What clinicians and researchers mean by a sub-group is not always clear. Sometimes it may refer to clustering of effect modifiers within a distinct diagnostic group, what may be referred to as a 'clinical sub-group'. Other times the term is used to indicate groups with a unique mechanism or cause underlying the disorder, the terms 'aetiological' or 'mechanistic sub-group' probably best capture this perspective. Both are seen in the LBP literature, for example the three major categories in the McKenzie approach 34 can be thought of as mechanistic sub-groups, whereas the sub-classifications of 'derangement' by directional preference and symptom distribution represent clinical sub-groups under this single mechanistic umbrella. Likewise, O'Sullivan 35 has proposed three broad mechanistic sub-groups for CLBP. These have similarly been divided into clinical sub-groups; the broad aims of treatment are the same for each mechanistic sub-group though the specifics of intervention vary within the clinical sub-classifications. It is mechanistic sub-grouping that is under discussion in this paper. Patients' presentations and responses to treatment will always vary and require personalisation of care. However, it is important to consider if these variations need be interpreted as indicating distinctly different mechanisms underlying the disorder.

The statistical problems and pitfalls of sub-grouping have been reviewed extensively in the literature 363738. However, other forms of bias in the construction of mechanistic sub-groups have been less often discussed. In particular, it is possible that a form of confirmatory bias pervades aspects of the sub-grouping argument 39. That is, explanations for the phenomena of LBP other than sub-grouping are not considered. There are a number of characteristics of the LBP experience that have prompted the proposition of a sub-grouping approach, and while sub-grouping is consistent with these, it is worthwhile to consider other potential explanations.

There is no doubt that CNSLBP represents a complex, multi-dimensional problem. The chronic LBP experience is characterised by a vast array of physical 404142 and psychosocial 4344 features. It is this complexity that is cited as a compelling argument for the existence of distinct mechanistic sub-groups within the CNSLBP population 3145. Complexity in presentation does not have to be interpreted as evidence of distinct mechanisms. The literature is replete with descriptions of the same degree of physical and psychological intricacy within discrete diseases 46474849505152. In these examples, the presence of a single disease gives rise to a host of physical and psychological characteristics. As with LBP, these characteristics contribute to the consequences of the problem to that individual. The understanding of this complexity is vital in helping personalise treatment and determining best how treatment goals may be obtained for each patient. However, there is a distinction between varying and personalising individual care within a single framework and seeking distinctly different modes of care.

The multi-dimensional nature of CNSLBP may also be interpreted in this way. While complexity in behaviour can be explained by distinct mechanistic sub-groups, it can equally be interpreted as the individualisation of response to a single underlying problem, and while the complexity of the presentation requires that intervention be broad and personalised it may still have as its basis a single idea.

Not only is CNSLBP complex, it is also possible to observe clustering of various characteristics in this population 31335354. Historically, diagnosis has been driven by labelling clusters of signs and symptoms 55 and in CNSLBP it may be that these clusters represent different mechanisms requiring different interventions 28. However, it is again worth considering other explanations. Researchers have repeatedly demonstrated grouping of characteristics in patients with specific disease processes. For example, patients with Parkinson's disease may be stratified by rate of clinical progression 56 and individuals with sickle cell disease can be grouped by level of fear avoidance 57. Clustering of signs and symptoms will almost always be possible within single disease entities. One reason for this is variability in the severity of the condition. It is a reasonable prediction that those patients whose condition is less severe will exhibit different signs, symptoms and behaviours to those whose condition is moderate or those whose condition is severe. Clustering in presentation does not necessarily represent evidence of separate mechanisms. Different mechanisms may explain clustering, but equally it is possible to see and explain this phenomenon in problems with a single aetiology.

Another consideration is the variability seen in clinical trials of treatment for CNSLBP. Most commonly applied interventions have been subject to numerous studies, and for any given intervention it is possible to identify some studies in which there is a demonstrated effect and others in which there is no effect. Failure to account for sub-groups within the LBP population has again been cited as a potential explanation for these findings 3832.

