The researchers surgically implanted the recording devices into four people who had been living with pain for more than a year and had found no relief through medications. For three of the patients, the pain began after a stroke. The fourth had so-called phantom limb pain after losing a leg.

At least three times a day, patients would rate the pain they were feeling and then press a button that would spur their implants to record brain signals for 30 seconds. By following patients daily, at home and at work, “this is the first time ever chronic pain has been measured in the real world,” Dr. Shirvalkar said.

The researchers placed electrodes in two brain areas: the orbitofrontal cortex, which hasn’t been studied much in pain research, and the anterior cingulate cortex, a region involved in processing emotional cues. Many studies have suggested that the anterior cingulate cortex is important for perceiving both acute and chronic pain.

The scientists fed the data on the patients’ pain scores and the corresponding electrical signals into machine learning models, which could then predict high and low chronic pain states based on brain signals alone.

The researchers found that certain frequency fluctuations from the orbitofrontal cortex were the best predictors of chronic pain. Although that brain signature was common among patients, Dr. Shirvalkar said, each patient also showed unique brain activity. “Every patient actually had a different fingerprint for their pain,” he said.

Given these variations and just four study participants, Tor Wager, a neuroscientist at Dartmouth College who was not involved in the study, suggested caution in dubbing orbitofrontal cortex signatures as biomarkers just yet.

“We definitely want to corroborate this with other studies using other methodologies that can provide systematic coverage of the whole brain,” he said.