Accuracy of physical examination for chronic lumbar radiculopathy

Background Clinical examination of patients with chronic lumbar radiculopathy aims to clarify whether there is nerve root impingement. The aims of this study were to investigate the association between findings at clinical examination and nerve root impingement, to evaluate the accuracy of clinical index tests in a specialised care setting, and to see whether imaging clarifies the cause of chronic radicular pain. Methods A total of 116 patients referred with symptoms of lumbar radiculopathy lasting more than 12 weeks and at least one positive index test were included. The tests were the straight leg raising test, and tests for motor muscle strength, dermatome sensory loss, and reflex impairment. Magnetic resonance imaging (n = 109) or computer tomography (n = 7) were imaging reference standards. Images were analysed at the level of single nerve root(s), and nerve root impingement was classified as present or absent. Sensitivities, specificities, and positive and negative likelihood ratios (LR) for detection of nerve root impingement were calculated for each individual index test. An overall clinical evaluation, concluding on the level and side of the radiculopathy, was performed. Results The prevalence of disc herniation was 77.8%. The diagnostic accuracy of individual index tests was low with no tests reaching positive LR >4.0 or negative LR <0.4. The overall clinical evaluation was slightly more accurate, with a positive LR of 6.28 (95% CI 1.06–37.21) for L4, 1.74 (95% CI 1.04–2.93) for L5, and 1.29 (95% CI 0.97–1.72) for S1 nerve root impingement. An overall clinical evaluation, concluding on the level and side of the radiculopathy was also performed, and receiver operating characteristic (ROC) analysis with area under the curve (AUC) calculation for diagnostic accuracy of this evaluation was performed. Conclusions The accuracy of individual clinical index tests used to predict imaging findings of nerve root impingement in patients with chronic lumbar radiculopathy is low when applied in specialised care, but clinicians’ overall evaluation improves diagnostic accuracy slightly. The tests are not very helpful in clarifying the cause of radicular pain, and are therefore inaccurate for guidance in the diagnostic workup of the patients. The study population was highly selected and therefore the results from this study should not be generalised to unselected patient populations in primary care nor to even more selected surgical populations.


Background
Lumbar radiculopathy is a common reason for physician consultations and imaging referrals [1][2][3]. Typical symptoms are radiating pain, often with numbness, paraesthesia, and/or muscle weakness [1,4]. Clinical examination aims to clarify whether there is mechanical impingement of a nerve root [5]. The most common clinical diagnostic tests are the straight leg raising test, and tests for tendon reflexes, motor weakness, and sensory deficits [6]. An inaccurate clinical diagnosis may lead to unnecessary imaging and healthcare expenditure, and additional concerns for patients [7][8][9][10][11][12].
The aim with imaging is to confirm or disprove a clinical suspicion, and to provide a roadmap for planning of surgical or other intervention procedures, if indicated. Mechanical nerve root impingements demonstrated with magnetic resonance imaging (MRI) or computer tomography (CT) is an accepted reference standard [13].
Systematic reviews on the diagnostic properties of clinical diagnostic tests for lumbar radiculopathy report variable accuracy, with sensitivities ranging from 0.14 to 0.61 for sensory deficits and impaired tendon reflexes [14,15], 0.27 to 0.62 for motor weakness [14,16], and 0.35 to 0.81 for the straight leg raising test [17]. Most studies report likelihood ratios (LRs) suggesting negligible differences between pre-and post-test probabilities for presence of nerve root impingement as the target condition, indicating limited value of the tests in clinical decision-making. A recent Cochrane review confirmed poor diagnostic performance of diagnostic tests in 18 studies from specialised care [13].
This review raised concern that none of the reported studies specifically discriminated between nerve root impingement and just the presence of a disc herniation when using imaging as a reference standard. This could be a major bias, since the prevalence of disc bulging or herniation in unselected populations without radiculopathy symptoms is high [18].
The aims of this study are to investigate the association between findings at clinical examination and nerve root impingement, to evaluate the accuracy of clinical index tests in a specialised care setting, and to see whether imaging clarifies the cause of chronic radicular pain.

