Musculoskeletal physiotherapists routinely assess lumbar spinal segmental motion and choose interventions on the basis of the findings of those assessments. However, the validity of clinical tests used to assess segmental motion has not been established. When physiotherapists examine the lumbar spine, common assessments include passive accessory intervertebral motion tests (PAIVMs) and passive physiological intervertebral motion tests (PPIVMs) 12. Movement abnormalities, such as hypermobility, are believed to be detected by these assessments 1.

To date, the only evidence for the concurrent validity of manual testing for the presence of lumbar segmental instability (LSI) comes from two studies in which the presence of spondylolysis was considered a proxy for the presence of segmental hypermobility. The first comprised of a very small subgroup analysis (6 patients) of patients with spondylolysis, within a sample of 62 patients with non-specific LBP 3. The results of that investigation indicated that PAIVMs and PPIVMs could identify the symptomatic level with 83% sensitivity and 98% specificity 3. In the second study, manual assessment (combined information from both PPIVMs and PAIVMs) was 69% sensitive and 96% specific for detection of the lytic segment 4. When analysis was restricted to subjects who reported visual analogue pain scores of greater than 4/10, sensitivity and specificity rose to 100% 4. In addition, some preliminary evidence indicates that PAIVM testing may have predictive validity for the purpose of classifying patients in a 'stabilisation' category, who respond better to an exercise intervention intended to increase lumbar segmental stability 5.

As there is currently no evidence in the literature to establish the concurrent validity of manual therapy tests for the detection of excessive sagittal planar motion of the lumbar spine, the aims of this study were to estimate the accuracy of three common clinical assessment items for the detection of lumbar segmental hypermobility (PAIVMs, flexion PPIVMs, and extension PPIVMs), compared to a criterion standard of radiographic measurement of sagittal segmental rotation and translation.

One hundred and thirty eight (138) consenting patients were recruited for clinical examination. One hundred and eight (108) were recruited in primary care; the remaining 30 presented to a hospital outpatient physiotherapy department. Ten patients failed to present to radiology for flexion-extension radiographs. Five sets of radiographs were of insufficient quality for analysis. Of the 123 included participants, 68 (55%) were males and 55 (45%) females. Further characteristics are described in Table 1. A STARD flow chart is provided in figure 4. No adverse events were reported.

Nine males and 24 females were available for recruitment into the asymptomatic sample. Three participants violated the exclusion criteria with regard to low back pain history, and were therefore ineligible. The asymptomatic sample therefore comprised of 9 males and 21 females, aged 23 to 60 years (mean 41.3, sd 12.8).

The 27 clinicians who collaborated on this study graduated with their first professional physiotherapy qualification between 1974 and 1996 (mean years since graduation 17, range 6 to 29). All had gained at least one post-graduate qualification in musculoskeletal physiotherapy which included training in manual therapy procedures for the spine, between 1983 to 2000 (mean years since graduation 8.7, range 2 to 19). They spent an average of 31 hours (interquartile range 21 to 40) per week treating patients, with LBP patients comprising, on average, 30% of their patient load (interquartile range 20 to 40).

Despite their widespread use, the validity of PAIVMs and PPIVMs for assessing abnormal sagittal planar motion has not been previously established. We have found PAIVMs and PPIVMs to have high specificity, but poor sensitivity, for the diagnosis of both rotation LSI and translation LSI.

Like sensitivity and specificity, the likelihood ratio for a positive test (LR+) is more powerful when its value is high. Because of the many factors which must be taken into account when applying a diagnostic test to an individual patient (such as the setting the test is used in, purpose of applying the test, prevalence of the disorder, consequences of missing a diagnosis, and risk of harm from the indicated therapy), there are no set cut-off values for sensitivity, specificity, or likelihood ratios, however some authors provide general guidelines 26. Tests returning LR+ values of 2 to 5 produce small but often useful changes in probability 26, while LR+ values of 5 to 10 (and greater) are more powerful. A test with a likelihood ratio of one is of no clinical utility. The results of this study indicate that a segment testing positive with a PAIVM test is approximately two-and-a-half times more likely to be hypermobile than not 27. The results for PPIVMs were higher, indicating that a segment testing positive with an extension PPIVM test is approximately seven times more likely to be hypermobile than it is to be normal or hypomobile.

