Four Pain Profiles and Their Health Implications in chronic/persisting pain [long] – HealthSkills Blog

I came across this really interesting paper by Barron and colleagues (Barron et al., 2024) about ‘function-based pain profiles’ and their relevance to chronic disease management. This paper is a study analysing data from the UK Biobank to understand how well the typical ‘body part’ diagnostic approach fits with data-derived profiles drawn from the Biobank. It explores whether data-derived profiles capture useful differences in ‘medical’ history.

In brief, the authors drew data from a subset of the UK Biobank respondents involving 34, 337 participants. They used high-resolution structural brain scans and other data collected from these participants, did some seriously geeky statistical analysis (more on this shortly). There are essentially two groups of data: (1) ‘brain’ stuff drawn from MRI (brain volume, mainly); and (2) pain data drawn from the databank. This pain data included phenotypic, diagnostic and medication history, including items from the PROMIS-29 Scales and Pain Intensity Items such as pain interference, physical function, pain intensity, social role participation, anxiety, depression, fatigue and sleep disturbance. The statistical analysis then identified relationships between these two groups of data.

Now let’s very briefly (and simplistically!) break down the analysis. Partial Least Squares (PLS) is a pattern-learning algorithm that examines the relationships or correspondence between two sets of highly variable measures, in this case, one set of brain structure measures and one set of pain measures, setting out to identify shared variation across both sets of data.

My understanding of this analysis is that if we have two groups of data, one called X and the other called Y, and we ‘centre’ them (ie find out the dominant dimensions of those datasets), fiddle with how to view those dimensions so we can examine the relationships between them (look at the latent variables expressed by those measures), we can then find out what’s revealed. A ‘latent variable’ in statistics is a random variable that is not directly measured but not necessarily unmeasurable. It is introduced into a model to represent features of interest that are not directly measurable (https://www.sciencedirect.com/topics/social-sciences/latent-variable). Latent variables are inferred from data, rather than directly measured.

The researchers linked real-world reports of pain from individuals, with individual differences in brain structure, and identified four profiles. It’s important to note that sex, age and BMI were relevant to these profiles but didn’t distinguish between them, and neither did pain intensity or location. This is important given how frequently these conventional variables are used in research when slicing and dicing the data.

The four profiles?

1. Pain interference
2. Depression
3. Medical Pain
4. Anxiety

Note that the names given to these profiles were developed by the authors – so they’re names these authors think represent the groups, but they’re not set in stone! The four groups represent latent associations between ‘brain stuff’ and ‘pain measures’. You’ll also see there are many, many overlaps between the variables in each group.

Group 1: Pain interference describes an association between symptoms, medications, diagnoses and phenotypes of various chronic diseases. People in this group reported difficulty doing daily life activities, experienced low mood, and pain in legs/knees/osteoarthritis/migraine/gout/diabetes, and relevant brain regions of bilateral default mode network, the right control network, the right somatomotor network and bilateral visual systems (p. 12 of 22). Participants in this group were more likely to use opioids, blood pressure modifiers, metabolic disease modifiers, reflux and migraine; and were also likely to have arthropathies (anywhere in the body), cardiovascular disease and metabolic diseases. This group of participants were more likely to have higher body mass, lower household income, receive disability allowance, were less likely to work, had a slower walking pace, more substance use, poorer cardiovascular indices, and more health impairment in their blood with things like higher lipid and cholesterol levels, higher A1C (glycated haemoglobin), creatinine, and changes associated with poorer renal function and gout. This pain profile wasn’t associated with a specific body part, and instead tended to represent associations with chronic diseases and management. It was the most statistically significant group, and the authors note that many of the associations are with potentially modifiable chronic diseases.

Group 2: Depression describes a group characterised by difficulty concentrating, lack of enjoyment in life, slowed up movement (psychomotor slowing), feeling inadequate and not being able to do their normal roles. Pain intensity and interference influenced this group, particularly in terms of how pain affected mood, enjoyment in life, sleep, overall activity, work and relationships. Brain regions associated with this group included the bilateral default mode network, the right control network, the right somatomotor network and bilateral visual systems associated with the first pain profile, as well as the bilateral somatomotor network and the left salience/ventral attention system. Participants in this group were more likely to use antidepressants, cardiovascular drugs, metabolic drugs, and have reflux. They were also more likely to have cardiovascular disease, metabolic diseases, and some of the similar factors found in group 1 such as body mass, substance use, and hypertension. This group included those with cancer pain.

