630

PAPER

Diagnostic value of history and physical examination in

patients suspected of lumbosacral nerve root
compression
P C A J Vroomen, M C T F M de Krom, J T Wilmink, A D M Kester, J A Knottnerus
.............................................................................................................................

J Neurol Neurosurg Psychiatry 2002;72:630–634

Objective: To evaluate patient characteristics, symptoms, and examination findings in the clinical
diagnosis of lumbosacral nerve root compression causing sciatica.
See end of article for Methods: The study involved 274 patients with pain radiating into the leg. All had a standardised
authors’ affiliations
....................... clinical assessment and magnetic resonance (MR) imaging. The associations between patient charac-
teristics, clinical findings, and lumbosacral nerve root compression on MR imaging were analysed.
Correspondence to: Results: Nerve root compression was associated with three patient characteristics, three symptoms,
Dr Patrick C A J Vroomen,
Department of Neurology,
and four physical examination findings (paresis, absence of tendon reflexes, a positive straight leg
Maastricht University raising test, and increased finger-floor distance). Multivariate analysis, analysing the independent
Hospital, P Debyelaan 25, diagnostic value of the tests, showed that nerve root compression was predicted by two patient char-
PO box 5800, 6202 AZ acteristics, four symptoms, and two signs (increased finger-floor distance and paresis). The straight leg
Maastricht, Netherlands;
[email protected]
raise test was not predictive. The area under the curve of the receiver-operating characteristic was 0.80
for the history items. It increased to 0.83 when the physical examination items were added.
Received 10 April 2000 Conclusions: Various clinical findings were found to be associated with nerve root compression on
In revised form MR imaging. While this set of findings agrees well with those commonly used in daily practice, the tests
7 August 2001
Accepted tended to have lower sensitivity and specificity than previously reported. Stepwise multivariate analysis
22 January 2002 showed that most of the diagnostic information revealed by physical examination findings had already
....................... been revealed by the history items.

D
isc herniation often does not cause symptoms.1 On the tion in the diagnosis of nerve root compression. We studied
other hand, it may cause sciatica by compressing the patients presenting to a primary care physician with a new
nerve roots. Diagnostic procedures such as magnetic episode of pain radiating into the leg, and we used MR imag-
resonance (MR) imaging help to establish disc herniation as ing to establish the presence of nerve root compression.
the anatomical basis for sciatica. Several treatments, especially
discectomy, have been advocated for patients with sciatica, but
are indicated only if nerve root compression is considered to be METHODS
the cause.2–4 Therefore, an accurate initial clinical diagnosis of Study population
nerve root compression is highly desirable.5 6 Fifty general practitioners in Maastricht and surrounding vil-
In one systematic review of the value of history and physi- lages referred patients with a new episode of pain radiating
cal examination in radicular syndromes it was concluded that into the leg below the gluteal fold to the neurology
none of the tests used had a high sensitivity or specificity.7 department of the Maastricht University Hospital. In this
Unfortunately, various symptoms and some commonly study we only included patients with pain sufficiently severe
assessed clinical signs were not reviewed in that study. It was to justify further action (defined as pain that might justify bed
concluded in two other reviews8 9 that the diagnostic value of rest for 14 days, because part of the study population was
many signs and symptoms had not been thoroughly involved in a trial of bed rest15). Exclusion criteria were previ-
investigated. ous spinal surgery, pregnancy, severe comorbidity, or contrain-
The most common methodological drawbacks in previous dication to MR imaging (metal containing prostheses,
studies have been the choice of a suboptimal gold standard, pacemakers).
selection bias, and univariate analysis.9 In many studies disc
herniation rather than nerve root compression was used as the Investigations
gold standard. However, disc herniation need not result in Eligible patients gave their written informed consent for the
nerve root compression and has been found to be asympto- study within two days of the first general practitioner consul-
matic in many subjects.1 10 Also, in several previous studies the tation. Clinical findings were then established in a standard-
history and physical examination findings probably affected ised fashion based on good clinical practice, standard textbook
the selection of the patients for the study.9 This so called sections, and published reports. The methods and interob-
“sequential ordering bias” occurs when patients with positive server consistency of the history and physical examination
findings are more likely to be selected for study than patients have been reported elsewhere.16
with negative findings.11–14 Finally, previous analytical methods The gold standard in all patients was MR imaging of the
have used univariate analysis; multivariate analysis, however, lumbar spine within 24 hours after the clinical examination.17
allows one to assess the contribution of separate clinical find- This was performed using a 0.5 T system according to the fol-
ings to the overall diagnosis, taking into account the high lowing scanning protocol: first, sagittal and transverse T1
degree of correlation between clinical findings. weighted sequences (TR/TE 400–600/20 ms) with 4 and 5 mm
With these factors in mind, we undertook a cross sectional slice thickness, respectively; and second, a sagittal dual echo
study examining the value of history and physical examina- proton density and T2 weighted sequence (fast spin echo

