Rating Scale Analysis of the Berg Balance Scale

Transcription

1 1128 Rating Scale Analysis of the Berg Balance Scale Diana L. Kornetti, MA, PT, Stacy L. Fritz, MSPT, Yi-Po Chiu, MHS, PT, Kathye E. Light, PhD, PT, Craig A. Velozo, PhD, OTR ABSTRACT. Kornetti DL, Fritz SL, Chiu Y-P, Light KE, Velozo CA. Rating scale analysis of the Berg Balance Scale. Arch Phys Med Rehabil 2004;85: Objectives: To examine, using Rasch analysis, the rating scale performance of the Berg Balance Scale (BBS) and to describe the 45/56 cutoff score in functional terms. Design: Retrospective chart review of BBS scores. Rasch rating scale analysis was performed on these data. Setting: Outpatient Veterans Affairs medical center. Participants: One hundred (99 men, 1 woman) communitydwelling veterans referred for balance deficits (age range, 64 88y). Interventions: Not applicable. Main Outcome Measure: The BBS. Results: Condensing item-rating categories allowed the elimination of underutilized categories and constructed categories that better separated people of differing abilities. Rating pivot points were developed for each item to represent a transition between passing and failing. Following pivot-point development and rating scale rescoring, person and item measures became more evenly distributed across the BBS and resulted in changes in item difficulty order. In our sample, functional indicators of a score of at least 45/56 were a rating of passing the item tandem stance, as well as passing 2 of the following 3 items: alternating foot, standing on one leg, and look behind. Conclusions: Our findings provide direction for improving the rating scale structure for each of the items and establish a connection between the BBS cutoff score of 45/56 and functional ability. Key Words: Balance; Rehabilitation by the American Congress of Rehabilitation Medicine and the American Academy of Physical Medicine and Rehabilitation WITH THE AGING of the population, the incidence of falls and fall-related injuries is expected to rise. Approximately 33% of people over the age of 65 fall at least once a year, as do approximately 50% of people over the age of 85. 1,2 In addition, falls continue to be the leading cause of injury in From the Departments of Physical Therapy (Kornetti, Fritz, Chiu, Light) and Occupational Therapy (Velozo), College of Public Health and Health Professions, University of Florida, Gainesville, FL; Office of Research and Development, Brain Rehabilitation Research Center (Fritz, Light), and Rehabilitation Outcomes Research Center (Velozo), Department of Veterans Affairs Medical Center, Gainesville, FL. Presented in part at the American Physical Therapy Association Combined Sections Meeting, February 15, 2003, in Tampa, FL. Supported in part by the Foundation for Physical Therapy and by the Office of Research and Development, Rehabilitation R&D Service, Department of Veterans Affairs, Gainesville, FL. No commercial party having a direct financial interest in the results of the research supporting this article has or will confer a benefit upon the author(s) or upon any organization with which the author(s) is/are associated. Reprint request to Diana L. Kornetti, MA, PT, Dept of Physical Therapy, College of Public Health and Health Professions, University of Florida, Box , UFHSC, Gainesville, FL 32610, dkornett@ufl.edu /04/ $30.00/0 doi: /j.apmr/ the elderly. Twenty to 60% of people who fall lose their independence. 2 Falls are a threat to quality of life and independence that is secondary to impaired mobility and loss of function. Emphasis on early identification of prevention for and intervention for elderly individuals who are at risk for falls is becoming increasingly important in the fields of physical therapy (PT) and rehabilitation. 3-9 There are 2 general categories of tools that are commonly used to assess balance. First, computerized instruments such as balance and force-perturbation platforms are frequently used in laboratory and research settings. 3,10-12 Although these devices measure a number of specific parameters that correlate with balance, their use in a clinic is limited because of their expense, time demands, and the technical expertise required to use them. Second, clinical standardized assessments are widely used in both research and clinical settings. A plethora of clinical instruments designed to assess balance have been developed over the last 20 years. 3-6,9,10,13,14 In general, these tests have been readily adopted because of their simplicity and low cost. Based on extensive clinical use and frequency of comparison with other balance measures in both clinic and laboratory and research settings, the Berg Balance Scale (BBS) is accepted as the clinical criterion standard by which to measure balance. Currently, more than 100 articles have cited the BBS since ,15-29 The BBS was specifically designed to assess balance ability of the elderly, to monitor changes in balance over time, to screen patients for rehabilitation therapy services, and to predict falls in community-dwelling and institutionalized older adults. 