Rasch Analysis and Clinical Interpretation of the Modified Oswestry Low Back Pain Disability Questionnaire

Ying-Chih Wang^1*, Jay Kapellusch¹, Bhagwant Sindhu¹, Leigh Lehman², Xiaoyan Li³ & Sheng-Che Yen⁴

¹Department of Occupational Science & Technology, University of Wisconsin-Milwaukee, Milwaukee, WI, USA
²Department of Occupational Therapy, Augusta University, Augusta, GA, USA
³Department of Physical Medicine and Rehabilitation, University of Texas Medical School at Houston, and TIRR Memorial Hermann Research Center, Houston, TX, USA
⁴Department of Physical Therapy, Northeastern University, 360 Huntington Ave, Boston, MA, 2115, USA

*Correspondence to: Dr. Ying-Chih Wang, Department of Occupational Science & Technology, University of Wisconsin-Milwaukee, Enderis Hall 955, 2400 E Hartford Ave, Milwaukee, WI 53211.

Copyright © 2018 Dr. Ying-Chih Wang, et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Abbreviations
m-ODI = modified Oswestry Low Back Pain Disability Questionnaire
ICC = intraclass correlation coefficient
DIF = differential item functioning

Introduction
Lumbar spine impairments affect a substantial proportion of individuals. It is estimated that 2.06 million people went to the emergency room for low back pain between 2004-2008 in the US, which approximates 3.2% of all emergency visits [1]. Low back pain is associated with abnormal alignment of vertebrae and abnormal muscle, ligament and bone structures. Weakness or injury to any of the spine structures (e.g., sprains and strains, intervertebral disc degeneration, herniated discs, radiculopathy, spondylolisthesis, traumatic injury, spinal stenosis, skeletal irregularities) at the lumbar region may cause functional disabilities associated with lumbar spine impairments.

The Oswestry Disability Index (ODI) [2,3] also known as the Oswestry Low Back Pain Disability Questionnaire, is commonly used in both clinical and research settings to measure functional disability related to low back impairments. The questionnaire has been applied to patients with various etiologies [4-7] and in diverse settings (e.g. community-dwelling adults [8], registry database [9], outpatient patients) [10]. Presently, more than 27 versions of ODI adaptations have been applied in 24 different languages/cultures [11]. In particular, Fritz and Irrgang (2001) [4] have reported a modified version that replaces the item of ‘sex life’ with ‘employment/homemaking’. This modification emphasizes the aspect of activity/participation that is otherwise absent from the original ODI. Here, we abbreviated this modified version as m-ODI.

Previous studies have examined the reliability, validity, and sensitivity of ODI to changes of low-back functional disability. For example, moderate to high reliability coefficients have been reported in a number of studies including test-retest reliability (intraclass correlation [ICC] = 0.92 [12], 0.90 [4], 0.88 [13], 0.70) [14], intrarater reliability (ICC = 0.93) [15], and internal consistency (Cronbach’s alpha = 0.97 [15], 0.96 [12], 0.88 [13], 0.85 [10], 0.82 [14], 0.78) [16]. The concurrent validity of ODI is evaluated by correlating the ODI with other measures such as visual analogue pain intensity scale (r = 0.67), Roland-Morris Disability Questionnaire score (r = 0.71 [13], r = 0.76) [12], and short-form-36 (r = 0.25 to 0.46) [6].

The unidimensionality of ODI is also supported by factor analysis [9,10,12]. Meanwhile, responsiveness studies reported several cutoff thresholds to classify patient’s improvement: minimum clinically important difference (score range for m-ODI 0-50 points) = 7.5 points [5], 6 points [4], Guyatt’s responsive index = 3.49 points [4], minimal detectable change = 4.23 points [5], and standardized response mean = 1.25 [5].

