Use of clinical and impairment-based tests to predict falls by community-dwelling older adults. (Research Report).

An estimated 25% to 35% of adults aged 65 years and older fall each year. (1-4) The high fall incidence in older adults is costly in terms of both health care dollars and quality of life. Most older adults hospitalized for falls are discharged to long-term care facilities. (5) Fear of falling is an important factor that affects the willingness of many older adults to participate in physical activity and exercise. (6,7) Cessation of physical activities, whether due to fear of falling or due to declining mobility and balance, accelerates the decline in muscle, force production and function, further increasing fall risk and further decreasing quality of life. (6,7) Low fall-related self-efficacy has been associated with decreased function, increased fall risk, and risk of admission to assisted living or long-term care institutions. (6,7)

A number of interventions to improve balance and decrease fall risk have been found to be effective. (8-10) Interventions may be performed in the hospital, clinic, retirement facility, or community. Some researchers (11-14) have found multifaceted intervention programs that include exercise to increase muscle force, flexibility, and balance to be effective approaches.

A number of fall-risk screening tests have been used to identify people at risk for falling ("fallers") among residents of nursing homes; however, these tests are less predictive of falls in older adults who have fewer health problems, live independently, and are more active. (15,16) The Berg Balance Scale (BBS), (17,18) the Tinetti Performance-Oriented Mobility Assessment (POMA), (19) the Tinetti Balance Subscales, (19) and the Timed "Up & Go" Test (TUGT) (20) were developed for, and validated primarily on, residents of nursing homes.

The usefulness of a test in predicting falls may vary depending on the health status and level of function of the older adults being tested. In a study of community-dwelling older adults who were in good health, O'Brien et al (16) found the BBS was less sensitive in predicting falls than did Berg et al (17) who studied residents of a nursing home. Other researchers (21) studying community-dwelling older adults found BBS scores to be predictive of falls. The fallers, however, were only those who had a history of recurrent falls, which, in our opinion, indicates that the group was at greater fall risk. Thus, the BBS may better identify older adults who have greater impairments and who are at risk for falls than older adults who are in good health and more active but who also maybe at risk for falls.

Many factors affect a person's likelihood of falling. Age,. vision, muscle force, flexibility, sensation, balance, number and type of medications, cognitive impairment, and concomitant medical problems have all been associated with fall risk. (2-4,19) The purpose of our study was to determine whether data from a combination of 5 balance assessment tests--combined with data regarding fall history, number of medications, dizziness, visual problems, use of an assistive device, physical activity level, sex, and age--could predict fall risk in a group of community-dwelling older adults who were independent. The balance tests that we used were the BBS, (17,18) the Dynamic Gait Index (DGI), (11,21) the TUGT, (20) the 100% Limits of Stability Test (100% LOS), (22) and the Modified Clinical Test of Sensory Interaction for Balance (Modified CTSIB). (23,24)

Method

Subjects

One hundred six community-dwelling older adults (age range=65-90 years) who were independent were recruited from retirement communities, senior centers, the 50-Plus Wellness Program, and the general community in Sacramento, Calif. Ninety-nine subjects (60 women and 39 men) completed the study. One subject died, 1 subject declined to continue the study, and 5 subjects could not be reached for continued follow-up.

In order to participate in the study, subjects must have been able to stand for at least 5 minutes without an assistive device and to walk a minimum of 12 m (40 ft) at a time with or without an assistive device. Inclusion criteria were communicated through recruiting materials, confirmed by telephone when appointments were scheduled, and reconfirmed in the medical history questionnaire at the time of testing. People with cognitive deficits or medical or neurological problems were excluded only if the condition prevented them from meeting the inclusion criteria. No screening for dementia was done, but all subjects were able to follow directions, give appropriate responses to survey questions, and participate in the interview process without assistance. Subjects with conditions such as heart or pulmonary problems, in which mild activity could cause medical risk during the testing, were excluded from the study. Before participating in the study, all subjects signed an informed consent form that summarized the purpose of the study, explained risks and discomforts, indicated that all information gathered would remain confidential, and assured subjects that they could withdraw at any time.

Table 1 summarizes the subjects' demographic information. Subjects had a mean age of 74.02 years (SD=5.64, range=65-90). The number of medications the subjects took ranged from 0 to 10 ([bar]X=2.74, SD=2.26). Fifty-six subjects (56.6%) reported problems with dizziness, 5 subjects (5.1%) reported vision problems when using corrective lenses, 10 subjects (10.1%) reported using an assistive device, and 87 subjects (87.9%) reported being involved in regular physical exercise. Seventeen subjects (17.2%) recalled falling 3 or more times, 33 subjects (33.3%) recalled falling 1 or 2 times, and 49 subjects (49.5%) recalled no falls in the year prior to the study.

Procedure

Subjects were tested one time for the study. They were then followed for 12 months to track falls. Testing was performed in the Kinesiology/Physical Therapy Laboratory at California State University, Sacramento, by physical therapy and kinesiology faculty and by graduate and undergraduate students in physical therapy and kinesiology.