3 Populations for which this scale is appropriate include community-dwelling elderly, 9 institutionalized elderly, 3,4,14 urban-dwelling elderly, 5 and patients with stroke. 10,28,30-33 There is extensive literature supporting its psychometric properties. 2,3 Furthermore, the acceptance of the BBS is evidenced by its frequent use as a validation standard for other balance measures, 16,19,20,21,27,29 including the Timed Up & Go (TUG) test, 3,27 Dynamic Mobility Assessment, 3,14 Emory Functional Ambulation Profile, 16,29 and Barthel Index balance subscale. 14 The BBS is intended to measure a subject s ability to maintain balance while doing functional tasks. It is based on the principle that a person s balance is challenged by diminishing his/her base of support. 3,15 The test has 14 items that are common in everyday life. These tasks progress from sitting to comfortable standing, to standing with feet together, and, finally, to tandem standing and single-leg standing (noted by Berg et al as the 2 most difficult items). 3,15 Each item of the BBS is rated on a 5-point scale. Lower ratings are given for supervision, cueing, assistance, and/or whether time or distance requirements are met. 3,15 When used as a standardized assessment, a cutoff score of 45/56 is an established criterion. When used to evaluate risk of falling, the same cutoff score has the greatest specificity in identifying people who do not fall. 23 When calculating a total score for the BBS, a point of concern is that test items have different operational definitions for each of their rating categories. For example, for the item transfer, a rating category of 2 indicates able to transfer with verbal cueing or supervision, whereas for the item standing on one leg the same rating category of 2 indicates able to lift

2 RASCH ANALYSIS OF THE BERG BALANCE SCALE, Kornetti 1129 leg independently and hold equal to or greater than three seconds. From this example, the question arises about the appropriateness of combining scores from ratings that have different meanings. A second concern with the rating scale is that a passing score on each item is not always the same numeric value. Although the BBS does not specifically define passing for each item, it could be defined as successful completion of the task, independent of qualifiers for example, time and speed. When this definition of passing is used for the item standing on one leg, a rating category of 2 must be achieved to successfully pass ( able to lift leg independently and hold equal to or greater than 3 seconds ). In contrast, for the item sitting unsupported, a rating category of 4 must be achieved to pass ( able to sit safely and securely 2 minutes ). Using this interpretative framework, more points are earned by passing the item sitting unsupported than would be earned by passing the item standing on one leg. With this definition of passing, all items should share the same numeric rating. In addition to challenges in interpreting the rating scales, similar challenges arise in interpreting total BBS scores. Although it is well established that a BBS score of 45/56 or greater indicates a low probability of falls, it provides limited information about the functional capability of the patient. 4 For example, a total score of 45/56 does not tell us whether a person can sit unsupported (the easiest item on the instrument) or stand on 1 foot (the most difficult item). Ironically, although the patient s score indicates the probability of falls, it fails to describe the patient in functional terms. The total score provides limited information with which to guide clinical intervention. Nevertheless, scores are still used to make important clinical decisions about screening, intervention, progress, and discharge planning. Based on the importance of the BBS as a clinical tool and research instrument in identifying people at risk for falls and assessment of balance, it is worthwhile to further evaluate it with the Rasch measurement model. Although traditional psychometric approaches focus on the total score of a given instrument, the Rasch measurement model makes possible analysis of instruments at the item and rating-scale level. First, Rasch analysis converts raw-score, ordinal data, such as the 0 to 4 scale, into interval-based measures the logs-odds metric or logit. Second, item goodness-of-fit statistics provided by the analysis determines the extent to which each item fits the construct it is intended to measure. Third, the analysis makes it possible to determine whether the rating scale is used in the expected manner; for example, people with lower balance ability would be expected to receive lower item ratings, whereas people with higher balance ability would be expected to have higher item ratings. Finally, Rasch measurement provides a connection between total score and the instrument s items. This is accomplished by placing the person ability (person measure) and item difficulty (item measure) on the same linear continuum. Ceiling and floor effects are revealed when person ability