On the other hand, there are limited studies on the psychometric properties of the ODI questionnaire using Rasch analysis at the item level [8,17-21]. Despite studies supporting unidimensionality of the ODI, the match of item difficulty to patient functioning, and general fit to the Rasch model, they revealed some disordered thresholds [17] and underused response category (especially category 2 from the 0-5 rating scale structure) [8,19,21] Inconsistent results from the Rasch analysis are also observed. For example, several studies [8,17-20,22] suggest lifting is the most challenging item; however, one study categorizes lifting items as having average difficulty [21]. The person separation index - an estimate of how well the scale can differentiate persons into statistically distinct person strata - demonstrates a wide range from 0.88 [17], 1.77 [8] to 2.83 [19]. Inconsistencies in the studies suggest that previous findings may be affected by specific settings (e.g., outpatient vs spine center, work-related vs spine deformity) and limited by their relatively small sample size (n = 42 [22], 95 [8], 100 [17], 133 [21], 408 [19], 942 [20], 1,610) [18]. Presently, there are only two studies [17, 22] examining the psychometric properties of m-ODI. To validate previous studies and achieve more reliable results, it is necessary to increase the sample size studied.

Furthermore, as norms (e.g., percentile rank distribution) and functional staging (function status classification) approach [23-26] has been applied to assist in clinically meaningful interpretations, such results have not yet been available for the m-ODI questionnaire.

Therefore, the purposes of this study were to: (1) examine the psychometric properties of 10-item m-ODI questionnaire using the Rasch analysis, (2) establish the percentile rank of a m-ODI score and (3) develop a functional staging to guide clinical interpretation of the patient’s improvement by interpreting m-ODI scores.

Methods

Data Collection
This study uses previously collected clinical patient data administered by clinical staff. Patients seeking rehabilitation care entered their demographic data and completed computer-based, self-report surveys using a Patient Inquiry® software developed by FOTO, Inc (Knoxville, TN, USA) prior to receiving their initial evaluation and therapy. The same survey was administered again at discharge.

Data for this study were selected from the FOTO database if patients met the following criteria: (1) were 18 years old and older; (2) received outpatient physical therapy; (3) received orthopedic care due to lumbar spine impairments; and (4) completed the m-ODI questionnaire between January 2012 and May 2016. IRB approval was waived as this was a secondary data analysis using de-identified data free of personal identifiers.

Setting and Participants
Data analysis was based on the survey of 28,751 patients with orthopedic lumbar spine impairments from 652 clinics in the US. Table 1 presents the demographic information of the patients.

Abbreviations: SD=standard deviation; min=minimum; max=maximum, m-ODI = modified Oswestry Low Back Pain Disability Questionnaire

Outcome Measures
The m-ODI questionnaire was designed to give clinicians information about how back or leg pain is affecting a patient’s ability to manage in everyday life. The questionnaire includes 10 items: pain intensity (item 1), personal care (item 2), lifting (item 3), walking (item 4), sitting (item 5), standing (item 6), sleeping (item 7), social life (item 8), traveling (item 9), and employment/homemaking (item 10). Each item consists of six statements correlating to scores of 0 through 5. Patients are instructed to choose the statement that best matches his or her functional ability (i.e., the patient reported that he/she was able to complete the task). Final scores are calculated as follows: [total score/(5 x number of questions answered)] x 100% and range from 0 to 100. A score of 0 indicates the least disability, and a score close to 100 represents the greatest disability (i.e., lower scores are better). Recommended by Fairbank et al., five levels of the score were suggested: (1) 0% to 20%: minimal disability, (2) 21% to 40%: moderate disability, (3) 41% to 60%: severe disability, (4) 61% to 80%: crippled, and (5) 81% to 100%: bed-bound [2,3]. This classification has been applied in routine clinical practice and medical research [27-29]. For patients with minimal disability (0%-20%), this group can cope with most living activities, apart from requiring advice on lifting, sitting posture, physical fitness, and diet. For patients with moderate disability (21%-40%), travel and social life are more difficult and they may be disabled from work. Personal care and sleeping are not grossly affected and the patient can usually be managed by conservative means. For patients with severe disability (41%-60%), pain remains the main problem in this group but activities of daily living are affected. For patents in the crippled disability (61%-80%), back pain impinges on all aspects of the patient’s life. For patients with the most disability (81%-100%), these patients are either bed-bound or exaggerating their symptoms.

Rasch Analysis
The m-ODI data collected at admission were analyzed using the Rasch partial credit model (PCM) to account for the m-ODI’s polytomous rating scale structure in which each item has a unique rating scale structure. The Rasch model assumes the probability of endorsing an item (i.e., be able to completing a task) depends on the relationship between a person’s ability and an item’s difficulty. The item difficulty is similar to the concept of the percentage of patients who can complete a particular task successfully (or respond to an item correctly). The larger the percentage completing a task or getting an item right, the easier the item. Rasch analyses were performed with the Winsteps software [version 3.90] [30].

