Diagnostic utility of patient history, clinical examination and screening tool data to identify neuropathic pain in low back related leg pain: a systematic review and narrative synthesis

To evaluate the diagnostic utility of patient history, clinical examination and screening tool data in order to identify NP in adults presenting with LBLP.

A systematic review was completed in accordance with a published study protocol [20]. The protocol was informed by the The Cochrane Handbook for Diagnostic Test Accuracy studies and the Centre for Reviews and Dissemination [21] and the Preferred Reporting Items for Systematic Reviews and Meta-Analysis-Protocols (PRISMA-P) checklist [22]. The systematic review is registered with PROSPERO (CRD42019140861). No changes were made to the original protocol [20].

The Sample, Phenomenon of Interest, Design, Evaluation, Research Type (SPIDER) guidelines were adopted to format and structure the eligibility criteria [23].

Two independent authors (JM, TN) independently searched pre-identified electronic databases (searched from inception to 31 July 2019), key journals and grey literature.

Searches comprised of:

The search was highly sensitive, devised in collaboration with all authors and a specialist librarian [20]. The key terms used for the search were: Diagnostic validity, Patient history, Clinical examination, Screening tool, Neuropathic pain and LBLP.

For the above search terms a list of synonyms and truncations were generated to maximise search inclusion. Key terms were formatted as per the requirements of each specific database in order to retrieve the maximum number of relevant articles. See example of search terms inputted into database (Box 1).

The Quality Assessment of Diagnostic Accuracy Studies 2 (QUADAS-2) tool was used as it is a recognised tool for assessing risk of bias (RoB) in diagnostic accuracy studies [26]. The four domains of the QUADAS-2 tool (patient selection, index test, reference standard and flow and timing) were independently assessed and judged by each reviewer (JM, TN) as ‘high’, ‘low’ or ‘unclear risk’. Reviewers then provided an overall verdict regarding bias of the studies assessed, ‘at RoB’ or ‘low risk’, if a study was judged as “high risk” or “unclear risk” in one or more domains then an overall judgement of “at RoB” was made [26]. The third reviewer (AR) was used to settle disagreements if consensus was not achieved between the two reviewers (JM, TN) on discussion. Furthermore, agreement was assessed between the reviewers (JM, TN) using Cohen’s k.

Summary measure tables were developed using the primary diagnostic data (sensitivity, specificity, LRs and PVs) retrieved from the included studies. Where data were not available the lead author (JM) used the raw data to calculate the missing results, using the formulae recommended by Akobeng [27]. Sensitivity and specificity cut of points were graded as low (≤50%), low/moderate (51–64%), moderate (65–74%), moderate/high (75–84%) and high (≥85%) as highlighted in the study protocol [20].

Heterogeneity was explored in relation to study design, population, comparable diagnostic data and reference standard to dictate the possibility of doing a meta-analysis. The data extraction form was used to compare study design, population and reference standards between studies and the summary measure tables were used to explore comparable diagnostic data. As stated in the study protocol [20], in the event that a meta-analysis was not possible a narrative synthesis would be conducted [28].

The Grading of Recommendations, Assessment, Development and Evaluations (GRADE) was used to assess the level of evidence; the GRADE method was adapted for diagnostic accuracy research (Table 1) [29]. The reviewers (JM, TN) assessed each included study according to five downgrading factors (RoB, inconsistency of evidence, indirectness of evidence, imprecision of results and publication bias) in order to assign GRADE ranking. The GRADE ranking process started with assigning an initial level of quality of evidence based on study design (cross sectional and cohort design considered high quality, any other design considered low quality) and then assessing the study against the downgrading factors to assign a final judgement on level of evidence [20]. Publication bias was suspected in situations where evidence was derived from a number of small studies.

Initial searches using electronic databases and additional sources resulted in 762 studies being retrieved. Following duplicate removal and title and abstract screening, 16 studies remained for full text review. On completion of full text screening, 5 studies were excluded and a subsequent 11 studies were included for analysis (Fig. 1).

Table 2 depicts the characteristics of the 11 included studies.

Complete agreement was achieved between the two reviewers for assessment of RoB, and thus the third reviewer was not required. Two studies were assessed as low RoB [12, 19], the remaining nine studies were considered at RoB (Table 3, Fig. 2). The primary concerns in relation to the at RoB studies were blinding of index and reference tests/insufficient description of procedures involved in index and reference test (six studies), flow and timing between tests (four studies) and patient selection (three studies). For all of the eleven included studies the reference standard for RoB and applicability was scored largely as unclear. This is because in the absence of a gold standard or clear recommendations/guidelines to diagnose NP in LBLP it is unclear whether the reference standards used in the studies correctly classify the target condition. Table 3 and Fig. 2 depict RoB and applicability concerns for each of the 11 included studies.

