Rationalization of outcome scores for low back pain: the Oswestry disability index and the low back outcome score

Transcription

1 ORTHOPAEDICS ANZJSurg.com Rationalization of outcome scores for low back pain: the Oswestry disability index and the low back outcome score Vivek Eranki,* Kongposh Koul* and Andrew Fagan *Royal Perth Hospital, Perth, Western Australia, Australia and Department of Orthopaedic Surgery and Trauma, The Queen Elizabeth Hospital, Adelaide, South Australia, Australia Key words low back outcome score, low back pain, low back pain questionnaire, Oswestry disability index. Correspondence Dr Vivek Eranki, Royal Perth Hospital, PO Box 218, Perth, WA 69, Australia. V. Eranki MBBS; K. Koul MBBS; A. Fagan MBBS, FRCS (Glas.), FRACS (Orth.). Accepted for publication 28 November 12. doi:.1111/ans.143 Abstract Background: The two commonly used questionnaires to assess low back pain are the low back outcome score (LBOS) and the Oswestry disability index (ODI). This study aims to identify unique questions and remove redundant questions to develop a composite questionnaire. Methods: Eighty-seven consecutive patients attending the practice of a single spinal surgeon completed both the ODI and the LBOS as part of their initial assessment. Both questionnaires were analysed to eliminate questions that exhibit floor ceiling bias and questions that are interdependent and correlate strongly. Total scores and the scores obtained for each question were then compared (Spearman s rho). A principal axis factor analysis using a varimax rotation was performed to reduce data and identify questions that were interdependent. Using these data, a composite questionnaire was proposed that would minimize overlap in clinical data. Results: Eighty-seven patients completed the LBOS and ODI. The mean age is 4, with a range between 18 and 8. The male to female ratio was :37. By eliminating questions that contain biases and overlap in clinical data, the composite questionnaire contains 11 questions. From LBOS; housework, dressing, sleeping, sitting, walking and travelling. From the ODI; pain, standing, social life and lifting. Conclusion: Analysis of the questionnaires identified eight questions that were similar in both questionnaires. Two questions were included that were unique to each questionnaire. The proposed composite questionnaire is of similar size as the original questionnaires and retains questions that are unique to each other while eliminating questions that are redundant and exhibit bias. Introduction Despite the publication of consensus documents, a variety of disease-specific questionnaires continue to be used by researchers to measure disability from low back pain (LBP). Reasons for the lack of consensus over the best instrument to use include the different questions used for each instrument, differing ease of administration and differing degrees of validation of measurement properties. Consensus over the use of a single instrument is desirable as it would permit meaningful comparisons of results of treatment across different populations. However, using a single instrument results in the loss of data that would be gained from completion of two separate questionnaires. This study examines the degree of overlap in data acquired by the Oswestry disability index (ODI) and the low back outcome score (LBOS) with the aim of proposing a single composite questionnaire that retains the strengths of both questionnaires but eliminates redundancy. The ODI 1,2 consists of questions. The questions are centred on the amount of pain experienced with different activities. Version 2. was used for this study. The LBOS consists of 13 questions. It was developed in Adelaide in and includes questions about objective measures of loss of function (such as the amount of rest needed each day and the number of times a doctor is consulted) as well as symptoms. The average time taken to complete each of the questionnaires is min, but this increases with age. 4 The correlation coefficients between the two questionnaire total scores has been reported as.6 and.87, 6 respectively. The test/retest reliability for the ODI has been reported to be 83% 7 and that for the LBOS is 84%. 6 The internal reliability of both the ANZ Journal of Surgery 13 Royal Australasian College of Surgeons ANZ J Surg 83 (13)