There are several alternative explanations that are worthwhile contemplating. Firstly, statistical power may underpin some of the variability seen. The inconsistency does not necessarily reflect variability in treatment efficacy, just variation in the statistical efficiency of studies to detect treatment effects. Secondly, there is likely to be some random error associated with a large number of studies. One would expect to see most studies yield results representative of the average effect of that intervention. Likewise, there would be some variability in the response, with a few studies showing effects greater than the average and a few studies showing effects lower than the average. The variation may simply be noise, rather than variability induced by "innovative trial design" 32. Finally there is some suggestion that authorship may influence the interpretation of results. Variability in the outcome of clinical trials may simply reflect different interpretations of effect sizes by authors 5859, rather than evidence of bias in trial design or greater attention to sub-grouping.

A further commonly applied argument for the pursuit of mechanistic sub-groups is that NSLBP is too broad a category. Researchers suggest that since the majority of low back pain patients fall into this category, it's value in characterising a patient's presentation is only marginally more helpful than simply stating that they have pain in the back 3140. NSLBP is indeed a broad category, representing the vast majority of those who develop a complaint of back pain. However, CNSLBP represents only a small percentage of this population, probably less than 10% 121314. It is a highly stratified group, representing that collection of patients whose symptoms were severe enough to warrant consultation, who had neither specific spinal pathology nor nerve root pain and whose symptoms failed to settle in the usual time frame.

This section concludes that a number of the features of CNSLBP can be explained by mechanisms other than the inculcation of mechanistic sub-groups. This does not invalidate sub-grouping, it merely recognises that it is one of several alternatives. A more robust critique of the sub-grouping perspective requires consideration of the predictions that a sub-grouping model makes, and investigation of data related to these predictions. This is the aim of the next section.

Supporters of the sub-grouping model contend that the small effect sizes seen in clinical trials result from the large positive effects seen in some patients being 'washed out' by the smaller or negative effects seen in others 32. The solution offered for this problem is the matching of a diagnostic profile with the type of intervention best suited to that profile 32. If this is the case there are a number of reasonable predictions that can be made from the clinical literature. Firstly, it would be expected that a treatment process that is derived from a patient examination and tailored to the individual's presentation would be superior to a treatment process that is not. Secondly, multimodal care is likely to be superior to unimodal care as there is less chance of 'washout' of treatment effect. Thirdly it should be possible to identify groups of patients who respond favourably to one form of treatment and less favourably to alternative treatments.

Several different protocols have been used that compare tailored to non-tailored treatment programmes, and all reach similar conclusions. Comparisons between treatment programmes based on individual patient assessment and more general and generic treatment programmes demonstrate equivalence in outcome 606162636465666768. Providing therapists with the opportunity to assess patients and provide them with the care they consider best addresses the mechanism underpinning their disorder does not seem to be better than treatment that is generically applied. Patients participating in general physical activity appear to do better than those engaged in specific back exercises 69 and level of training and experience of the therapist, presumably allowing more appropriate matching of treatment with the patients' profile, seems not to impact on outcome 70. Overall, there seems to be little support for the view that specific, individualised treatment produces better outcome.

Kent and colleagues 71 systematically evaluated this perspective for manual therapy using meta-analysis. Their review compared the effects of manual therapy when therapists were free to select the intervention used, to studies in which the manual therapy treatment was prescribed. They were unable to detect any difference in outcome between these two conditions. Using a robust methodology, this group found no support for the perspective that individualised and clinically reasoned interventions are superior to generically applied interventions.

Hayden et al 21 compared outcome from individually tailored exercise programmes, which presumably sought to address the mechanisms underlying a patient's disorder, to standardised programmes. Individually designed programmes had no effect on function in comparison to general programmes and only a small and clinically unimportant influence on pain.

If the CNSLBP population contains distinct mechanistic sub-groups requiring different intervention strategies, then a reasonable prediction would be that multimodal interventions demonstrate superior outcomes to unimodal care, particularly when the unimodal components have been shown to impart some benefit to patients. The sub-grouping model would predict that subjects in multimodal care have a greater likelihood of receiving their appropriate intervention, thus decreasing the chance of 'washout'. While there is a possibility that treatments may interact negatively, or in other complex ways, we consider this a reasonable prediction. While some papers show small differences with the addition of treatments 727374, the majority of studies report no difference when unimodal and multimodal approaches are compared 6875767778798081828384. Increasing the likelihood that patients are exposed to the 'correct' treatment does not seem to enhance outcome.

Hayden et al 21 again provide systematic and methodologically rigorous information on this perspective. They compared exercise only groups with groups that received exercise and additional therapy. The addition of other conservative care to exercise therapy for CLBP patients produced only small and clinically insignificant changes in pain and function.