www.jnnp.com
History and examination in nerve root compression 631

Table 1 Baseline patient characteristics and their relation to nerve root compression
(n=274)
Characteristic MRI+ (n=152) MRI− (n=122) Odds ratio (95% CI)

Age (years) (mean (SD), 46 (12) years)

16–40 43 48
41–50 50 44 1.3 (0.7 to 2.3)
51–81 59 30 2.2 (1.2 to 4.0)
Male sex 67 72 1.8 (1.1 to 3.0)
Duration of disease (median 19 days)
14 days or less 70 55
15–30 days 41 19 1.7 (0.9 to 3.2)
>30 days 41 48 0.7 (0.4 to 1.2)
Living alone 29 19 1.3 (0.7 to 2.4)
High education level 45 25 1.6 (0.9 to 2.9)
Job type
No job 66 43
Cognitive 42 26 1.1 (0.6 to 2.0)
Standing/walking/lifting 28 36 0.5 (0.3 to 0.95)
Physically demanding 16 17 0.6 (0.3 to 1.3)
Any comorbidity 64 44 1.3 (0.8 to 2.1)
Smoking 65 63 0.7 (0.4 to 1.1)
Sports activities 67 51 1.1 (0.7 to 2.2)
Exercised back/abd muscles 60 54 0.8 (0.5 to 1.3)
Family history sciatica 51 38 1.1 (0.7 to 1.9)
Quetelet index (mean (SD)) 25.5 (5.1) 25.6 (4.0) p=0.79

Values in bold indicate significant effect.

Abd, abdominal; CI, confidence interval; MRI+/−, magnetic resonance imaging positive or negative.

TR/TE 4500/25–150 ms) with 5 mm slice thickness. We also The clinical investigator was unaware of the MR findings,
carried out MR radiculography, consisting of two heavily T2 while the neuroradiologist was unaware of the clinical
weighted fast spin echo sequences (TR/TE 6000/450 ms) with findings.
4 mm slice thickness, oriented parallel to the emerging L5
root, about 20° left and right oblique to the coronal plane, Statistical analysis
reformatted with a maximum intensity projection protocol.18 For the analysis the statistical package for the social sciences
A standardised assessment of all MR imaging studies was (SPSS) was used. A bivariate analysis was based on 2 × 2
done by a neuroradiologist (JW) experienced in spinal tables relating history and physical examination items to
magnetic resonance imaging. The assessment focused on the nerve root compression on MR imaging. Sensitivity, specificity,
presence or absence of nerve root (sleeve) compression. Crite- and likelihood ratios for positive and negative test results
ria used for grading included the presence of a protruding (LR+ and LR−) were calculated, as well as the diagnostic odds
annulus or extruded nucleus material, or lateral recess ratio (DOR, with DOR = LR+/LR−).19 20 The statistical signifi-
narrowing on standard MR imaging, and flattening and com- cance of the association was assessed using Pearson’s χ2
pression of the ventrolateral border of the dural sac or emerg- statistic.
ing nerve root sleeve, with obliteration of the surrounding In a backward stepwise logistic regression analysis,21 22 the
epidural fat. In the MR radiculography images, positive crite- likelihood ratio test was used for each step at a significance
ria included dural impressions, kinking, and swelling of the level of 0.05. First, a set of general patient characteristics (all
intradural nerve root and cut off of the root sleeve filling. variables in table 1) was modelled, leading to model A.