and item difficulty fail to match at the extremes of the continuum. Using the analytic features of Rasch measurement, the overall purpose of our study was to address challenges in scoring of the BBS. First, we investigated how each item s rating scales are being used. That is, are all ratings of each item endorsed? For example, for the item tandem stance, is the rating 1 needs help to step but can hold 15 seconds, being used by raters? If a particular rating is not being used, is it appropriate to include that rating in a scale? Alternatively, if a rating is being used, is that particular rating used as expected? That is, we would expect people of lower ability to be more likely to use lower ratings on an item and people of higher ability to be more likely to use higher ratings. Second, we addressed conceptual and mathematical inconsistencies in the BBS. This speaks to the concern that a passing rating on each item is not always of the same numeric value. By applying Rasch analysis to the BBS, pass-fail rating transition points can be developed. Establishing these pivot points provides a theoretical point at which we can identify the ability of a person to pass or fail an item. The purpose of pivot points is to equalize all BBS item scores (ie, passing an item would have the same numeric value across all items). Our final objective was to address the limitation of the BBS in describing a patient in functional terms. Rasch measurement makes it possible to connect an individual s total score with the content of the BBS. By placing person and item measures on the same continuum, it should be possible to describe in functional terms a patient who scores a 45/56. METHODS Participants Data were retrieved retrospectively from the records of 100 community-dwelling veterans who were receiving care at the Malcom Randall North Florida/South Georgia Veterans Affairs Medical Center in Gainesville, FL. The subjects were participants in a gait and balance rehabilitation program, after being referred by their primary care physician. The BBS was a component of a 90-minute, comprehensive, initial PT assessment. Subjects (99 men, 1 woman) ranged in age from 64 to 88 years. They were divided into 4 primary diagnostic categories: neurologic, orthopedic, cardiopulmonary, and other. All subjects had multiple comorbidities. In addition to the BBS, a standardized history of falls interview revealed that the subjects had had from 0 to 12 or more falls over the last 12 months (mean, 5.56; median, 5). The TUG scores 9,27 ranged from 7.58 to seconds (mean, 19.98s) and Falls Efficacy Scale scores ranged from 14 to 100 (mean, 27.19). Rating Scale Analysis We used the Rasch measurement model with the Winsteps computer program, 33,a because it offers distinct advantages over traditional psychometric approaches. First, the Rasch model converts raw scores into interval data. Although ordinal data are often used for statistical analysis, equal interval data are fundamental for even the most basic mathematical operations. Second, Rasch analysis permits evaluation of the rating scale of an instrument by providing the probabilities of using each value of the rating scale (ie, ratings 0 4) relative to the abilities of the people in the sample. Third, because the Rasch model places item measures and person measures on the same continuum, it is possible to determine if the difficulty level of the instrument matches the ability level of the intended population. The connection of person ability and item difficulty permits identification of the items a person of a particular ability would be expected to pass or fail. A particular focus in this study was the investigation and modification of item rating scales. To determine whether the rating scale for each BBS item was being used in an expected manner, we examined the probability of each rating (0 4) based on the subjects overall performance on the BBS. We expected that people of low ability would use lower parts of the rating scale (eg, 0 or 1) and people of higher ability would use higher parts of the rating scale (eg, 3 or 4). When this relationship was not apparent, we combined rating categories (eg, collapsing rating 0 and 1 to a new category 0) and reanalyzed

3 1130 RASCH ANALYSIS OF THE BERG BALANCE SCALE, Kornetti Fig 1. (A) Original scale for item tandem stance is a graphic presentation of the probability of responses for each of the 5 ratings (0 4) as a function of the overall performance of individuals on the BBS as originally written. (B) Revised scale presents the probabilities of the revised rating categories for tandem stance. the rating scales. We continued this process until all revised rating categories were more probable to emerge at different ability levels (eg, combined ratings of 0 and 1 emerged as more probable for persons of low ability than any other rating categories). Sample-Size Determination for Rasch Analysis The sample size needed to obtain stable person, item, and rating-scale calibrations for use with the Rasch model is based on the level of error expected in