Use of the Berg balance test to predict falls in elderly persons.

Key Words: Assessment, Balance, Geriatrics, Falls, Validity.

One in three persons over 65 years of age and almost one in two persons over 80 years of age will fall at least once each year.[1] This incidence increases to 66% for ambulatory residents of nursing homes.[2] Although only 5% of all falls result in serious injury, the psychological effects can lead to impaired mobility, loss of function, and an overall decrease in a person's quality of life.[3] Falls among elderly persons have a multifactorial etiology, and considerable research has been devoted to determining the variables that correlate with the incidence and severity of falls. Although no consistent relationship has been found between a person's medical status, diagnosis, and functional ability,[4] Most researchers[5,6] agree that the characteristic most predictive of falling is a past history of falls. Several neurologic and musculoskeletal factors such as decreased ankle dorsiflexion strength, reduced sensation, and increased reaction time contribute to postural instability and falls.[7,8] Tinetti and Speechley[5] identified three factors that correlate highly with falling in community-dwelling and institutionalized elderly persons: lower-extremity disabilities, foot problems, and gait and balance abnormalities.

Impaired balance has been correlated with an increased risk for falls and a resulting increase in the mortality rate of elderly persons who are prone to falling compared with those who are not prone to falling.[9,10] A reliable and valid clinical measure for balance abilities increases the physical therapist's ability to predict who is at risk for falls. In our study, validity was assessed by examining sensitivity and specificity. A test is determined to be sensitive if it accurately identifies persons who should have a positive test result; specificity indicates how efficiently a test identifies those persons who should have a negative result.[11]

Several researchers[12-14] have measured balance abilities following externally generated perturbations by using a force platform. Research performed on elderly persons[15] indicates that sway increases with increased age. Overstall et al[16] noted, however, that sway did not differentiate between people who were not prone to falling and people who tripped but that sway was increased in people who fell because they lost their balance.

Hu and Woollacott[17,18] have documented improvements in postural stability and decreases in electromyographic response time following a training program with a balance platform. Berg et al[19] found that speed of sway, which is thought to measure the amount of postural adjustment necessary to maintain a neutral position, was not sensitive enough to differentiate between people based on assistive-device use. A faster speed of sway is assumed to identify a person with greater difficulty balancing. No correlation between increased sway amplitude and frequency of falls has been documented.[3,13]

Several studies have been conducted on the sensory organization balance test (SOT).[8,20-22] This test is designed to assess balance abilities during six conditions of altered vision and altered support surface (dense foam). Di Fabio and Badke[21] reported interrater agreement ranging from 68% to 100% for items on the SOT.

Anacker and Di Fabio,[23] in a study of elderly persons with a recent history of falling, concluded that there was a reduction in stance duration when the subjects stood on a compliant surface as compared with when they stood on a firm surface. There is some doubt, however, as to whether this test is sensitive enough to enable clinicians to predict who is at risk for falls.[24]

A-nother test, the postural stress test, was developed for use with elderly people.[25-27] Interrater reliability estimates for this test have ranged from .83 (intraclass correlation coefficient [ICC]) to .99 (Cronbach's alpha).[12,26,27]

Three tests assess functional abilities and balance (ie, tests described by Tinetti,[4] Mathias et al,[28] and Berg et al[29]). Tinetti4 reported 85% agreement on test items when the test items were simultaneously scored by two raters. It is difficult to compare Tinetti's interrater reliability finding with those of other authors because she did not use nonparametric statistical analysis for her ordinal data. Tinetti et all also found that the risk of falling increases linearly with the number of risk factors identified.

