Pain Medicine 2011; 12: 276–281 Wiley Periodicals, Inc.
NEUROPATHIC PAIN SECTION Original Research Articles Can Neuropathic Screening Tools Be Used As Outcome Measures? pme_1037
276..281
Robert D. Searle, MBChB,* Michael I. Bennett, MD,† and Alan Tennant, PhD‡ *St James’s University Hospital, Leeds; †
International Observatory on End of Life Care, School of Health & Medicine, Lancaster University, Lancaster; ‡
Faculty of Medicine & Health, The University of Leeds, Leeds, UK Reprint requests to: Robert D. Searle, MBChB, Academic Unit of Anaesthesia, Level 8 Clinical Sciences Building, St James’s University Hospital, Leeds LS9 7TF, UK. Tel: +44 (0) 113-20-65282; Fax: +44 (0) 113-20-64140; E-mail:
[email protected].
struct) and there is no evidence of differential item functioning across gender and age groups. The scale only has reliability consistent with use at the group level. Conclusions. Neuropathic screening tools such as the LANSS have been used as outcome measures in clinical studies. Rasch analysis demonstrates that the LANSS can be used as such in specific populations of patients with neuropathic pain, however it’s reliability in this context does not support use at the individual level and it cannot be used as a generalised measurement tool across pain diagnostic groups. The LANSS remains primarily a diagnostic tool. Key Words. LANSS; Rasch; Neuropathic Pain; Outcome Measure
Abstract Introduction Objective. To examine whether the Leeds Assessment of Neuropathic Symptoms and Signs (LANSS) screening tool can satisfy Rasch model expectations and therefore be transformed into an interval level measurement scale, suitable for use as an outcome measure in clinical studies. Design. Rasch analysis (using the software RUMM2020) of LANSS data from both a random selection of all chronic pain patients and also specific chronic pain diagnostic categories. Patients. Original LANSS data from a previous study of 2,480 patients with chronic pain were used. Specific diagnostic groups examined included diabetic neuropathy, postsurgical pain, osteoarthritis, posttraumatic injury, and low back pain. Outcome Measures. The following assessments were made: fit to the Rasch model, scale reliability, scale multi-dimensionality, and differential item functioning. Results. The overall fit to the Rasch model was only acceptable in two groups; diabetic neuropathy and chronic postsurgical pain. For these groups, the scale is unidimensional (measures a single con276
Neuropathic pain is common in many diseases or injuries of the peripheral or central nervous system, and has a substantial impact on quality of life and mood [1]. Consequently, the accurate measurement of neuropathic pain is essential, both for clinical practice, and research. The Leeds Assessment of Neuropathic Symptoms and Signs (LANSS) was initially designed as a diagnostic aid to help distinguish neuropathic from nociceptive pain and was not conceived of as a measurement tool [2]. However, there is support for the theoretical construct that pain can be more or less neuropathic, and that pain scales are sensitive to changes in the degree of neuropathic pain [3]. Clinical trials have demonstrated corresponding falls in both pain intensity and LANSS scores with treatment but not in controls [4–6]. As such, it is valid to ask can the LANSS be considered a potential outcome measure for interventions concerned with neuropathic pain? Most outcome measures in healthcare, including the LANSS are ordinal in nature. This is sufficient to separate patients into groups based upon the magnitude of the construct being measured, as for example do many depression scales. It is also sufficient to determine if a patient has improved or worsened. Although ordinal data can be subjected to a substantial body of nonparametric
Neuropathic Screening Tools as Outcome Measures statistics, it is important to realize this does not constitute measurement as it is known in the physical sciences [7]. Although ordinal pain intensity scores are ranked, the distance between each point does not necessarily reflect equidistant steps in the underlying trait [8]. This becomes a problem when more complex statistical interpretation is required such as calculating change scores or “effect size” in clinical trials, which requires normally distributed, interval level measurement [7]. Recently this problem has, to a certain extent, been overcome by the application of Rasch analysis to ordinal scales. This process can transform ordinal scales into interval level measurement, providing the data fit the Rasch mathematical model to an acceptable degree [9]. The Rasch model is underpinned by the theory of conjoint measurement, which demonstrates that when analysing data the important indication of a measurement structure is in the relationship between variables, not the physical values themselves [7]. This becomes important in the human sciences where traits such as pain are not directly measurable. In the case of neuropathic pain, scales that satisfy the expectations of the Rasch model can be used to specify not only that a patient’s neuropathic pain has improved, but also by how much it has improved on an interval scale. This has important implications for research and clinical practice, allowing calculation of the magnitude of the effect of interventions to reduce neuropathic pain. To our knowledge, no existing neuropathic pain scales have been developed to Rasch model standards such that their ordinal raw scores can be transformed to interval level measurement. In addition, the LANSS was developed as a screening tool, and these are often not designed to perform as measurement scales, rather to focus on a cut point, which determines, in this case, neuropathic pain. Thus, we aimed to investigate whether the LANSS can satisfy Rasch model expectations as a summed score that can be transformed to an interval level measurement scale, suitable for use as an outcome measure in clinical studies.