the measure. Linacre 34 indicated that 0.3 log-equivalent units (logits) is the best that can be expected for most variables, and that 1 logit stability at a 99% confidence interval is recommended for obtaining useful, stable estimates. The number of subjects needed to obtain this level of item stability is We had 100 subjects in this study to further minimize error. RESULTS Investigation and Modification of Rating Scales To determine whether the rating scale for each BBS item was being used in the expected manner, we examined the probability of each rating (0 4) based on overall performance on the BBS. For example, one would expect that subjects with low performance would tend to use lower ratings on all items (ie, for the item tandem stance using the rating 0 equals loses balance while stepping or standing ). One would also expect that subjects with high performance would tend to use higher ratings on all items (ie, for the item tandem stance, using the rating 4 equals able to place foot tandem independently and hold for 30 seconds ). Figure 1A shows the probability of responses for each of the 5 ratings (0-4) for the item tandem stance as a function of the overall performance of individuals on the BBS as it was originally written. That is, where a person s ability level is far below the average item difficulty level (person [minus] item measure is a large negative number), the probability is high that he/she will be rated with 0 (lowest rating available), and the probability is low that he/she will be rated with a 4 (highest rating available). In contrast, where a person s ability level is far above the average item difficulty level (person [minus] item measure is a large positive number), the probability is low that he/she will be rated with 0 (lowest rating available), and the probability is high that he/she will be rated with a 4 (highest rating available). Across ability levels, we would expect that particular ratings would be more probable than other ratings. That is, we would expect 5 prominent curves, representing each of the 5 rating values (0, 1, 2, 3, 4). Figure 1A shows that only 3 parts of the rating scale were more probable to emerge at different ability levels: 0, 3, and 4. Ratings 1 and 2 were never more probable than the ratings 0, 3, and 4. As seen from the key below the figure, ratings 1 and 2 were only used a total of 15% of the time.

4 RASCH ANALYSIS OF THE BERG BALANCE SCALE, Kornetti 1131 Fig 2. (A) Original scale for item standing on one leg and (B) revised scale for standing on one leg present another example of rating-scale modification, this time using the item standing on one leg. In the original rating scale (A), although ratings 0, 1, and 4 emerge as more probable than other ratings, ratings 2 and 3 do not. To compensate for the low probability of using these 2 rating categories, rating 3 was combined with rating 2. (B) The resultant probability curves from this revision. Because ratings 1 and 2 were not used as expected, these 2 ratings were combined into a new category, 0, for the next phase of the analysis. Figure 1B presents the probabilities of the revised rating categories for tandem stance. In contrast to figure 1A the original rating scale figure 1B shows that all revised rating categories are now more probable to emerge at different ability levels. For person minus item measures between 8 to 6 logits, receiving a0ismore probable than receiving any other rating. This indicates that people of low ability on the BBS will have the greatest probability of receiving the lowest rating. Between 6 and 0 logits, receiving a 3 is more probable than receiving any other rating. This indicates that people of middle ability ( 6 to 0 logits) on the BBS will have the greatest probability of receiving a middle rating. Finally, between 0 and 8 logits, receiving a 4 is more probable than receiving any other rating. This indicates that subjects of high ability will have the greatest probability of receiving the highest rating. The purpose of the above modifications is 2-fold. First, the modifications eliminate the redundancy of rating categories that are underutilized in the original scale. Second, it preserves the intent of the rating scale by ensuring that each rating is distinctly separated from every other rating category. That is, as person ability changes, a different rating category is used (low ability people use low ratings, higher ability people use higher ratings). Figures 2A and B present another example of rating-scale modification, this time using the item standing on one leg. In the original rating scale (fig 2A), ratings 0, 1, and 4 emerge as more probable than other ratings, while ratings 2 and 3 do not. To compensate for the low probability of these 2 rating categories being used, rating 3 was combined with rating 2. Figure 2B presents the resultant probability curves. As with the ratingscale conversion for tandem stance, the conversion for standing on one leg also results in all revised rating categories becoming more probable to emerge at different ability