2 872 Eranki et al. Table 1 Spearman s correlations coefficients between all the questions in the low back outcome score (LBOS) and Oswestry disability index (ODI) ODI Pain Personal care Lifting Walking Sitting LBOS Pain Work Housework Sports Rest Health services Medication Sex life Sleeping Sitting Walking Travelling Dressing Standing Sleeping Sex life Social life Travelling Pain Work Housework Sports Rest Health services Medication Sex life Sleeping Sitting Walking Travelling Dressing Similar questions are outlined in bold. Correlations.6 are highlighted in green and.4 are highlighted in red. The correlation of the total score between LBOS and ODI is.8. scores is comparable and good. Version 2. of the ODI has been shown to have a Cronbach s alpha between.76 and For LBOS, Holt et al. reported a Cronbach s alpha between.79 and.8. 6 Greenough and Fraser found the LBOS to be more discriminating by providing an increased spread of results compared with ODI. 3 Materials and methods Eighty-seven consecutive patients attending the practice of a single spinal surgeon (ABF) completed both the ODI and the LBOS as part of their initial assessment. Questions were analysed for bias, similarity and correlation to eliminate redundant questions. The scores obtained for each question were compared using Spearman s rho correlation after correcting for the opposite polarity between ODI and LBOS. It was hypothesized that questions measuring similar variables would obtain a correlation coefficient of >.6, while questions measuring dissimilar variables would obtain a correlation coefficient of <.4. 9 The relationship between similar and dissimilar questions is displayed in a matrix (Tables 1,2). Multivariate factor analysis with varimax rotation was used to identify questions that were answered similarly. All factors with eigenvalues of >1 were retained. For each factor, questions with a loading of >.4 were identified. Questions loading onto the same factor were thought of as being similar as they shared common variance. Question selection was based on three steps: (1) Elimination of questions that exhibit floor ceiling effect and response bias (2) Retention of questions that are unique (3) Eliminate questions that duplicate each other (ask about similar sensations or activities) The first step was to exclude all questions exhibiting floor ceiling effect or response. Floor ceiling effect was defined where % or more of the responses were biased towards the lowest or highest response, respectively. Response bias was defined where the response rate for a question was less than 9%. The next step was to retain all unique questions that did not exhibit show bias. Unique questions were selected using two steps: (1) Questions that are unique to each questionnaire. These questions have been identified by visual inspection of the question and answer choices (Table 3). (2) Unique questions that load onto the same factor (Table 4). Questions within the same factor correlating poorly with their counterpart. Questions in a factor with no similar questions. The final step is to exclude questions that duplicate clinical data. This is carried out with the aid of Spearman s rho correlation, factor analysis and analysis of the spread of answer options. Two questions are thought of as duplicating each other when they correlate strongly and load onto the same factor (Table 4, highlighted in green). In the absence of bias, the question with the most central distribution of responses (i.e. exhibiting the greatest responsiveness to change) was selected. The histogram of the distribution of responses of both ANZ Journal of Surgery 13 Royal Australasian College of Surgeons

3 The Oswestry disability index and the low back outcome score 873 Table 2 Corresponding questions from both the Oswestry disability index (ODI) and the low back outcome score (LBOS) Expectation LBOS ODI Correlation Good correlation (similar questions) Pain VAS Pain intensity.6 Sex life Sex life.9 Sleeping Sleeping.8 Sitting Sitting.6 Walking Standing.6 Travelling Travelling.6 Good correlation (dissimilar questions) Housework Personal care.6 Dressing Personal care.7 Good correlation (unexpected: dissimilar questions) Rest Social life.6 Poor correlation (unexpected: similar questions) Sports Social life.3 Rest Sleeping.4 Medication use Pain intensity.4 Sex life Sleeping.4 These questions were selected on visual inspection of the question and answer options. They were all expected to correlate strongly due to this similarity. Questions highlighted in green display a strong correlation of greater than.6 (.6). Questions highlighted in red display a correlation of less than.4 (.4). VAS, visual analogue scale. Table 3 There are three questions unique to the Oswestry disability index (ODI) and six unique to the low back outcome score (LBOS) questions was qualitatively assessed and the question displaying the most central distribution of responses was selected (Figs 1 3). If both questions display central distributions, the question with between 6 and 8 Likert items is selected. Results LBOS ODI Sensation Pain Pain Activities Dressing Personal care Lifting Walking Walking Sitting Sitting Standing Sleeping Sleeping Sex life Sex life Travelling Travelling Rest Social life Work Housework Sports Health services