A Biomarker for Chronic Back Pain?

Pauline Anderson

December 19, 2012

Researchers are investigating what they believe is an objective biomarker for chronic pain that could not only eventually help monitor pain therapies but actually separate out patients with real chronic pain from those more intent on getting compensation.

They used advanced computer algorithms that predicted, with an accuracy of 76%, the presence of chronic lower back pain.

The algorithms are based on "mind reading" technology that has been used for some time in the field of cognitive neuroscience but is now being used to a greater extent in medicine, study author Sean Mackey, MD, PhD, chief, Division of Pain Management, and professor, anesthesiology, Stanford University, Palo Alto, California, told Medscape Medical News. "It's an area that is kind of exploding."

But he cautioned that the technology "is not yet ready for clinical application."

The study is published online December 17 in Cerebral Cortex.

Machine Learning

Previous studies of back pain involved having patients undergo imaging to see which areas of their brain are activated, said Dr. Mackey. "What we never did until now is take a brain image and ask whether this brain image represents someone who has low back pain or not."

To do this, researchers enrolled 47 patients with non-neuropathic lower back pain and 47 healthy controls matched for sex and age (within 2 years). "This is fairly large for a neuroimaging study," commented Dr. Mackey.

Investigators used a novel type of multivariate "machine learning" algorithm called support vector machines (SVM) to analyze structural MRI scans and identify patterns of gray matter density that best distinguished persons with chronic low back pain from controls. They also used whole brain voxel-based morphometry (VBM) to determine whether there were regional differences.

Dr. Sean Mackey

The algorithms used by the researchers grew out of artificial intelligence research and are used on a daily basis. Dr. Mackey cited the example of a typical Google search that incorporates "pattern classifiers" and "facial recognition." "This has been around for a couple of decades, but the tools and techniques are really getting refined to the point now where we can apply it in novel ways," he said.

Previous studies, for example, have used the technique to noninvasively identify Huntington's disease with 83% accuracy (Kloppel et al, 2009), Alzheimer's disease with up to 89.3% accuracy (Vemuri et al, 2008), and acute painful stimuli with 81% accuracy (Brown et al, 2011), they write.

The SVM analysis indicated that lower back pain is characterized by a pattern of structural changes in the gray matter of patients with pain. Key regions of increased gray matter density included the left medial orbital gyrus and in the right cuneus (secondary visual cortex), whereas regions of decreased gray matter density were in the right cerebellum, areas of the temporal lobe, left primary and secondary somatosensory cortices, left primary motor cortex, right calcarine sulcus (primary visual cortex), and right dorsolateral prefrontal cortex. The VBM analysis did not reveal any areas in which gray matter density was significantly different between the 2 groups.

Average Accuracy

The SVM classifier obtained an average accuracy of 76%. Sensitivity and specificity were 76% and 75%, respectively, and positive and negative predictive values were 75% and 76%, respectively. All values were significant at P < .001, as determined by a permutation test. A receiver-operating characteristic curve showed that the classifier performed better than random chance, with an area under the curve of 0.91.

That level of accuracy was much better than Dr. Mackey had envisioned. "I first went into this detection area fairly convinced this wasn't going to work," he said, adding that he was pleased at the tool's "resounding" positive predictive ability.

But the accuracy of the technology is even more accurate — over 80% — when detecting acute pain, according to the previously mentioned paper by Brown et al, published last year, that involved inducing pain using a heat stimulus.

The accuracy of the technology in chronic pain may improve over time. "The hope is that with further refinement of the computer algorithms, we can get better and better accuracy and better detection," said Dr. Mackey.

In their study, the researchers didn't look at the intensity or duration of pain. "We didn't specifically try to take on the issue of sex, gender, or intensity of pain; it was determining whether we could, with some degree of accuracy, do this," said Dr. Mackey.

Overlapping Regions?

Although the tests clearly confirm that the central nervous system is involved at some level in chronic pain, it doesn't determine the cause of the pain, noted Dr. Mackey. However, he thinks there's probably an "overlap" of brain regions involved in different types of chronic pain. "We're still trying to identify signatures of different chronic pain states — back pain, pelvic pain, headache, etc — and trying to determine what areas are common and what areas are distinctive."

Self-reported pain remains the "gold standard," but this new tool could be useful for vulnerable populations not able to vocalize their discomfort, including the very young, the very old, and those in intensive care units, said Dr. Mackey.

The refined technology could be used as a kind of pain biomarker that targets and monitors pain therapies, he said. "We will really try to fine-tune this and turn it into a really useful tool that will ultimately help us to target therapies and then to individualize those therapies to a particular person with high degree of safety and efficacy."

But Dr. Mackey was quick to point out the limitations of the study. "We want to make sure people don't overread what we're doing," he said, stressing that the research participants were "very carefully chosen," weren't receiving any medications, and had no psychiatric disorders or other chronic pain condition. In addition, their back pain was localized.

"The truth is, they don't really represent the vast majority of patients I see in my clinic who are on multiple medications and have some depression or anxiety," said Dr. Mackey. "People need to be careful about extrapolating this to other conditions or even to general low back pain."

Ethical Issue

He also raised the ethical issue of using the technology to determine whether patients are really suffering pain and are eligible for insurance coverage, or are enhancing their pain to collect disability.

"Whenever you have a new technology, there will be enterprising people who want to make a buck off of it," he said. "There is a huge medical legal industry that spends probably hundreds of millions of dollars trying to determine whether patients are experiencing pain and suffering or not."

An estimated 100 million Americans have chronic pain, with chronic low back pain being the most common cause of limiting activities in those younger than 45 years of age, according to the study. The prevalence of lower back pain has risen significantly in the United States, from 3.9% in 1992 to 10.2% in 2006.

No specific abnormality can be identified in up to 90% of chronic lower back pain cases, said the authors. In many of these cases, spinal imaging shows no abnormalities.

This work was supported by grants from the National Institutes of Health, IASP International Collaborative Research, and the Redlich Pain Research Endowment. Dr. Mackey has disclosed no relevant financial relationships.

Cerebral Cortex. Published online December 17, 2012. Abstract