Mathias et al[28] reported interrater agreement of W=.85 (Kendall coefficient of concordance test) when elderly subjects were tested by physical therapists and W=.69 when elderly subjects were tested by senior physicians. Mathias and colleagues' "get up and go" test has been compared with several other balance assessments, but a relationship between the test results and risk of falling has not been determined.[19,23]

For the test described by Berg et al,[29] interrater reliability was reported as.98 (ICC). To address validity, Berg et al[6] determined that the balance score obtained on the test correlated with a global rating of balance made by the treating therapist (r=.81). The balance test also correlated with platform-based measures of postural sway. Both spontaneous sway (r=.55) and sway in response to pseudorandom perturbations (r=.38) were assessed. Validity of the balance test has been demonstrated by determining that any subject scoring in the impaired range on the assessment is 2.7 times more likely to experience multiple falls as compared with subjects not scoring in the impaired range.[6] The Berg balance test also was used to differentiate between subjects based on their use of assistive devices. The Berg balance test scores of those people who used a walker or a cane indoors were different from each other and lower than than those of individuals who used a cane outdoors only or who walked without an assistive device. The Berg balance test has also been shown to correlate with both the Tinetti mobility index (r=.91) and the "get up and go' test (r= -.76). That is, a person who scores low on the Berg balance test will have a higher test score (time for completion) on the "get up and go" test.

We chose to study the Berg balance test because it takes approximately 20 minutes to complete and requires no sophisticated equipment, making it useful in clinical settings. Measurements obtained with this test have demonstrated excellent interrater reliability and a tendency toward at least moderately strong concurrent validity.[6,19,29]

The purpose of our study was to determine whether the Berg balance test would demonstrate sensitivity and specificity[11] in a population of elderly persons residing in life-care communities. Each life-care community provided its residents with three different levels of care: independent living, assisted living or personal care, and skilled nursing. The research questions were: (1) Is the Berg balance test predictive of current and future risk for falls in elderly persons residing in life-care communities, as measured by self-report on a questionnaire, and (2) Do balance score results achieved on the Berg balance test correlate with the individual's self-perceived activity level and overall balance ability? A fall was defined as unexpected contact of any part of the body with the ground. Near-falls were not included in our analysis because they are more difficult to define, less memorable, and less consequential.

Method

Subjects

A sample of convenience was used. Any independent-living resident in two life-care communities who volunteered was tested. Seventy-one subjects were initially tested, and only the 5 persons who were unable to follow the directions for the test after three repetitions were excluded. No subject was excluded on the basis of age, gender, or disability. The remaining group of 16 men and 50 women had a mean age of 79.2 years (SD=6.2, range=69-94).

Instrumentation

The equipment used for the Berg balance test was a step stool, a mat table, a chair with arms, a tape measure, a stopwatch, a pen, and a table. The balance assessment consists of 14 subtests performed in a standard order (Tab. 1). Each task is scored on a five-point scale (0-4) according to the quality of the performance or the time taken to complete the task, as ranked by the test developers. The maximum score for this assessment is 56. Based on clinical experience, Berg et al[6] contend that scores below 45 indicate that someone is impaired, with an increased risk for falls.

Table 1.
Berg Balance Test Subtests

Item                  Description

1                      Sitting to standing
2                      Standing unsupported
3                      Sitting unsupported
4                      Standing to sitting
5                      Transfers
6                      Standing with eyes closed
7                      Standing with feet together
                       Reaching forward with an
8                        outstretched arm
9                      Retrieving object from floor
10                     Turning to look behind
11                     Turning 360o
12                     Placing alternate foot on stool
                       Standing with one foot in front of
13                      the other foot
14                     Standing on one foot

Raters

At least two raters out of a pool of six raters were used each day the data were collected. All raters were physical therapists with at least 1 year of experience in the treatment of patients with neurologic disorders. The raters were not given any formal instr-uction on test administration, but they were given the test instructions and the descriptors for each test item, as written by and obtained from the test developers, the day prior to the balance clinic. They were given the opportunity to ask questions of the first author (LDBT) prior to administering the test.