A meta-analysis was not completed since there were inconsistencies in the reference standard used between studies. Even amongst the studies that used the same reference standard, differences were highlighted in how it was measured [16, 17]. Furthermore, the majority of studies were considered at RoB making any further statistical analysis equally at RoB. Finally, the number of studies retrieved for screening tools (n = 2) and patient history taking (n = 3) were limited and the studies investigating a clinical examination test used a wide variety of different tests resulting in insufficient data for pooling. A narrative synthesis was therefore conducted.

One study, at RoB, investigated patient history data [34] in relation to diagnosing nerve root compression or herniated disc in patients with LBLP. This study investigated 20 separate patient history items (Table 4). Of the 20 items, moderate/high and high sensitivity values in both herniated disc and nerve compression groups were observed for health-related absenteeism (81 and 80% respectively) and in subjective sensory loss (89 and 90% respectively). Having had pain in the same leg previously demonstrated the highest specificity, in both herniated disc and nerve compression groups (90 and 91% respectively). Indirectness of evidence was highlighted as a highly selective population of patients were recruited (Table 5). Using GRADE, there is low quality of evidence to support the use of Verwoerd et al’s [34] patient history indicators in diagnosing nerve root compression or herniated discs (Table 5).

Two studies investigated both patient history taking and clinical examination findings together [18, 19], in relation to diagnosing peripheral NP in LBP with or without leg pain [18] and in suspected lumbosacral nerve root compression [19]. Smart et al. [18] identified a cluster of three signs and symptoms (pain referred in dermatomal or cutaneous distribution, history of nerve injury, pathology or mechanical compromise, pain/symptom provocation with mechanical/movement tests) which demonstrated a high sensitivity (86.3%) and specificity (96%) (Table 4). This study was considered at RoB as clinicians were aware of the reference standard before issuing the index test. Furthermore, indirectness of evidence was highlighted due the use of clinical judgement as a reference standard without specifying what criteria were used to make this judgement. Using GRADE, there is a low level of evidence to support Smart et al’s [18] cluster of signs and symptoms in diagnosing peripheral NP in LBP (Table 5).

Vroomen et al’s [19] study was deemed to be low RoB. Vroomen et al. [19] identified 8 signs (including patient history and clinical examination signs) which were predictive of lumbosacral nerve root compression demonstrating moderate sensitivity (72%) and moderate/high specificity (80%) (Table 4). This study shared one common item with Smart et al’s [18] cluster; pain referred in a dermatomal distribution. In both instances this indicator was used in association with other indicators, raw data were not available to assess this indicator in isolation. Vroomen et al. [19] used MRI as a reference standard, which has been questioned for its diagnostic validity [36], furthermore this study was investigating lumbosacral nerve root compression which does not necessarily infer NP. Using GRADE, there is a moderate level of evidence to support Vroomen et al’s [19] eight signs in diagnosing lumbosacral nerve root compression (Table 5).

Six studies investigated the use of clinical examination tests in isolation. All six studies were considered at RoB [14, 16, 17, 31‐33]. Two studies investigated the diagnostic accuracy of the SLR test for identifying sciatica. Capra et al. [16] found a low sensitivity (36%) and moderate specificity (74%) whilst Poiraudeau et al. [31] found the opposite, moderate/high sensitivity (79%) and low specificity (37%) (Table 4). Indirectness was highlighted in both studies partly due to the use of imaging as a reference standard (Table 5). Using GRADE, there is low level evidence to support the use of the SLR test in diagnosing sciatica. Poiraudeau et al. [31] also investigated three other tests (Bell’s test, HE test, Crossed lasegue test) all of which demonstrated low or low/moderate sensitivity and specificity values (Table 4), expect for moderate/high specificity found for the crossed lasegue test (83%). Using GRADE, there is low level evidence to support Bell’s test, HE test and Crossed lasegue test in diagnosing sciatica (Table 5).

The slump knee bend [32] was found to have high and moderate/high sensitivity and specificity values (100, 83%) diagnosing upper/mid lumbar nerve root compression, similarly the slump test [33] had high and moderate/high sensitivity and specificity values (91, 78%) diagnosing NP in LBLP (Table 4). Low sample sizes were characteristic of both these studies, with one being a pilot study [32]. Using GRADE, there is very low evidence to support the diagnostic utility of the slump knee bend and slump test in diagnosing upper/mid lumbar nerve root compression and NP in LBLP respectively (Table 5).