Before administering the tests, students and faculty completed 6 hours of training and practice in the correct use of the NeuroCom Balance Master 6.1, * administration and recording of all tests, and interview techniques. All were tested for the reliability of their data collection skills. We were concerned about the reliability of the data collected by all testers because the BBS and DGI require raters to make a judgment about movement quality. Interrater reliability for these tests was established by having all testers view videotapes of 3 subjects and judge movement quality using the BBS and the DGI. An 80% or better agreement between testers was achieved before proceeding with subject testing. Reliability was not assessed with the use of any statistics, such as those that correct for chance agreement or are based on probabilistic models.

Subjects were interviewed about their medical history, history of falls, and physical activity. Medical history included questions about vision (with and without corrective lenses, in poor light), dizziness experienced in the year before the study, number of medications, cardiac and pulmonary problems, use of an assistive device, and cancer. Fall history was determined by self-report of the number of falls that the subject recalled from the previous year, including the time of day, location, conditions, and injuries. A fall was defined as any disturbance of balance during routine activities that resulted in a person's trunk, knee, or hand unintentionally coming to rest on the ground, wall, table, chair, or some other surface. Physical activity was determined by self-report of the activity type and how often and how regularly the activity was performed in the year before the study. Subjects were classified as "active" if they exercised regularly at least one time per week.

After the interview, 2 subjects were tested at a time. Subjects' heights were measured to the nearest half inch. One subject was given performance-based tests (the BBS, TUGT, and DGI) followed by impairment-based tests (100% LOS and Modified CTSIB) on the NeuroCom Balance Master 6.1. The other subject was given impairment-based tests, followed by performance-based tests. Performance-based tests are functional tests that we believe identify functional limitations without necessarily identifying their causes. Impairment-based tests, which were done using the Balance Master 6.1, are tests that attempt to specifically identify which sensory systems are impaired or how motor control is impaired (eg, speed, direction of movement, accuracy of movement). Two test administrators were present for each subject's testing. One tester administered the test, and the other tester assisted primarily by standing by the subject to prevent falls during testing. All subjects wore gait safety belts throughout the balance testing.

Each subject was given a booklet at the end of the balance test session. The booklet contained a calendar for recording falls, the definition of a fall, and instructions for use of the calendar. Subjects were asked to record daily whether a fall occurred. Details of any falls, including time, location, and circumstances, were recorded on a form provided on the back of each month's page.

Telephone Survey

Follow-up contact was made by telephone or e-mail every 2 to 4 weeks during the 12 months following the balance assessment to track the subjects' fall history. Subjects referred to their fall calendars to report losses of balance. If a fall had occurred, the subjects were asked whether they were injured and whether they sought medical attention. Subjects were asked if they were using the fall calendar on a daily basis and were encouraged to continue daily recording of fall status.

Balance Tests

The balance assessment consisted of 3 performance-based balance tests (BBS, TUGT, and DGI) and 2 impairment-based tests (Modified CTSIB and 100% LOS). Both impairment-based tests were conducted using the NeuroCom Balance Master 6.1.

Berg Balance Scale. Reliability of data obtained with the BBS has been established in a previous study of 35 residents of nursing homes and 35 patients with stroke (intraclass correlation coefficient [ICC]=.97-.98). (18) Criterion validity was established in a study of 31 subjects with a mean age of 83 years. The BBS scores were correlated to the Tinetti Balance Subscale and the TUGT (r=.76-.91). (17) The BBS is designed to challenge subjects to keep their balance with an increasingly narrow base of support; the initial level is sitting, and the final level is one-leg standing. Weight shifting, turning, and reaching also are measured. Points for each item are totaled. The highest possible score is 56 points. In a clinical setting, the cutoff score to separate fallers from people who are not at risk for falling ("nonfallers") is usually 45 points. (15) We used the BBS score for statistical analysis, rather than as a cutoff value for distinguishing fallers from nonfallers.

Dynamic Gait Index. The DGI uses 8 test items to measure a person's ability to accommodate to changes in environment, speed, and head position during gait. Tasks are rated on a 3-point scale from 0 (unable) to 3 (normal execution). The highest possible score is 24. The rating is based on the person maintaining normal gait pace (a person's natural walking pace) and staying within a 38.1-cm-wide (15-in) pathway without stumbling or staggering during walking. This test is designed to demand many of the adjustments to gait that should occur when walking in the community or home, including walking with horizontal and vertical head turns, walking while speeding up and slowing down, walking over and around objects, and ascending and descending stairs. No studies measuring intrarater and interrater reliability of data obtained with the DGI as a single test have been found. We chose the DGI for our study because we believed it to be the most challenging gait mobility test available for older adults. For statistical analysis, we used the DGI score rather than a cutoff value to distinguish fallers from nonfallers.