Procedure

The purpose of the study and the subject of confidentiality were discussed with all volunteer subjects. Confidentiality was maintained by using randomly assigned numbers on all questionnaires and scoring sheets. After signing a consent form, the following demographic data were collected for each subject: age, gender, and the presence of any major orthopedic or neurologic diagnoses. Each participant was given a packet containing a questionnaire, a balance test score sheet, and an envelope. Each person was asked to address an envelope that was later used to mail the 6-month follow-up questionnaires. Each participant completed a modified Activity Index questionnaire.[30,31] Each subject 4vas then randomly assigned to one of the raters, who completed the Berg balance test (Tab. 1). Every fourth subject; after a brief rest, was asked to complete the test again with another rater. Seventeen of the 71 subjects were retested. These results were used to calculate interrater reliability.

A Spearman rho ([r.sub.s]), a nonparametric version of the Pearson product-moment correlation coefficient, was used to analyze interrater reliability because the scores of the Berg balance test are ordinal or ranked data. The scores obtained from the 17 subjects who were tested twice by two different raters were compared ([r.sub.s]=.88).

A letter was sent to all participants 3 months after completion of the initial balance tests thanking them for their participation and reminding them that a follow-up questionnaire would be forthcoming. Six months after the initial balance tests, the same questionnaire used prior to the tests was sent to the participants with an addressed and stamped envelope. The returned questionnaires were used to assess the predictive validity of the balance assessment.

Data Analysis

To answer the first research question, the relationship between each subject's balance score and his or her reported frequency of falls was analyzed by a chi-square test. A Yates correction was not used because 2 degrees of freedom was used in the computation, the significance of the relationship was not borderline, and the correction itself tends to be overly conservative.[32,33]

To answer the second research question, a multiple-regression analysis was run to determine how each factor (ie, age, gender, activity level, frequency of falls, self-perception of balance, and use of an assistive device) contributes to the composite score achieved on the balance test. A chi-square test also was used to determine whether the score on the Berg balance test successfully differentiated among subjects based on their use of an ambulatory assistive device. The original and 6-month follow-up questionnaires were compared to determine whether any major changes had occurred in the sample under study. Any changes were to be reported.

Activity level was analyzed using information obtained from the questionnaire on the amount of participation in regular exercise. Exercise levels were given a rank score of 0 to 8. A higher number corresponded to increased frequency of participation in at least one of three levels of activity: vigorous sports activity, light physical activity, and social and recreational programs. A limitation of this rating system was that the scale did not allow for differentiation between subjects who frequently participated in one type of activity and subjects who infrequently participated in multiple levels of activity. For the four questions regarding performance of activities of daily living (ADLs), each subject was given a rank score of 0 to 6. The score given depended on how much assistance the subject reported needing to complete daily tasks such as cooking, cleaning, shopping, and obtaining transportation. Subjects who scored higher on the scale reported needing less assistance than those who scored lower.

Results

Twenty-five out of 66 subjects (38%) reported some type of neurologic or orthopedic impairment. The four most common orthopedic diagnoses reported were arthritis (n=4), hip arthroplasty (n=2), status postlaminectomy (n=2), and fractured vertebrae (n=2). The two most common neurologic diagnoses reported were stroke (n=6) and Parkinson's disease (n=5). Fifty-two percent of the subjects rated their own balance as good or excellent, and 86% of the participants felt that their overall health was good or excellent. Seventy-five percent of the subjects reported not using an assistive device (Tab. 2), and most subjects did not report experiencing any falls either initially (83%) or 6 months after taking the Berg balance test (80%) (Tab. 3).

