The quality of evidence of psychometric properties of three-dimensional spinal posture-measuring instruments

Background Psychometric properties include validity, reliability and sensitivity to change. Establishing the psychometric properties of an instrument which measures three-dimensional human posture are essential prior to applying it in clinical practice or research. Methods This paper reports the findings of a systematic literature review which aimed to 1) identify non-invasive three-dimensional (3D) human posture-measuring instruments; and 2) assess the quality of reporting of the methodological procedures undertaken to establish their psychometric properties, using a purpose-build critical appraisal tool. Results Seventeen instruments were identified, of which nine were supported by research into psychometric properties. Eleven and six papers respectively, reported on validity and reliability testing. Rater qualification and reference standards were generally poorly addressed, and there was variable quality reporting of rater blinding and statistical analysis. Conclusions There is a lack of current research to establish the psychometric properties of non-invasive 3D human posture-measuring instruments.


Background
Postural assessment is a standard and essential component of examining individuals with neuromusculoskeletal disorders [1,2]. Prolonged static postures are widely recognised as a risk factor of neuromusculoskeletal pain among children, adolescents and adults [3][4][5][6][7][8][9]. No uniform definition for "ideal" posture exists and therefore researchers and clinicians continue to seek the best way of assessing and describing posture. Ideal spinal posture is proposed as neutral spinal alignment, however the relationship between spinal segments in a normal population remains unknown [10,11]. The spine is a complex three-dimensional (3D) anatomical structure, whose segmental position in space should be described in all three planes (sagittal, frontal and transverse) [12][13][14]. Precise positional data can be derived from a number of biomechanical measurement tools, of which non-invasive 3D instruments are preferred.
It is essential that a spinal posture-measuring instrument is shown to be reliable and valid. Without this assurance, it cannot facilitate diagnosis, chart variability in 'usual' posture or assist objective monitoring of patient progress with treatment [1]. Researchers and clinicians should therefore be familiar with the psychometric properties of spinal posture-measuring instruments, and choose the ones with the best evidence of performance [15].
Two core elements of psychometric properties are reliability and validity [16]. Reliability and validity are interlinked of which reliability is a prerequisite to validity. A measurement tool cannot be recommended with confidence if there is a lack of evidence about its reliability and validity [17]. Reliability, refers to being able to estimate the inherent variability of posture, as well as error that can be attributed to the rater and the measurement instrument [17]. Error can relate to the consistency with which measurements are taken by the same or different raters, or over multiple occasions of testing [16]. Reliability is variously classified as test-retest reliability, inter-and intra-rater reliability. Test-retest reliability describes the stability of the measurement instrument in obtaining the same results with repeated measurements using the identical test on two or more separate occasions, keeping all testing conditions as constant as possible [17]. Intra-rater reliability is defined as the stability of data recorded by one observer across two or more test occasions. Inter-rater reliability is the extent to which two or more observers obtain similar scores when rating the same individuals [16,17].
Validity is the extent to which an instrument measures what it is intended to measure [18]. Criterion-related validity is the ability of one test (index test) to predict results obtained on an external criterion (gold standard/ reference standard) which is assumed to be valid. When both tests are performed on the same subjects, the scores from the index test are correlated with those achieved by the criterion measure. Construct validity is the ability of an instrument to measure an abstract concept, which cannot be observed directly and which has been constructed to represent an abstract trait [17]. There are two types of criterion-related validity. Concurrent validity is evaluated when the index test and the criterion measure are taken at the same time so that it reflects the same incident of behaviour while predictive validity is tested when the index test is performed and measured prospectively to ascertain the relationship between the index test and the criterion scores to determine whether the index test is a valid predictor of the outcome [17]. There are three types of construct validity. Convergent validity indicates that two measures, which are believed to reflect the same construct, will have similar results or will correlate highly [17]. Whereas divergent validity indicates that two measures, which are believed to measure different constructs, will correlate poorly [19]. Convergent and divergent validity assess the sensitivity and specificity of a measurement respectively [19]. Discriminative validity is the extent to which measures from a measurement instrument distinguishes between individuals or populations that would be expected to differ [19].
Establishing the psychometric properties of spinal posture-measuring instruments is not a trivial task, given the complex nature of human posture. Thus, convincing evidence of reliability and validity of any posture-measuring instrument can only be established by assessing the methodological quality of the underpinning developmental studies. Specific psychometric study design features are therefore essential to establish and assess, for instance, controls that are put in place for systematic bias, non-systematic bias and inferential error. An important requirement for psychometric testing of posture measurement is that the instrument be tested under a given set of conditions on a specific population within the context of the instrument's intended use. Therefore it is essential that posture-measuring instruments be tested on humans at some stage of development, and not just on inanimate objects [17].
The purpose of the systematic review reported in this paper was 1) to identify the non-invasive 3D tools which measure human static sitting or standing spinal posture and 2) to review the quality of the evidence of reliability and validity of the identified 3D posture-measuring instruments.