Timed "Up & Go" Test. The TUGT is measured with a stopwatch. The subject is instructed to move from a seated position in a chair to a standing position, walk 3 m (10 ft) at a normal and safe pace, turn around, walk back to the chair, and sit down. The subject is given a practice trial followed by 2 timed trials. The 2 timed trials are averaged for each subject's score. Excellent intertester and intratester reliability of data obtained with the TUGT were established (ICC=.99 for both) in a study of 60 older adults who were frail and 10 older adults who were in good health. (20) Trueblood et al (24) felt that a cutoff time of 10 to 12 seconds separated fallers from nonfallers in a group of community-dwelling older adults. A 20-second cutoff time had previously been used when testing elderly people who were frail for independence in functional mobility. (20) In another study in which investigators chose a 14-second cutoff time, the TUGT was found to have 87% sensitivity of correctly predicting fallers and 87% specificity for correctly predicting nonfallers. (25) In our study, we did not use a cutoff time, but we used total time in the logistic regression equation.

Modified Clinical Test for Sensory Interaction on Balance. The Modified CTSIB was conducted on the NeuroCom Balance Master 6.1. (26) The Modified CTSIB examines postural sway during the 4 conditions assessed for the CTSIB: "standing on a firm surface with eyes open" (FEO), "standing on a firm surface with eyes closed" (FEC), "standing on a foam surface with eyes open" (FOEO), and "standing on a foam surface with eyes closed" (FOEC). Composite sway is the mean sway speed averaged over the 4 conditions. Each condition is tested 3 times. Although visual examination of amplitude and speed of sway is used as a measure in the CTSIB, testing with the Modified CTSIB on the Balance Master uses dual force platforms to measure amplitude, direction, and speed of movement of a person' center of pressure.

Subjects stood straight and still on a force platform during three 20-second trials in each of the 4 conditions. For each condition, each subject's feet were placed in the standard position recommended by the manufacturer of the Balance Master. (26) Foot position was monitored throughout the test. If foot placement changed, the feet were again placed in the correct position. The Modified CTSIB gives 2 sets of data collected by the computer from the 4 conditions. Data include mean center of pressure sway speed (which is measured in degrees per second) and average center of pressure position (which measures deviation of the center of pressure in degrees over 20 seconds). We used center-of-pressure speed for the 4 conditions and composite sway for statistical analysis. In a study of 12 subjects from 24 to 68 years of age ([bar]X=42.2), test-retest reliability for the Modified CTSIB using only FEO and FEC conditions was found to be high (ICC [3,4]=.91 for FEO, ICC [3,4] =.97 for FEC). (27) It should be noted, however, that the sample studied was much younger than the sample used in the present study.

100% Limits of Stability Test. A theoretical 100% LOS is established for each person by the Balance Master 6.1 software based on the person's height. This theoretical 100% LOS is the maximum angle a person of a given height should be able to sway the body over the feet without losing balance and having to take a step. It is measured by a person's ability to shift his or her center of pressure from a center point to 8 targets viewed on a computer screen that are placed around the center of pressure: front, sides, back, and 4 diagonal points. The targets represent the maximum distance the person should be able to sway the center of pressure in any direction without losing balance and having to change foot position.

The 100% LOS provides 5 sets of information as a person shifts his or her center of pressure from the center toward individual visual targets on the computer screen. Reaction time, measured in milliseconds, is the time from the computer's command to move and the initiation of movement. Movement speed, measured in degrees per second, is the average speed of movement of the center of pressure. End-point excursion is the greatest distance reached by the center of pressure in the first sustained attempt to reach the target and is expressed as a percentage of a straight line from center to target. Readjustments in position after the initial movement are not calculated in end-point excursion. Maximum excursion is the greatest distance reached by the center of pressure toward the target during the target's entire trial period. Maximum excursion is expressed as a percentage of a straight line from center to target. Directional control is a ratio of the distance of a straight line from center to target to the total distance that the subject moved. Deviation from a straight path will increase the total distance moved. Directional control is given as a percentage, with a higher percentage showing better directional control. A directional control score of 100% would mean that the subject did not deviate from a straight path. (26)

Data from all four 100% LOS measures were used in the statistical analysis. In a study of 38 community-dwelling older adults without histories of falling, test-retest reliability estimates of the 100% LOS, using the 8 targets over 3 test days, were found to be moderately high to high for movement speed, maximum excursion, and end-point excursion. (22) No differences in measurements across the 3 test days were found for movement speed (F=2.07; df=2,23; P>.10), maximum excursion (F=l.02; df=2,29; P>.25), and end-point excursion (F=4.50; df=2,17; P>.025). A study of 12 subjects between 24 and 68 years of age (X=42) demonstrated what we would consider moderate test-retest reliability of movement time to targets (the measure of speed on older versions of Balance Master programs) and path length to targets (the measure of end-point excursion on older versions of Balance Master programs) (ICC=.83 and .78, respectively). (27)