Table 2.
Assistive Device Use as Reported on Questionnaire

When Used                                     No. of Subjects
Never                                         49
Outside or for long distances                  6
All the time                                  10
Missing data                                   1
Table 3.
Number of Subjects Reporting Falls

                         Initial               Final
No falls                  54                    43
1-2 falls                  8                     9
3-4 falls                  1                     0
>4 falls                   2                     0
Missing data               1                    12
Total                     66                    66

Eighty-five percent of the 6-month follow-up questionnaires were returned. The comparison of number of falls reported by the subjects on the initial questionnaire and their performance on the Berg balance test is shown in Table 4. When the information was compressed into a 2x2 chi-square table (Tab. 5), a relationship was found between the actual values and the expected frequencies. The sensitivity of the Berg balance test as compared with the initial fall frequency was 53% (9/17), whereas the specificity of the measure was 96% (46/48). When the scores achieved on the Berg balance test were applied to the 6-month follow-up fall frequency, the sensitivity remained at 53% (8/15) and the specificity was 92% (36/39).

[TABULAR DATA 4 OMITTED]

Table 5.
Chi-square Test Results for Comparison of Number of Falls Reported
on Initial Questionnaire and Performance on Berg Balance Test(a)

          Subjects-impaired       Subjects-nonimpaired

No falls   8 (13.9)               46 (40.1)
Falls      9 (2.8)                 2 (8.2)

(a) Expected frequencies shown in parentheses. Subjects with impairments
scored below 45 on the Berg balance test; subjects without impairments scored
45 or above on the Berg balance test.

The multiple regression indicated that performance of ADLS predicted 43% of the total score achieved on the Berg balance test (P<.001). Age accounted for an additional 5% of the total score achieved (P<.02). Table 6 shows the numbers of subjects who scored above the cutoff score of 45, separated by age in decades. There was no relationship between scores achieved on the balance test and each subject's self-perception of his or her balance abilities.

Table 6.
Performance on Berg Balance Test as Analyzed by Ages in Decades
(N = 66)

                                        No. of
                   No. of               Subjects Who
Age (y)           Subjects              Scored >45
60-70                3                    2
71-80               34                   27
81-90               25                   19
91-100               4                    1

Scores on the Berg balance test also were examined in relation to the use of assistive devices (Tab. 7). For the purpose of the chi-square test, the three groups were compressed into two groups: assistive device used and no assistive device used. The chi-square test (Tab. 8) indicated that a relationship existed (P<.001) between the use of an assistive device and the score on the Berg balance test. The sensitivity of the Berg balance test for predicting use of an assistive device was 76% (13/17), whereas the specificity was 94% (45/48).

Table 7.
Effect of Use of Assistive Device on Performance on the Berg Balance
Test (n=65)

Frequency of             Berg Balance Test Score (Frequency)
Use                    <20   21-30   31-45   >45   Total
Do not use an
assistive device                      4       45     49
Use an assistive
device outside
and for long
distances                             3         3      6
Use an assistive
device all the
time                   2      3       5               10

Discussion

Using the test developers' cutoff score of 45 (out of 56) as a guideline, the Berg balance test is designed to identify those individuals who are classified as "fallers" (those who score below 45) and those individuals who are classified as "nonfallers" (those who score 45 or above). When analyzed with a chi-square test (Tab. 5), our two groups demonstrated a difference that would not occur by chance. In addition, the Berg balance test was demonstrated to be highly specific (96%). For clinicians, this finding indicates that those persons who score 45 and above on this assessment have a high probability of not falling. The sensitivity of the Berg balance test was low (53%). We attributed this result to a variety of factors. We agree with Tideiksaar[10] and Studenski et al,[34] who commented on the unclear relationship between risk of falls and physical impairments. The subjects who fell most frequently were those who scored closer to the cutoff score (Tab. 4), not further away. A few of the subjects who scored in the most impaired range on the balance test appeared to adopt strategies for minimizing their risk of falling (eg, use of companions and assistive devices). This finding emphasizes the difference between risk of falling and the presence of some physical impairments. Subjects who were the most physically impaired (limited in either ability to ambulate or ability to transfer) did not have the highest risk for falling because of their use of external supports.