Rating Scale Structure
We examined the rating scale structure using Linacre’s three essential rating scale criteria: [31] (1) at least 10 observations should be in each rating scale category; (2) average measures (i.e., Rasch-Thurstone thresholds) should advance monotonically (i.e., the step calibrations become successively more positive as category number increases); and (3) outfit mean-squares should be less than 2.0. Because our sample size was relatively large, we increased the first criterion to be at least 300 observations should be in each rating scale category. Categories with low frequency counts or disordered rating scale structure may suggest that the operational definition of the rating scale category can be assigned to the respondent only in rare situations, with a narrowly defined scope, or was redundant when other response categories were present.

Item Difficulty Hierarchy
Based on how hard or easy patients perceived (i.e., have the functional ability or are able to complete the task), the Rasch mathematical model computes each item’s difficulty estimate and each item’s rating scale thresholds before establishing the empirical item difficulty hierarchy in logits (the logit of the probability is the logarithm of the odds). The empirical item hierarchies are supposed to reflect the underlying theoretical basis of the construct. To simplify data interpretation, item difficulty parameters and personal measures were expressed as higher scores (in logits) representing more difficult items or persons with higher (lumbar) functioning.

Item Fit
Fit statistics were performed to investigate whether the response patterns on the lumbar functioning fit the Rasch’s probabilistic model. A fit statistic index calculates the ratio of the observed variance divided by the expected variance. The index ranges from 0 to positive infinity with an ideal value of 1. Outfit is based on a sum of squared standardized residuals, the differences between the observations and their expected values according to the Rasch model. Infit is an information-weighted form of outfit and is more sensitive to item fit around the person’s ability level. Items with fit statistics greater than 1.4 were considered misfit items [32].

Person-Item Match / Targeting
In Rasch analysis, both person ability and item difficulty estimates are expressed on a common metric in logits. To determine whether the items are of appropriate difficulty for the sample, comparison of the personal measure distribution to the item difficulty parameter distribution were performed using a visual inspection of the two distributions. Ceiling and floor effects, which could reduce the chance to detect true change, were also examined.

Separation Index
The person separation index (G) is an estimate of how well the scale can differentiate persons into statistically distinct person strata [strata = (4*separation + 1)/3] [33]. Statistically distinct levels of item difficulty refers to difficulty strata with centers three calibration errors apart [33]. For example, a strata equal to 3 indicates that the scale can statistically distinguish patient lumbar functioning at least 3 (e.g., high, middle and low) levels.

Differential Item Functioning (DIF)
DIF occurs when groups have different probabilities of endorsing a given item on a polytomous rating scale after statistically controlling for overall personal abilities, and could become a source of bias in person measurement. In other words, DIF analysis can be used to examine whether the item difficulty hierarchy is similar across different groups after controlling for the ability levels of the compared groups. In this study, the DIF [34] was determined by calculating the mean difference in item difficulty of the item between the two groups, and then dividing the result by the pooled standard deviation.

Suggested by Linacre, there should be at least 0.5 logits for DIF to be noticeable. We performed DIF analysis on m-ODI items by gender, age group (18-44, 45-64, >=65 years old), and acuity (acute: <22 days, subacute: 22-90 days, chronic: >90 days). The DIF analysis is a series of pairwise t-tests, which equals item difficulty estimates contrast divided by the joint standard error. Since large sample size may result in trivial differences showing significant effect, we a priori determined that DIF items were those with: (1) statistically significant t-test (p-value =<0.01), and (2) difference in item difficulty estimates of 0.35 (logits) or greater.

Unidimensionality
To assess unidimentionality, we conducted exploratory factor analyses (EFA) of latent trait variables, followed by confirmatory factor analyses (CFA), utilizing Mplus (Muthe´n & Muthe´n, Los Angeles, CA) [35]. We reported: (1) the factor loadings, and (2) variances explained by each factor. As suggested by Nunnally [36], items with factor loadings above 0.40 in one factor are preferred. One-factor model fit was evaluated using the comparative fit index (CFI), the Tucker-Lewis index (TLI), and the root-mean-square error of approximation (RMSEA). The TLI and CFI range from 0 (poor fit) to 1 (good fit). Values of CFI and TLI greater than 0.90, and RMSEA values less than 0.08 are indicative of good model fit [37].

