Grey correlations: A commentary on Chehadi et al.

In this issue, Chehadi et al. are presenting the results of a brain imaging study on the neural correlates of thought suppression (TS) in patients with chronic low back pain (Chehadi et al., 2018). Correlations between regional grey matter volume (GMV) and TS scores revealed a positive association in the right medial temporal gyrus (MTG) and negative associations in the left superior and middle temporal gyri (STG/MTG), right postcentral gyrus, and left precuneus. This study adds to an ever-growing body of studies examining brain characteristics associated with chronic pain, as well as chronic pain-related variables such as personality traits or attitudes towards pain. Unfortunately, the results of these different studies have so far been highly variable to the point that even meta-analyses present important discrepancies (e.g. Smallwood et al., 2013; Cauda et al., 2014). We believe that this problem derives from a lack of theory: data accumulates, but in the absence of an overarching model, the current state of the field resembles much the tale of the blind men and the elephant. This state of affairs is not uncommon in scientific fields dominated by mainly exploratory observational approaches, and has led to the historical demonstration of falsification's logical superiority over verification: it is easier to prove that something is false than that something is true. In the absence of a clear theory, all results have an equal importance and it is difficult to make progress. While we recognise the dangers of erecting premature theories as straw men for falsification tests, we believe that it may be possible to agree on a small number of principles that could help judge the relative importance of new findings pertaining to the neural correlates of chronic pain. First, most chronic pain syndromes probably follow a general diathesis-stress model of illness. In that perspective, it would seem advisable to try to discuss results in terms of diathesis and stress and to favour study designs that could help shed some light on this important question. Although prospective longitudinal designs are ideally suited to answer that question, useful information can also be derived from comparisons with healthy controls or patients with different types of chronic pain or related conditions. For instance, given the relatively high prevalence of chronic pain, predisposing factors should be relatively widespread amongst the general population, and hence differences between patients and healthy controls should be relatively small. Conversely, consequences of pain chronicity should be relatively specific to patients with CP. Moreover, several traits could also fall between predispositions and consequences. For instance, certain psychological traits may only become meaningfully expressed when people are subjected to real-life stressors. For instance, it may be difficult to measure TS when there are not any disturbing thoughts to suppress! Accordingly, Chehadi et al. found no significant correlations between TS and GMV in healthy individuals, suggesting that TS scores in healthy individuals may not be a risk factor for chronic pain because they probably reflect a slightly different construct in healthy controls not exposed to chronic pain. Still, in the absence of a prospective longitudinal design, it remains unclear whether the reported TS-GMV associations in patients with chronic pain reflect predispositions that can only be measured in the context of chronic pain, or if they reflect plastic consequences of the constant challenge chronic pain imposes on TS processes. A second principle to keep in mind when investigating associations between brain structure and psychological traits is that highly specific psychological constructs like TS are unlikely to map onto brain structure in a simple one-to-one manner. As mentioned previously, individual predispositions towards pain chronicity are probably fairly common and are therefore likely to be related to broad psychological constructs. Indeed, one of the most robust findings in Psychology is that there is only a limited number of personality traits, usually around five (Abram and DeYoung, 2017), that can be derived from questionnaire data. In one of the few large-scale prospective studies of pain chronicity, Fillingim et al. (2013) used principal component analysis (PCA) to identify psychological factors predicting the incidence of chronic temporomandibular disorder (TMD) pain. From the 26 different scales used in the study, only one factor related to global psychological and physical symptoms predicted the incidence of TMD. Consequently, the search for brain correlates of psychological factors associated with pain chronicity should start by examining these broad dimensions rather than individual variables taken in isolation. At the very least, new variables should be related to these broad constructs to help build a synthesis of the available knowledge. For instance, how does TS and STG/MTG volume relate to broad personality traits like neuroticism, or executive functions like inhibition or flexibility? Without such constraints, there is a risk to observe a combinatorial explosion in the number of papers reporting ‘original’ findings with new questionnaires in new sub-populations using new methodological techniques. Another challenge more specific to brain imaging relates to the difficulty in interpreting the directionality of associations with GMV, Indeed, the actual metric used for such measures, that is intensity gradients on T1-weighted scans, can be driven by a variety of underlying processes loosely related to actual GMV or density (Zatorre et al., 2012). In general, regional GMV positively correlates with several types of learned or innate skills, but there are many exceptions, which complicates interpretation. For instance, why would higher TS scores be associated with lower GMV in the left STG/MTG? In those circumstances, triangulating with other important variables may facilitate the interpretation. For instance, Chehadi et al. cleverly used mediation analyses to show that the portion of variance in STG/MTG GMV that negatively correlates with TS is also negatively related to pain (negative mediation term). Thus, mediation can help interpret the meaning of GMV-TS associations in terms of pain outcomes: low STG/MTG GMV is detrimental here, but only because it is partly associated with high TS. Still, we should note that the analysis does not allow to infer causality: it could be that higher pain calls for more TS, or that TS paradoxically increases pain (i.e. ‘white bear effect’). Follow-up studies manipulating the alleged mediator are therefore needed to infer causality. For instance, if the following model is right (MTG/STG GMV -> TS -> pain), then therapy aiming at reducing TS should also reduce pain, but should not affect MTG/STG GMV. Conversely, if the alternative model is right (pain -> TS -> MTG/STG GMV), therapy should also increase MTG/STG GMV, but should not affect pain. Ideally, studies allowing new hypotheses to be generated and tested in subsequent studies should be favoured. One last caveat regarding the directionality of GMV effects is related to the use of covariates. After controlling for depression, the relationship between STG/MTG GMV was found to be negative, but without controlling for depression the association was probably positive. It is important to always ask ourselves how the use of covariates main change the process we are interested in. For instance, which process is reflected by the portion of TS that does not correlate with depression (e.g. maybe successful TS)? Or perhaps the part of TS that we should be interested in is the one that covaries with depression (e.g. unsuccessful rumination)? What is important to realise here is that adding covariables may change the nature of what is being measured, that is TS controlling for depression may probably something much different from the original TS score. Under those circumstances, it may sometimes be preferable to maintain the original construct intact and simply discuss the relationship of the construct with other similar variables. In any case, serious thought should be given about the consequences of adding covariates. In conclusion, we believe that there is value in exploratory studies like that of Chehadi et al., but findings should ideally lead to the formulation of new testable hypotheses that could impact on our understanding of chronic pain. In the absence of clear hypotheses, results from exploratory studies should be guarded against false positives to provide solid empirical foundations for theories to emerge. Ideally, novel findings should be cross-validated to generate new hypotheses that should then be tested in openly available independent dataset (longitudinal and transversal chronic pain brain imaging data are available from collaborative initiatives, such as open pain http://www.openpain.org/). This strategy may be the ultimate solution to circumventing the quintessential ‘chicken or the egg’ problem of chronic pain neural correlates. Using this approach, any given regions, such as the MTG/STG identified by Chehadi et al. on the basis of TS scores, can easily be tested as a predisposing factor or a consequence of persistent pain. This way, isolated findings can be integrated to build a more unified theory of pain chronicity that will help us generate more targeted and productive hypotheses.

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.