Predicting Chronic Pain States in Humans
Post by Lani Cupo
The takeaway
The authors developed a neural biomarker to predict chronic pain in patients, with the goal of facilitating diagnosis and treatment of neuropathic pain.
What's the science?
Neuropathic chronic pain (e.g. after a stroke or amputation of a limb) is the cause of great suffering in patients, however it can be difficult to develop objective biomarkers to aid diagnosis and treatment. It is also still not fully clear how brain activity changes with fluctuations in chronic pain levels, and how these changes differ from activity associated with acute pain. This week in Nature Neuroscience, Shirvalkar and colleagues presented a neural biomarker for chronic pain using implanted electrodes in patients, successfully predicting pain ratings.
How did they do it?
The authors enrolled four adult participants in their study (two women), three of whom had post-stroke chronic pain, and one who had phantom limb pain. The authors implanted electrodes into two brain regions important in the processing of pain: the orbitofrontal cortex (OFC) and anterior cingulate cortex (ACC). The study took place over 2.5 - 6 months, during which time participants were asked to record their pain at least 3 times per day. After recording their pain rating (which was inherently subjective, as pain is by definition a subjective, individualized experience), they pushed a button on a remote control which triggered a 30 second recording from the implanted electrodes. This in-depth recording method allowed researchers to track fluctuations in pain over the day as well as over the weeks.
Next, the authors trained a machine learning model to predict subjective pain scores with the neural activity from the implanted electrodes. They compared models trained on data from only one brain region versus models combining data from both electrodes to see which brain region best predicted chronic neuropathic pain.
Finally, the authors sought to compare the neural mechanisms underlying chronic pain with those underlying acute pain in a laboratory experiment. They brought the patients into the lab and presented thermal stimuli (heat at five different temperatures) to both the most painful part and side of the body and the same region on the other side. During the experiment, they recorded neural activity from the electrodes and trained a machine learning algorithm to predict subjective acute pain ratings on the neural activity alone.
What did they find?
First, the authors observed patients had diurnal fluctuations in pain levels (over the 24-hour period), however, they also found cycles of pain in some participants every 3 days. Second, the authors successfully trained an algorithm (linear discriminant analysis) to classify subjective pain states as high vs. low. For three participants, the best prediction resulted in combining data from the ACC and OFC, however, overall the best subregion to predict neuropathic pain was the contralateral OFC — the OFC on the opposite brain hemisphere of the perceived pain. For example, if pain was felt in the left leg, the right OFC was the most effective region to predict pain. The results were stable across the months of the study, suggesting the model was robust in its predictions. Finally, the authors successfully trained a model to distinguish high-vs-low pain states in the acute pain experiment, but importantly, only models that included data from the ACC were successful, unlike the chronic pain state. This suggests the ACC is more centrally involved in acute pain, rather than chronic pain.
What's the impact?
This study is the first to successfully predict subjective recordings of chronic pain from intracranial recordings over a period of months. In time, their findings may be used to develop patient-specific metrics to aid in diagnosis of chronic pain states. Further, implanted electrodes may be used to stimulate regions integral to chronic pain processing, reducing the pain that patients experience and improving their quality of life.