Nerve palpation was found to have moderate/high sensitivity (83%) and moderate specificity (73%) in identifying LBLP [35], the SLR and slump tests were used as reference standards which led to serious indirectness being highlighted in this study. Low quality of evidence supports the use of nerve palpation in diagnosing LBLP, following the use of GRADE (Table 5). SQST was found to have low/moderate sensitivity (62%) and high specificity (95%) when detecting lumbar lateral stenosis of the L5 nerve root [17] (Table 4). However, indirectness of evidence was highlighted as the participants recruited into this study were all surgical patients and therefore not fully representative of the target population for this review. Using GRADE low level of evidence supports the use of SQST in diagnosing lumbar lateral stenosis of the L5 nerve root (Table 5).

One study investigated four screening tools; S-DN4 (Self-completed douleur neuropathique, ID Pain, PDQ (painDETECT questionnaire) and S-LANSS (Self-completed Leeds Assessment of Neuropathic symptoms and Signs) [30] to identify NP in LBP. Three of the screening tools were identified as having a range of low/moderate to high sensitivity and specificity values; 58.5% & 98% (S-DN4), 70.7% & 84.3% (ID Pain), 76.8% & 78.4% (PDQ) (Table 4). However, the S-LANSS was identified as having a low specificity of 13% (Table 4). This study was deemed at RoB as patient applicability was compromised, this was partly due to the recruitment of patients with LBP with or without leg pain which is not consistent with the target population for this review. Furthermore, the reference standard, clinical judgement, was not adequately described and thus subject to bias. Additionally, this study was completed in a different language and cross-cultural validation cut of points used are yet to be validated. Using GRADE, there is low level of evidence to support the diagnostic utility of the S-DN4, ID Pain, painDETECT and S-LANSS tools in diagnosing NP in LBP (Table 5). The StEP tool [12] was found to have a high sensitivity (92%) and specificity (97%) when diagnosing lumbar radicular pain, this evidence was found to be of low RoB. Using GRADE, there is moderate level of evidence to support the diagnostic utility of the StEP tool in diagnosing lumbar radicular pain (Table 5).

Patient history indicators to diagnose lumbosacral nerve root compression have been investigated by Verwoerd et al. [34] (low level of evidence), this study found moderate/high sensitivity in; “health-related absenteeism”, high sensitivity in “subjective sensory loss” and high specificity in “having had pain in the same leg previously.” However, there is no further evidence to support these patient history indicators in diagnosing NP in LBLP. Clusters of patient history and clinical examination indicators have been highlighted by two studies in this review demonstrating high sensitivity and specificity in one study [18] and moderate sensitivity and moderate/high specificity in the other [19]. Low quality evidence supports a cluster of three signs and symptoms in diagnosing peripheral NP (pain referred in dermatomal or cutaneous distribution, history of nerve injury, pathology or mechanical compromise and pain/symptom provocation with mechanical/movement tests) [18]. Moderate level of evidence supports the diagnostic utility of a cluster of eight signs in diagnosing lumbosacral nerve root compression (two patient characteristics - age and duration of disease, four symptoms from the history - paroxysmal pain, pain worse in leg than back, typical dermatomal distribution, worse on coughing/sneezing/straining and two signs from the physical examination - finger to floor distance and Paresis) [19]. These two studies share only one common indicator; pain referred in a dermatomal distribution. However, this indicator was included as part of a cluster of signs/symptoms in both studies and therefore the diagnostic validity of this indicator alone is unclear. The 2016 Neuropathic Pain Special Interest Group (NeuPSIG) grading system highlights that in order for NP to be probable or definite pain/sensory signs must follow a neuroanatomically plausible distribution [37], which would encompass a dermatomal pattern, supporting the use of this clinical indicator. Conversely, research investigating entrapment neuropathies has demonstrated an extraterritorial spread of symptoms following mild sciatic nerve compression [38], disputing the use of this indicator. Due to the lack of clarity of the performance of this indicator in isolation and the uncertainty in the literature, the diagnostic utility of this patient history indicator remains unclear.

The SLR was found to have moderate/high sensitivity and low specificity when diagnosing sciatica [31], however on the contrary Capra et al. [16] found the opposite in their study investigating sciatica (low sensitivity and moderate specificity). Overall low level of evidence supports the use of the SLR in diagnosing sciatica. The slump knee bend [32] and slump test [33] were found to have high sensitivity and moderate/high specificity in diagnosing upper/mid lumbar nerve root compression and peripheral NP in the lower limb respectively. Very low level of evidence associated with both these tests were largely due to the small sample sizes used in each study. Evidence to support the use of neurodynamic testing to identify NP in LBLP is conflicting with an increasing body of evidence highlighting the low diagnostic validity of these tests [38].

