holes' in the vaccine and then refine the vaccine to fill those holes," explains Gilbert. Sieve analysis helps biostatisticians make sense of those holes and some educated guesses about the underlying biological reasons for them.

It has been a point of contention in AIDS vaccine research whether the current crop of vaccines is really ready for large-scale testing. Cautious scientists say too little is known about the immune response to the vaccines, and to HIV itself. But other researchers argue that small studies in humans, as well as "challenge studies" looking for analogous immune responses in animals, can be uninformative, if not downright misleading. The only way to ever identify an efficacious vaccine will be with large-scale trials, they say. So from their standpoint, sieve analysis was a major blessing. "To my way of thinking, it gives you a rich analysis of what protects and what doesn't protect. It helps define what a vaccine can and cannot do," said Mary Lou Clements-Mann, a prominent AIDS vaccine researcher at Johns Hopkins, in an interview with the Review before her tragic death in the Swissair crash in early September. Max Essex, chair of the Harvard AIDS Institute, says that sieve analysis helps answer objections that too little is known about aids immunology for large-scale trials to be useful. "It provides you with a statistical way to learn by doing trials," he says.

Gilbert came to the School in January 1997 as a postdoctoral student after eight years at the University of Washington, a school with a biostatistics department that competes with the School's department for best-in-the-nation status. Drawn to AIDS research partly by his Christian faith, Gilbert says it was his faculty adviser at Washington, Steven Self, who steered him toward sieve analysis.

Gilbert coauthored one of the seminal papers on the subject, though he generously gives Don Francis credit for coining the term. VaxGen and sieve analysis are closely linked: it was a sieve analysis that guided VaxGen's reformulation of its vaccine for the large-scale trials in this country so that it now includes two varieties of HIV1-B antigens.

A modest person, Gilbert acknowledges that some of the underlying concepts of sieve analysis have been around "ever since people started trying to make vaccines against pathogens that are diverse in their genetic sequence or, more importantly, in their antigenic properties." The key difference, notes Essex, is that those earlier analyses were often done after the fact and in reaction to trial results: "Sieve analysis is different because it is part of the design of the trial."