Likelihood ratios for negative tests from this research were less impressive than were the LR+ values, with values between 0.76 and 0.96. None were statistically significant. A LR- closer to zero is more powerful, whereas a LR- of one has no discriminative power. Tests returning LR- values of 0.2 to 0.5 produce small but useful changes in probability, while those with values less than 0.2 are more powerful 26. This research indicates that a negative result for hypermobility with PAIVM or PPIVM tests is clinically uninformative.

The low prevalence of rotation LSI in this non-surgical, mostly primary care cohort indicate that sagittal rotation hypermobility does not appear to be associated with R/CLBP, as the number of segments hypermobile in rotation is less than the number that would be expected in a sample from a normally distributed asymptomatic population. Sagittal translation hypermobility was found in a significantly higher than expected proportion of patients with R/CLBP (12.0%), and therefore using a Gaussian definition of abnormality (i.e. beyond 2sd from a reference mean) 28 can be considered a valid clinical disorder. Only a small proportion of segments (3.6%) satisfied this Gaussian definition for sagittal translational LSI, however, indicating that it is neither common in this population nor strongly associated with C/RLBP. This may be considered surprising in the light of the emphasis on sagittal translation in the LSI literature 2930. This proportion does, however, compare well with clinicians' judgement using PAIVM tests. In the present study, therapists considered 5% of lumbar segments to have manual tests findings positive for LSI. This figure compares well to the 12% of patients with LBP reported to be hypermobile by therapists using PAIVM testing in other research 5. With regard to the physical examination, though, it is also recognised that assessment of displacement kinematics alone may not be a sufficient basis for the diagnosis of LSI 3132.

This study has a number of limitations which limit the interpretation of these results. Firstly, while the assessments were nested within a comprehensive clinical examination, and performed in the physiotherapists' own clinical setting, only these three physical assessments were studied in isolation. No attempt was made to identify clusters of assessments that may multiplicatively improve diagnostic accuracy. It is likely that these assessments would have much greater clinical utility within a cluster of other valid signs, symptoms, and history items 1619. Furthermore, it may be necessary to adjust the likelihood ratios of these and other tests researched in the future, to remove the influence of conditional dependence, using statistical methods such as logistic regression 33. Secondly, the prevalence of LSI (using a Gaussian definition of abnormal motion) in this population is low. Defining LSI using a statistical model other than the Gaussian definition used here may result in different prevalence rates. We derived our cut-point for the definition of LSI from the results of our asymptomatic sample; validating the cut-points in another, independent sample would make these results more robust. Sensitivity and specificity, and hence likelihood ratios, may differ in a population with different prevalence rates, such as gymnasts or other athletes, patients with spondylolysis, or surgical candidates 34. It is also well known that diagnostic tests achieve higher values in the secondary and tertiary care populations, where severity of disease is generally higher 34. For this reason, too, values may differ in a population with a different spectrum of the target disorder(s), such as patients with spondylolisthesis or higher pain or disability scores. In the primary care low back pain population, the severity of low back conditions is generally low, making differential diagnosis more difficult. In the context of the present population, however, because mechanical low back pain is not life-threatening and the risks of physiotherapeutic interventions are very low 35, moderate index values are acceptable and may still be useful in the diagnosis of low back pain subgroups. Thirdly, analysis of segmental motion from flexion-extension radiographs was limited to sagittal segmental planar rotation and translation. These are properties of displacement kinematics, and as such identify only abnormalities in the quantity of motion. Other parameters of displacement kinematics, such as ratio of translation to rotation 36, instantaneous axis of rotation, and centre of reaction 6 may better characterise abnormalities of movement quality, rather than quantity. Motion abnormalities may also occur in the mid-range of movement and thus cannot be captured on flexion-extension radiographs, but may be detectable by videofluoroscopy. Furthermore, displacement kinematics are only one aspect of segmental motion (and may not be the most important aspect). The physical examination procedures employed by physiotherapists may assess important parameters other than displacement kinematics 32. This study has not attempted to examine physical assessment of spinal motion velocity, acceleration, or temporal patterns of displacement, nor has it examined physical assessment of kinetics relevant to spinal segmental motion, such as stiffness, viscoelasticity, or force-displacement characteristics. Further research is warranted to fill in the gaps in the literature addressing these limitations.