Our data provide evidence that falls are primarily multifactorial in nature.[1,10] A complete assessment of risk of falling also must examine the environment in which a person functions, not just a score achieved on this or any other balance assessment. Studenski et al,[35] using a home evaluation tool, determined that the risk of falling, as identified by defined hazards, increased when the number of hazards increased. Modification of the environment may reduce the frequency of falls.[1] A home evaluation, such as that developed by Chandler et al,[36] could be used in conjunction with the Berg balance test to provide a clearer picture of the client's functional risk of falling.

Our research did not agree with the test developers'[6,19] determination that the Berg balance test is sensitive in identifying fallers but did provide support for the claim of specificity to identify individuals in the nonfaller category. The Berg balance test demonstrated a moderately strong ability to predict a person's use of an assistive device (Tab. 8).

Table 8.
Chi-Square Test Results for Assistive Device Use and Performance on
Berg Balance Test(a)

       Subjects-Impaired Subjects-Nonimpaired

No device        4 (12.6)                45 (36.4)
Device used     13 (4.1)                  3 (11.4)

(a) Expected frequencies shown in parentheses. Subjects with impairments
scored below 45 on the Berg balance test: subjects without impairments scored
45 or above on the Berg balance test.

A linear relationship between increased frequency of falls, either initially or at follow-up, and decreased scores on the Berg balance test was not established. Increasing age also did not correlate with decreasing balance ability in our study. In this respect, we disagree with some authors[37] and agree with others.[38] Additional research is needed to determine the effect aging has on an individual's balance ability. The reason for the disagreement among researchers may be due to the different nature of the groups of subjects studied.[38]

Activity level, did not appear to contribute to performance on the Berg balance test, as determined by multiple-regression analysis. This finding may in part be due to the way activity level was measured. A standardized index,[30] was used, with some modifications,[31] as a questionnaire. The scoring system, however, may not have been sensitive enough to the variations in individual activity patterns. This hypothesis is supported by the contribution ADL performance (43%) made to the composite scores achieved on the Berg balance test. It could be argued that those individuals who are capable of performing the daily activities of cooking, cleaning, shopping, and obtaining transportation are more active. More research is needed in this area to determine the contribution that activity level makes to overall balance ability.

Our interrater reliability ([r.sub.s]=.88) is lower than the reliability determined by the test developers ICC=.98).[29] We used a nonparametric test to determine interrater reliability because of what we considered to be the ordinal nature of the scale used in the Berg balance test. Raters disagreed the most on three items: reaching for-ward with an outstretched arm, standing with one foot in front of the other foot, and standing on one foot. These are the most difficult of the subtests for the subjects to perform. A determination of why the raters disagreed the most on these items is beyond the scope of this article, but a careful reading of the descriptors for each test item is recommended prior to using the assessment.

Seventy-three percent of the subjects scored above the cutoff score of 45, and 11% of the subjects achieved a perfect score of 56. There may be a ceiling effect to the Berg balance test, making it insensitive to differences among persons with very high levels of balance ability. These individuals, however, are probably those who would not require physical therapy for balance problems.

Although problems have been noted with the self-report method of collecting data on falls,[1,27] we attempted to control for this limitation by assigning each subject a random number and sending a reminder halfway through the follow-up period. By doing so, we hoped to increase the honesty of responses as well as improve the recall of fall frequency.

Additional research is needed on individuals who score between 31 and 45 on the balance test. Fifty-five percent (6/11) of those individuals who fell scored in this category (Tab. 4). Further analysis of this group also may lead to improved sensitivity of the instrument. We hypothesize that this group engaged in more risk-taking activities and did not use external supports to compensate for their deficits as compared with the group who scored in the most impaired range. Additional research is needed to answer this question and to examine whether environmental factors contributed to this group's higher risk of falling compared with the group who scored lower on the balance assessment. A more refined weighing system also is needed to discriminate between levels of activity. This weighing system is necessary to determine whether participating in a rigorous exercise program is more beneficial than simply maintaining an active lifestyle.