levels. Our second objective was to address the inconsistencies in the rating values representing successful item completion. That is, summative ratings are difficult to interpret if the description of the rating value varies across the items. On the original BBS, a rating representing passing is not always the same numeric value across items. As noted earlier, a rating of 2 must be achieved to successfully pass the item standing on one leg, whereas a rating of 4 must be achieved to pass the item sit unsupported. To improve the interpretation of the BBS we created conceptual transition points, or pivot points, for each item s rating scale. Pivot points provide a theoretical transition that defines a person s ability to pass a specific item, thereby equalizing all item ratings. We define the term passing as the ability to do a specified task (item description) without assistance or verbal cueing, and we define failing as the need for assistance to pass the task. Pivot points anchor each item around a theoretical midpoint. We established pivot points after modifying each item s rating scale (figs 1, 2). As an example of pivoting, in the key below figure 1B, the arrow identifies the pivot point between assistance needed and no assistance needed to complete the specified task. For example, the pivot point for the item tandem stance is between 3 ( able to place

5 1132 RASCH ANALYSIS OF THE BERG BALANCE SCALE, Kornetti Fig 3. Berg Balance Scale. The column marked old refers to the original BBS. The column marked revised refers to the revised scale based on Rasch analysis. foot ahead of other independently and hold 30 seconds ) and 4 ( able to place foot tandem independently and hold 30 seconds ). A rating of 4 is the only rating at which the individual accomplishes a tandem-stance position. The rating scale modifications and the pivot-point decisions for all items are presented in figure 3. Once the pivot points were designated, the statistical software was programmed to anchor all items across their pass-fail threshold. Comparison of Person Ability to Item Difficulty Figures 4A and B represent the original and the modified BBS scoring (after rating-scale revision and pivot-point anchoring). Rasch analysis places person ability and item difficulty (logits) on the same linear continuum. Figures 4A and B place person ability measures on the left and item difficulty on the right of the vertical line. Person ability measures represented by X increase from lowest to highest value (bottom to top of each figure). Similarly, item difficulty measures are mapped on the same continuum and increase from easiest to hardest (bottom to top of each figure). In figure 4A (the original BBS scoring), person abilities range between 2.35 and 6.0 logits, forming a bell-shaped curve with a slight skew toward the higher measures. Item measures range from 4.98 to 2.65 logits. Several items calibrate at the same difficulty level, that is, pickup item from floor, stand to sit, transfer, reach forward, and stand with feet together. There is a tendency for a ceiling effect with person measures above 2.65 logits but no items with similar calibrations. Although there is a gap between the 2 easiest items, sit unsupported and stand unsupported, there are no person measures within this calibration range. The scale rescoring and pivot-point anchoring results in a noteworthy change in the person and item measure distributions. Figure 4B, representing the revised BBS scoring, shows a greater spread in person ability measures (range, 5.66 to 6.0 logits). Although person measures in figure 4A concentrate around the mean, these measures in figure 4B are more evenly distributed across item difficulties. Item measures in the revised scoring also cover a larger range ( 5.33 to 5.85 logits) and are also more evenly distributed with less redundancy (ie, only items stand to sit and stand unsupported are

6 RASCH ANALYSIS OF THE BERG BALANCE SCALE, Kornetti 1133 Fig 4. (A) Original item map and (B) revised item map place person ability measures on the left and item difficulty on the right of the vertical line. Mean (M), 1 standard deviation (S), and 2 standard deviations (T) are indicated for both person ability and item difficulty measures. The horizontal line indicates the 45-cutoff score. at the same calibration). With the revised scoring, there is a decreased tendency for a ceiling effect, because the item tandem stance is approximately at the same calibration level as the 2 highest person ability measures. Although there are similar gaps across the easiest items, again there are few persons at this low ability level. The rescoring and pivot-point anchoring modifications also resulted in major changes in item difficulty ordering. Three items that show the largest changes in ordering are tandem stance, stand eyes closed, and turn 360. Tandem stance, a middle-difficulty level item in the original scale, becomes the most difficult item after scale revision. Stand eyes closed, a relatively easy item on the original scale, becomes a middle-difficulty item. Turn 360, among the most difficult items on the original scale, also becomes a middle-difficulty item after scale modification. These shifts in item difficulty are discussed below. Although the item maps show person/item spread and hierarchy, they also link person ability to item difficulty. By placing person ability and item difficulty on the same linear continuum, person ability values reflect what a person can or cannot do on the basis of item content. For example, in figure

