Diagnostic work‐up of neuropathic pain: Computing, using questionnaires or examining the patient?

In the diagnostic work-up of neuropathy, painful or painless, classical neurological criteria apply, i.e., an etiology linked to neuroanatomically distributed (projected) sensory symptoms and signs. The notion that characteristic somatosensory descriptors, with an emphasis on pain descriptors, could be used to aid in the diagnosis of neuropathic pain has gained increased footing over the past decade (Table 1). From such studies it is clear that the overlap of descriptors between neuropathic and suggested nociceptive pain conditions is extensive and that no pathognomonic term exists (Bouhassira et al., 2005). It deserves to be emphasized that conditions labeled as nociceptive in such comparative studies are only tentatively so since positive diagnostic criteria for the included conditions are lacking (e.g., low back pain, musculoskeletal pain). Surveying such studies led an earlier editorial on the matter by Dr. Paolo Marchettini (2005) to the unwarranted conclusion that the search for "burning" as a pain descriptor might be crucial from a diagnostic point of view when the available time spent with the single patient is limited. The time issue is, to us, never a valid argument to turn to imprecise diagnostic approaches. Behrman et al., 2006 (this issue) set out to assess the diagnostic accuracy of two pain scoring systems, the neuropathic pain scale (NPS; Galer and Jensen, 1997) and the neuropathic pain questionnaire (NPQ; Krause and Backonja, 2003), and two methods for statistical analysis, i.e., logistic regression (LR) and artificial neural networks (ANNs). NPS includes ten pain descriptors (e.g. 'sharp', 'hot', 'cold', 'itchy') rated on a 0–10 scale and a temporal domain assessed categorically. NPQ consists of 12 descriptors (e.g. 'shooting', 'squeezing', 'overwhelming' and 'increase of pain due to weather changes') rated on a 0–100 scale. The clinical pain diagnosis arrived at by an experienced pain clinician on the basis of history, clinical examination and, when needed, laboratory tests served as the reference for the results of the NPS, NPQ, LR and ANN. Patients with a combination of nociceptive and neuropathic pains were asked to concentrate on the most severe component of their pain when completing the two questionnaires, and patients with pain of unknown origin were excluded. Altogether 155 patients (91 with predominately neuropathic pain and 64 with predominately non-neuropathic pain) were included in the analyses. The NPQ includes a calculation formula for the score to classify the pain as neuropathic or non-neuropathic. For the NPS, which was not designed as a diagnostic tool and hence does not provide a cutoff point value for the score to assess the pain as neuropathic or non-neuropathic, the authors set the cutoff point on the basis of data from another study (Argoff et al., 2006). The raw data obtained by the questionnaires were analyzed by linear LR and non-linear ANN, which classified the pains as neuropathic or non-neuropathic. The primary objective of the study was to evaluate if ANN (as a more complicated statistical procedure) could improve upon the scoring systems of the NPS or NPQ. The main result of the study was that the diagnostic accuracy of the NPS, NPQ, LR and ANN were 55.5%, 68.4%, 67.7% and 69.0%, respectively, which is far from acceptable in the clinic. Another essential finding was that the ANN provided only slight (1.3%) improvement compared to the results of the LR, i.e., a more complicated statistical procedure of the same data including various qualitative descriptors, did not improve the diagnostic accuracy. Pain type prediction by ANN was high in easy-to diagnose pain types such as causalgia (88.9%) and pain in polyneuropathy (81.3%) but low, actually about that of heads or tails when tossing a coin, in radicular neuropathic pain (55.6%) and myofascial pain (46.2%). As the diagnostic accuracy is so low, the methods studied here must not replace the traditional diagnostic work-up procedure in cases with suspected neuropathic pain. In the validation study of the NPQ, the sensitivity and specificity were 66.6% and 74.4%, respectively (Krause and Backonja, 2003), which means that one patient out of three with neuropathic pain was not identified as such and one patient out of four without neuropathic pain would receive a neuropathic pain diagnosis. An example of the risk of a false positive diagnosis is the suggestion of a neuropathic origin of fibromyalgia on the basis of the outcome of a screening tool (Martinez-Lavin et al., 2003). In the study of Behrman et al., there was a noticeable overlap in the descriptors between neuropathic and non-neuropathic cases, confirming that single descriptors have very little to offer, if anything, in the diagnosis of neuropathic pain (Rasmussen et al., 2004). The results of Behrman's study also demonstrate that the combination of descriptors of the NPS and the NPQ, even when analyzed combined with LR or ANN, should not replace the traditional diagnostic approach. CRPS-1 was regarded as a neuropathic pain condition both in the validation study of the NPQ and in this study, as was the case also in the validation studies of the NPS (Galer and Jensen, 1997) and the LANSS (Bennett, 2001), which lacks a valid rationale and contaminates the conclusions of these studies. CRPS-1 would more accurately be labeled pain of unknown origin since neither etiology nor pathogenetic mechanisms are known. Since the many different questionnaires available are not suitable for diagnosing neuropathic pain their future place in the clinical context, if any, must be proven by rigorous studies. It maybe suggested that they could prove useful for screening purposes in the clinic. A sensitive screening tool may increase the alertness of the physician to carefully examine the patient in search of neuropathic pain. On the other hand, every pain patient deserves to be examined thoroughly enough to identify the pain type and the potential disease or lesion causing it. The diagnosis of neuropathic pain should be followed by therapeutic considerations of both causative treatment, albeit possible only rarely, and symptomatic pharmacotherapy or neuromodulation. In this context, an over-simplified questionnaire-based diagnostic approach increases the risk of an over-simplified treatment approach. Questionnaires may also prove useful as an additional source of information in the most difficult-to-classify cases. If patients present symptoms and signs that point to a certain level of a lesion/disease but without fulfilled positive criteria to support underlying pathology, less firm additional support from questionnaires may provide treatment strategy guidance. In addition, the use of questionnaires may add to the detailed assessment of treatment effects (Galer and Jensen, 1997; Bouhassira et al., 2004). The commonly employed measures of pain intensity and pain relief may harbor some limitations and do not necessarily reveal the efficacy of a treatment on some components of neuropathic pain. The potential clinical significance of this waits to be gauged. In a recently suggested classification scheme (Rasmussen et al., 2004; Jensen and Hansson, 2006) pain descriptors were not included as part of the diagnostic criteria for neuropathic pain but rather factors such as etiology, pain distribution and somatosensory examination outcome, adding to the notion that the analysis of somatosensory descriptors, including pain, is not a recommended strategy to by-pass pivotal measures in diagnosing neuropathic pain. For now, classical neurological diagnostic criteria apply in the work-up of tentative peripheral and central neuropathic pains. Since our ignorance is still extensive in a number of areas related to the diagnostic accuracy in neuropathic pain states abundant work is still in demand. Computing sensory descriptor profiles is not part of that.