Methods Original data were collected from a previous study of the diagnosis and management of neuropathic pain in Belgium [10]. In this study, 177 general practitioners and 97 specialists were asked to complete LANSS questionnaires on patients presenting with pain of >3 months duration. Details of the possible underlying cause, age, gender and duration of pain were also recorded. A translation and back translation method was employed to adapt the LANSS into Dutch and French. Data were collected from 2,480 patients with chronic pain. The frequency of different underlying diagnostic groups is presented in Table 1. Where diagnostic uncertainty existed, clinicians were able to choose more than one diagnostic category for that patient.
Table 1 Frequency of likely underlying causes for chronic pain from the data studied Diagnosis
Number of Patients
Diabetic neuropathy Cancer Low back pain Osteoporosis Multiple Sclerosis Thalamic syndrome Post-herpetic neuralgia Post traumatic injury CRPS Alcohol abuse Syringomyelia Other Postsurgical lesion Carpal tunnel Osteoarthritis Post CVA
253 75 781 184 38 11 163 326 178 85 11 499 232 91 590 76
All data were formatted in SPSS before exporting as an ASCII file to RUMM2020 Rasch analysis software (RUMM Laboratory Pty Ltd.) for analysis. Rasch analysis on the following data groups were performed: 1. A random, representative sample of all LANSS responses for 400 patients. 2. Analysis of individual diagnoses separately where the number available for analysis was greater than 150 (following exclusion of extreme scores). This included Diabetic neuropathy, postsurgical pain, osteoarthritis, posttraumatic injury, and low back pain. The Rasch analysis included the following investigations listed below. Fit to the Rasch Model At the scale level, summary fit statistics included item and person residuals and a chi-square item–trait interaction fit statistic. With perfect fit to the model, item, and person residuals would have a mean of 0 and a standard deviation of 1, and the chi-square item–trait interaction should be nonsignificant (Bonferroni adjusted). At the individual item level, fit residuals should lie between ⫾2.5 and chisquare results should be nonsignificant (Bonferroni adjusted). Scale Reliability The person separation index indicates the power of the scale to discriminate among respondents. A person separation index of 0.7 is the minimum acceptable to differentiate between two groups, and 0.9 is needed for interpretation at the individual level [11]. 277
Searle et al.
Table 2 Summary Rasch statistics Item Residual
Person Residual
Chi-Square Interaction
Analysis Name
n
Value
SD
Value
SD
Value (df)*
P
PSI†
Random sample Osteoarthritis Postsurgical pain Trauma Diabetic neuropathy Low back pain
313 386 173 213 171 558
-0.071 0.053 0.222 0.109 -0.213 -0.05
1.54 1.103 1.245 1.427 1.065 1.495
-0.125 -0.1 -0.025 -0.029 -0.146 -0.112
0.720 0.707 0.718 0.7 0.743 0.684
34 28 21 21 28 28
0.00 0.00 0.06 0.00 0.05 0.00
0.7 0.7 0.7 0.8 0.7 0.7
* Degrees of freedom. † Person separation index.
Only LANSS data groups with acceptable summary fit statistics and a person separation index of at least 0.7 proceeded to further analysis:
from the analysis as there is no variation in their responses. Initial analysis was also performed on individual diagnostic categories with >150 participants once extreme scores were removed.
