In today’s Academic Minute, Dr. Andrew Goldfine of Weill Cornell Medical College takes a second look at a study that found awareness in some patients that were in a vegetative state.
Andrew Goldfine is an assistant professor of neurology and neuroscience at Weill Cornell Medical College in New York City. He is currently working in the labs of Nicholas Schiff and Jonathan Victor at Weill Cornell Medical College, studying the pathophysiology of disorders of consciousness, and the use of neurophysiological tools to track recovery of movement and large-scale cerebral networks.
Dr. Andrew Goldfine – Awareness in Vegetative Patients
Brain imaging studies over the past 10 years have shown that there can be a dissociation between level of consciousness and ability to move. This implies that there can be patients with history of severe brain injury who appear unconscious, yet are actually fully conscious but unable to move. In 2011, a team of researchers in London, Ontario and Cambridge, England claimed to have detected evidence of consciousness in 3 out of 16 vegetative state patients using a technique called electroencephalography, or EEG. Subjects were alternately asked to imagine moving their hand or their foot, and the EEG was used to determine if they did it.
This was an exciting finding because these patients showed absolutely no signs of consciousness, and the task they were asked to do required remembering a command for an entire minute – something difficult for many patients with only mild brain injuries. We decided to validate the methods and findings, and the investigators supplied us with the original data to do so. We first did straightforward statistics on the EEG, comparing the hand and foot movement periods, and we found no evidence that the vegetative patients performed the task.
In the original study, the authors had used an approach called machine learning to determine if the EEG was different between the hand and foot tasks. Machine learning is how Google differentiates your spam from real email – these algorithms are not smart, but simply learn to recognize patterns. It turns out that the EEG of the patients were contaminated by electrical activity of the muscles on the scalp, and the machine learning algorithm used the muscle activity rather than brain activity to differentiate the hand from foot tasks.
We concluded that their method was invalid and should not be used on patients. Our study shows the importance of scientists sharing data to validate findings, and that one needs to be careful with powerful machine learning tools because false results can lead to unsupported major changes in patient care.
Production support for the Academic Minute comes from Newman’s Own, giving all profits to charity and pursuing the common good for over 30 years, and from Mount Holyoke College.
