US experts say medication may do more harm than good.

Up to a fifth of patients being treated for depression with some medications may do worse than those given a placebo, according to a US study.

The findings, published in Archives of General Psychiatry, highlight the importance of quickly identifying how patients respond to certain drugs, said lead author Ralitza Gueorguieva, at the Yale University School of Health.

"Identifying variables that are associated with response is a very important question that we haven't quite tackled," she said.

The researchers combined data from seven studies that randomly assigned patients to receive Eli Lilly's drug Cymbalta, known generically as duloxetine, other antidepressants, or a drug-free pill for two months. The trials involved about 2,500 people with major depression.

People getting the placebo tended to report small, gradual improvements in depression symptoms. But those on Cymbalta or another antidepressant fell into one of two categories: most had steeper, steady improvement in depression symptoms, but a sizeable chunk didn't seem to get any better.

One of the study's authors is an employee of Eli Lilly and another is on the company's scientific advisory board.

Researchers found that patients' symptoms over the first few months of antidepressant use separated them into "responders", who got progressively better, and "non-responders", who didn't improve with treatment but may still have suffered side effects.

About four in five patients on all antidepressants were responders. For Cymbalta, about 84 percent of patients improved but 16 percent did not.

Medication responders saw significantly bigger improvements in their depression symptoms than patients assigned to the placebo. Non-responders, however, actually did worse.

"You know within the first couple weeks of starting a treatment who's the most likely to benefit because they're already starting to show improvement," said Michael Thase, a psychiatrist from the University of Pennsylvania, who wasn't involved in the study.

"I think this finding holds true for the antidepressants that are most commonly used today," he added.

The side-effects of antidepressants, such as stomach problems and poor sleep, could explain the worse symptoms seen in non-responders compared to placebo patients, he said.