Search Strategies
Two inter-related search strategies (A and B) were implemented to ensure that all eligible papers were included. Strategy A sought any primary research studies which reported the use of 3D non-invasive instruments measuring static sitting or standing spinal posture. Strategy B sought primary research into the psychometric testing of these instruments. One reviewer searched six electronic databases that were available at the Stellenbosch University Library. The databases were BioMed Central, CINAHL, PEDRO, PROQUEST, PUBMED and SCIENCE DIRECT. The publication date was restricted to papers published from 1980 to June 2010. The search was limited to full-text papers published in English. MESH terms were used in PUBMED. See additional file 1 for a detailed description of the database searches.
In addition, secondary searching was performed on the reference list of the included papers. Experts in this field of research, and authors who failed to provide references to studies which tested an instrument's psychometric properties, were contacted.

Keywords and synonyms
The following keywords were used: three-dimensional, measurement tool, assessment tool, instrument, measurement, assessment, spinal posture, posture, validity, reliability, accuracy and reproducibility.

Inclusion and exclusion criteria for selection of papers
Papers were included if they reported testing an instrument's psychometric properties, specifically reliability and/or validity, using humans, or the instrument's validity using objects. A core inclusion criteria was that static standing or sitting spinal posture had to be evaluated with an instrument that could quantitatively calculate 3D spinal posture without using a baseline reference value such as zero. This was because a reference value requires that the subject be required to first assume a neutral or resting posture at which point the instrument is zeroed before the instrument can measure static spinal posture. For the purpose of the review, static posture should be assessed instantaneously without any guiding from the researcher.
Papers were excluded if (1) they reported neither reliability nor validity testing (2) they did not report on static spinal posture (e.g. reported on the 3D motion of the spine, scapulo-humeral girdle or pelvis); (3) the study reported on the validity testing of an instrument using motion (as motion was not incorporated in this review, and we argue that validity be evaluated within the context of the instrument's intended use; (4) the instrument only measured cadaver or in vitro spinal posture; (5) the instrument was invasive e.g. biplanar radiography and stereoradiography; (6) only an algorithm or a mathematical formula were reported.

Study selection
One reviewer excluded papers by screening all the titles and reading the abstracts after which two independent reviewers selected the eligible papers after reading the full text version of the remaining papers. Figure 1 describes the procedures of study selection for each of the two search strategies.

Methodological Quality Appraisal
The full text eligible papers were then subjected to methodological critical appraisal. The Critical Appraisal Tool (CAT) applied in this review was purpose-built, in the absence of any other relevant CAT. It was adapted from the Quality Assessment of Diagnostic Accuracy Studies (QUADAS) [20] and the Quality Appraisal of Reliability Studies (QAREL) [21]. The purpose-built CAT has 13 items, however its data is not designed to be reported as a composite quality score (see additional file 2). The CAT was designed to assess the impact of each individual item on the quality of the methodological procedures implemented in each paper. Prior to critical appraisal of the included articles, three papers were randomly selected and assessed independently by three reviewers using the purpose-built CAT. Disagreements were discussed to ensure that interpretation of the CAT items were consistent.