Data Analysis

Logistic regression was performed using the SPSS 10.0 for Windows program.* Forward stepwise logistic regression was used, with multiple (more than 1) falls or no multiple falls (0 or 1) as the dichotomous dependent variable in order to analyze which tests, combinations of tests, or other variables predicted falls. Because of the sample size and number of variables, the entry probability for analysis was set at the .10 instead of .05 level of significance in an effort to avoid a type II error. Number of medications, fall history, dizziness, visual problems, previous falls, physical activity level, and balance test scores were included as independent variables in the logistic regression. Kendall tau correlations were calculated to determine which dependent and independent variables were related (Tab. 2). Variables that were correlated were not placed in the same logistic regression model to prevent interaction that could confound the analysis. Variables that seemed logical based on previous research, correlation values, and clinical reasoning were used for different models. If variables were correlated--and, therefore, not placed into regression models together--different combinations of variables were used until all possible combinations of noncorrelated variables were entered into the regression models. After models were suggested by the forward stepwise regression, logistic regression calculations were run, and each variable was entered one at a time to determine the contribution of each variable to the predictive value of the model. Logistic regression also was used to determine whether various balance tests or combinations of tests could predict injurious falls.

Results

Reported Falls

Because of a lack of personnel, 30 subjects were not called for a 3-month period. When contact was resumed, subjects were asked to refer to their calendars to report fall status.

In the 12 months following testing, 22 of the 42 fallers (52%) fell only once, 11 (26%) fell twice, and 9 fell 3 or more times (Tab. 3). Fifty-seven subjects (57.6%) reported no falls as compared with 42 subjects (42.4%) who reported 1 or more falls (mean for the entire group=0.91 falls per person, SD= 1.67) (Tabs. 3 and 4). Of those who fell in the year following testing, the mean number of falls per person was 2.14 (SD=2, median=1) (Tab. 3). Multiple falls (2 or more falls) were reported by 20 (20.2%) of the entire subject group (Tab. 4). Only 4 subjects (4.0%) sustained falls with injuries serious enough to seek medical care (Tab. 4). Because of the small number of subjects who fell more than twice (9.1%), we used a multiple falls category (2 or more falls). For logistic regression analysis, multiple falls was defined as 2 or more falls, and nonmultiple falls was defined as 0 or 1 fall.

Table 2 shows the Kendall tau correlations between the test variables. Highly correlated test scores were not placed together into logistic models. In fall prediction calculation, both forward stepwise and the enter method logistic regression gave similar models. Logistic regression revealed that the FEC portion of the Modified CTSIB, the speed of center of pressure sway while a subject stands on a firm surface with eyes closed, was predictive of multiple falls (Tab. 5). A model containing FEC and the covariates of age and sex also resulted in a predictive model, although adding age and sex to FEC did not improve the predictive ability of the model (Tab. 5). Closer scrutiny of FEC scores revealed one score of 1.3[degrees]/s that fell more than 2 standard deviations (SD=0.186) from the mean score of 0.381[degrees]/s. After this extreme score was removed from the data, we found no model of significance. The FEC score, however, was consistent with the other Modified CTSIB scores for this subject; therefore, the score was included in the data for analysis. Logistic regression analysis revealed no other variables or combinations of variables that predicted multiple fall or nonmultiple fall status.

Use of assistive device. Logistic regression analysis did not indicate that the use of an assistive device was predictive of falls. The small number of people using assistive devices in our sample (10 subjects) may have affected the results.

Injurious foils. Only 4 subjects reported injurious falls in the study (Tab. 4). Dizziness problems were found to be predictive of injurious falls in the logistic regression. A 96% correct prediction rate was found, although none of the 4 fallers were correctly identified.

Discussion

The rate of multiple falls reported by our subjects was consistent with previous reports. (21,24,28) Trueblood et al (24) reported a 16.5% fall rate compared with the 20.2% rate that we found, but Trueblood et al followed their subjects for only 6 months. Our results suggest the possibility that older adults are falling even more often than is generally reported. In the literature, one third to one half of community-dwelling older adults fall 1 or more times in a year. (1-3) Forty-two percent of the subjects in our study reported at least 1 fall, which is a fairly high proportion of the sample population. Whether our sample was unique in terms of fall status or whether the large number of reported falls was due to the frequency with which we tracked falls is not known. Use of a fall report calendar and the frequent follow-up may have resulted in a more accurate report of falls than in studies with more infrequent follow-up.

In another prospective study of falls in which subjects were contacted weekly for 12 months, researchers found a similar fall rate: 40% of the subjects reported at least 1 fall. (29) Older adults may fall and forget that they have fallen. Weekly follow-up of 304 older adults over a 12 month period by Cummings and colleagues (30) showed that 13% to 32% of those with confirmed falls did not recall falling 3, 6, or 12 months after the fall, and only a weak correlation was found between falls and recalled falls. We believe that the possibility of recall errors underscores the importance of prospective study design to predict fall risk.