One approach to investigating sub-grouping in LBP is to consider which cluster of patient factors identify those most likely to benefit from a particular treatment. Research on ALBP, for example, has identified distinct and theoretically plausible clinical profiles that predict success from manipulation 85 and stability training 86 and preliminary evidence suggests that grouping patients this way makes a significant difference to outcome 8788.

If CNSLBP is similarly made up of different mechanistic subgroups, and a particular intervention is applied to the whole group, those patients that are matched to that treatment should obtain benefit and their clinical profile ought to emerge from analyses of treatment moderators. There have been many attempts to elucidate the features that influence outcome in CNSLBP patients receiving care. These data are difficult to interpret. The results are influenced by the type of outcome investigated and the follow-up interval 89. In addition, there is great variability in the potential predictors that have been investigated, making comparison and synthesis difficult 90 and interpretation is often confounded by lack of a comparison group 91.

While acknowledging these problems, there is little evidence from clinical studies of a patient profile that responds uniquely to a given intervention. Pain intensity, duration of the problem and distress appear repeatedly as important predictors of outcome regardless of the type of treatment 30909293949596. There is a sense that the characteristics of patients who do poorly or well with treatment are the same regardless of the intervention 9293 and may simply be a reflection of severity. Unlike the situation in ALBP 858697 it has not yet been demonstrated that the chronic population contains unique groups that respond differently to different interventions.

In the previous sections we saw that the features of CNSLBP that have prompted clinicians and researchers to consider sub-grouping may have alternate explanations and that clinical data on CNSLBP does not strongly support the presence of sub-groups. Despite its popularity, sub-grouping may not be an important explanation for the disappointing results seen in clinical trials. An alternative perspective is that treatment has been ineffective because it has been misdirected. One possible scenario is that the problem of CNSLBP does not lie within the back, but within the brain. Thinking of CNSLBP as a problem of cortical reorganisation and degeneration may increase our understanding of the problem and more appropriately direct intervention.

There is growing evidence that the brains of patients with CNSLBP are different to those of normal subjects 9899100101102 and there are corresponding alterations in brain function 99103104. Herta Flor's group 99 provided the initial data in this area. They recorded evoked magnetic fields in the brain in response to electrical stimulation of the back. They demonstrated greater reactivity in the primary somatosensory cortex (S1) and an expansion of the S1 representation of the back in CLBP patients. Altered S1 representation is also a feature of other complex pain problems 105106.

Recently investigators have looked more widely at brain structure and function, and a consistent pattern of brain changes has begun to emerge. Siddall et al. 101 used magnetic resonance spectroscopy (MRS) to monitor biochemical changes in the anterior cingulate cortex, prefrontal cortex and thalamus of 32 CLBP patients. This method was highly accurate in distinguishing CLBP patients from normal volunteers. Grachev et al. 100 also studied the biochemical profile of these three brain regions. They looked at the concentrations of nine different metabolites using proton MRS. Again there was strong evidence for alterations in the biochemical profile of CLBP patients, most notable was evidence of neurodegeneration within the dorsolateral prefrontal cortex (DLPFC), a result replicated more recently in a group of depressed CLBP patients 107.

Brain morphometry has been used to investigate both general and regional gray matter density changes in CLBP patients. Apkarian et al. 98 reported a greatly decreased neocortical gray matter volume in patients compared to matched controls. Regional analysis supported the biochemical data, with the DLPFC showing the greatest extent of loss. Less extensive gray matter loss was also noted in the thalamus. Schmidt-Wilke et al. 102 also found gray matter loss in the DLPFC of CLBP patients as well as decreased volumes in S1 and the brainstem. This group also reported small increases in density in the thalamus and basal ganglia. Finally Honda et al. 108 looked at cerebral blood flow in CLBP patients as part of a larger study on diverse chronic pain states. Again the DLPFC featured, demonstrating decreased blood flow at rest. Decreased blood flow was also observed in the orbitofrontal cortex and anterior cingulate cortex.