Conclusion

The Berg balance test continues to show strong interrater reliability ([r.sub.s]=.88) when raters are provided with written instructions and are allowed to ask questions prior to administering the balance test. Measurements obtained with the Berg balance test show high specificity but poor sensitivity,for identifying people with increased risk of falling. Improved sensitivity of the instrument is needed, particularly for those older adults scoring closer to the cutoff score of 45. The Berg balance test, however, shows sensitivity and specificity to predict use of assistive devices in the older adult. The Berg balance test is easy to administer and requires no special equipment. We believe that the determination of patients' risk of falling may be substantially improved by also examining their environment and how well they complete their activities of daily living.

Acknowledgments

We thank the residents of The Quadrangle, Haverford, Pa, and White Horse Village, Newtown Square, Pa, for their dedication, without which this study would not have been possible. We also thank Dean Paulson and the staff at Bryn Mawr Rehabilitation Hospital for their invaluable assistance and support.

References

[1] Tinetti ME, Speechley M, Ginter SF. Risk factors for falls among elderly persons living in the community. N Engl J Med. 1988;319:1701-1707. [2] Tinetti ME. Factors associated with serious injury during falls by ambulatory nursing home residents. J Am Geriatr Soc. 1987;35:644-648. [3] Berg KO. Balance and its measure in the elderly: a review. Physiotherapy Canada. 1989;41:240-246. [4] Tinetti ME. Performance-oriented assessment of mobility problems in elderly patients. J Am Geriatr Soc. 1986;34:119 -126. [5] Tinetti ME, Speechley M. Assessment of risk and prevention of falls among elderly persons: role of the physiotherapist. Physiotherapy Canada. 1990;42:75-79. [6] Berg KO, Wood-Dauphinee SL, Williams JT, Maki B. Measuring balance in the elderly: validation of an instrument. Can J Public Health. 1992;83:S7-S11. [7] Whipple RH, Wolfson LI, Amerman PM. The relationship knee and ankle weakness to falls in nursing home residents: an isokinetic study. J Am Geriatr Soc. 1987;35:13-20. [8] Lord SR, Clark RD, Webster IW. Postural stability and associated physiological factors in a population of aged persons. J Gerontol 1991;46:M69-M76. [9] Wild D, Nayak USL, Issacs B. Prognosis of falls in old people at home. J Epidemiol Community Health. 1981;35:200-204. [10] Tideiksaar R. Falls and instability in the elderly. Neurorehabilitation. 1993;3:51-61. [11] Rothstein JM, Echternach JL. Primer on Measurement: An Introductory Guide to Measurement Issues. Alexandria, Va: American Physical Therapy Association; 1993:68-73. [12] Badke MB, Duncan PW. Patterns of rapid motor responses in normal and hemiplegic subjects during postural adjustments in standing. Phys Ther. 1983;63:13-20. [13] Nashner LM. Fixed patterns of rapid postural responses among leg muscles during stance. Exp Brain Res. 1977;30:13-24. [14] Lehmann JF, Boswell S, Price R, et al. Quantitative evaluation of sway as an indicator of functional balance in post-trauma6c brain injury. Arch Phys Med Rehabil. 1990;71:955-962. [15] Sheldon JH. Effect of age on control of sway. Gerontol Clin. 1969;4: 129-138. [16] Overstall FW, Exton-Smith AN, Imms FJ, Johnson AL. Falls in the elderly related to postural imbalance. BMJ. 1977;29:261-264. [17] Hu MH, Woollacott MH. Multisensory training of standing balance on older adults, 1: postural stability and one4egged balance. J Gerontol. 1994;49:M52-M61. [18] Hu MH, Woollacott MH. Multisensory training of standing balance in older adults, II: kinematic and electromyographic postural responses. J Gerontol 1994;49:M62-M71. [19] Berg KO, Maki BE, Williams JI, et al. Clinical and laboratory measures of postural balance in an elderly population. Arch Phys Med Rehabil. 