Functional Staging
Functional staging [38] refers to a set of hierarchical outcome levels that can be used to classify a patient into different stage levels and describe an individual’s expected functional status within each level. The functional staging was developed using the following steps: [23-26] (1) built a conceptual model based on clinical judgment, (2) determined the cut-scores between stages identified in step 1, and (3) specified expected performance at each level. As described [2,3], the m-ODI has an existing score interpretation that aggregates the index into 5 hierarchical functional levels with 4 cut-scores along the functional status continuum: (1) 0% to 20%: minimal disability, (2) 21% to 40%: moderate disability, (3) 41% to 60%: severe disability, (4) 61% to 80%: crippled, and (5) 81% to 100%: bed-bound. As a result, we primarily specified the expected performance by merging the probabilistic model in which each response category has a unique threshold for each item, and we aligned these thresholds to develop the functional staging.

Results

Rating Scale Structure
All rating categories of the m-ODI items had at least 300 observations, except three response categories: ‘5’ in personal care item (n = 90), ‘5’ in sitting item (n = 121), and ‘5’ in walking item (n = 100). No items had disordered rating scale thresholds, but 2 thresholds had outfit mean-squares greater than 2.0: (1) response category ‘5’ “I do not get dressed, wash with difficulty, and stay in bed” in personal care item, and (2) response category ‘5’ “pain prevents me from sleeping at all” in sleeping item. The category probability curves (Figure 1) show how probable the observation of each category (y-axis) is for measures relative to the item measure (x-axis). Results revealed that few response categories were less likely to be observed.

For example, response category ‘2’ in pain intensity, traveling, lifting, standing; ‘1’ in sitting; ‘4’ in employment/ homemaking, and narrow scope in middle response categories in the sleeping item. These response categories were suggested as being underused or having narrowly defined scope [e.g., for the employment/ homemaking item, if patients’ lumbar spine condition was worse than response category ‘3’ (pain prevents me from doing anything but light duties), patients tended to report ‘5’ (pain prevents me from performing any job or homemaking chores) rather than response category ‘4’ (pain prevents me from doing even light duties)].

Item Difficulty Hierarchy
Table 2 presents the estimated item statistics of the m-ODI items in difficulty order. Item ‘lifting’ (on the top) appeared to be the most difficult item, followed by ‘pain intensity’. Item ‘personal care” (at the bottom) was the easiest.

*Rasch-Thurstone thresholds (50% Cumulative Probability) is the thresholds where a person at the boundary between “1” and “2” would have a 50% chance of selecting a rating category of ‘1’ or below, and a 50% chance of selecting ‘2’ or above.

Item Fit
All items showed good infit and outfit statistics, except the pain intensity item which had a marginal misfit (outfit MNSQ = 1.53 > 1.40).

Person-Item Match/Targeting
Comparing to the mean (SD) of item difficulty estimates of 0.0 (0.49) logits, the patient ability level, on average, had a slightly higher mean (SD) of 0.9 (1.0) logits, suggesting that items were slightly easier for the patient sample. However, only 179 (0.6%) patients who obtained the maximum score (i.e., ceiling) and 0 (0%) patients who had the minimum score (i.e., floor), suggesting sufficient evidence for the lack of floor or ceiling effects.

Separation Index
With a separation index equaled to 2.1, the m-ODI items can differentiate patients into 3.1 statistically distinct person strata.

Differential Item Functioning (DIF)
DIF analysis supported that m-ODI items were free of DIF by gender and symptom acuity, but 3 items were suggestive of DIF by age group. Sitting was easier for patients >= 65 years old as compared to patients 18-44 years old (difference between item difficulty estimates was 0.71 logits). Standing, and walking was more challenging for patients >= 65 years old as compared to patients 18-44 years old (differences were 0.54 and 0.83 logits).

Unidimensionality
Results supported the unidimensionality of the m-ODI questionnaire. The factor loadings ranged from 0.80 (employment/homemaking) to 0.45 (pain intensity). The CFI was 0.88, the TLI was 0.95, and the RMSEA was 0.13, suggesting marginally good model fit. The first factor explained 48.6% of the total variance, followed by 10.7% variance explained by the second factor, and 8.3% by the third factor.