Alteration in brain function may be the mechanism that underpins the problem of CNSLBP. The patterns of central changes are consistent across a number of studies and methodologies and the extent of the central nervous systems changes are proportional to the duration and severity of the condition 9899102. Indeed, Baliki et al. 109 suggest that brain changes account for over 70–80% of the variance for intensity and duration of CLBP. Cross-sectional studies always raise the problem of causality, and while there is no equivalent information on CNSLBP, some small longitudinal studies on other pain problems characterised by altered cortical function suggest that interventions directed at cortical processing may be effective 110111112113, in two studies more effective than conventional care 111113. Furthermore, good outcome is strongly related to normalisation of cortical function 114115.

A cortical model of CNSLBP must be able to explain the complex clinical presentation of the problem. Critical challenges include accounting for the innumerable physical and psychological manifestations of CNSLBP and offering an explanation for the equivalence in outcome of very disparate treatments. Though speculative, alteration in brain function may offer some interesting perspectives and insights into these issues.

There is now strong evidence that structural changes within the spine have little meaning in the context of CNSLBP 18116117. Recent evidence suggests that structural abnormalities have little impact on the outcome of conservative treatment 118 and invasive interventions, specifically designed to address putative peripheral pathology, appear to offer little benefit to patients 16116119. Pain beyond the normal healing time or in the absence of any meaningful peripheral pathology has long been a challenge in the understanding of CNSLBP.

Sensitization of the nociceptive system is one suggested explanation for this problem. Enhanced synaptic efficiency of nociceptive networks may facilitate the perception of pain with little or no peripheral nociceptive input. Evidence indicates that there is some degree of sensitisation in CNSLBP patients 99120121. Theoretically, this may be the result of changes in the periphery, the spinal cord 122123, the brain 124 or in a combination of these areas.

Studies of inflammatory and neuropathic pain models in animals demonstrate extensive structural and functional alterations in the dorsal horn of the spinal cord that represent a relatively specific degeneration of the inhibitory interneurone system, leading to central sensitization and increases in pain-related behaviours (for review see 122). Similar changes in the dorsal horn have been demonstrated post-mortem in human subjects who had suffered from post-herpetic neuralgia 125. While it must be considered that CNSLBP develops in the absence of demonstrable peripheral nerve injury it is possible that the initial afferent nociceptive barrage that follows acute spinal injury may induce lasting alterations in dorsal horn function that lead to the maintenance of pain after tissue healing has occurred.

Methodological barriers to studying cord structure and function in humans make it difficult to ascertain the contribution of such changes to CNSLBP. Furthermore, in rodents, the spinal cord represents a far greater proportion of the CNS than in humans, and animal models demonstrate that while cord changes are prevalent in younger animals, brain changes dominate in older animals 126. Nevertheless, the animal data is important as it suggests a common aetiology of neurodegeneration. While there is currently no evidence of spinal cord changes in CNSLBP patients, there is significant data on altered brain function 99103121124127. It could be argued that supraspinal changes may be more likely to drive the sensitised state seen in human subjects with CNSLBP.

The work on brain changes in CLBP patients provides speculative support for this view. Schmidt-Wilke et al. 102 demonstrated gray matter degeneration of the brainstem, an area highly associated with inhibitory pain control, potentially leading to a loss of anti-nociception and the promotion of a sensitised state. There is also strong evidence for the role of the DLPFC in pain modulation and inhibition. Activity in the DLPFC has been shown to be negatively correlated with pain intensity and unpleasantness 128. This area appears to have a role in affective disengagement from pain 129 and activity in this region is negatively correlated with the degree of pain related catastrophising 130. These functions are likely to be affected by the extensive DLPFC degeneration seen in CLBP. In support of this idea, Baliki et al. 109 have shown that the sustained spontaneous pain of CLBP is strongly associated with increased activity of the medial pre-frontal cortex (mPFC) and suggest that this is the result of a loss of inhibitory control of the mPFC from the DLPFC.

A second proposition for the perception of pain in the absence of spinal pathology is the presence of central pain memories. Flor 131 has suggested that the alterations in sensory cortical representation found in chronic pain represent the neuronal substrate of implicit pain memories that serve to maintain and enhance pain perceived in the affected body region. As well as providing a mechanism for the maintenance of ongoing back pain, this may help to explain how previous episodes of low back pain lead to the development of chronicity, by contributing to and reinforcing these pain memories.