Pain medication All unique questions are highlighted in yellow. The mean age of all patients was 4 (range 18 8). The male to female ratio was :37. Fifty-eight (67%) patients presented with LBP without radiation and 29 (33%) presented with low back and leg pain. Nine questions with floor ceiling effect and response bias are displayed in Table (highlighted in red). After the exclusion of these questions, we are left with 14 questions. Floor effect was evident in ODI personal care and walking and LBOS sex life and dressing. Ceiling effect was evident in LBOS pain, sport, health services and medication. Response bias was evident in both the ODI and the LBOS sex life questions. The correlation of the total ODI and LBOS score was.8. Spearman s rho univariate correlations between responses to all the questions are displayed in Table 1. On the whole, there was a good correlation between similar questions (Table 2). The strongest correlation was observed between the two questions about sex life (.9) and sleeping (.8), and the ODI question about personal care and the LBOS question about dressing (.7). There was a good correlation between dissimilar questions that addressed similar activities (the ODI question on social life and the LBOS question on housework). Four pairs of questions we expected to behave similarly had a poor correlation (Table 2). Four main factors were identified by multivariate factor analysis. The first contains questions that are mainly about activity. These are ODI personal care, ODI walking, ODI sex life, ODI social life, LBOS work, LBOS housework, LBOS sport, LBOS rest, LBOS walking and LBOS sex life. The second contains questions that are mainly about pain and activities of daily living. These include OFI pain, ODI sleeping, LBOS pain, LBOS medication, LBOS sleeping and LBOS dressing. The third contains questions that are mainly about posture and includes ODI sitting, ODI standing, ODI travelling, LBOS sitting and LBOS travelling. The fourth contains two unrelated questions (lifting and the use of health services). The first step of selecting questions for the composite questionnaire is to exclude questions exhibiting floor ceiling effect and response bias. These questions are highlighted in red in Table. The next step is to include questions that are unique. There are several questions that are unique to each questionnaire. Unique questions were identified by visual inspection of the question and the answer options. ODI lifting, standing and social life and LBOS rest, working, housework, sport, health services and pain medication have been found to be unique (Table 3). ODI pain and LBOS walking can be considered unique and included into the composite questionnaire as their counterparts were excluded due to bias. Questions unique within each factor are unique and hence are included into the composite questionnaire. These are the questions highlighted in yellow in Table 4. The factor analysis reduced the questions to four principle factors. Within these factors, two questions can be considered similar if they load within the same factor ANZ Journal of Surgery 13 Royal Australasian College of Surgeons

4 874 Eranki et al. Table 4 Factor analysis Activity Pain Posture Miscellaneous ODI personal care, LBOS housework ODI pain, LBOS pain ODI sitting, LBOS sitting ODI lifting ODI walking, LBOS walking ODI sleeping, LBOS sleeping ODI travelling, LBOS travelling LBOS health services ODI sex life, LBOS sex life ODI pain, LBOS medication ODI standing ODI social life, LBOS sport LBOS dressing LBOS rest LBOS work The different factors are labelled in the first row. Similar questions loading to the same factor are arranged in pairs underneath each label. Pairs with a correlation.6 are highlighted in green. Pairs with a correlation.4 are highlighted in red. LBOS, low back outcome score; ODI, Oswestry disability index. ODI si ng 4 ODI sleeping LBOS si ng 4 LBOS sleeping Fig. 1. Comparison of frequency plot of the answers to the Oswestry disability index (ODI) and low back outcome score (LBOS) questions on sitting. Both distributions have a central peak. The ODI question is closer to the ideal number of Likert items and hence preferable. and demonstrate a strong correlation. Questions were coupled within the same factor based on assessment of clinical data. In Table 4, the pairs of questions highlighted in green demonstrate strong correlation. The pairs highlighted in red have a poor correlation and questions highlighted yellow are unique in each factor. The pairs of questions in red can be considered unique and are included into the composite questionnaire with the exception of LBOS sport and medication, which have been eliminated due to bias. The questions highlighted in yellow are unique and are included into the composite; however, LBOS