Table 2 Diagnostic accuracy of the history (n=274)

Characteristic MRI+ (n=152) MRI− (n=122) Odds ratio (95% CI)

Pain worse in leg than in back 125 56 5.5 (3.2 to 9.4)

Typically dermatomal distribution 136 84 3.8 (2.0 to 7.3)
Pain worse on coughing/sneezing/
straining 76 40 2.1 (1.3 to 3.4)
Dermatomal cold sensations in leg 47 24 1.8 (1.0 to 3.2)
More pain on sitting 82 69 0.9 (0.6 to 1.5)
Less pain standing or walking 31 26 0.9 (0.5 to 1.7)
Less pain on lying down 91 69 1.1 (0.7 to 1.9)
Sudden onset 60 51 0.9 (0.6 to 1.5)
Paroxysmal pattern 67 47 1.3 (0.8 to 2.0)
Pain worse at night 47 42 0.9 (0.5 to 1.4)
With known cause 38 37 0.8 (0.5 to 1.3)
Subjective muscle weakness 41 46 0.6 (0.4 to 1.0)
Subjective sensory loss 60 51 0.9 (0.6 to 1.5)
Dermatomal paraesthesiae 78 72 0.7 (0.5 to 1.2)
Previous back pain episodes 113 85 1.3 (0.7 to 2.1)
Previous sciatica 55 52 0.8 (0.5 to 1.2)
History indicating root compression
according to investigator 118 64 3.1 (1.9 to 5.3)

Values in bold indicate significant effect.

CI, confidence interval; MRI+/−, magnetic resonance imaging positive or negative.

www.jnnp.com
632 Vroomen, de Krom, Wilmink, et al

Table 3 Diagnostic accuracy of the physical examination (n=274)

Characteristic MRI+ (n=152) MRI− (n=122) Odds ratio (95% CI)

Paresis 41 8 5.2 (2.4 to 11.7)

Finger-floor distance >25 cm 68 32 2.4 (1.4 to 4.0)
Absence of ankle/knee tendon reflex 22 8 2.4 (1.0 to 5.6)
Positive SLR 97 53 2.3 (1.4 to 3.7)
Sensory loss
Hypesthesia 43 42 0.8 (0.5 to 1.3)
Hypalgesia 26 19 1.2 (0.6 to 2.1)
History and examination indicating root
compression according to investigator 123 59 4.5 (2.6 to 7.8)

Values in bold indicate significant effect.

CI, confidence interval; MRI+/−, magnetic resonance imaging positive or negative; SLR, straight leg raise.

Second, a set of history items (all variables in table 2) was RESULTS

modelled (model B), also including the variables in model A In all, 338 patients were referred for the diagnostic study. Sixty
that were significantly predictive. Third, all physical examina- four were excluded (35 because of insufficient pain and 29
tion findings in table 3 and the significant predictors of model because of previous spinal surgery or contraindications to MR
B were modelled into the final reduced model C (presented in imaging). Thus the investigations were carried out in 274
table 4). patients. The intraobserver consistency of the MR imaging
The neuroradiologist assessed 60 scans twice to allow diagnosis of nerve root compression was high (κ value 0.83).
calculation of a κ value for intraobserver variability.21 Table 1 shows the most important patient characteristics.
The probability of nerve root compression on MR imaging Root compression was found on 56% of the MR imaging stud-
predicted by the various reduced models was calculated for ies. The pain radiating into the leg was clinically diagnosed as
each patient. Subsequently, to evaluate the predictive power of sciatica in 67% of patients, while other diagnoses included
each model, receiver-operating characteristic (ROC) curves coxarthrosis, spondylolisthesis, regional neoplastic growth,
were plotted, displaying the sensitivity v (1 − specificity) for peripheral neuropathy, and pain of tendomyogenous or other
each cut off point of predicted posterior probability.23 24 The origin. The sample included sciatica patients both with and
areas under the ROC curves were calculated. The probability without various degrees of neurological deficit. Two patients
cut off with the maximum expected gain was determined as
suffered from urinary incontinence and saddle anaesthesia.
the point with the highest sum of sensitivity and specificity.25
In the univariate analysis, there was an association between
As a reference, the sensitivity and specificity of the observer’s
nerve root compression on MR imaging and three patient
summary statement on the presence of nerve root compres-
sion are shown (at the bottom of tables 2 and 3) .

1.0
Ethics
The procedures of this study were in accordance with the
ethical standards of the revised Helsinki declaration of 1983 0.8
and were approved by the Maastricht University Hospital eth-
ics committee.
0.6
Sensitivity

Table 4 Predictors of nerve root compression on 0.4

magnetic resonance imaging: results of multiple logistic
regression analysis (n=274)
Adjusted 0.2
Test diagnostic OR 95% CI