1992;73:1073-1080. [20] Horak FB. Clinical measurement of postural control in adults, Phys Ther. 1987;67:1881-1885. [21] Di Fabio RP, Badke MB. Relationship of sensor-y organization to balance function in patients with hemiplegia. Phys Ther. 1990;70:542-548. [22] Di Fabio RP, Badke MB. Stance duration under sensory conflict conditions in patients with hemiplegia. Arch Phys Med Rehabil 1991;72: 292-295. [23] Anacker SL, Di Fabio RP. Influence of sensory inputs on standing balance in community-dwelling elders with a recent history of falling. Phys Ther. 1992;72:575-581. [24] Horak FB. Commentary on "Influence of sensory inputs on standing balance in community-dwelling elders with A recent history of falling." Phys Ther. 1992;72:581-582. [25] Hill K, Vandervoort A, Kramer J. Reproducibility of performance on a test of postural responses in healthy elderly females. Physiotherapy Canada. 1990;42:61-67. [26] Chandler J. Duncan PW, Studenski SR. Balance performance on the postural stress test: comparison of young adults, healthy elderly, and fallers. Phys Ther. 1990;70:410-415. [27] Wolfson LI, Whipple R, Amerman P, et al. Stressing die postural response: a quantitative method for testing balance. J Am Geriatr Soc. 1986;34:845-850. [28] Mathias S, Nayak USL, Issacs B. Balance in elderly patients: the "Get-up and Go" test. Arch Phys Med Rehabil. 1986;67:387-389. [29] Berg KO, Wood-dauphinee SL, Williatns JT, Gayton D. Measuring balance in the elderly: preliminary development of an instrument. Physiotherapy Canada. 1989;41:304-311. [30] Duke Multidimensional Functional Assessment. In: Pfeiffer E, ed. Multidimensional Functional Assessment: The OARS Methodology. Durham, NC: Duke University Center for the Study of Aging; 1975. [31] Leuhring S. Component Movement Patterns of Two Groups of Older Adults in the Task of Standing From the Floor. Richmond, Va: Medical College of Virginia; 1989. Thesis. [32] Daniel WW. Biostatistics: A Foundation for Analysis in the Health Sciences. New York, NY: John Wiley & Sons Inc; 1974:320-321. [33] Bahn AK. Basic Medical Statistics. New York, NY: Grune & Stratton; 1972:66-67. [34] Studenski S, Duncan PW, Weiner D, et al. The role of instability in falls among older persons. In: Duncan PW, ed. Balance: Proceedings of the American Physical Therapy Association Foram. Alexandria, Va: American Physical Therapy Association; 1990:57-60. [35] Studenski S, Duncan PW, Chandler J, et al. Predicting falls: the role of mobility and nonphysical factors. J Am Geriatr Soc. 1994;42:297-302. [36] Chandler J, Prescott B, Duncan PW, Studenski S. Reliability of a new instrument: the functional environmental assessment. Phys Ther. 1991; 71 (suppl):574. Abstract. [37] Reichel W, ed. Clinical Aspects of Aging. 3rd ed. Baltimore, Md: Williams & Wilkins; 1989:551-554. [38] Woollacott MH. Changes in posture and voluntary control in the elderly: research findings and rehabilitation. Topics in Geriatric Rehabilitation. 1990;5(2):1-11. Chandler

LD Bogle Thorbahn, PT, NCS, is Program Coordinator of Neuroscience, Department of Physical Therapy, Bryn Mawr Rehabilitation Hospital, 414 Paoli Pike PO Box 3007, Malvern, PA 19355 (USA). This study was completed in partial fulfillment of Ms Bogle Thorbahn's Advanced Master's Degree at Temple University.

RA Newton, PhD, PT, is Professor in Physical Therapy, College of Allied Health, Temple University, N Broad St, Philadelphia, PA 19140.

The study protocol was approved by the institutional review boards of Temple University and Bryn Mawr Rehabilitation Hospital.

This article was submitted November 8, 1993, and was accepted March 6, 1996.