Fall Prediction Tests.

We expected that falls would be predicted by some combination of demographic and health covariates and the 5 commonly used tests that were selected. In our opinion, the poor ability to predict falls from data obtained with these tests suggests that new screening tests are needed for community-dwelling older adults who are active. Although logistic regression tests showed a weak ability to predict multiple falls using FEC, the FEC, after removal of an extreme score, did not predict falls. The score, 1.3[degrees]/s of sway, fell greater than 2 standard deviations from the mean FEC score. All other FEC scores ranged from 0.1[degrees]/s to 0.9/s, with 85% of scores falling between 0.2[degrees]/s and 0.5[degrees]/s ([bar]X=0.371, SD=0.1618). The ability of the model to predict falls was weak even with the extreme score included (Tab. 5). Although logistic regression attributes were statistically significant, clinical relevance is doubtful because only 1 of 20 falls was correctly predicted.

Trueblood et al (24) found FEO to be predictive of falls. The mean FEO score was 0.47/s for fallers and 0.136[degrees]/s for nonfallers. The standard deviation was 0.304[degrees]/s for fallers and 0.118[degrees]/s for nonfallers (N=179). With a standard deviation of 0.118[degrees]/s to 0.304/s, a difference of 0.11[degrees]/s between fallers and nonfallers seems quite small. In our opinion, differences as small as 0.10[degrees]/s (the difference between means for FEC for multiple and nonmultiple fallers in our study) and 0.11[degrees]/s (the difference between means for FEO for fallers and nonfallers in the study by Trueblood et al) are not clinically meaningful.

We selected the BBS for this study because it is designed to narrow the subject's base of support to a single-leg stance. Some researchers (17,21) found better fall prediction using the BBS than we did, but subjects in those studies were known to be frail, were residents of a nursing home, or had an unknown physical activity status. In other studies of community-dwelling older adults, (15,16) the BBS was not found to be a good predictor of falls. O'Brien et al (16) tested 49 subjects, 13 of whom reported 1 or more falls in the year before testing. Their subjects were recruited from family medicine clinics, geriatric day hospitals, senior centers, and home care programs. Because some of the locations where they recruited subjects tend to be used by older adults who need assistance, these subjects may have been more frail than the subjects involved in our study. The mean BBS score for fallers who fell 1 or more times in the study by O'Brien et al was 45 (range-21-54) as compared with a mean BBS score of 53.18 (range=46-56) in our study. The mean BBS score of nonfallers in the study by O'Brien et al was 55 (range=46-56) as compared with a mean score of 53.15 (range=34-56) in our study. The range for nonfallers was much narrower (10 points) in the study by O'Brien et al than in our study (22 points). We found virtually no difference in BBS scores between the groups, whereas O'Brien et al found a 10-point difference. Even with a less fit population and a greater difference in scores between fallers and nonfallers than our study, O'Brien et al found poor sensitivity (54%) for correctly predicting fallers using the BBS.

Examination of the individual scores of the multiple fallers in our study showed that many of them did quite well on the BBS and other performance-based tests. For instance, one subject scored 55/56 on the BBS but fell 9 times. Two subjects scored 54/56 and 56/56 on the BBS, and each subject fell 4 times. Some subjects may have been very active and engaged in more risky activities. The 3 subjects just mentioned were very active people in their early to middle 70s. One woman did weight training and used a stationary bicycle and a ski machine regularly. One subject swam competitively and participated in the Senior Olympics. The BBS was not sensitive enough to uncover factors that contributed to falls in these older adults who were active. Conversely, 3 subjects who scored lowest on the BBS--1 who scored 34/56 and 2 who scored 39/56--had no falls. At least one of these subjects participated only in activities of daily living. She did not exercise and was less likely to go into high-risk situations. Again, we are faced with the multifactorial nature of falls. Although physical activity is important in maintaining balance function, people who are active are more likely to engage in activities that put them at greater risk for falls. Speechley and Tinetti (4) found that older adults who were vigorous had a lower incidence of falls (17%) compared with older adults who were frail (52%), but they were more likely to fall on stairs and away from home, situations encountered less frequently by older adults who are frail.

Scores on the TUGT were not predictive of falls. A 10- to 12-second cutoff time to differentiate fallers from nonfallers has been recommended (24) as has a 13.5-second cutoff time. (25) The times of nonfallers in our study clustered around 8 to 10 seconds, and the times of multiple fallers clustered around 9 to 13 seconds. Many of the multiple fallers had times that were below commonly used cutoff times. Although a cutoff time would be difficult to find in this population, investigators should try to determine whether a lower (10-second) cutoff time should be established for fall risk in community-dwelling older adults who are active.

In a prospective study similar in design to our study, 100 older adults were tested using the BBS, the Functional Reach Test, a step-up test, lateral reach, FEO and FEC measures, and an LOS test in an effort to predict falls over a 1-year period. (28) The best impairment-based test, step time during a step-up task, gave only 66% predictive ability. The best combination of factors gave a 77% correct prediction.