7 1134 RASCH ANALYSIS OF THE BERG BALANCE SCALE, Kornetti 4B, the person with the lowest ability measures, that is, X located at 5.6 logits, should at best be only able to sit unsupported. Furthermore, this person should have a lower probability of successfully completing turn 360 and have 0 probability of successfully completing tandem stance. In reviewing this subject s response, he/she achieved a passing score on the easiest item, sit unsupported, and failed all but 1 other item on the instrument sit to stand. Conversely, the 2 people with the highest ability measures, that is, 2 X s located at 6.0 logits, have a 50% probability of successfully completing tandem stance because it is located at their ability level. Additionally, these subjects should be able to successfully pass all items below their ability level (ie, all other items on the instrument), which a review of their responses showed that they did. Of particular relevance to clinical practice and research is the 45-cutoff point. The horizontal lines in figures 4A and B represent a division between subjects who have scored at least 45 and those who have not. The letter X in boldface type represents those who scored 45 or higher. Both maps identify items that subjects scoring 45 or higher should be able to complete successfully. For example, in figure 4B, persons scoring above 2.25 logits have a 50% probability or greater of successfully completing items alternating foot, standing on one leg, and look behind, because these items are located near the 45-cutoff point. In reviewing these item scores, 100% (33/33) of the subjects who scored 45 or higher successfully completed alternating foot, 55% (18/33) successfully completed standing on one leg, and 76% (25/30) successfully completed look behind. DISCUSSION Our purpose in this study was to evaluate critically the rating scale of the BBS. First, Rasch analysis indicated that the probability of using each rating was not consistent across items. Two distinct features of the BBS are (1) some ratings were not used at all, and (2) many did not distinguish between people of different abilities. By condensing item-rating categories, we were able to eliminate underutilized categories and to construct categories that better separated people of differing abilities. Second, to address the inconsistencies in the rating scales, pivot points were developed for each item to represent a transition between passing and failing. The scale rescoring and pivot-point anchoring resulted in changes in the person and item measure distributions. Person and item measures became more evenly distributed across the BBS continuum and resulted in changes in item difficulty order. Finally, by placing person ability and item difficulty on the same linear continuum, the analysis made possible a description, in functional terms, of what a person should be able to accomplish to score a 45. Our first objective was to use Rasch analysis to determine whether the rating scale was being used in the expected manner. Analysis showed that some rating categories were not used at all and other rating categories were underutilized. This underutilization could be due to several factors. First, a particular rating category may not serve as a useful point of differentiation. For example, for the item sit unsupported, everyone in the sample could sit without support for 10 seconds (rating category 1). Second, the evaluator may be unable to distinguish across similar rating categories. For example, in the item pick up item from floor, it may be difficult to distinguish between rating categories 1 and 2 (1 unable to pick up and needs supervision while trying, 2 unable to pick up but reaches 2 5cm from slipper and keeps balance independent). Considering the lack of reliability in distinguishing these categories, collapsing across rating categories simplifies the evaluation process, results in fewer decisions for the therapist to make, and thus reduces the time demands on the patient. After collapsing the rating scales, the next modification involved the development of pivot points to anchor each item at its theoretical midpoint. This procedure addressed the concern that passing each item of the BBS is not always of the same numeric value. Bode 35 introduced the pivot-point procedure to align rating score categories. When using a polychotomous rating scale, the middle of the scale, if it is not specifically designated, will be the mathematical anchor, that is, rating category of 2 on the BBS. However, because the mathematical value of 2 does not represent a consistent