Group 3: Medical conditions is a group associated with diagnoses commonly known to have pain as a key feature such as osteoarthritis, cancer pain, gout, carpal tunnel, diabetes and diabetic neuropathy. The group were also more likely to have musculoskeletal pain (of both upper and lower limbs), poor sleep, abdominal pain, high pain interference and high pain intensity. Brain regions included above-average weightings for the bilateral default mode network, the right control network, bilateral visual networks and bilateral somatomotor networks (p. 13 or 22).

Group 4: Anxiety is a group characterised by difficulty with worrying, restlessness, trouble relaxing and feelings of doom. While depression was part of the picture, this group reported different aspects of low mood – mainly thoughts of death and poor appetite. These participants had more problems with fatigue, sleep and pain associated with medical conditions (as in Group 3), but had fewer pain-related variables (unlike Groups 1 & 2). Similarly to the other groups, the bilateral default mode network, the right control and right somatomotor networks were involved, but uniquely, this group showed more involvement of the bilateral dorsal attention networks. Participants were more likely to use antidepressants, meds for rheumatological conditions, metabolic disease, and cardiovascular disease. Diagnostic associations were with metabolic dieases, and both rheumatologic and sex-related disease (menopause – we could argue about this being defined as a disease, BTW!, but also vaginal prolapse). Other associations were with socioeconomic status, lifestyle (sexual partners, balding, exercise, sleep schedule), and similar risk factors for chronic disease such as body mass, substance use, and blood pressure. Metabolic, endocrine disorders were also indicated in blood test results.

So what question

I try not to write about things that won’t have any effect in practice, or don’t make a decent contribution to our understanding of pain in people. This study is really useful for showing that pain location (body part) doesn’t have a clear value when profiling people with pain. In other words, ‘spine’ pain or ‘headache’ weren’t revealed as clear-cut groups in the data. Instead, it was more common for people to have pain in many different parts of the body and associated with chronic diseases, often metabolic and cardiovascular, and for there to be associations with modifiable factors such as substance use, body mass and socioeconomic variables.

Another important aspect of this study is the finding that people with different profiles of pain were prescribed different medications. Opioids and antidepressants, both separately and prescribed together, were associated with reports of depressive symptoms but not pain interference – in other words, worsening mood and not pain interference is associated with opioid prescribing in this group of 167, 203 participants in the UK.

Caveats must be applied to any implications drawn from this research. This is correlational work, not causal. We don’t know which came first: are those brain factors present before people develop pain problems and the other associated health issues or did they develop afterwards? We don’t know the direction of any associations, and we can’t say things like ‘Group 4 have pain from anxiety’ because this kind of study cannot determine this.

Neither can we draw conclusions about ‘medical pain’ and ‘other pain’ in terms of ‘authenticity’ or whether they are ‘psychological’ or otherwise in origin. The authors clearly state that ‘it is possible that the medical pain profile simply captures the portion of the [Biobank] cohort that suffers from chronically painful conditions, albeit with less functional deficits related to pain interference or mood dysfunction.’ They go on to say “While our sample likely overlaps with that used by Johnston et al., our results align with the conclusion that the nervous system remains the common organ [italics mine] that predisposes, processes, and perceives painful stimuli, independent of mood dysfunction or specific medical condition and, critically here, independent of whether a patient reports a specific body part in relation to their pain.”(p. 17 of 22).

Interestingly for brain geeks, three important areas of the brain were strongly associated with pain – the default mode network, control, and attention networks. These areas (and the coordination between them) keep coming up as areas of interest when looking at how our conscious experience and attention to nociceptive stimulation emerge. To argue that pain (our conscious unpleasant, sensory and emotional experience associated with, or resembling that associated with actual or potential tissue damage, IASP) is not intimately influenced by how our cortical structures process information flies in the face of consistent research findings.

What do we do with this information?

Well, this is a long post, lots to chew on, yet the research is quite new and we need to know more. My take-homes are that the location of pain is probably less critical than the overall burden of pain on individuals and how they live their lives. It’s unlikely, on the basis of this study and many others, that we’ll find the One Exercise to Fix Back Pain, or the One Pill to Treat Pain, or indeed, any single therapy of any kind that will have a solid impact across the board. And please don’t suggest ‘exercise’ and ‘diet’ will be The Things. We’re talking very complex relationships between many variables in a context of real world environments, both social and structural – and making a difference at a population level is very different from how we work with individuals.

Barron, D. S., Saltoun, K., Kiesow, H., Fu, M., Cohen-Tanugi, J., Geha, P., Scheinost, D., Isaac, Z., Silbersweig, D., & Bzdok, D. (2024). Pain can’t be carved at the joints: defining function-based pain profiles and their relevance to chronic disease management in healthcare delivery design. BMC Medicine, 22(1), 594. https://doi.org/10.1186/s12916-024-03807-z