This research has focussed on the diagnostic accuracy of PAIVMs and PPIVMs, and the multi-centre, pragmatic design of the study precluded assessment of their reliability. The reliability of these clinical assessments has been debated in the literature for many years 3738. While many studies have found reliability to be poor 3940, others have reported considerably better reliability 4142. Contrary to popularly held opinion 4344, it is not easy to conduct a valid and rigorous reliability study. The biostatistical literature points out quite clearly that there numerous difficulties and pitfalls to the study of reliability 4546474849505152 which may threaten the validity of research results. Common methodological problems include violation of the assumptions necessary for the statistical tests used, selection of an inappropriate sample of subjects, lack of true variance in the levels or categories within the sample tested, low prevalence of results across the full spectrum of test scores, and skewed or assymetrical distribution of data. These factors all have a very large impact on the validity and interpretation of much of the literature available on the reliability of these physical examination items: much of the published research regarding reliability may be biased toward the null. It has been argued that tests can be useful for clinical decision-making, in spite of ostensibly low reliability 53, and that it is more important to establish validity of a test or measure 46. For these reasons, it can be argued that reliability should only be studied in the context of validity 53. Further research is warranted into these issues.

The first research published in the peer-reviewed literature to test the concurrent validity of these manual assessments for the detection of abnormal segmental rotation appeared in the literature only recently 54, and addressed lumbar segmental hypomobility. The findings of that research indicated that PAIVMs were moderately sensitive (75%) but not specific (35%) for the detection of hypomobility, while flexion PPIVMs were found to be specific (89%) but not sensitive (42%), with a LR+ of 3.9 54. Those findings, and others from the literature on predictive validity of hypomobility 165556, are generally consistent with the present results, and represent a gathering body of evidence supporting the validity and clinical utility of these manual clinical assessments.

While the LR+ values reported in the present research are only of moderate strength, they may have some clinical utility. If a patient returns a positive test using the extension PPIVM, this would increase the probability that the lumbar segment being tested has translation LSI from 3.6% (the proportion of lumbar segments found to have LSI in this study) to 20.9%. Even assuming conditional independence of the tests, if the patient then returns a positive test using the central P-A PAIVM, post-test probability that the segment is hypermobile would rise to only 40%. This is, however, still too low for clinical or research usefulness, without further improvement in diagnostic certainty being available from other components of the clinical examination (such as the patients history and interview findings, other patient-derived information, and other physical signs). Research investigating the predictive validity of clinical examination findings has found that manual assessments of a similar nature to be a significantly useful addition to a clinical prediction rule, when combined in a test item cluster with other findings 165556. These factors mean that the LR+ values found in this study may be of a magnitude sufficient to be useful in clinical practice when combined with other information from the clinical examination.

This study provides the first evidence reporting the concurrent validity of manual assessments for detecting the excessive sagittal planar motion associated with LSI in vivo. PAIVMs and PPIVMs were specific, but not sensitive, for the detection of rotation LSI and translation LSI. Positive PAIVM and extension PPIVM tests had statistically significant likelihood ratios for identifying translational LSI. The validity of the manual therapists' assessments of excessive sagittal planar motion was only moderate, but as these results do not take into account other important parameters of segmental mobility, such as stiffness or viscoelasticity, this level of validity is still encouraging. Further investigation into the validity of the clinical examination for the detection of lumbar segmental motion disorders is warranted, such as whether greater accuracy may be achieved from clinical examination when manual assessments are combined with other information from the patients' history and physical examination.

The authors declare that they have no competing interests.

JHA conceived, designed and coordinated the study, recruited the clinicians, recruited and examined some of the patients, carried out data analysis and prepared the manuscript. JHA retains copyright on all contents. BMcC assisted with measurement technology & data analysis, and manuscript preparation. PH provided statistical support. GM, CC, and TH assisted in clinician recruitment, patient recruitment and examination, data collection, and provided facilities. All authors read and approved the final manuscript.