How can general-health symptoms classify the severity of chronic pain and support individualized pain management?

Chronic pain is a global healthcare crisis that causes suffering for millions of people, substantial medical expenses, and lost productivity. Chronic pain is commonly classified and diagnosed based on where the pain is experienced in the body — from low-back pain, headaches, or pelvic pain, to widespread pain in multiple body locations.

However, patients with all types of chronic pain conditions experience physical symptoms such as fatigue, psychological symptoms such as depression, and social symptoms such as social isolation. This commonality inspired us to develop a new way to classify patients with chronic pain based on their self-reported general-health symptoms — regardless of their formal diagnosis or the potential underlying cause of their pain. By doing so, we are incorporating their subjective and personal experience as part of the diagnostic and prognostic process.

As recently reported in Science Advances, we developed our novel classification system by analyzing patient survey data using machine learning methods. These data were collected as part of routine clinical care at Stanford’s Pain Management Centers via an open-source learning health system – CHOIR – used to characterize patients at each clinic visit.  Our overall approach aimed to classify patients into groups based on their general-health symptoms and then use their pain symptoms to confirm these groups’ diagnostic properties in terms of their pain symptoms.

We focused on nine general-health symptoms to identify the patient groups, such as fatigue, depression, anxiety, and social isolation. These symptoms are commonly used to assess patients' health who have a wide variety of chronic health conditions, such as cancer and cardiovascular diseases, among others.

For validation, we used various pain-specific measures such as the intensity of experienced pain; the body regions in which chronic pain is experienced; the extent pain interferes and restricts daily life activities; and maladaptive pain cognitions.

Initially, we trained our algorithm on a CHOIR data subset of 11,448 patients. Our analysis of this dataset identified an optimal solution of three distinct groups of patients. These groups differed in severity levels across all general-healthy symptoms, which is an expected result of the algorithm. Importantly, we validated the groups by demonstrating they also differed in all of the pain-specific measures.

The differences between groups in each of the measures were the same – one group always had the lowest level of severity, another group always had the worst level of severity, and the third group of patients was always in the middle. In other words, there was a linear scale of severity between these three groups. We demonstrated these same results in two additional CHOIR subsets of 3817 and 1273 patients, supporting the robustness of our findings.

One key finding related to the importance of each of the nine general-health symptoms to the group assignment process. In line with our previous perspective on the strong relationship between pain and emotions, we confirmed that the severity of depression, anxiety, and anger were the most critical factors in assigning patients to their groups. This suggests that negative emotion-related health factors might hold the key for understanding the underlying mechanisms that differ between the three patient groups. It also suggests that clinicians should consider treatments to reduce emotion-related symptoms when treating patients, in addition to the traditional approaches to target peripheral sources of pain—especially those with the most severe chronic pain symptoms.

Another important finding was that patients with pain in multiple body locations suffered more than those with a more localized pain condition. However, pain in specific body locations — such as the lower back, head, or pelvis — were not associated with any of the assigned groups. This highlights that instead of using the location of pain in the body for diagnosing chronic pain conditions, clinicians should consider the particular patterns of more general-health symptoms, and this may also guide their treatment recommendations to reduce pain-specific symptoms.

Finally, the CHOIR subset of 1273 patients included follow-up assessments that occurred 3 to 12 months after their initial CHOIR survey. Using this longitudinal dataset, we demonstrated that the initial assignment of patients into groups predicted the severity in all general-health and pain-specific measures of the follow-up assessments. A linear relationship was apparent again – the group with the lowest level of symptom severity at initial assessment continued to have the lowest level of severity at follow-up, and likewise for the other two groups.

However, after running the group assignment algorithm on the follow-up assessments, we observed that about 30% of the patients changed their group assignment between the initial and follow-up surveys. This suggests that various factors can improve or worsen an individual patient’s chronic pain condition. Thus, the challenge for future clinical and research efforts will be to personalize treatments to increase the likelihood of an overall reduction in general health and pain severity.

Overall, our study suggests that classifying patients with chronic pain based on their general-health symptoms may help improve their diagnosis, prognosis, and pain management. These findings are particularly important during the COIVD-19 pandemic when clinicians cannot always physically examine their patients.

Further research is needed to generalize our results to larger, more diverse groups of patients who are followed for a longer period of time. However, in the future, we hope our simple and cost-effective classification system based on general-health symptoms can support clinical decision-making to relieve the burdens of people who suffer from chronic pain — and even those who suffer from other diseases, such as cancer and cardiovascular diseases.

Dr. Gadi Gilam is a post-doctoral research fellow in the Stanford Division of Pain Medicine’s Systems Neuroscience and Pain Laboratory. The paper’s co-authors included lab members Eric Cramer MS, Dr. Kenneth A. Weber, Dr. Maisa Ziadni, as well as Dr. Ming-Chih Kao, Associated Chief of Stanford’s Division of Pain Medicine, and Dr. Sean C. Mackey, Chief of Stanford’s Division of Pain Medicine.