An alternative mechanism has been proposed by which cortical reorganisation may lead to the generation of ongoing, movement related pain from within the brain. Harris 132 suggests that altered cortical representation of somatic input may falsely signal incongruence between motor intention and movement. The generation of motor activity within the central nervous system is closely coupled to sensory feedback systems, which are monitored to detect any deviation from the predicted response 133. If there is conflict between motor output and sensory feedback, it is hypothesised that pain is produced as a warning signal to alert the individual to abnormalities within information processing 133. Alteration of the lumbar spine's representation in S1 99 may be the substrate of abnormal proprioceptive representation of the back and therefore a possible source of sensory-motor incongruence.

McCabe et al. 133 provided some experimental support for this hypothesis by inducing pain in healthy volunteers moving in an environment that created a degree of sensory-motor conflict. These results have attracted some criticism 134 and research using an alternative methodology to disrupt cortical proprioceptive representation was unable to evoke pain 135. It is worth considering that CNSLBP develops following an initial painful event and it is arguable that sensory-motor conflict may have more potential for generating pain when the pain 'matrix' is in a sensitised state.

Lack of visual input of the moving part is a critical component of the incongruence hypothesis 132. Vision has a dominant role over other senses in self-recognition, and many studies demonstrate that body position sense is re-calibrated to conform to visual information (for review see Jeannerod 136). Thus the lumbar spine, which is not directly visualised during task performance, may be more susceptible to this problem as abnormal cortical proprioceptive representation cannot be 'corrected' by visual feedback.

Interestingly, conditions that generate sensory-motor conflict are associated with increased activation in the right dorsolateral prefrontal cortex (DLPFC) 137. As mentioned above structural neurodegeneration and altered activity of the DLPFC is a consistent finding in patients with CNSLBP 98102107, additionally a number of studies have shown these changes are localised to the right hemisphere 102107. Apkarian et al. 98 suggest excitotoxicity secondary to long-term potentiation as a possible cause of this degeneration. One possible candidate for inducing these changes is over-activation secondary to sensory-motor incongruence.

In summary, there are numerous ways whereby reorganisation within the brain may generate persistent pain in the absence of ongoing peripheral pathology. These include, reduced cortico-cortical and descending inhibition producing an abnormally sensitised state, the reactivation of implicit pain memories, and centrally generated pain in response to sensori-motor incongruence when the patient moves the back. It is plausible that these mechanisms coexist. It is currently unknown to what extent that these brain changes are reversible. Similar but distinct brain changes are noted in phantom limb pain and CRPS I, and both these conditions appear amenable to treatment directed towards the brain 110111112113114. If some of the changes noted are not reversible it might still be possible to alter pain by affecting processing within this altered system.

The biomechanical aspects of LBP have been widely investigated 138, and the range of mechanical manifestations seen in CNSLBP patients is vast. The back is moved less during functional tasks 139140141142143 and there is greater asymmetry and variability in performance 140144145. There is a consistent decrease in the velocity of movement 146147148. Muscles are activated in a dys-coordinated manner in static and dynamic situations 149150. When lifting, the back muscles are recruited earlier and stay on for longer and there is a greater amount of co-contraction 151152153. There are altered patterns of muscle recruitment with limb movements 154, with sudden loading and unloading 155156157158, and balance is impaired 157159160161162. There is some suggestion of poorer proprioception 163164165 and there are delays in reaction times to a variety of tasks 166167168.

The congruence seen within these data is surprising. The overwhelming sense is a loss of confidence, control and coordination within the back. While Moseley and Hodges 169 contend that it is unclear if these abnormalities are compensatory, causative, neither or both, a common approach is to interpret these findings as causative. Poor stability and control is thought to influence loading of the lumbar spine and sustain the production of peripheral nociceptive input 35154170171. Within this framework treatment is directed towards normalising motor performance based on a peripheral biomechanical model. However, it may be that these physical abnormalities are not causative but result from altered higher centre representation and the individual's attempts to maintain function with an altered body schema. It is conceivable that these features are epiphenomena of cortical dysfunction rather than the underlying mechanism of CNSLBP. If this is the case treatment guided by a biomechanical explanatory model might be misdirected.

Support for this view can be seen in a number of areas. There appears to be little association between changes in patient centred outcome and changes in any aspect of low back related physical performance 117. Lumbar stabilisation training programmes, which interpret these problems as causative and specifically address them from a biomechanical perspective 172 and other active 173 and passive 174 interventions aimed at changing the mechanical behaviour of the lower back demonstrate limited clinical effectiveness.