Percentile Rank Distribution
To assist in clinical interpretation, Table 3 presents the percentile rank distribution of the raw score and Figure 2 shows the (raw) score distributions of the m-ODI questionnaire at admission and at discharge. On average, the mean (SD) m-ODI at admission was 34.9 (17.0) and improved to 23.8 (18.0) at discharge.

Functional Staging
Figure 3 displays the functional staging of the m-ODI questionnaire and shows the expected response (horizontal bars) to a given item as a function of the underlying ability (i.e., functional disability) estimated by the m-ODI questionnaire. In this figure, the m-ODI items are listed in descending order of difficulty in the left column - more challenging items are listed on the top and the easier items are listed at the bottom. Beneath the figure is the m-ODI score continuum ranging from 0 to 100 (higher values represent greater disability) separated by different levels of functional staging ranged from level 1 (bed-bound) to level 5 (right or minimal disability). The cut-scores distinguishing between neighbored levels were 20%, 40%, 60%, and 80%.

Using the functional staging method, we can compare the patient’s m-ODI score to the functional stages and better interpret the patient’s score. The expected responses of each item at each functional level can be obtained by drawing a vertical line over an m-ODI score (x-axis) on the figure.

A Clinical Example
One patient record and her responses on m-ODI items was selected (not randomly) from the data set. The patient, Mrs. Jane Doe (female, age 44), came to the clinic due to subacute lumbar spine pathology. Her initial m-ODI score was 56 at admission, and was 18 at discharge. To visualize her responses, we plotted all her responses using the keyform in Figure 4. Note that the keyform is Rasch model based and the x-axis is the person measures estimated by the Rasch model in logits, not the m-ODI (raw) scores. In the functional staging (Figure 3), all the scoring are based on m-ODI scores.

At admission, the functional staging classified Mrs. Jane Doe as having ‘severe disability’ (level 3). By drawing a vertical line at 56 (on the x-axis), she was expected to report a ‘4’ (I can lift only very light weights) on lifting, ‘3’ on most items, ‘2’ on walking (pain prevents me from walking more than ½ mile) and ‘1’ on personal care item (I can take care of myself normally, but it increases my pain). The real responses of items 1 to 10 from Mrs. Jane Doe and the expected responses ‘()’ based on the staging were 3(3), 1(1), 4(4), 2(2), 3(3), 4(3), 2(3), 2(3), 4(3), and 3(3), respectively. Infit and outfit statistics for Mrs. Jane Doe were 0.41 and 0.39 (both <1.40). The real responses were very close to the expected responses.

At discharge, the functional staging classification suggested Mrs. Jane Doe improved to level 5 (minimal disability). The real responses of items 1 to 10 and the expected responses ‘()’ were 4(1), 0(0), 4(2), 0(0), 1(1), 0(1), 0(1), 0(1), 0(1), and 0(1), respectively. Infit and outfit statistics for Mrs. Jane Doe were 2.35 and 1.66 (both >1.40), indicating a mis-fit response pattern (e.g., item 1 pain intensity with observed ‘4’ vs expected ‘1’). Specifically, Mrs. Jane Doe had no difficulty in almost all daily activities, except that the pain medication provides her with little relief from pain, and the pain prevented her from lifting medium to heavy weights.

Discussion
As compared to prior studies [8,17-21], this paper examined the psychometric properties of the m-ODI questionnaire with a substantially larger sample size and specific for patients seeking outpatient physical therapy services. Overall, results showed that the m-ODI questionnaire has sufficient psychometric properties. The coverage of m-ODI items matched well with the patient functional abilities, and m-ODI items can differentiate persons into at least the statistically distinct person strata of: high, middle, and low.

Despite these different samples, several findings were similar. Results supported the unidimensional onefactor structure. Item ‘lifting’ appeared to be the most difficult item, and ‘sleeping’, ‘walking’ and ‘personal care” tended to be the easy items. Similar to two other studies [17,22] employment/homemaking turned out to be on the 3rd (our study) or 4th (Choi’s study [22], Davidson’s study [17]) difficult item. Interestingly, Davidson et al. [17] also showed walking item consistently exhibited uniform DIF by age (18-44, 45-64, 65+) across the original and modified ODI questionnaires.