Factors Affecting Fall Prediction and Fall Risk

Falls are known to be multifactorial, (4,19) and a complex interaction among different factors determines whether a person is at risk for falls. (19,31) A model that included the covariates age and sex with FEC did not improve predictive ability of the model (Tab. 5). Although the sample size was too small for analysis of subgroups based on age, sex, and number of medications, exploratory analysis of subgroups showed that FEC was more predictive of falls in men, younger subjects, and those taking 2 or more medications (see footnote in Tab. 5). This preliminary information suggests that an interaction between FEC and the covariates of age, sex, or number of medications may be important in predicting falls. Based on the data we gathered, we used nQuery Advisor 3.0 software ([double dagger]) to estimate that a sample size of 182 subjects would be necessary to evaluate a model with 80% power, assuming an odds ratio of 2.0 for FEC, while controlling for age (alpha=.05, 2-sided test). In the future, if this or a similar study were done with a larger sample size, subjects could be grouped by sex, age, and number of medications to determine whether multiple-fall prediction results vary in the different groups. The fall screening tests that effectively predict falls may differ from group to group.

Many factors affecting fall prediction remain to be identified. If these factors can be identified, health care professionals might be able to discern what tests should be used in different situations. Some tests, for example, may be appropriate for an active rather than an inactive population or a relatively younger rather than older population.

In a study of the effect of physical activity on balance and falls, (32) the researchers found that older adults who were active performed better than older adults who were inactive on some balance tests, including the sharpened Romberg test (active subjects: [bar]X=59.4-60.0 seconds, standard error [SE] =0.0-0.5 seconds; inactive subjects: [bar]X=41.5-41.8 seconds, [SE]=6.1-7.2 seconds) and one-leg stance time (active subjects: [bar]X=40.0-55.1 seconds, SE=3.4-4.5 seconds; inactive subjects=27.5-33.0 seconds, SE=6.1-7.1 seconds). In a study related to ours that used the same sample, physical exercise correlated with improved scores on many balance tests (Boulgarides et al, unpublished data, December 2001). The lack of significance in fall prediction of these balance tests may have been influenced by more high-risk activities that were engaged in by some of the higher scoring subjects. Although the subjects who were active scored well on balance tests, they also engaged in more activities that might result in falls. Some descriptions of falls after testing included falling off a bicycle, falling while running backward in a tennis game, and falling while getting off a bus while touring Europe. Only 4 subjects had injurious falls. Falls without injuries may not be of as great a concern as falls with injuries. Whether the long-term effects of noninjurious falls will be harmful is not known. Falls of any type we believe should not be underestimated, particularly because the effect may be to decrease confidence and thus a person's willingness to continue an active lifestyle. (6,7)

In general, multiple fallers and nonmultiple fallers both scored very well on many of the balance tests. The small sample size, the large number of subjects who were physically active, and the high level of subject performance may have affected the results of the statistical tests. Multiple falls did occur in this population, but these falls could not be predicted by the tests. Because many subjects scored very high on the BBS, the DGI, and TUGT, a ceiling effect may have occurred, indicating that these performance-based tests are not suitable for older adults who are high functioning, even when they are at risk for falls. Different, possibly more challenging, performance-based tests might reveal balance deficits that could cause falls in people who are high functioning. The development of new tests for this population is indicated.

Limitations of the Study

With only 99 subjects, the power of the logistic regression calculations was determined to be 38% for the FEC model using nQuery Advisor 3.0 software. Statistical power of logistic regression tests would have been greater with a larger sample size. Similar studies should be done with larger sample sizes. Because the population studied was quite homogeneous, a larger sample size is necessary to give the power necessary to show significance. A larger sample size would allow tests of the interaction of factors such as age, number of medications, and sex with independent variables in a population of older adults who are active.

The physical activity level of the subjects in our study probably does not represent the current population of older adults in this country, and this may be part of the reason that results of our study differ from some other studies. Eighty-four percent of the subjects in our study reported performing some type of physical exercise, which included exercising at least one time per week. The percentage of adults over 65 in the general population who participate in regular exercise is reported to be between 30% and 50%, decreasing to between 15% and 20% of people over the age of 85 years. (8) Because of the recruitment of many subjects from a 50 Plus Wellness program, activity and fitness levels might be expected to be higher than in the general population. Many subjects were very active--dancing, walking, swimming, playing golf or tennis, and competing in sports.

Conclusions

Five currently used balance tests combined with health and demographic factors did not predict falls in a sample of community-dwelling older adults who were active and independent. Two conclusions can be drawn from the current study. The results reinforce the notion that fall risk depends on an interaction of many factors and is particularly difficult to predict in older adults who live active independent lives. The factors contributing to falls may interact differently at different ages and activity levels. A larger sample size than the one used in this study may reveal the interaction among factors contributing to fall risk. It is also evident that current tests are not as successful in predicting fall risk in active older adults as they have been found to be in more frail populations. This study indicates the need for development and testing of new assessment tools for an increasingly active aging population.