dressing and the health services have been excluded already due to bias. Questions highlighted green in Table 4 display a correlation of >.6 within the questions loading into the same factor. Several Fig. 2. Comparison of frequency plot of the answers to the Oswestry disability index (ODI) and low back outcome score (LBOS) questions on sleeping. The LBOS sleeping question is preferred as it has a more central peak. questions were already excluded due to bias. Out of the questions highlighted in green loading into the factor assessing activity, LBOS housework and walking. From the questions assessing pain, ODI pain can be selected due to its counterpart being excluded from bias. There are three pairs of questions that demonstrate strong correlations with neither question in the pair demonstrating bias. These are ODI and LBOS sleeping, sitting and travelling. LBOS sleeping and travelling were found to have the most central distribution (Figs 2,3). They were therefore retained for the composite questionnaire. Both ODI and LBOS sitting questions have a central distribution (Fig. 1). In this case, ODI sitting was selected because it has ANZ Journal of Surgery 13 Royal Australasian College of Surgeons

5 The Oswestry disability index and the low back outcome score closer to an ideal number of Likert items and hence preferable. Arguments exist regarding the ideal number of Likert items to maximize sensitivity. LBOS work and housework have a strong correlation and load onto the same factor, suggesting that the two questions are answered similarly. These two questions can therefore be combined into one. The composite questionnaire therefore contains questions. From the ODI, the questions on pain, lifting, sitting, standing and social life were included. From the LBOS, questions on work/ housework, rest, sleeping, walking and travelling were included (Appendix I). Discussion ODI travelling LBOS travelling Fig. 3. Comparison of frequency plot of the answers to the Oswestry disability index (ODI) and low back outcome score (LBOS) questions on travelling. LBOS question on travelling has a more central peak and therefore preferable. This study includes 87 patients with a mean age of 4 who were asked to complete both the ODI and the LBOS questionnaires as a part of their initial assessment. The data from the questionnaires were analysed to propose a composite questionnaire that minimizes bias, yet remains easy to administer. We eliminated 13 of the 23 questions used in the ODI and the LBOS, gauged redundancy on the basis of bias and duplicating clinical data. This left us with questions suitable for use. Elimination of questions exhibiting bias intends to improve the accuracy of clinical data gained. Most notable floor ceiling effect was evident in questions in LBOS. The greater presence of floor ceiling effect in LBOS could be attributed to the number of answer options available. ODI has a 6-point Likert scale, while LBOS employs a 4-point with an 11-point visual analogue scale (VAS) for pain. Application of scales with a larger number of Likert items has been found to minimize floor ceiling bias. Several studies, however, consistently show a lack of benefit in employing scales with greater than 7 Likert items. Questions such as the LBOS VAS pain question employing a -point Likert scale would therefore carry no advantage to the ODI pain question, which has been found to be clear of bias Based on this, we preferentially selected questions with responses between 6 and 8 Likert items (all other factors being equal) to minimize floor ceiling bias and maximize accuracy. Questions exhibiting floor ceiling bias are also more likely to exhibit response bias. LBOS sex life question, for example, had a 21.7% floor effect and a response bias. The selective participation of patients to respond to the LBOS sex life question reflects patients either being unable to participate secondary to LBP or deeming the question irrelevant or embarrassing. 14,1 In this case, if greater number of patients answered the question, floor effect would be significantly higher. This relationship between floor ceiling effect and response bias was also evident in Mannion et al. 16 where they found the selective participation of patients that were highly satisfied with their treatment to increase floor ceiling effects. Factor analysis with Spearman s rho was used to identify similar question. Questions loading within the same factor either demonstrating poor correlation with its counterpart or having no similar questions can be considered unique. Two sets of questions were identified which appear to have similar clinical parameters and comparable answer options yet displayed poor correlation. These are ODI pain/lbos pain medication and ODI social life/lbos sport. In both cases, the unique questions