Prior studies show great variations of item difficulty hierarchical order [8, 17-21] and similarly, our study suggests a different hierarchical order of items at the middle difficult level. In our study, we report person separation index of 2.10, which is higher than 0.88 (100 patients seeking physical therapy treatment) [17] and 1.77 (95 community-dwelling adults) [8], but lower than 2.83 (408 patients in an academic spine center) [19]. We did not observe the ceiling and floor effects as described by Brodke et al. [18] who reported ceiling effect of 3.9% and floor effect of 29.9%.

It has been reported that low back pain demonstrates a bimodal distribution with peaks between 25 and 29 years of age and after 65 years of age without differentiation by underlying etiology [1,39]. Some studies have shown no gender difference [1], while others have shown women to have significantly increased incidence rate for low back pain or degenerative spondylolisthesis compared with men [40,41]. In our sample, we found that the m-ODI had a unimodal distribution with one clear peak around 65 to 75 years of age, and women had a higher incidence rate with a ratio of approximate 1.46 (i.e., 17,039 vs 11,712) visiting outpatient physical therapy due to orthopedic lumbar spine impairments.

In confirmatory factor analysis of ODI items, Gabel et al. [9] verified a one-factor solution accounting for 54% of the total variance structure. In our study, the factor analysis showed similar results with the first factor explaining 49% of the total variance. During the exploratory factor analysis of two-factor solution, we found the CFI and TLI improved to 0.96 and 0.98, respectively, suggesting a better model fit. Seven out of 10 m-ODI items had factor loadings above 0.40 on the first factor. The remaining 3 items: standing and walking items had high factor loadings (0.67 and 0.86, respectively) on the second factor and lifting item had almost even loadings (0.32 on the first factor and 0.37 on the second factor). Such a finding suggested some practical concerns of pulling items with different aspects together to capture the functional disability. Nonetheless, the m-ODI items have been shown to be clinically useful.

After inspecting rating scale structure, some studies [8,20,21] decided to collapse categories (e.g., merge response categories 1 and 2, as well as 4 and 5). We decided not to collapse categories for couple of reasons. First, unlike other measures with Likert scale rating scale structure, each of m-ODI item has its own operational definition. Therefore, response categories need to be inspected by each item and merging the same response categories across all 10 items may obscure useful information. Second, Page et al. [8] found that the person separation and reliability did not improve after collapsing response categories. Third, it did not provide practical benefits as clinicians or researchers still use the original version of the questionnaire, rather than the revised questionnaire. Last, if indeed the response categories were revised, validation and duplication studies would be needed to support the practical use of the questionnaire.

There are several clinical application of the functional staging. For example, clinicians can set short-term or long-term goals by inspecting expected responses between (1) m-ODI score at admission and (2) m-ODI score at admission plus minimal detectable change points. Meanwhile, patients’ unexpected responses (e.g., observed response is deviated from the expected response) are easy to identify and these may be useful to help clinicians manage patients who may consider whether there is a logical reason why the client had an unexpected response.

There are several limitations of this study. Since this study was a secondary analysis of prospectively collected data via a proprietary database management company, the researchers were not in control of the data collection procedure. Missing values (e.g., symptom acuity and impairment codes) were inevitable as data were collected in routine outpatient clinics, and the database does not link to the electronic medical records. Specific impairment codes were missing for 37% patients. There were 199 (0.7%) patients with missing m-ODI scores at discharge. Selection biases associated with patients choosing a participating clinic are unknown and might affect the generalizability of results. Clinics, which subscribe services from FOTO, Inc., might be different from other clinics that did not participate. We reported the clinical interpretation of the m-ODI based on a sample of patients with varies etiologies. Similar procedures should be validated across different specific diagnostic groups within patients seeking outpatient rehabilitation therapy.

Conclusion
This study provided psychometric analysis of the m-ODI questionnaire and supported its use in outpatient rehabilitation settings. By deriving more clinically meaningful interpretations of outcomes measures, we hope that it may facilitate use of these measures by clinicians in their routine clinical practice.

Conflict of Interest Statement
The authors declare no conflict of interest.