transition point across all items, we formulated a theoretical pass-fail midpoint. The intent of the pivot-point anchors was to establish a homogeneous transition point among all BBS item ratings. Numeric ratings should imply a common reference point across all items. Without a common reference point, further mathematical operations such as the summation of scores could lead to distorted meanings. This is especially relevant when additive scores on the BBS are used to measure progress in rehabilitation, or to predict falls. Pivot anchoring, in addition to collapsing rating scales, can have a major effect on item calibrations. As shown by Bode, 35 these modifications result in an alteration in item calibrations. In this study, we show that the item tandem stance becomes a much more challenging item. According to Linacre, Overall, the ordering of items makes more sense than before pivot anchoring. 34 The hierarchy of items provides the operational definition of construct. When a rating scale is used within an instrument, pivot anchoring makes possible consistent ordering of items based on comparable rating scale categories. 34 Because these manipulations (rating scale reduction and pivot anchoring) can strongly influence item calibrations, decisions regarding such changes should be based on strict operational definitions. Finally, by utilizing the Rasch measurement model, functional characteristics of a 45/56 score on the BBS were defined for this sample. That is, a functional definition of a 45 score was achieved by placing person ability and item difficulty on the same linear continuum. Four items appeared to be important for subjects to achieve a score at or near the 45 cutoff. Three of these items ( alternating foot, standing on one leg, look behind ) served as markers to differentiate people at or near the cutoff score of 45. In figure 4B, these person ability levels are at the item difficulty level that represents the 45 cutoff. As noted in the Results section, subjects scoring at or above this cutoff showed a 50% or greater probability of successfully completing these 3 tasks. A fourth item, tandem stance, emerged as the most difficult item on the BBS. Four of 5 subjects who successfully passed tandem stance scored above the 45 cutoff for this sample. On further evaluation of the 3 marker items at the 45 cutoff, 3 of the 4 subjects passed 2 of the 3 items. The remaining 2 successfully passed tandem stance and achieved overall scores of 47 and 44. Passing tandem stance, in association with passing at least 2 of the 3 marker items, results in scoring above the 45 cutoff in this sample. These findings suggest that when using the BBS as a clinical tool for determination of falls, it is unnecessary to administer all items. If a person passes the most challenging item on the instrument, there is a high probability that he/she will score at or above the 45 cutoff. Success in at least 2 of the other 3 marker items would improve one s confidence in designating a subject as having a low probability of falling. Modification of the BBS test procedure should be considered. Testing solely on the 4 aforementioned items can provide substantial information about a persons score in relation to the

8 RASCH ANALYSIS OF THE BERG BALANCE SCALE, Kornetti cutoff. Additional BBS items appear to add to the length of the test without providing information about a person s probability for falls (item difficulty level below person ability level). Although, the BBS is primarily used as a determinant of probability of falls, completion of other items scores can help clinicians identify the functional status of people at lower ability levels. CONCLUSIONS Traditional scoring of the BBS measure disconnects the content of the instrument from the ability level of the person. Item response methodologies, such as Rasch measurement, allow a closer look at the item performance of the BBS at the level of the rating scale. The present findings provide direction for improving the rating scale structure for each of the items. More importantly, placement of person measures and item difficulty on the same continuum allows attachment of functional meaning to a subject s score on the BBS. Of special importance to the clinician, our study provides a functional definition to the numeric 45-cutoff point. Our study is limited in that the BBS was analyzed in its original form, artificially modifying item scales and creating pass-fail points based on the original item rating-scale descriptions. These modifications require further investigation. For example, a logical extension of our study would be to redefine the BBS items and create rating scales that have conceptually consistent transition points across items. Such modifications are likely to improve the connection between our clinical assessments and the observed functional outcomes of our clients. Acknowledgments: We acknowledge the assistance of the physical therapists who collected the data for this study at the North Florida/South Georgia Veterans Health System, Gait and Balance Clinic. References 