Results from the search strategies
One hundred and thirty possible papers were considered, of which 30 papers were deemed to be eligible. Nine additional papers were identified after searching the reference lists of these papers. Two further papers were included after experts and authors had been contacted. Figure 2 provides a consort diagram to demonstrate the selection of papers.

Volume of literature
Eighteen instruments were identified from the two literature searches, 15 from Search A, one from Search B and two from author contacts. The instruments are listed in the first column of Table 1, the papers addressing aim one appear in the second column and those addressing aim two are in the third column. Papers reporting these instruments, are identified by bold script if from strategy A, italics if from strategy B, normal script if from author search and with a * if from secondary searching. The Automatic Scoliosis Analyser System (Auscan) (Italy), the Elite system (Italy), the Optotrak 3020 (Canada), the Peak Motus (USA), the PosturePrint (Canada), the Qualysis Proreflex Motion Capture Unit system (Sweden), the Vicon 370 (England) and an Optoelectronic camera system (Canada) are optoelectronic analysis systems. The Seventeen papers reported on reliability and/or validity of the included instruments and were thus assessed to address Aim two (see Table 1 third column). One paper by Smidt et al. [22] reported on both reliability and validity, and was therefore reviewed as if it was two separate papers, due to the nature of this review. Drerup et al. [23] tested a new algorithm for processing data presented in a previous paper [24]. These papers were reviewed as if they were one paper, because the previous paper reported on the study procedure in more detail whereas the latter paper discussed the latest improvement made on the data processing procedure.

Aim of the reliability studies
The aim of six studies was to test the reliability of a 3D instrument in assessing the spinal posture of humans [22,[25][26][27][28][29].

Aim of the validity studies
The aim of eleven studies was to test the validity of a 3D posture instrument. Four studies [23,[30][31][32] used human subjects to measure 3D spinal posture and to compare the results with those obtained from a reference standard. The other seven studies either used mannequins [33][34][35], wooden wedges [36], a steel frame [22], parallelograms [37] or other objects with known parameters [38] to test the validity of an instrument that could be used to assess 3D spinal posture of humans in future.

Study design for reliability and validity studies
The type of reliability and validity tested, as well as the time interval for the reliability studies and the reference standard for the validity studies, are reported in Table 2. Table 3 summarizes the statistical procedures implemented in the reliability and validity studies. Comparing the findings in this table with the types of reliability and validity testing reported in Table 2, highlights the variability in choice and application of statistical tests to assess the same constructs. Table 4 reports the findings from the critical appraisal of the papers, related to reliability and validity testing.

