The implant involves connecting a wire to the left vagus nerve in the side of the neck; a battery is implanted high in the left chest or under the armpit, and the amount of current can be regulated externally. Typically, the implant sends a 30-second pulse of current followed by a five-minute pause, 24 hours a day.

Karmen McGuffee of Garland, Tex., told the panel she had tried virtually every antidepressant drug on the market before getting the device. Improvement came within weeks, she said.

“My mother said she wasn’t looking into the eyes of a dead person anymore,” she said. When people asked why she was willing to get an implant, she replied, “I had nothing to lose.”

McGuffee was paid by Cyberonics to testify. Rush, who was paid to conduct some of the clinical trials, said there are 30,000 suicides a year in America, about 80 percent of them attributable to depression. Half of those are among patients with treatment-resistant depression who had tried multiple therapies and were still not better, he said.

In response to suggestions by some panel members that the company conduct another trial, Rush warned, “In the time to do another trial, we will lose another 1,000 patients a month, 36,000 if the trial takes three years.”

Still, at least two members of the eight-member panel said the company had not provided sufficient evidence.

“It’s possible this is a viable treatment; it is also possible it is not a treatment,” said Richard P. Malone, a psychiatrist at Philadelphia’s MCP Hahnemann University. “I’m not sure it is ethical to give a treatment for which there is not substantial evidence.”

Malone and another panel member, statistician Jonas Ellenberg, nearly convinced other panelists that the company should conduct another trial, but the proposal was shot down after a tense vote, with one panelist changing her mind. The chairwoman of the panel, Kyra Becker, a neurologist at the University of Washington School of Medicine in Seattle, cast the deciding vote.

The panelists mostly agreed that the data presented by Cyberonics had problems but were swayed by the lack of alternative treatments and the evidence that the device is generally safe, based on its widespread use to reduce epileptic seizures.

Cyberonics’ central study was a randomized trial involving 221 patients. All received implants, but the power was turned on for only half the group.

Of 111 patients getting stimulation, 17 showed an improvement of at least 50 percent on a psychiatric scale of depression after 12 weeks. Of 110 patients getting “sham” treatment, only 10 showed similar improvement. The difference between the groups was not statistically significant, however.

The company conducted a separate analysis with patients in another study. This group involved 124 patients who were similarly ill but were not getting vagus nerve stimulation. The company showed trend data that over 12 months, patients with the device were significantly better compared with those getting drugs or electro-convulsive therapy.

Malone, Ellenberg and the FDA’s statisticians expressed worry that the two groups of patients were not comparable and the apparent benefit might be statistically spurious. Cyberonics officials argued that conducting a new trial would be problematic because of the special needs of this patient group.