not displaying bias were included. Poor correlation between these questions can be attributed to the differences in the patient s frame of reference when answering these questions. An example of this is the relationship between ODI pain and LOBS pain medication questions. The pain intensity/medication relationship should in theory have a high correlation. LBOS pain medication question offers four options being: (i) never; (ii) occasionally; (iii) almost every day; and (iv) several times each day. When pain and pain medication are compared, matching the questions is not as straightforward because perception of pain is subjective and the need for analgesia is determined by factors other than the level of pain such as access to medication, fear of side effects and tolerance to analgesia. Compared to questions that correlate strongly, no clear pattern between the intensity of pain and frequency of pain medication could be established. Out of similar questions with no bias, the question with the most ideal distribution was selected. Currently, there is no agreement on the ideal distribution of responses for questionnaires. It could be argued that an equal distribution of responses across all answer options could be ideal, but this may better approximate a healthy population, especially in questions assessing function. It could be argued that a centrally distributed curve of responses will result in greater responsiveness to change and minimize floor ceiling effect, ANZ Journal of Surgery 13 Royal Australasian College of Surgeons

6 876 Eranki et al. Table Floor and ceiling effect and response bias Questions Floor effect (%) Ceiling effect (%) Response rate (%) ODI pain ODI personal care ODI lifting ODI walking ODI sitting ODI standing.8 9 ODI sleeping ODI sex life ODI social life ODI travelling LBOS pain LBOS work LBOS housework LBOS sport LBOS rest LBOS health services LBOS medication LBOS sex life LBOS sleeping LBOS sitting LBOS walking LBOS travelling LBOS dressing Items highlighted in red exhibited bias. LBOS, low back outcome score; ODI, Oswestry disability index. which is a desirable characteristic for an outcome assessment of individuals with LBP. A centrally distributed pattern of responses has the advantage of measuring changes to pain and function more effectively compared to an even distribution. To maintain statistical integrity, the wording of the questions remained unaltered. The structure of the composite questionnaire thus runs the risk of reading inconsistently. As expected, the questions have either a 4- or a 6-point Likert scale. Seven of the questions have a single word as their question followed by response options, while three questions, focusing on function from the LBOS series of questions, provide a statement as a question followed by the answer options. The question on work/housework, walking and pain could easily be reduced to the key word, but alteration of the question format compared to the original could run the risk of affecting the statistical integrity of the question itself and the one preceding it. 17 Further work is required to standardize the wording. This needs to be preceded by an analysis of the effect of the rewording on the performance of the question. The orientation of the scoring was altered to remain consistent with the LBOS and reward function rather than disability. The total score of the composite questionnaire sums to, which provides the benefit of easy scaling to if desired, a feature of the ODI that has been proven popular. A total score of can be achieved by altering the work/housework question to have a total score of 7. The effect of changing the weighing of one or other question on the performance of the questionnaire as a whole would need further testing for responsiveness and validity before the final selection is made. Conclusion By eliminating questions showing bias and choosing questions that are statistically unique to each questionnaire, we were able to reduce the 24 questions found in two popular outcome scores for LBP to one questionnaire containing questions. We recommend a similar approach be pursued in the ongoing attempt to rationalize the use of outcome instruments for LBP. Acknowledgement The authors would like to thank Mr Aditya Eranki for the assistance with data analysis and writing of the manuscript. Thanks are also due to the Faculty of Health Science, University of Western Australia, Western Australia (72831@student.uwa.edu.au). References 1. Fairbank JC, Couper J, Davies JB, O Brien JP. The Oswestry low back pain disability questionnaire. Physiotherapy 198; 66: Fairbank J. To the Editor. Spine 199; : Greenough CG, Fraser RD. Assessment of outcome in patients with low-back pain. Spine 1992; 17: Dixon S, Bunker T, Chang D. Outcome scores collected by touchscreen: medical