Study participants
The study was performed as part of a multicentre randomised controlled trial on the treatment effect of caudal epidural injections [19]. Eligible patients with suspected chronic lumbar radiculopathy, aged between 20 and 60 years, referred to outpatient multidisciplinary back clinics of five Norwegian hospitals, were consecutively assessed for inclusion. All patients were referred with a history suggesting chronic lumbar radiculopathy, and the clinical diagnosis was verified with at least one corresponding positive clinical test (index test) consistent with affection of a specific lumbar nerve root. These inclusion criteria ensured a homogenous patient population with clinically verified lumbar radiculopathy and a high pre-test probability of nerve root impingement. MRI or CT was used to specifically clarify whether the nerve root in question was impinged or not. The reference standard was set to be disc herniation causing impingement (compression and/or dislocation) of a spinal nerve root. Written informed consent was obtained, and the Regional Committee for Medical and Health Research Ethics in North Norway approved the study.
We assessed 461 patients with suspected lumbar radiculopathy for inclusion ( Figure 1). 376 (81.6%) were referred from general practitioners, and 85 (18.4%) were internally referred in the participating hospitals. The inclusion criteria were unilateral lumbar radiculopathy lasting for more than 12 weeks and one or more positive index tests consistent with nerve root affection. The intensity of the leg pain, radiating from the back to below the knee, had to be comparable to or worse than the back pain. Whilst obtaining the patient's history, enquiries were made about the intensity of leg and low back pain on a visual analogue scale, the possible dermatome distribution of the pain, the presence of paraesthesia in the leg, whether the pain was aggravated by forward flexion or sitting, and whether there was any muscle weakness in the lower extremity.

Physical examination
The physical examination was performed according to the recommendations given by the American Spinal Injury Association [20][21][22]. It consisted of the following index tests: the straight leg raising test, the femoral nerve stretch test, testing of muscle power in seven muscle groups on a five-point scale, dermatome sensory loss using light touch and pin prick classified on a threepoint scale, and reflex impairment testing on a fourpoint scale. Each index test was dichotomised as being normal or abnormal according to the standard neurological classification. The straight leg raising test was considered abnormal when pain occurred before 60 degrees passive elevation from horizontal, and the femoral nerve stretch test was considered positive when the patient experienced radiating pain [23].
Specialists in neurology or physical medicine and rehabilitation did the examination in cooperation with a physiotherapist. Prior to the study, they were trained to perform the tests in a standardised way.
Based on an overall evaluation of the patient history and results of all the index tests, a clinical decision was reached for each patient concerning the suspected level and side of nerve root affection [24][25][26][27]. The clinical decision for a nerve root involvement required a history of radicular pain accompanied by one or more corresponding positive index tests. The clinicians were blinded to the results of the imaging until this decision had been reached. To diagnose an L4 radiculopathy the clinician placed emphasis on the femoral nerve stretch test, the straight leg raise test, the knee reflex, sensory loss in the L4 dermatome and the muscle power for the ankle dorsiflexion. To diagnose an L5 radiculopathy the clinician focused on the straight leg raise test, sensory loss in the L5 dermatome, and the muscle power for the hip abduction, ankle dorsiflexion, ankle eversion, and the big toe extension. For an S1 radiculopathy the clinician emphasized the straight leg raise test, the ankle reflex, sensory loss in the S1 dermatome, and the muscle power for hip extension, knee flexion, ankle plantarflexion, and ankle eversion.
Imaging reference standard MRI in 109 (94.0%) patients or CT in 7 (6.0%) patients was performed. Experienced radiologists evaluated the images, and a written report from the radiologists was available for the clinicians to be able to exclude patients with severe intra-spinal pathology obviously demanding surgery [19,28].
All the MRI and CT scans were re-evaluated by two independent neuroradiologists using the Nordic Modic Classification [29]. They were blinded regarding patient history and clinical findings. The locations of the disc herniation were identified in the axial plane, and were categorised as being localised centrally or to the left or right in the spinal canal [30]. In cases of disagreement, a consensus was reached emphasising the most experienced.