History
0.0
Age (years) 0.0 0.2 0.4 0.6 0.8 1.0
41–50 v 16–40 1.8 1.3 to 2.6 1 – Specificity
51–81 v 16–40 2.8 1.9 to 4.2
Duration of disease (days) Figure 1 Upper curve: diagnostic value of history and physical
15–30 v <15 2.2 1.5 to 3.3 examination (variables in table 4; area under the curve = 0.83).
>30 v <15 0.8 0.6 to 1.1 Lower curve: diagnostic value of history (area under the
Paroxysmal pain 1.8 1.3 to 2.5 curve = 0.80). • Maximum diagnostic gain of model according to
Pain worse in leg than in back 4.5 3.3 to 6.2 Connell and Koepsell.25 n Sensitivity and 1 − specificity for the
Typical dermatomal distribution 3.2 2.2 to 4.7 observer’s diagnosis after history. m Sensitivity and 1 − specificity
Pain worse on for the observer’s diagnosis after history and physical examination.
coughing/sneezing/straining 2.0 1.4 to 2.7
A physician makes a diagnosis of nerve root compression when he
feels that the clinical findings indicate disease with a high enough
Physical examination
probability. The latter is the physician’s intrinsic cut off probability of
Finger-floor distance (cm)
disease. For example, the intrinsic cut off probability for the observer
5–24 v 0–4 1.1 0.7 to 1.6
is shown by the black triangle in the figure. The logistic models
>25 v 0–4 2.8 1.9 to 4.3
Missing v 0–4 1.0 0.4 to 2.1
allows an estimation of diagnostic properties for all possible prob-
Paresis 5.2 3.3 to 11.6
ability cut off points. Plotting these properties results in the receiver-
operating characteristic. The area under the curve represents the
Intercept −3.511 overall diagnostic value of the model regardless of cut off probabil-
ity. Maximum diagnostic gain might be estimated as the point in the
CI, confidence interval; OR, odds ratio. upper left hand corner or as the highest sum of sensitivity and
specificity as proposed by Connell and Koepsell.19 25

www.jnnp.com
History and examination in nerve root compression 633

characteristics: age, sex, and having a job with a predomi- nerve root compression. It is possible that subjective weakness
nance of standing, walking, or lifting activities (table 1). MR and sensory loss have no neurological basis in most patients.
imaging was also associated with three symptoms: typical Table 3 shows that four physical examination items were
dermatomal pain, coldness in the leg, and increased pain on related to nerve root compression, and that even before the
coughing, sneezing, or straining (table 2). Four physical physical examination, the a priori likelihood of disease is
examination items were discriminative: paresis, a finger-floor altered by the history. Rather than considering only one or two
distance of more than 25 cm, absence of knee or ankle tendon tests, one should use history and physical examination as two
reflex, and a positive straight leg raise test (table 3). batteries of tests, and interpret the items in each battery
However, in diagnosing nerve root compression clinical simultaneously. The independent and simultaneous contribu-
findings are not considered separately but in combination. tions of tests to the diagnostic work up of the sciatica patient
This is reflected in multiple logistic regression models, which have not been studied previously. Table 4 and the ROC curve in
allow calculation of the independent predictive value of fig 1 indicate that, after the history has been taken, few signs
patient characteristics and clinical findings and predict the in the physical examination contribute to a more accurate
probability of nerve root compression. Table 4 shows that two diagnosis. Typically, a dermatomal distribution of pain, an
general patient characteristics (age and duration of disease), increase in pain on coughing, sneezing, or straining, paroxys-
four symptoms from the history, and two signs from the mal pain, and predominant leg pain are indicators of nerve
physical examination were independent predictors of nerve root compression. This is in agreement with clinical experi-
root compression. The straight leg raise test was not a signifi- ence. The predictive value of paresis also accords with general
cant predictor, while an increased finger-floor distance and beliefs.8 A surprising predictor was an increased finger-floor
paresis were significant predictors. distance. The action of bending over to touch the floor with the
The area under the curve of the ROC for the model using fingers not only stretches the nerve roots and the sciatic nerve
history findings alone was 0.80. This increased to 0.83 when but also cancels the lordosis of the lumbosacral spine; the lat-
the examination findings were added (fig 1). Maximum ter action may accentuate the mechanical effect of the disc
predictive gain of the model occurred at a predicted probabil- herniation on the nerve root. We were struck by the fact that
ity value of 0.625 with a sensitivity and specificity of 72% and the straight leg raise test was not a predictor of root compres-
80%, respectively. sion. This test may indicate nerve root tension or irritation, but
not necessarily nerve root compression.