Scale Multi-Dimensionality The possibility that the scale is measuring more than one latent construct is examined by two methods, through a Confirmatory Factor Analysis based upon a tetrachoric correlation, and a method recommended by Smith for use within Rasch analysis [12]. The former is used as the latter is likely to be underpowered with just seven dichotomous items. The latter involves examining the first principal component of the residuals, and then using t-tests to compare person-locations based on different subsets of items located on the same scale. If less than 5% of the t-test comparisons are significant, then the scale is considered unidimensional. Differential Item Functioning (DIF) Another key aspect of measurement is the invariance of the scale by external groups, such as age and gender. At any specific level of neuropathic pain, the response to an item should be the same, irrespective of group membership. This is tested though DIF analysis across gender and age groups (0–35, 36–70, 71–99). Finally, Rasch transformed interval level scores were created for data sets that fit the Rasch model to an acceptable degree. Results Unidimensionality and Fit to Rasch Model Confirmatory factor analysis on the random sample of 400 cases indicated a strong unidimensional construct with a Root Mean Square Error of Approximation of 0.00; CFI and TLI of 1.0. Data from the random sample were then fitted to the Rasch model. Extreme scores (representing the maximum or minimum score on the LANSS scale) were excluded 278
A summary of the test of fit statistics for the random LANSS selection and diagnostic categories with >150 participants available for analysis are presented in Table 2. The person separation index was 0.7 across the groups analyzed, suggesting the scale only has reliability consistent with use at the group level. The overall fit to the Rasch model (reflected by nonsignificant Chi-square interaction statistics) was only acceptable in two groups; diabetic neuropathy and chronic postsurgical pain. These two groups were tested further for unidimensionality, DIF, and local dependency. Independent t-test was used to compare person locations that had been estimated using two different subsets of items from the final scale. In practice, this meant that the three highest positive loading items on the first residual component were compared with the four highest negative loading items (with both sets calibrated on the same metric scale). One hundred and seventy (diabetic neuropathy) and 173 (postsurgical pain) t-test comparisons were made; none were significant, supporting the unidimensionality of the scale in these groups. DIF Age and gender as person factors were used to examine for both uniform and nonuniform DIF. No items demonstrated evidence of DIF following Bonferroni adjustment in either the diabetic neuropathy or postsurgical pain groups. This suggests the LANSS scale items are measuring the same construct amongst different gender and age groups. Local Dependency Local dependency occurs when an individual’s response to one item will influence their response to another, different item within the scale. Local dependency can be identified by searching for positive correlations among item
Neuropathic Screening Tools as Outcome Measures
Figure 1 Item–person threshold map for postsurgical pain patients.
residuals. No items had correlations greater than 0.3 in both groups analyzed, indicating an absence of local dependency. Item–Person Threshold Distributions The item–person threshold distribution map for both diabetic neuropathy and postsurgical pain groups are presented in Figures 1 and 2, respectively. These graphs display distributions of person “ability” (in this case level of neuropathic pain) and LANSS question difficulty on the same logit scale. In both groups the item difficulties are spread through the middle of the ability range. There is some evidence of a floor and ceiling effect, meaning that the scale is less discriminating at lower and higher levels of neuropathic pain. The raw scores for the LANSS scale were transformed into interval scale scores and are presented in Table 3.
Discussion Our analysis shows that for patients with diabetic peripheral neuropathy and chronic postsurgical pain the LANSS fits the Rasch model to an acceptable degree, demonstrating unidimensionality, and no evidence of DIF between gender and age groups. LANSS raw scores can therefore be transformed into interval level measurement for these groups. It is important to note that the reliability of the LANSS only allows statistical interpretation of results at the group level, and cannot be reliably used to measure change in individuals. In addition, the item–person distribution maps show that the LANSS is unable to discriminate between persons at either end of the ability spectrum. This is likely to reflect the relatively low number [7] of questions in the LANSS scale. However, it is also important to note that classical tests of reliability are not well suited to the
Figure 2 Item–person threshold map for diabetic neuropathy pain patients. 279
Searle et al.
Table 3 Raw LANSS score to Rasch transformed interval score conversion table Interval Score (Logits) LANSS Raw Score
Diabetic Neuropathy
Postsurgical Pain
0 1 2 3 4 5 6 7
0.00 1.18 2.12 2.91 3.65 4.50 5.61 7.00
0.00 1.23 2.19 2.99 3.75 4.58 5.65 7.00
LANSS = Leeds Assessment of Neuropathic Symptoms and Signs.