Statistical analysis
We calculated means and standard deviations (SD) for continuous variables, and frequencies and proportions for categorical variables. The prevalence of nerve root impingement based on the reference standard and the post-test probabilities for a positive and negative test were calculated. Diagnostic accuracy was quantified by calculating sensitivities, specificities, and positive and negative likelihood ratios (LR), including 95% confidence intervals (CI), for each clinical test. In a multivariable logistic regression model we included all index tests as independent variables. The estimated model was used to predict the probability of a positive MRI/CT for each patient. These probabilities were used to produce a receiver operating characteristic (ROC) curve and an estimate for the area under the curve (AUC). All analyses were performed using the Statistical Package for the Social Sciences software (SPSS), version 19 (IBM Software, NY, USA).

Results
In total, 116 patients with unilateral chronic lumbar radiculopathy were included. Their clinical and demographic  characteristics are summarised in Table 1. Mean age was 42.0 (SD 10.3) years, 68 (58.6%) were males, and the mean duration of symptoms on inclusion was 42.0 (SD 99.0) weeks. Figure 1 shows the results of MRI or CT for the included patients. The overall prevalence of disc herniation at any of the studied lumbar levels (L2 to S1) was 77.8%. Table 2 shows the frequencies of positive index tests, the overall clinical evaluation, and the imaging findings. Table 3 shows the diagnostic accuracies for the different index tests for detection of the level and side of the nerve root impingement. None of the individual tests were highly accurate, as both sensitivities and specificities were low with wide CIs. All positive LRs were ≤4.0, and all negative LRs ≥0.4. Table 4 shows that the clinicians' overall evaluations using information from all relevant index tests to predict nerve root impingement were slightly more accurate than each of the individual index tests. ROC analysis of the diagnostic properties of the overall clinical evaluations showed AUCs of 0.95 (95% CI 0.90-1.00) for L4, 0.67 (95% CI 0.56-0.77) for L5, and 0.66 (95% CI 0.54-0.77) for S1 nerve root impingement.