Psychological factors are an important part of the chronic LBP experience. They contribute to the progression to chronicity, explain a significant amount of the variance among CLBP patients and have been identified as important barriers to resolution 44175176. Alterations in brain function offer some additional insight into the role psychological features play in the development of CNSLBP and the maintenance of the problem once it is established.

Pincus et al. 175176 cite distress/depressive mood as integral psychological factors in the transition from acute to chronic LBP and several psychosocial models have been postulated for the role of distress/depression in this transition 176177. Although speculative, brain reorganisation may also be significant. It appears that there is considerable overlap in the neural circuitries of the brain in both depression and CLBP 107. It is of particular interest that alteration of DLPFC function has been strongly implicated in depressive disorders 178179. Although its precise role is currently not clearly defined 180181, evidence points to primary abnormal activity within the DLPFC 107. Depressed CNSLBP patients have been shown to have evidence of DLPFC degeneration and the level of degeneration was highly correlated with the level of depression 107. One mechanism by which depressed mood influences chronicity may be via contributing to DLPFC degeneration.

Psychological variables also account for a significant amount of the variance in the CNSLBP condition 44. Despite their explanatory value, the evidence for the efficacy of psychologically based treatments is not strong. Psychologically focused treatment only delivers small effects on pain and function when compared to no treatment 7. They seem to be no more effective than other active interventions 7117182 and provide little additional benefit when added to other conservative treatment approaches 84117183. The brain changes seen in CNSLBP may help explain this situation.

The fear-avoidance model is seen as a central psychological mechanism in the maintenance of CNSLBP 184185. Fundamental to this model are the opposing behavioural responses of confrontation and avoidance. Confrontation is seen as adaptive and likely to lead to resolution. Alternatively, avoidance is thought to be maladaptive and contribute powerfully to the development of chronicity and numerous associated biological and psychological sequelae 186. Catastrophic beliefs and pain related fear are primarily what drives avoidance 184, and are consequently seen as the primary targets for intervention. The origins of these beliefs are viewed as a major challenge in developing the fear-avoidance model 186187 and numerous possibilities have been suggested 184185186187.

One possible candidate for the genesis of these beliefs is alteration in cortical representation of the back and resultant distorted body schema. Keefe et al 187 state that fear may be fuelled by unexpected and novel bodily sensations. Both McCabe et al. 133 and Moseley et al. 135 demonstrated that strangeness, foreignness and peculiarity are features of moving when there is sensory-motor incongruence. As with the biomechanical features of CNSLBP, fear and catastrophising may also be viewed as responses to an altered body schema. These beliefs would be a reasonable response to an individual with an altered cortical representation of their back who experiences unexplained pain and unexpected and novel sensations when moving. Treatments that seek to address these problems from a cognitive perspective may be less than ideal, as the fundamental substrate for the genesis of fear and catastrophisation would still be present.

It is also possible that degeneration within the DLPFC may contribute directly to greater catastrophic interpretation of pain. The DLPFC seems to be important in emotional disengagement from pain 128129 and it has been proposed that degeneration of the DLPFC may contribute to the production of spontaneous pain 109, which may fuel catastrophic thoughts, and, as discussed above, there is evidence that DLPFC activity is negatively correlated with catastrophic thinking about pain 130.

Individuals' coping strategies are also considered to be important contributors to the disability associated with CNSLBP. Active coping strategies, characterised by efforts to function in spite of pain or to distract oneself from pain are thought to be important in adaptive functioning 188. Psychological management of CNSLBP seeks to use distraction and encourages engagement in activity despite pain, as part of the treatment process 189. This requires that the individual is able to let task relevant stimuli to dominate over pain 128, which can be particularly difficult given the high attentional demands of pain 190. One suggested role of the DLPFC is in governing efficient performance in the presence of interfering stimuli 128. Degeneration of the DLPFC will obviously interfere with this ability. Patients may not respond to this component of treatment as the neural substrate for engagement in this process is not available to them. These views do not challenge the biopsychosocial approach to the management of CNSLBP. It is abundantly clear that the thoughts, feelings and beliefs of an individual impact significantly on the problem 444191. What may require reconsideration is the origins of some maladaptive thoughts and beliefs and how these may be best modified.