Table 1.

Subject Characteristics (N=99)

                                           n       %       [bar]X

Age (y)                                                     74.02
No. of medications                                           2.74
Sex
  Female                                   60     61
  Male                                     39     39
Reported dizziness
  No                                       43     43.4
  Yes                                      56     56.6
Reported vision problems (corrected)
  No                                       94     94.9
  Yes                                       5      5.1
Use of assistive device
  No                                       89     89.9
  Yes                                      10     10.1
History of falls
  0                                        49     49.5
  1-2                                      33     33.3
  3 or more                                17     17.2
Regular physical activity
  No regular exercise                      12     12.1
  Regular exercise                         87     87.9

                                          SD     Median    Range

Age (y)                                  5.64      74      65-90
No. of medications                       2.26       2       0-10
Sex
  Female
  Male
Reported dizziness
  No
  Yes
Reported vision problems (corrected)
  No
  Yes
Use of assistive device
  No
  Yes
History of falls
  0
  1-2
  3 or more
Regular physical activity
  No regular exercise
  Regular exercise

Table 2.

Kendall Tau Correlation Matrix of Balance Variables With
Multiple Falls (a)

            M Falls      MVL          EPE          MXE

M Falls     1.000        -.017        -.132        -.083
MVL                      1.000         .326 (b)     .219 (b)
EPE                                   1.000         .732 (b)
MXE                                                1.000
DCL
RT
FEO
FEC
FOEO
FOEC
Comp
Dizzy
Glauc
Meds
Fall H
P Act
BBS
FRT
TUGT
DGI

            DCL          RT           FEO          FEC

M Falls     -.064         .000         .038         .145
MVL         -.069        -.296 (b)     .151         .066
EPE          .463 (b)    -.217 (b)    -.058        -.101
MXE          .523 (b)    -.234 (b)    -.049        -.093
DCL         1.000        -.003        -.224 (b)    -.178 (c)
RT                       1.000        -.103        -.083
FEO                                   1.000         .612 (b)
FEC                                                1.000
FOEO
FOEC
Comp
Dizzy
Glauc
Meds
Fall H
P Act
BBS
FRT
TUGT
DGI

            FOEO         FOEC         Comp        Dizzy

M Falls      .060        -.089         .112        -.067
MVL         -.025        -.094         .032         .055
EPE         -.255 (b)    -.214 (c)    -.112         .118
MXE         -.207 (b)    -.196 (c)    -.076         .094
DCL         -.312 (b)    -.219 (b)    -.122         .015
RT           .036         .122         .045        -.006
FEO          .323 (b)     .132         .451 (b)    -.110
FEC          .218 (b)     .231 (b)     .392 (b)    -.191 (c)
FOEO        1.000         .311 (b)     .381 (b)    -.053
FOEC                     1.000         .217 (c)    -.050
Comp                                  1.000        -.226 (c)
Dizzy                                              1.000
Glauc
Meds
Fall H
P Act
BBS
FRT
TUGT
DGI

            Glauc        Meds         Fall H       P Act

M Falls      .072         .054        -.009        -.121
MVL          .094        -.121         .128         .124
EPE          .074        -.102         .176c        .103
MXE          .082        -.073         .127         .044
DCL          .088        -.024         .048         .065
RT          -.084         .012        -.123        -.002
FEO         -.158         .075        -.044        -.145
FEC         -.135         .102        -.078        -.122
FOEO        -.146         .093        -.156        -.194 (b)
FOEC        -.206 (c)     .132        -.159        -.107
Comp        -.126         .162 (c)    -.127        -.201 (c)
Dizzy        .164        -.169         .216 (c)     .237 (c)
Glauc       1.000        -.109         .042         .387 (b)
Meds                     1.000        -.260 (b)    -.087
Fall H                                1.000         .098
P Act                                              1.000
BBS
FRT
TUGT
DGI

            BBS          FRT          TUGT         DGI

M Falls     -.013         .013         .143        -.024
MVL          .129         .130        -.177 (c)     .195 (b)
EPE          .308 (b)     .180 (c)    -.293 (b)     .311 (b)
MXE          .313 (b)     .182 (c)    -.263 (b)     .278 (b)
DCL          .347 (b)     .138        -.279 (b)     .225 (b)
RT          -.082         .019         .099        -.091
FEO         -.256 (b)    -.101         .138        -.193 (c)
FEC         -.234 (b)    -.125         .138        -.211 (b)
FOEO        -.376 (b)    -.277 (b)     .328 (b)    -.317 (b)
FOEC        -.202 (c)    -.051         .252 (b)    -.373 (b)
Comp        -.093        -.070         .173 (c)    -.227 (b)
Dizzy        .065        -.075        -.196 (c)     .149
Glauc        .181 (c)     .106        -.170         .175 (c)
Meds        -.171 (c)    -.138         .213 (b)    -.163 (c)
Fall H       .210 (c)     .086        -.207 (c)     .264 (b)
P Act        .167         .068        -.201 (c)     .182 (c)
BBS         1.000         .399 (b)    -.431 (b)     .457 (b)
FRT                      1.000        -.313 (b)     .222 (b)
TUGT                                  1.000        -.435 (b)
DGI                                                1.000