Discussion
This study included patients with symptoms suggesting lumbar radiculopathy. Patients were recruited by screening and referral from general practitioners, and those with large disc herniation obviously requiring surgery were excluded. The sample emerging from these criteria is typical for the chronic radiculopathy population seen in specialised care. Results from the study are relevant for our understanding of diagnostic accuracy in the common clinical setting where specialists have access to imaging findings prior to the clinical examination, and often are challenged by having to evaluate which of  Ankle plantarflexion (Gastro-cnemius and Soleus S1,S2) 45 3.9 Ankle eversion (Peronei L5,S1) 80 6.9 Big toe extension (Extensor hallucis longus L5,S1) 25    Clinician concluded S1 nerve root impingement numerous positive imaging findings are to be considered clinically relevant. The main finding is that individual clinical index tests lack diagnostic accuracy for predicting whether a lumbar nerve root is impinged or not at a specific level in patients with chronic lumbar radiculopathy in specialised care. The overall clinical evaluation, consisting of the specialists' combined interpretation of the patients' history and all index tests, was somewhat more accurate. For L5 and S1 nerve root impingement, however, LRs did not reach the levels usually considered necessary to influence post-test probability and thereby clinical decision-making (positive LR >5.0 and negative LR <0.2) [31]. Accuracy was better (positive LR 6.28, negative LR 0.70) for L4 nerve root impingement. This was probably because L4 nerve root involvement occurred only in 3 (2.6%) cases, and was suspected after the overall clinical evaluation only in 7 (6.0%) cases. This resulted in a high number of true negatives, and thereby high specificity. Clinically, the low pre-test probability for L4 nerve root involvement is well known [32], and these test properties are therefore not very useful. Accordingly, clinical examination is inaccurate both for predicting the presence or absence of nerve root impingement, and for clarifying the relevant level and side in patients with multiple positive imaging findings.
Our findings are mainly in accordance with other studies of selected populations from specialised care [13]. Most previous studies have, however, aimed for a generalised understanding of test properties from such selected materials [13]. This approach is confusing, as the pre-test probability always must be taken into consideration. Recently, a study aimed to specifically investigate the accuracy of clinical index tests from the neurological examination for identification of the level of disc herniation in patients with the target condition already confirmed by MRI [33]. Unfortunately the study did not find evidence to support this. The results were disappointing, with no single test reaching an AUC >0.75, and only slightly better results (AUC = 0.80) for the neurologists' overall evaluation.
It has been a weakness of most previous studies that interpretation of the imaging findings has been limited to categorising the target condition (usually a disc herniation) as present or not, without considering whether a nerve root actually was impinged at the relevant spinal level and side [34]. We therefore improved the study design by specifically addressing findings relevant for clinical decision-making: correspondence between index tests and impingement of specific nerve roots as revealed by MRI [32]. Disappointingly, this did not improve diagnostic accuracy, neither for individual tests nor for the clinicians' overall evaluation. AUCs for L5 and S1 nerve root impingement did not reach levels above 0.66, which are even lower than those observed by Hancock et al. in an almost similar specialised care setting [33]. This could be because we used one or more positive index tests as an inclusion criterion, which probably increased both the proportion of false positives and false negatives. The false negatives increased because the index tests are not independent of each other, implying that inclusion based on one or more positive tests entails an increased proportion of false negatives, since many tests are performed in each patient. We do not consider the selection of patients in our study a methodological weakness, but rather an expression of clinical reality in specialised care. There should, however, be concern about both the definition of the target condition and the reference standard being subjects to bias. First, neuroanatomical overlap between spinal segments influences accuracy when the analysis is done on the level of each single nerve root [35][36][37]. Patients may have radiculopathy from causes other than ongoing nerve root impingement, and even when an impingement is present, this is not necessarily the cause of the pain. Imaging showed no sign of nerve root impingement in 56 (48.3%) of the included cases despite a clear history and clinical findings suggesting lumbar radiculopathy. This confirms that radiculopathy may have other causes, such as neuropathic and inflammatory conditions, or be mimicked by myofascial pain [6,[38][39][40]. Moreover, disc herniation without nerve root impingement was demonstrated in 25.9% of the included patients, and in 73.8% of those excluded due to symptoms classified as unspecific low back pain with referred leg pain. This is not surprising, since the prevalence of disc herniation revealed by MRI in the general population is known to be as high as 30% [3,18,[41][42][43][44].
We suggest that our findings reflect clinical reality very well: in a population selected by referral from primary care and exclusion of the most obvious surgical cases, co-morbidity bias and imaging findings not related to the symptoms are common. Diagnostic imaging combined with clinical tests is therefore inaccurate for clarifying the cause of radicular pain. This is probably one of the reasons why these patients are so difficult to treat, and the same inaccuracy may cause significant inclusion bias in clinical trials evaluating treatments for lumbar radiculopathy.
The present study has weaknesses. We did not register inter-tester variability for the clinical tests and image interpretations. However, all clinicians were trained to perform the tests in a standardised manner, and agreement should thus be superior to that achieved between clinicians in daily practice [22]. MRI was substituted with CT in 7 (6.0%) of the study subjects. A few cases of nerve root impingement may have been missed, but this is unlikely to have influenced the results significantly.
Further, the duration of symptoms (average 42 weeks) was relatively long. Development of chronic centralised pain followed by regression of nerve root impingement may have occurred in some patients, and our results may not be generalisable to situations with shorter symptom duration.
Finally, it must be emphasised that the index tests work differently when applied in other settings. In unselected primary care populations, the proportion of false positives will be lower and the specificity of the tests higher. Accordingly, the tests may be useful in primary care to reduce the post-test likelihood of lumbar radiculopathy, and thereby restrict unnecessary referrals for imaging and specialised care. On the other hand, when applied in a highly selected surgical patient population with shorter duration of symptoms and a large disc herniation obviously corresponding with the symptoms, the proportion of true positives will be high and the proportion of false positives low, resulting in high sensitivity and specificity. The results from the present study should therefore not be generalised to unselected patient populations in primary care nor to even more selected surgical populations.

Conclusions
In conclusion, the accuracy of individual clinical index tests used to predict imaging findings of nerve root impingement in patients with lumbar radiculopathy is low when applied in specialised care, and clinicians' overall evaluation does not improve diagnostic accuracy significantly. Accordingly, the tests are not very helpful in clarifying the cause of radicular pain, and are therefore inaccurate for treatment guidance of patients who often have multiple positive imaging findings. These results suggest that previous belief in the benefit of combining different neurological tests to accurately diagnose the level of nerve root affection has been exaggerated [45,46]. Co-morbidity and imaging findings not related to the symptoms are probably the most important causes for diagnostic inaccuracy in chronic lumbar radiculopathy [3,28,39,[47][48][49].