Methodological Quality Appraisal
Item 1: If human subjects were used, did the authors give a detailed description of the sample of subjects used to perform the (index) test?
Item 3: Was the reference standard explained?   [32] scored "yes" as they provided references for the methods used to digitize the radiographs. Pazos et al. [35] and Pearcy et al. [36] scored "yes" because the authors named and stated the accuracy of the instruments used as the reference standard. Norton et al. [38] scored "no" because the ruler or tape measure was inappropriately used as a reference standard for calculating 3D coordinates of a point in space. Harrison et al. [33], Janik et al. [34], Normand et al. [37] and Smidt et al. [22] scored "no" because the authors used an object with known 3D parameters as reference standards, but the methods to measure these 3D locations, angles or distances were not explained.
Item 4: If interrater reliability were tested, were raters blinded to the findings of other raters?
Normand et al. [26] and Smidt et al. [22] scored "yes" because subjects were evaluated separately by the different raters. Geldhof et al. [25], Warren et al. [28] and Whittle and Levine [29] only tested intrarater reliability and scored "not applicable". Pazos et al. [26] scored "not applicable" because no rater reliability was evaluated but instead test-retest reliability of the instrument, when using different postures, was evaluated.
Item 5: If intrarater reliability were tested, were raters blinded to their own prior findings of the test under evaluation?
Geldhof et al. [25], Normand et al. [26] and Smidt et al. [22] scored "yes" because the raters were sufficiently blinded to their own prior measurements as either repeated digitizing of the anatomical landmarks took place one week apart, all photographs were numbered and were not identifiable by subject name, occasion or characteristics, and no skin markings were made on subjects. Warren et al. [28] and Whittle and Levine [29] scored "no" because passive and skin markings respectively were placed only once on the subject and were not removed between repeated measurements. Pazos et al. [27] scored "not applicable" because they did not test rater reliability.
Item 6: Was the order of examination varied? Normand et al. [26] scored "yes" because subjects were evaluated in random order. Warren et al. [28] and Whittle and Levine [29] scored "no" because repeated measurements were performed consecutively without changing the order of subjects during testing. Geldhof et al. [25] scored "no" as the order of testing was kept the same for the repeated measurements one week apart. Smidt et al. [22] scored "no" as insufficient information was provided. Pazos et al. [27] scored "not applicable" because no rater reliability was tested.
Item 7: If human subjects were used, was the time period between the reference standard and the index test short enough to be reasonably sure that the target condition did not change between the two tests?
Item 8: Was the stability (or theoretical stability) of the variable being measured taken into account when Table 4 Summary of the methodological quality appraisal results of the studies (n = 17)

Pearcy et al (1989)
n/a x √ n/a n/a n/a n/a n/a √ √ √ n/a √ Smidt et al (1992) (validity) n/a x x n/a n/a n/a n/a n/a Norton et al (1993) n/a x x n/a n/a n/a n/a n/a √ √ √ n/a x Drerup et al (1994; x x √ n/a n/a n/a √ n/a √ √ √ √ √ Whittle et al (1997) √ x n/a n/a x x n/a √ n/a √ n/a √ √

Normand et al (2002)
n/a x x n/a n/a n/a n/a n/a

Pazos et al. (2005)
n/a x √ n/a n/a n/a n/a n/a √ √ √ n/a √

Harrison et al (2007)
n/a x x n/a n/a n/a n/a n/a x √ x n/a √

Janik et al (2007)
n/a x x n/a n/a n/a n/a n/a

Pazos et al (2007)
√ x n/a n/a n/a n/a n/a √ n/a √ n/a x √ determining the suitability of the time-interval between repeated measures? Six papers scored "yes" because repeated measurements of posture were either taken on the same day [22,[27][28][29] one week [25] or one day apart [26].

Normand et al (2007)
Item 9: Was the reference standard independent of the index test?
Item 10: Was the execution of the (index) test described in sufficient detail to permit replication of the test?
Item 11: Was the execution of the reference standard described in sufficient detail to permit its replication?
Seven papers scored "yes" because clear descriptions of how the reference standard were used on the subjects [23,32] or on the inanimate objects [35,36,38] or citations for the methodology [30,31] were provided. Harrison et al. [33], Janik et al. [34], Smidt et al. [22] and Normand et al. [37] scored "no" for the reasoning provided for item 3.
Item 13: Were the statistical methods appropriate for the purpose of the study?
All but one paper by Norton et al. [38] implemented appropriate statistical analysis and thus scored "no". Although the other sixteen papers reported appropriate statistical analysis only six papers [23,30,31,26,28] provided a justification or motivation for using their chosen statistical measures.

Discussion
This review attempted to evaluate the quality of reporting of psychometric properties of 18 3D human posture measuring instruments. It identified a lack of well-documented studies testing the psychometric properties of these instruments, as papers describing the development of only eight instruments were found (see Table 1 column C). The review suggests that the PosturePrint and rasterstereography had relatively more psychometric testing than the other tools included in this review. However, the methodological quality of the testing procedures for all instruments was flawed, when considering the methodological criteria applied in this review.