Bibliography

1. Waterman, B. R., Belmont, P. J. Jr. & Schoenfeld, A. J. (2012). Low back pain in the United States: incidence and risk factors for presentation in the emergency setting. Spine J., 12(1), 63-70.
2. Fairbank, J. C., Couper, J., Davies, J. B. & O'Brien, J. P. (1980). The Oswestry low back pain disability questionnaire. Physiotherapy, 66(8), 271-273.
3. Fairbank, J. C. & Pynsent, P. B. (2000). The Oswestry Disability Index. Spine (Phila Pa 1976)., 25(22), 2940-2952.
4. Fritz, J. M. & Irrgang, J. J. (2001). A comparison of a modified Oswestry Low Back Pain Disability Questionnaire and the Quebec Back Pain Disability Scale. Phys Ther., 81(2), 776-788.
5. Vanti, C., Ferrari, S., Villafane, J. H., Berjano, P. & Monticone, M. (2017). Responsiveness and minimum important change of the Oswestry Disability Index in Italian subjects with symptomatic lumbar spondylolisthesis. J Orthop Traumatol., 18(2), 145-150.
6. Ko, S. & Chae, S. (2017). Correlations Between the SF-36, the Oswestry-Disability Index and Rolland-Morris Disability Questionnaire in Patients Undergoing Lumbar Decompression According to Types of Spine Origin Pain. Clin Spine Surg., 30(6), E804-E808.
7. Pasha, I. F., Qureshi, M. A., Farooq, M., Talha, M., Ahmed, N. & Ismail, J. (2015). Assessment with Oswestry disability index in surgically treated patients with lumbar spondylolisthesis: experience in 96 patients. J Pak Med Assoc., 65(11 Suppl 3), S166-S170.
8. Page, S. J., Shawaryn, M. A., Cernich, A. N. & Linacre, J. M. (2002). Scaling of the revised Oswestry low back pain questionnaire. Arch Phys Med Rehabil., 83(11), 1579-1584.
9. Gabel, C. P., Cuesta-Vargas, A., Qian, M., Vengust, R., Berlemann, U., Aghayev, E., et al. (2017). The Oswestry Disability Index, confirmatory factor analysis in a sample of 35,263 verifies a one factor structure but practicality issues remain. Eur Spine J., 26(8), 2007-2013.
10. Saltychev, M., Mattie, R., McCormick, Z., Barlund, E. & Laimi, K. (2017). Psychometric properties of the Oswestry Disability Index. Int J Rehabil Res., 40(3), 202-208.
11. Yao, M., Wang, Q., Li, Z., Yang, L., Huang, P. X., Sun, Y. L., et al. (2016). A Systematic Review of Cross-cultural Adaptation of the Oswestry Disability Index. Spine (Phila Pa 1976), 41(24), E1470-E1478.
12. Shah, S. & Balaganapathy, M. (2017). Reliability and validity study of the Gujarati version of the Oswestry Disability Index 2.1a. J Back Musculoskelet Rehabil., 30(5), 1103-1109.
13. Joshi, V. D., Raiturker, P. P. & Kulkarni, A. A. (2013). Validity and reliability of English and Marathi Oswestry Disability Index (version 2.1a) in Indian population. Spine (Phila Pa 1976), 38(11), E662-E668.
14. Yu, E. M., Nosova, E. V., Falkenstein, Y., Prasad, P., Leasure, J. M. & Kondrashov, D. G. (2016). Validation of a Russian Language Oswestry Disability Index Questionnaire. Global Spine J., 6(7), 636-639.
15. Aiyegbusi, A. I., Akodu, A. K. & Agbede, E. O. (2017). Reliability and validity of the Yoruba version of the Oswestry disability index. Niger Postgrad Med J., 24(2), 103-106.
16. Lee, C. P., Fu, T. S., Liu, C. Y. & Hung, C. I. (2017). Psychometric evaluation of the Oswestry Disability Index in patients with chronic low back pain: factor and Mokken analyses. Health Qual Life Outcomes, 15(1), 192.
17. Davidson, M. (2008). Rasch analysis of three versions of the Oswestry Disability Questionnaire. Man Ther., 13(3), 222-231.
18. Brodke, D. S., Goz, V., Lawrence, B. D., Spiker, W. R., Neese, A. & Hung, M. (2017). Oswestry Disability Index: a psychometric analysis with 1,610 patients. Spine J., 17(3), 321-327.
19. Lu, Y. M., Wu, Y. Y., Hsieh, C. L., Lin, C. L., Hwang, S. L., Cheng, K. I., et al. (2013). Measurement precision of the disability for back pain scale-by applying Rasch analysis. Health Qual Life Outcomes, 11, 119.