non-normal distributions typically found in population screening instruments [13]. In these circumstances, the emphasis is upon the cut point, not the fact that large numbers of people are well below this level. Consequently, the test information function needs to be maximised at the cut point, giving the greatest degree of precision at this level of the scale. Nevertheless, just seven dichotomous items do not present a high degree of precision, nor are they suitable for outcome measurement as they have extremely poor precision at the margins of the scale. The analysis showed that the LANSS does not fit the Rasch model to an acceptable degree when used across diagnostic categories. Of the diagnostic categories with sufficient numbers to allow analysis, the categories more associated with nociceptive pain such as osteoarthritis and low back pain did not fit the Rasch model. In contrast, diagnostic categories associated with neuropathic pain (diabetic neuropathy and postsurgical pain) did fit the Rasch model. Although Rasch analysis has been used to examine the clinical and experimental responses of individuals to ordinal pain intensity scales, there is little work examining scales that assess neuropathic pain [8,14,15]. A number of neuropathic pain screening tools have been developed in the last 10 years in addition to the LANSS, and although our findings may extend to include these, they should be examined individually to confirm this. Similarly, our findings do not extend to scales developed to measure change in neuropathic pain (in contrast to screening tools), such as the Neuropathic Pain Scale [16,17]. Such scales are developed differently, and any interval scale properties should be examined individually. One such scale, the Pain Quality Assessment Scale, has recently been examined using item response theory, revealing a lack of interval scaling [18]. That screening tools such as the LANSS have been used as outcome measures suggests there is a need for 280
outcome tools that reflect changes in neuropathic pain in response to treatment. Although we have demonstrated that the LANSS can be used as such in specific populations of patients with neuropathic pain, it’s reliability in this context does not support use at the individual level and it cannot be used as a generalized measurement tool across pain diagnostic groups. The LANSS remains primarily a screening tool, properties that were not investigated by the current study. This study does demonstrate however, that the utility of the LANSS does not necessarily translate to use as an outcome measure in clinical trials. Future work should concentrate on expanding the number of questionnaire items to aid differentiation between neuropathic pain “ability” levels and improving scale reliability. Developing a reliable and generalised neuropathic pain measurement tool, which fits the Rasch measurement model, would have an important impact on both research and clinical work. Acknowledgment We are grateful to Pfizer for allowing us access to the original LANSS data used in this study. References 1 Finnerup NB, Sindrup SH, Jensen TS. Chronic neuropathic pain: Mechanisms, drug targets and measurement. Fundam Clin Pharmacol 2007;21(2):129–36. 2 Bennett M. The LANSS pain scale: The Leeds assessment of neuropathic symptoms and signs. Pain 2001;92(1–2):147–57. 3 Bennett MI, Smith BH, Torrance N, Lee AJ. Can pain can be more or less neuropathic? Comparison of symptom assessment tools with ratings of certainty by clinicians. Pain 2006;122(3):289–94. 4 Solak O, Metin M, Esme H, et al. Effectiveness of gabapentin in the treatment of chronic postthoracotomy pain. Eur J Cardiothorac Surg 2007; 32(1):9–12. 5 Khedr EM, Kotb H, Kamel NF, et al. Longlasting antalgic effects of daily sessions of repetitive transcranial magnetic stimulation in central and peripheral neuropathic pain. J Neurol Neurosurg Psychiatry 2005; 76(6):833–8. 6 Hans G, Joukes E, Verhulst J, Vercauteren M. Management of neuropathic pain after surgical and non-surgical trauma with lidocaine 5% patches: Study of 40 consecutive cases. Curr Med Res Opin 2009;25(11):2737–43. 7 Tennant A, McKenna SP, Hagell P. Application of Rasch analysis in the development and application of quality of life instruments. Value Health 2004;7(suppl 1):S22–6.
Neuropathic Screening Tools as Outcome Measures 8 Pesudovs K, Noble BA. Improving subjective scaling of pain using Rasch analysis. J Pain 2005;6(9):630–6. 9 Tennant A, Conaghan PG. The Rasch measurement model in rheumatology: What is it and why use it? When should it be applied, and what should one look for in a Rasch paper. Arthritis Rheum 2007; 57(8):1358–62. 10 Hans G, Masquelier E, De Cock P. The diagnosis and management of neuropathic pain in daily practice in Belgium: An observational study. BMC Public Health 2007;7:170. 11 Fisher JW. Reliability statistics. Rasch Meas Trans 1992;6(3):238. 12 Smith EV Jr. Detecting and evaluating the impact of multidimensionality using item fit statistics and principal component analysis of residuals. J Appl Meas 2002;3(2):205–31. 13 Wright B. Separation, reliability and skewed distributions. Rasch Meas Trans 2001;14:786.
14 Decruynaere C, Thonnard JL, Plaghki L. Measure of experimental pain using Rasch analysis. Eur J Pain 2007;11(4):469–74. 15 Decruynaere C, Thonnard JL, Plaghki L. How many response levels do children distinguish on faces scales for pain assessment. Eur J Pain 2009; 13(6):641–8. 16 Bennett MI, Attal N, Backonja MM, et al. Using screening tools to identify neuropathic pain. Pain 2007;127(3):199–203. 17 Galer BS, Jensen MP. Development and preliminary validation of a pain measure specific to neuropathic pain: The Neuropathic Pain Scale. Neurology 1997; 48(2):332–8. 18 Waterman C, Victor TW, Jensen MP, et al. The assessment of pain quality: An item response theory analysis. J Pain 2010;11(3):273–9.
281