audit as it should be in the 21st century? Ann. R. Coll. Surg. Engl. 7; 89: Taylor SJ, Tylor AE, Foy MA, Fogg AG. Responsiveness of common outcome measures for patients with low back pain. Spine 1999; 24: Holt AE, Shaw NJ, Shetty A, Greenough CG. The reliability of the low back outcome score for back pain. Spine 2; 27: Grönblad M, Hupli M, Wennerstrand P et al. Intercorrelation and testretest reliability of the pain disability index (PDI) and the Oswestry disability questionnaire (ODQ) and their correlation with pain intensity in low back pain patients. Clin. J. Pain 1993; 9: Fisher K, Johnston M. Validation of the Oswestry Low back pain disability questionnaire, its sensitivity as a measure of change following treatment and its relationship with other aspects of the chronic pain experience. Physiother. Theory Pract. 1997; 13: ANZ Journal of Surgery 13 Royal Australasian College of Surgeons

7 The Oswestry disability index and the low back outcome score Ferrer M, Pellisé F, Escudero O et al. Validation of a minimum outcome core set in the evaluation of patients with back pain. Spine 6; 31: Cummins RA, Gullone E. Why we should not use -point Likert scales: the case for subjective quality of life measurement. Proceedings of the Second International Conference on Quality of Life in Cities National University of Singapore, Singapore, ; Jacoby J, Matell MS. Three-point Likert scales are good enough. J. Mark. Res. 1971; 8: Wyatt RC, Meyers LS. Psychometric properties of four -point Likerttype response scales. Educ. Psychol. Meas. 1987; 47: Dixon PN, Bobo M, Stevick R. Response differences and preferences for all-category-defined and end-defined Likert formats. Educ. Psychol. Meas. 1984; 44: Rodgers JL, Billy JO, Udry JR. The rescission of behaviors: inconsistent responses in adolescent sexuality data. Soc. Sci. Res. 1982; 11: Ford K, Norris A. Methodological considerations for survey research on sexual behavior: urban African American and Hispanic youth. J. Sex Res. 1991; 28: Mannion AF, Elfering A, Staerkle R et al. Outcome assessment in low back pain: how low can you go? Eur. Spine J. ; 14: Choi BCK, Pack AWP, Purdham JT. Effects of mailing strategies on response rate, response time, and cost in a questionnaire study among nurses. Epidemiology 199; 1: Appendix I Composite questionnaire (1) Pain intensity (a) I have no pain at the moment 4 (b) The pain is very mild at the moment 3 (c) The pain is moderate at the moment 2 (d) The pain is fairly severe at the moment 1 (e) The pain is very severe at the moment (f) The pain is the worst imaginable at the moment (2) At present are you able to do your regular work/household chores 9 (a) Normally 6 (b) As much as usual but more slowly 3 (c) A bit, not as much as usual (d) Not at all (3) Social life (a) My social life is normal and gives me no extra pain 4 (b) My social life is normal but increases the degree of pain 3 (c) Pain has no significant effect on my social life apart from limiting my more energetic interests, e.g. sport 2 (d) Pain has restricted my social life and I do not go out as often 1 (e) Pain has restricted my social life to my home (f) I have no social life because of pain (4) Do you have to rest during the day because of pain 7 (a) Not at all (b) A little 3 (c) Half the day (d) Over half the day () Sleeping 3 (a) Not at all 2 (b) Mildly/not much 1 (c) Moderately/Difficult (d) Severely/Impossible. (6) Standing (a) I can stand as long as I want without extra pain 4 (b) I can stand as long as I want but it gives me extra pain 3 (c) Pain prevents me from standing for more than 1 h 2 (d) Pain prevents me from standing for more than min 1 (e) Pain prevents me from standing for more than min (f) Pain prevents me from standing at all (7) Sitting (a) Pain prevents me from sitting at all 4 (b) Pain prevents me from sitting more than min 3 (c) Pain prevents me from sitting more than min 2 (d) Pain prevents me from sitting more than 1 h 1 (e) I can only sit in my favourite chair as long as I like (f) I can sit in any chair as long as I like (8) Lifting (a) I can lift heavy weights without extra pain 4 (b) I can lift heavy weights but it gives me extra pain 3 (c) Pain prevents me lifting heavy weights off the floor but I can manage if they are conveniently placed e.g. on a table 2 (d) Pain prevents me lifting heavy weights but I can manage light to medium weights if they are conveniently positioned 1 (e) I can only lift very light weights (f) I cannot lift or carry anything (9) How much does your back pain affect your ability to walk? 3 (a) Not at all 2 (b) Mildly/not much 1 (c) Moderately/Difficult (d) Severely/Impossible. () Travelling 3 (a) Not at all 2 (b) Mildly/not much 1 (c) Moderately/Difficult (d) Severely/Impossible. TOTAL SCORE / ANZ Journal of Surgery 13 Royal Australasian College of Surgeons