1. Tinetti ME, Speechley M, Ginter SF. Risk factors for falls among elderly persons living in the community. N Engl J Med 1988;319: Rubenstein LZ, Josephson KR, Robbins AS. Falls in the nursing home. Ann Intern Med 1994;121: Berg KO, Maki BE, Williams JI, Holliday PJ, Wood-Dauphinee SL. Clinical and laboratory measures of postural balance in an elderly population. Arch Phys Med Rehabil 1992;73: Bogle Thorbahn LD, Newton RA. Use of the Berg Balance Test to predict falls in elderly persons [see comments]. Phys Ther 1996;76: Newton R. Balance screening of an inner city older adult population. Arch Phys Med Rehabil 1997;78: Shumway-Cook A, Woollacott MH. Motor control: theory and practical applications. Baltimore: Williams & Wilkins; Cumming RG, Kelsey JL, Nevitt MC. Methodological issues in the study of frequent and recurrent health problems. Falls in the elderly. Ann Epidemiol 1990;1: Cummings SR, Nevitt MC, Kidd S. Forgetting falls: the limited accuracy of recall of falls in the elderly. J Am Geriatr Soc 1988;36: Steffen TM, Hacker TA, Mollinger L. Age-and gender-related test performance in community-dwelling elderly people: Six-Minute Walk Test, Berg Balance Scale, Timed Up & Go Test, and gait speeds. Phys Ther 2002;82: Niam S, Cheung W, Sullivan PE, Kent S, Gu X. Balance and physical impairments after stroke. Arch Phys Med Rehabil 1999; 80: Stevenson TJ, Garland SJ. Standing balance during internally produced perturbations in subjects with hemiplegia: validation of the balance scale. Arch Phys Med Rehabil 1996;77: MacKnight C, Rockwood K. A hierarchical assessment of balance and mobility. Age Aging 1995;24: Franchignoni F, Tesio L, Martino MT, Ricupero C. Reliability of four simple, quantitative tests of balance and mobility in healthy elderly females. Aging (Milano) 1998;10: Harada N, Chiu V, Damron-Rodriguez J, Fowler E, Siu A, Reuben DB. Screening for balance and mobility impairment in elderly individuals living in residential care facilities. Phys Ther 1995; 75: Berg K, Wood-Dauphinee S, Williams JI, Maki B. Measuring balance in the elderly: validation of an instrument. Can J Public Health 1992;(83 Suppl 2):S Baer HR, Wolf SL. Modified Emory functional ambulation profile: an outcome measure for the rehabilitation of poststroke gait dysfunction. Stroke 2001;32: Brauer SG, Burns YR, Galley P. A prospective study of laboratory and clinical measures of postural stability to predict communitydwelling fallers. J Gerontol A Biol Sci Med Sci 2000;55:M Brown M, Sinacore DR, Binder EF, Kohrt WM. Physical and performance measures for the identification of mild to moderate frailty. J Gerontol A Biol Sci Med Sci 2000;55:M Creel GL, Light KL, Thigpen MT. Concurrent and construct validity of scores on the timed movement battery. Phys Ther 2001;81: Hsieh CL, Hsueh IP, Mao HF. Validity and responsiveness of the Rivermead mobility index in stroke patients. Scand J Rehabil Med 2000;32: Karlsson A, Frykberg G. Correlations between force plate measures for assessment of balance. Clin Biomech (Bristol, Avon) 2000;15: Newton RA. Validity of the multi-directional reach test: a practical measure for limits of stability in older adults. J Gerontol A Biol Sci Med Sci 2001;56:M Riddle DL, Stratford PW. Interpreting validity indexes for diagnostic tests: an illustration using the Berg balance test. Phys Ther 1999;79: Rosengren KS, McAuley E, Woods D, Mihalko S. Gait, balance, and self-efficacy in older black and white American women. J Am Geriatr Soc 2000;48: Rubenstein RL, Powers C. Falls and mobility problems: potential quality indicators and literature review (the ACOVE Project). Santa Monica: RAND Corp; Shumway-Cook A, Baldwin M, Polissar NL, Gruber W. Predicting the probability for falls in community-dwelling older adults. Phys Ther 1997;77: Shumway-Cook A, Brauer, S, Woollacott M. Predicting the probability for falls in community-dwelling older adults using the Timed Up & Go Test. Phys Ther 2000;80: Stevenson TJ. Detecting change in patients with stroke using the Berg Balance Scale. Aust J Physiother 2001;47: Wolf SL, Catlin P, Gage K, Gurucharri K, Robertson R, Stephen K. Establishing the reliability and validity of measurements of walking time using the Emory Functional Ambulation Profile. Phys Ther 1999;79: Geiger RA, Allen JB, O Keefe J, Hicks RR. Balance and mobility following stroke: effects of physical therapy interventions with and without biofeedback/forceplate training. Phys Ther 2001;81: Mao HF, Hsueh IP, Tang PF, Sheu CF, Hsieh CL. Analysis and comparison of the psychometric properties of three balance measures for stroke patients. Stroke 2002;33: Teasell R, McRae M, Foley N, Bhardwaj A. The incidence and consequences of falls in stroke patients during inpatient rehabilitation: factors associated with high risk. Arch Phys Med Rehabil 2002;83: Linacre JM, Wright BD. Winsteps [computer program]. Chicago: MESA Pr; Linacre JM. Sample size and item calibration stability. Rasch Meas Trans 1994;7: Bode RK. Partial credit model and pivot anchoring. J Appl Meas 2001;2(1): Supplier a. Winsteps, PO Box , Chicago, IL