20. White, L. J. & Velozo, C. A. (2002). The use of Rasch measurement to improve the Oswestry classification scheme. Arch Phys Med Rehabil., 83(6), 822-831.
21. Lochhead, L. E. & MacMillan, P. D. (2013). Psychometric properties of the Oswestry disability index: Rasch analysis of responses in a work-disabled population. Work, 46(1), 67-76.
22. Choi, B. S. (2014). Keyform analysis of Rasch measurement accessible to clinicians in rehabilitation clinics. Phys Ther Korea, 21(2), 74-81.
23. Wang, Y. C., Hart, D. L., Stratford, P. W. & Mioduski, J. E. (2009). Clinical interpretation of computerized adaptive test generated outcomes measures in patients with foot/ankle impairments. J Orthop Sport Phys., 39(10), 753-764.
24. Wang, Y. C., Hart, D. L., Stratford, P. W. & Mioduski, J. E. (2009). Clinical interpretation of computerized adaptive test-generated outcome measures in patients with knee impairments. Arch Phys Med Rehabil., 90(8), 1340-1348.
25. Wang, Y. C., Hart, D. L., Stratford, P. W. & Mioduski, J. E. (2009). Clinical interpretation of a lower-extremity functional scale-derived computerized adaptive test. Phys Ther., 89(9), 957-968.
26. Wang, Y. C., Hart, D. L., Werneke, M., Stratford, P. W. & Mioduski, J. E. (2010). Clinical interpretation of outcome measures generated from a lumbar computerized adaptive test. Phys Ther., 90(9), 1323-1335.
27. Aithala, J. P. (2015). Difficulties in using Oswestry Disability Index in Indian patients and validity and reliability of translator-assisted Oswestry Disability Index. J Orthop Surg Res., 10, 90.
28. Maataoui, A., Vogl, T. J., Middendorp, M., Kafchitsas, K. & Khan, M. F. (2014). Association between facet joint osteoarthritis and the Oswestry Disability Index. World J Radiol., 6(11), 881-885.
29. Shi, J., Wang, Y., Zhou, F., Zhang, H. & Yang, H. (2012). Long-term clinical outcomes in patients undergoing lumbar discectomy by fenestration. The Journal of international medical research, 40(6), 2355-2361.
30. Linacre, J. M. (2009). Winsteps (Version 3.90) [Computer Software]. Chicago: Winsteps.com.
31. Linacre, J. M. (2003). Size vs. Significance: Standardized Chi-Square Fit Statistic. Rasch Measurement Transactions, 17(1), 918.
32. Linacre, J. M. (2002). What do Infit and Outfit, Mean-square and Standardized mean? Rasch Measurement Transactions, 16(2), 878.
33. Wright, B. D. & Masters, G. N. (2002). Number of person or item strata. Rasch Measurement Transactions, 16(3), 888.
34. Linacre, J. M. (2017). Winsteps® Rasch measurement computer program User's Guide. Beaverton, Oregon: Winsteps.com.
35. Muthén, L. K. & Muthén, B. O. (2004). Mplus User’s Guide. Los Angeles: Muthén & Muthén.
36. Nunnally, J. (1978). Psychometric theory. New York: McGraw-Hill.
37. Hu, L. T. & Bentler, P. (1999). Cutoff criteria for fit indices in covariance structure analysis: conventional criteria versus new alternatives. Struct Equation Model, 6(1), 1-55.
38. Jette, A. M., Tao, W., Norweg, A. & Haley, S. (2007). Interpreting rehabilitation outcome measurements. J Rehabil Med., 39(8), 585-590.
39. Domino, F. J., Baldor, R. A., Golding, J. & Grimes, J. A. (2014). The 5-Minute Clinical Consult Premium 2015. Philadelphia, PA: Lippincott Williams & Wilkins.
40. Knox, J., Orchowski, J., Scher, D. L., Owens, B. D., Burks, R. & Belmont, P. J. (2011). The incidence of low back pain in active duty United States military service members. Spine (Phila Pa 1976), 36(18), 1492-1500.
41. Vining, R. D., Potocki, E., McLean, I., Seidman, M., Morgenthal, A. P., Boysen, J., et al. (2014). Prevalence of radiographic findings in individuals with chronic low back pain screened for a randomized controlled trial: secondary analysis and clinical implications. Journal of manipulative and physiological therapeutics, 37(9), 678-687.