Rater qualification
Both reliability and validity studies should provide descriptions of the qualifications of the rater(s) used in the studies because the rater(s) professional background, expertise and prior training operating these instruments will affect psychometric property assessment. Appropriate training of raters is important to minimise measurement error, and to facilitate interpretation of findings. These factors should therefore be considered when interpreting study findings, and extrapolating them for applicability and generalisability to other clinical and research settings [39].

Reference standard
Four studies, which used inanimate objects, did not identify the instruments used to obtain the known values of objects which provided the reference standard data. In order to test validity, it is important that the psychometric properties of the reference standard be known to confirm that the reference standard is suitable [39]. The most suitable non-invasive 3D reference standard for postural measurements has not been unanimously determined in this field of research. The validity studies that used humans also used stereoradiography as reference standard, as radiography remains the most accurate assessment for posture. This situation continues, even though there is a possible health risk for repeated X-ray exposure to healthy spines and organs [40].
Norton et al. [38] used a ruler or tape measure as a reference standard. The x, y, z coordinates obtained from the index test had to be mathematically transformed to distances between pairs of points before the reference data, obtained from the ruler or tape measure, could be used. It would have been better had these authors used a reference standard with known accuracy to measure 3D coordinates directly. The ruler or tape measure was also a poor reference standard to use when measuring the distance between pairs of points on the human skeleton.

Blinding for intra-or interrater reliability
The repeated measurements by Geldhof et al. [25] were performed one week apart however the order of the subjects was fixed. Therefore this enhances the possibility for the raters to recall the test outcomes of the previous measurements and potentially incurs increased bias. Warren et al. [28] and Whittle and Levine [29] tested intrarater reliability however the marking of the anatomical landmarks was only undertaken once before repeated measurements were taken, without allowing for removal and replacement of the markers between repeated measurements. Both raters in these studies were not blinded to their previous measurements of the same subjects. Consequently this potentially introduced bias and compromised the quality of the studies and findings.

Statistical analysis
Given the complexity of posture measurement and interpretation, no statistical strategy for psychometric property testing is without its disadvantages. Therefore it seems sensible to report the findings of two or more different statistical analysis approaches in order to validate findings [21]. This did not occur in any of the included papers. For example Pearcy et al. [36] used linear regression analysis to demonstrate that as the magnitude of the one variable increases so does the amount of error however there is no indication of a cut off value (e.g. 95% CI and SD) up to where the 3 Space Isotrak can be expected to accurately measure an angle.
As a variety of statistical measures were reported in this review, another method to improve reporting quality would be for authors to justify why they chose a particular statisical test, relevant to the purpose of testing. This would provide the reader with better insight into the results, and would perhaps guide future authors in choice, and interpretation of more appropriate statisical analysis. For example Norton et al. [38] used multiple analysis to determine whether there is agreement between measures. However Pearson product moment correlation only reports on the correlation between two different measurements and cannot quantify the amount of aggreement or indicate whether there is systematic error. Repeated t-tests are also inappropriate to test systematic differences, as this testing will inflate the type I error and compromise interpretation of significance.

Limitations
One limitation to this review comes from our inability to retrieve potentially eligible papers from authors who failed to respond to email inquiries. It could be that there are other relevant instruments which have been adequately evaluated for reliability and validity, however these papers were not available despite using multiple search methods (database, internet and author searches).

Conclusions
This review described 18 non-invasive ways of measuring static human 3D sitting or standing spinal posture, and the methodological procedures of testing reliability and validity of a subset of these instruments. The review concludes that further research into the reliability and validity testing of these instruments is required to improve the quality of reliability and validity evidence of posture-measuring instruments. Psychometric property testing should be improved by addressing rater qualification, clearer definitions of the reference standards, applying appropriate methodological procedures to enhance rater blinding and improving the quality of reported statistical analysis. By improving the methodological rigor of reliability and validity testing, it would consequently enhance users' confidence in the psychometric evidence of static human 3D sitting or standing spinal posture in clinical and research settings.