Amylin's recent woes with its new diabetes drug, Byetta—six people have died, possibly as a result of taking the drug—raises a crucial question that has long vexed companies and patients about the medications we take for what ails us.

What is an acceptable trade-off between drugs that help millions of people living with chronic conditions such as diabetes—or pain or heart disease—but hurt or kill a few?

It's the pharmaceutical industry's version of the old moral dilemma of whether it is acceptable to sacrifice a few to save many. It's akin to what a person would do if he saw a crowd of people about to be hit by a speeding train. It's too late to warn them off the tracks, but he could throw a switch and divert the train to a siding—where a single person is standing and would almost certainly be killed.

What would you do?

In the case of Byetta, the millions of people who use it won't die if they don't take it, and there are alternatives. Yet for many patients with non-insulin-dependent diabetes (type 2 diabetes), other medicines don't control their glucose levels as well as Byetta; that can lead to declining health.

Should they be denied this treatment to save a handful of people who contracted severe pancreatitis, an inflammation of the pancreas that is sometimes fatal?

A more spectacular recent case is the painkiller Vioxx, which was blamed in the heart-attack deaths of thousands of people with heart disease. Merck, its maker, voluntarily withdrew Vioxx from the market, even though millions of people had taken the drug safely and benefited from it.

Another example is cerivastatin, a cholesterol-lowering drug that Bayer introduced in the late 1990s to compete with Pfizer's Lipitor. Bayer voluntarily pulled the drug in 2001 after it was linked to 52 deaths due to renal failure and to other, nonlethal complications. This was after the Food and Drug Administration had ordered warnings issued to patients and physicians, warnings that failed to stem the fatalities.

Regulators took similar steps with Byetta. In 2007, the F.D.A. ordered Amylin to place information about the dangers of pancreatitis in the drug's insert. It warned that patients should immediately tell their doctors if they experience abdominal pain while taking the drug. This step was in reaction to a small number of people getting sick before anyone had died.

In the wake of the deaths, the F.D.A. says it wants the warning strengthened. The agency says it will monitor the drug closely over the next few months, looking for further deaths.

Amylin and its partner, Eli Lilly, insist that there is no direct proof the patients died because of Byetta. Diabetics have a three-times-greater risk of pancreatitis than nondiabetics.

Investors were spooked nonetheless. Amylin's share price dropped nearly 23 percent when news of the deaths was released. Byetta accounts for 80 percent of the company's revenues.

Drug-company scientists and executives complain that when balancing risks and rewards, the bar is often too high for drugs that are effective for many people but toxic for a few.

One executive, an expert in vaccines, once told me that scientists had come up with a way to inoculate people against H.I.V. but have declined to go public because the vaccines will kill five percent of those who take it.

"Is this worth it to the other 95 percent?" he asked, adding that he wasn't sure how he felt about such an equation. "What if it was one percent?" he mused.

Vaccines already in use cause a relatively small number of deaths each year; that is considered an acceptable risk given the alternative of widespread viral illness.

Other familiar drugs cause deaths now and then as well, including even aspirin if susceptible people take too much. Yet no one talks about removing aspirin from shelves.

Society also accepts much larger numbers of fatalities and injuries while using other products, such as automobiles, which annually kill more than 40,000 people.

Yet for those whose loved ones die because of a drug, any risk of death is too great, of course.

One way to improve drug outcomes is to better define who is susceptible to harm. This is one of the issues with Vioxx, which apparently was safe for many people to use that didn't have a heart condition.

There is evidence Merck was aware of the danger its drug posed to a subset of those taking it; in any case, the company did not follow through on that information, which might have shrunk its market—and profits.

In the wake of the Vioxx fiasco, the dilemma of what risk is too much prompted Congress to require that the F.D.A. more closely monitor drugs after approval. Some believe a fear of repeating that mistake is one reason that regulators are approving so few drugs. Last year, the F.D.A. approved only 17 new drugs, a near-record low.

So we will wait to see which way Byetta goes. Will these six deaths in three years be anomalies, or will others succumb? Have there already been too many deaths for a drug that may be better than others for some, but not dramatically? Or is the trade-off so far worth it?