DISCUSSION Conclusions
In our 274 primary care patients with leg pain, three patient The main component in the diagnosis of sciatica caused by
characteristics, three symptoms, and four signs were associ- disc herniation is the history. Few physical signs add useful
ated with nerve root compression on MR imaging. However, additional information or result in alteration of a diagnosis
the sensitivity and specificity values we found in this study made on the basis of the history.
were lower than previously reported.7 Different study popula-
tions and designs may explain this.13 .....................
Positive clinical findings were not the reason for referral in
Authors’ affiliations
our study and therefore we have avoided the verification P C A J Vroomen, M C T F M de Krom, Department of Neurology,
bias25 26 and spectrum bias27 of previous studies. There are also Maastricht University Hospital, Maastricht, The Netherlands
indirect clues that the participating patients were indeed an J T Wilmink, Department of Diagnostic Radiology, Division of
unbiased primary care population. For example, both sexes Neuroradiology, Maastricht University Hospital
were equally represented, while previous secondary care A D M Kester, Department of Methodology and Statistics, Maastricht
University
patient series have included nearly twice as many men as J Knotterus, Department of General Practice, Maastricht University
women. Also, neurological deficits were far less common in
our patient sample than in previous series.9 To avoid including
asymptomatic or minor conditions, we accepted only those REFERENCES
1 Jensen MC, Brant-Zawadzki MN, Obuchowski N, et al. Magnetic
patients who had severe pain radiating into the leg. We felt resonance imaging of the lumbar spine in people without back pain. N
that a clinical diagnosis of nerve root compression in patients Engl J Med 1994;331:69–73.
with no leg pain was probably rare and in any case should not 2 Mixter WJ, Barr JS. Rupture of the intervertebral disc with involvement of
the spinal canal. N Engl J Med 1934;211:210–15.
lead to discectomy. 3 Weber H. Lumbar disc herniation; a controlled, prospective study with
MR imaging was used as the state of the art procedure to ten years of observation. Spine 1983;8:131–40.
demonstrate nerve root compression.17 28–30 Sensitivity and 4 Hoffman RM, Wheeler KJ, Deyo RA. Surgery for herniated discs: a
literature synthesis. J Gen Intern Med 1993;8:487–96.
specificity values of MR imaging in the diagnosis of disc her- 5 Knutsson B. Comparative value of electromyographic, myelographic
niation range from 71% to 100% and from 50% to 86%, and clinical–neurological examinations in diagnosis of lumbar root
respectively.17 31 32 Several investigators have reported disc her- compression syndrome. Acta Orthop Scand Suppl 1961;49:1–134.
niation in asymptomatic subjects using MR imaging,1 33 com- 6 Kosteljanetz M, Espersen JO, Halaburt H, et al. Predictive value of
clinical and surgical findings in patients with lumbago-sciatica a
puted tomography,34 and caudography.35 However, none of prospective study (part I). Acta Neurochir 1984;73:67–76.
these related the disc herniation to the presence of nerve root 7 Hoogen HJMM, Koes BW, van Eijck JTM, et al. On the accuracy of
compression. Some false negative diagnoses may be explained history, physical examination and erythrocyte sedimentation rate in
diagnosing low-back pain in general practice. Spine 1995;20:318–27.
by changes in the pathological anatomy of the nerve root with 8 Andersson GBJ, Deyo RA. History and physical examination in patients
posture and over time.36 The additional value of MR radiculo- with herniated lumbar discs. Spine 1996;21:10–18S.
graphy has yet to be unequivocally demonstrated.18 9 Vroomen PCAJ, de Krom MCTFM, Knottnerus JA. Diagnostic value of
history and physical examination in patients with sciatica due to disc
The only history item that could be compared with a previ- herniation; a systematic review. J Neurol 1999;246:899–906.
ous report was the typical dermatomal pain distribution. This 10 Andersson GB, Weinstein JN. Disc herniation [editorial]. Spine
had a 90% sensitivity in our study, compared with 99% sensi- 1996;21:1S.
11 Feinstein AR. Clinical epidemiology: the architecture of clinical
tivity in a study by Kerr and colleagues.37 Strikingly, subjective research. Philadelphia: WB Saunders, 1985.
muscle weakness and sensory loss actually diminished the 12 Lachs MS, Nachamkin I, Edelstein PH, et al. Spectrum bias in the
likelihood of nerve root compression. These findings are at evaluation of diagnostic tests: lessons from the rapid dipstick test for
variance with current opinion and with our previous beliefs. urinary tract infection. Ann Intern Med 1992;117:135–40.
13 Knottnerus JA. The effects of disease verification and referral on the
The selection bias in previous studies may have affected our relationship between symptoms and disease. Med Decis Making
understanding of the true relation of these complaints to 1987;7:139–48.