SQST [17] demonstrated low/moderate sensitivity and high specificity when diagnosing lumbar lateral stenosis involving the L5 nerve root in a study at RoB. The population of patients used were all surgical and therefore not fully representative of the target population for this review, thus the applicability of these findings is poor. There is evidence to support the use of quantitative sensory testing (QST) in diagnosing small fibre nerve degeneration in entrapment neuropathies [39]. However, SQST differs to QST as it describes tests which are inexpensive and accessible within a clinical setting (e.g. coin for testing temperature). Evidence to support SQST to detect small fibre nerve degeneration is limited [40] and yet to be investigated in participants with LBLP. The sensory profiles of those with NP in LBLP is not known and therefore support for SQST in identifying NP in LBLP is inconclusive.

A range of low/moderate to high sensitivity and specificity values were found for S-DN4, ID Pain and PDQ in a study investigating CLBP with or without leg pain [30]. This study was found to be at RoB and indirectness was observed due to inconsistencies in cross cultural validation. Scholz et al. [12] found high sensitivity and specificity in their study investigating the use of the StEP tool in identifying lumbar radicular pain, this study was at low RoB. Moderate level of evidence supports the diagnostic utility of the StEP tool in diagnosing lumbar radicular pain. However clinical judgement was used as a reference standard which was not adequately described, furthermore clinical judgement is not a validated means to identify NP in LBLP. There is no further research to support the use of the StEP tool in identifying NP in LBLP, further research is needed to support its use.

Collective synthesis of patient history data, clinical examination data and screening tool data

Primary diagnostic data reported in these studies support the use of certain subjective history items, clinical examination items and screening tools, however due to the overall RoB assessment and low level of evidence supporting the use of clinical indicators these results must be observed with caution. Only two studies were reported as low RoB and demonstrated moderate level of evidence supporting the diagnostic utility of a cluster of eight patient history/clinical examination signs and the StEP tool in diagnosing lumbosacral nerve root compression and lumbar radicular pain respectively. However, due to the indirectness of these studies in relation to the central question of this review the diagnostic utility of these indicators in regards to identifying NP in LBLP remains unclear.

Indirectness highlighted in all of the included studies is largely related to the phenomena of interest being investigated and its consistency with the focus of this review in identifying NP in LBLP. Included studies investigated the diagnostic utility of clinical indicators in relation to identifying; lumbosacral nerve root compression, L5 lateral stenosis, sciatica, LBLP and chronic LBP, all of which may imply NP in LBLP but not explicitly. Without appropriately defining in the study that NP in LBLP will be investigated, the above-mentioned titles remain ambiguous. Furthermore, in studies where the phenomena of interest are termed as such that imaging is needed to confirm them, e.g. lateral stenosis, it could be questioned whether this an appropriate approach to identify NP. It is well established that structural abnormalities found on imaging are not always directly correlated with symptom presentation [36]. In cases where sciatica is the phenomena of interest, without specifying the interest of investigating the presence of NP within this presentation, sciatica could also encompass cases where NP is not present, as highlighted by Mahn et al., [41]. This is also the case for studies that investigate LBLP, as a manifestation of LBLP may be pain induced by activation of the nervi nevorum (connective tissue sheaths of the peripheral nerve) which result in increased mechanosensitivity which is deemed largely nociceptive in nature and can occur without NP [39]. Furthermore, pain into the leg originating from the back may also be as a result of non-nervous tissue in the lumbar spine being implicated (such as muscle, ligament, disc) which can follow a somatic referred pattern into the leg [5].

As a result of the indirectness highlighted regarding applicability concerns as well as the highly heterogenous data, the studies have been largely assessed individually and the limited synthesis made between studies have been suggested with caution. Due to the general low level of evidence, high RoB and indirectness of evidence we believe that further research is needed to address the title of this review.

A strength of this systematic review is it adhered to a pre-defined protocol which enabled robust identification and synthesis of the available evidence. Through the analysis that was carried out, recommendations for future research have been made. In the absence of a gold standard to diagnosis NP in LBLP there is no standardised commonly used reference standard in its place, this is a key limitation to this review. Therefore, the use of imaging, clinical opinion and clinical tests used within the included studies are questioned as it is unclear which reference standard is superior. This in turn results in the interpretation of the primary diagnostic accuracy data generated from these studies being contentious, as the reference standard is subject to debate. Another limitation to this study was that, due to the highly heterogeneous data obtained from the included studies, a meta-analysis was not possible. Furthermore, due to the general low level of evidence supporting the investigated clinical indicators and high RoB owing to a range of reasons (Table 6), the conclusions made from this systematic review are limited. Finally, the moderate level of evidence supporting the two clinical indicators (a cluster of eight patient history/clinical examination signs and the StEP tool) must be observed with caution. The evidence used to support this level of evidence is assessed from individual studies and therefore despite being deemed ‘moderate level of evidence’ (following the use of GRADE) the generalisability to a wider population is poor.