Dr. Judes Poirier remembers the drug company's reaction when he gave some Alzheimer's patients genetic tests to determine whether the medicine would do any good. "They were not very keen on slicing their market in half," he said.

Now Poirier's research is at the forefront of a new science called pharmacogenomics - developing simple tests to tell before patients ever swallow a pill if a drug will help them or just cause a serious side effect."This is an enormously exciting new frontier," said Dr. Francis Collins, genetics chief of the National Institutes of Health, which is launching a $30 million, three-year program to help drug companies hunt genetic information that lets doctors personalize drug prescriptions.

"This is the kind of thing we'd love to have for every drug in every disease," he added.

Doctors have always known that even the best medicines don't work for every patient, and some are dangerous. A recent study in the Journal of the American Medical Association found 2 million Americans are hospitalized annually for bad drug reactions, and 100,000 die.

Medicines are sold based on studies in which they prove an effect on the average disease sufferer, not every individual. So there's little way to tell who is wasting time and money on treatment that will fail, or to predict who's at risk of side effects.

"It's a crapshoot," George Poste, chief science officer of drug maker SmithKline Beecham, acknowledged Tuesday at the biotechnology industry's annual meeting, where about a dozen companies discussed early attempts to fix the problem.

Pharmacogenomics "should be the rational foundation of rational therapy," Poste added.

Scientists are deciphering the roughly 80,000 genes in people, hunting mutations that cause diseases so they can attempt to repair the defects.

People also have variations in genes, called polymorphisms, that are too subtle to cause disease. Until now, these variations were largely dismissed. But new evidence shows polymorphisms actually can determine how people metabolize and react to drugs.

Take Poirier's research at Canada's McGill Centre for Aging. He discovered that the main Alzheimer's drug tacrine doesn't help patients who have a gene subtype called ApoE-4.

That's about half of Alzheimer's patients, Poirier reported Tuesday. And it's important information because tacrine can cause serious side effects, so doctors don't want to give it to patients who won't benefit.

Collins expects another early success to be tests for women about to take birth control pills. A rare but serious side effect is blood clots. New studies suggest about 7 percent of the population has a genetic variation in a blood-clotting gene called factor V-leiden - and that women with that variation may be at greatest risk from the pill, he said.

But this information could rule out potential customers for drug manufacturers. How much will they research?

"Overwhelmingly, it's going to come back to economics," said SmithKline's Poste, noting that some older drugs are just too cheap to justify spending money on genetics work.