www.jnnp.com
634 Vroomen, de Krom, Wilmink, et al

14 Knottnerus JA, Leffers P. The influence of referral patterns on the rheumatoid arthritis: the relevance of multiple sources of knowledge for a
characteristics of diagnostic tests. J Clin Epidemiol 1992;45:1143–54. decision-support system. Med Decis Making 1992;12:250–8.
15 Vroomen PCAJ, Krom MCTFM, Wilmink JT, et al. Lack of effectiveness 27 Ransohoff DF, Feinstein AR. Problems of spectrum and bias in
of bed rest for sciatica. N Engl J Med 1999;340:418–23. evaluating the efficacy of diagnostic tests. N Engl J Med
16 Vroomen PCAJ, de Krom MCTFM, Knottnerus JA. The consistency of the 1978;299:926–30.
history and the physical examination in the patient suspected of sciatica. 28 Ellenberger C. MR imaging of the low back syndrome. Neurology
Spine 2000;25:91–6. 1994;44:594–600.
17 Jackson RP, Cain J, Jacobs RR, et al. The neuroradiographic diagnosis 29 Deyo RA. Magnetic resonance imaging of the lumbar spine; terrific test
of lumbar herniated nucleus pulposus: II. A comparison of computed or tar baby? N Engl J Med 1994;331:115–16.
tomography (CT), myelography, CT-myelography, and magnetic 30 Herzog RJ. The radiologic assessment for a lumbar disc herniation.
resonance imaging. Spine 1989;14:1362–7. Spine 1996;21:19–38S.
18 Hofman PAM, Wilmink JT. Optimising the image of the intradural nerve 31 Bischoff RJ, Rodriguez RP, Gupta K, et al. A comparison of computed
root: the value of MR radiculography. Neuroradiology 1996;38:654–7. tomography-myelography, magnetic resonance imaging, and
19 Sackett DL, Haynes RB, Guyatt GH, et al. Clinical epidemiology. A
myelography in the diagnosis of herniated nucleus pulposus and spinal
basic science for clinical medicine. Boston: Little, Brown and Co,
stenosis. J Spinal Disord 1993;6:289–95.
1991:439.
32 Szypryt ET, Twining P, Wilde GP, et al. Diagnosis of lumbar disc
20 Knottnerus JA, Volovics A. Medische statistiek en onderzoek in de
protrusion. J Bone Joint Surg Br 1988;70:717–22.
huisartsgeneeskunde. NHG-Publikaties, vol 4. Utrecht: Nederlands
Huisartsen Genootschap, 1990. 33 Boden SD, Davis DO, Dina TS, et al. Abnormal magnetic-resonance
21 Altman DG. Practical statistics for medical research. London: Chapman scans of the lumbar spine in asymptomatic subjects. A prospective
and Hall, 1996. investigation. J Bone Joint Surg Am 1990;72:403–8.
22 Kleinbaum DG. Logistic regression: a self-learning text. In: Dietz K, 34 Didry C, Lopez P, Baixas P, et al. Lumbar disk herniation with no
Krickeberg K, Singer B, eds. Statistics in the health sciences. New York: surgical treatment. Clinical and x-ray computed tomographic follow-up
Springer, 1994:282. study [in French]. Presse Med 1991;20:299–302.
23 Metz CE. Basic principles of ROC analysis. Semin Nucl Med 35 Hitselberger WE, Witten RM. Abnormal myelograms in asymptomatic
1978;8:283–98. patients. J Neurosurg 1968;28:204–6.
24 Hanley JA. Receiver operating characteristic (ROC) methodology. Crit 36 Bozzao A, Gallucci M, Masciocchi C, et al. Lumbar disk herniation: MR
Rev Diagn Imaging 1989;29:307–35. imaging assessment of natural history in patients treated without surgery.
25 Connell FA, Koepsell TD. Measures of gain in certainty from a Radiology 1992;185:135–41.
diagnostic test. Am J Epidem 1985;121:744–53. 37 Kerr RSC, Cadoux-Hudson TA, Adams CBT. The value of accurate
26 Bernelot Moens HJ, Hirschberg AJ, Claessens AA. Data-source effects clinical assessment in the surgical management of the lumbar disc
on the sensitivities and specificities of clinical features in the diagnosis of protrusion. J Neurol Neurosurg Psychiatry 1988;51:169–73.

www.jnnp.com