Figure 1. presents a speculative model for the development and maintenance of NSCLBP with altered cortical function at its core. In the model, current and previous episodes of back pain contribute to an altered cortical representation of the back. Conceivably, previous episodes of LBP may also increase distress about the problem. Alterations in proprioceptive representation, subsequent sensory-motor incongruence and pre-existing depressive mood lead to over-activation and subsequent neurodegenerative change in the DLPFC. Sensory-motor incongruence may also directly produce pain, sustain altered motor control strategies and contribute to fear and catastrophic thoughts. The resulting DLPFC dysfunction contributes to central sensitization and subsequent ongoing and exaggerated pain, and also decreases the patient's ability to disengage from the pain, thus feeding back into the negative psychological influences. The mechanisms presented in the model should be acknowledged as speculative but represent an attempt to interpret the alterations in brain function that have been demonstrated in CNSLBP in light of the clinical manifestations of the condition.

In summary, CNSLBP patients have back pain yet no conservative or surgical pain relieving measures directed at the back appear effective. They display a number of biomechanical abnormalities, however treatment directed at normalising lumbar biomechanics has little effect and there is no relationship between changes in outcome and changes in spinal mechanics. Finally, these patients demonstrate some psychological problems but psychologically based treatments offer only partial solution to the problem. A possible explanation for these findings is that they are epiphenomena, features that are incidental to a problem of neurological reorganisation and degeneration.

As discussed, the response of CNSLBP patients is very similar across a broad range of treatment options. A simple explanation would be to conclude that current interventions are ineffective or demonstrate only non-specific or placebo effects. However, the evidence does not support either of these perspectives. Most commonly applied conservative treatments demonstrate some level of effectiveness, albeit small, and this seems to be above what is produced by sham treatment 15. Alternatively, disparate treatments may demonstrate equivalence in outcome because they all work through the same mechanism. Massage, manipulation, exercise, acupuncture, education and other interventions may work by influencing cortical function. The effectiveness is comparable across treatments because the pathway by which they act may be the same.

This perspective may also explain the disappointing effect sizes seen with conservative care. A possible reason for the small effects seen is that none of these interventions are currently delivered in a way that focuses on central processing. Mobilisation of the lumbar spine might be made more cortically directed by having the patient respond verbally to the stimulus, trying to localise the level or determine direction of pressure, akin to the sensory discrimination task employed by Flor et al. 114 in managing phantom limb pain or by Moseley et al. 192 in CRPS. Exercises aimed at improving trunk muscle recruitment seem an ideal way to normalise cortical representation, but rather than emphasis being placed on performing a particular type of muscle recruitment pattern based on a peripheral biomechanical model 154170171 participation in a variety of challenging exercises may be more useful. Simply mastering any skill that the patient finds difficult may be all that is required.

Similar thinking can be applied to most commonly applied conservative interventions, and whilst changing the emphasis of existing treatments may be a viable option for enhancing clinical efficacy, it may also be useful to consider other possibilities. There is mounting evidence for the effectiveness of alternative and innovative strategies for "re-training" cortical function in other complex, long standing, pain problems 110111112113114193 as well as recent data that training of brain activation with functional MRI can favourably influence chronic pain 194. Advances in non-invasive brain stimulation techniques such as repetitive transcranial magnetic stimulation and transcranial direct current stimulation also offer approaches for altering cortical function to achieve pain relief 195196 and to enhance cognitive function 197198199. The challenge presented by this model for researchers and clinicians is to determine how therapy might be designed to address cortical function.

In conclusion there is little evidence to endorse current treatment for CNSLBP. Most treatments provide only small short-term changes and there is currently scant evidence that one form of treatment is superior to another. Sub-grouping of CNSLBP is a common response to these disappointing results. The sub-grouping perspective is not consistently supported by the available data and it is important that other research agendas are entertained. An alternative perspective is that treatment has been ineffective because it has been misdirected. CNSLBP may be a problem of abnormal cortical function. There is growing evidence of supraspinal abnormalities in CNSLBP, these mechanisms can potentially explain the complexity of the LBP experience and the outcomes of clinical trials on CNSLBP. Various research groups have demonstrated the potential value of 'training the brain' in other conditions characterised by abnormal cortical processing. Given the growing body of evidence indicating similar neural mechanisms in CNSLBP, the development of clinical strategies targeted at normalising neurological processing represents a challenging new direction for musculoskeletal clinicians and researchers involved in the management of CNSLBP.