(a) M Falls=multiple falls (2 or more falls), MVL=movement speed,
EPE=end-point excursion, MXE=maximum excursion, DCL=directional
control, RT=reaction time, FEO="standing on a firm surface with eyes
open," FEC="standing on a firm surface with eyes closed,"
FOEO="standing on a foam surface with eyes open," FOEC="standing on a
foam surface with eyes closed," Comp=composite sway, Dizzy=reports of
dizziness, Glauc=glaucoma, Meds=number of medications, Fall H=history
of falls, P Act=physical activity, BBS=Berg Balance Scale,
FRT=Functional Reach Test, TUGT=Timed "Up & Go" Test, DGI=Dynamic Gait
Index.

(b) Significant at P=.01.

(c) Significant at P=.05.

Table 3.

Number of Reported Falls per Subject

No. of Falls                No. of
per Subject                 Subjects      %

 0                           57          58
 1                           22          22
 2                           11          11
 3                            3           3
 4                            2           2
 5                            2           2
 9                            1           1
10                            1           1
                             99         100

Average falls/subject (N=99)=0.91, SD=1.67, Median=0
Average falls/subject who fell (n=42)=2.14, SD=2.0,
  Median=1

Table 4.

Summary of Fall Frequency (N=99)

Falls Occurrence               n          %

No falls                       57       57.6
1 or more falls                42       42.4
Total                          99      100

Multiple Falls
Zero or one fall               79       79.8
Two or more falls              20       20.2
Total                          99      100

Fall Severity
No or noninjurious falls       95       96
Injurious falls                 4        4
Total                          99      100

Table 5.

Logistic Regression: Prediction of Multiple Falls (a)

                                                           Correctly
Models Comprising                                          Predicted
Independent and      Level of            % Correct         Nonmultiple
Control Variables    Significance        Prediction        Fallers

Model                .042                80.8              79/79
  FEC

Model                .045                80.8              78/79
  FEC
  Age
  Sex

Models Comprising    Correctly
Independent and      Predicted           Coefficient of
Control Variables    Multiple Fallers    Terms (B)         SE of B

Model                1/20
  FEC                                    2.556             1.285

Model                2/20
  FEC                                    1.786             1.332
  Age                                    0.072             0.048
  Sex                                    0.822             0.540

Models Comprising                        95% CI for B
Independent and      Odds Ratio
Control Variables    (Exp B)             Lower             Upper

Model
  FEC                12.884              1.039             159.830

Model
  FEC                 5.967              0.438              81.146
  Age                 1.075              0.978               1.181
  Sex                 2.275              0.790               6.557

(a) In testing population subgroups, FEC was predictive of:
(1) multiple falls in men (P=.033, n=39) but not women (P=.840, n=60),
(2) multiple falls in those subjects who were 74 years of age or
younger (P=.005, n=55) but not in those subjects who were older than
74 years of age (P=.856, n=44), and (3) multiple falls in those
subjects taking 2 or more medications (P=.029, n=66) but not in those
subjects taking fewer than 2 medications (P=.169, n=33). Because of
the small sample sizes, caution is warranted in the interpretation of
subgroup results.

* NeuroCom International, 9570 SE Lawnfield Rd, Clackamas, OR 97015.

([dagger]) SPSS Inc, 233 S Wacker Dr, Chicago, IL 60606-6307.

([double dagger]) Statistical Solutions Ltd, South Bank, Crosse's Green, Cork, Ireland.

References

(1) Sattin RW. Falls among older persons: a public health perspective. Annu Rev Public Health. 1992;13:489-508.

(2) Tinetti ME, Doucette J, Claus E, Marottoli R. Risk factors for serious injury during falls by older persons in the community. J Am Geriatr Soc. 1995;43:1214-1221.

(3) Tinetti ME, Inouye SK, Gill TM, Doucette JT. Shared risk factors for falls, incontinence, and functional dependence: unifying the approach to geriatric syndromes. JAMA. 1995;273:1348-1353.

(4) Speechley M, Tinetti M. Falls and injuries in frail and vigorous community elderly persons. J Am Geriatr Soc. 1991;39:46-52.

(5) Ellis AA, Trent RB. Do the risks and consequences of hospitalized fall injuries among older adults in California vary by type of fall? J Gerontol A Biol Sci Med Sci. 2001;56:M686-M692.

(6) Legters K. Fear of falling. Phys Ther. 2002;82:264-272.

(7) Cumming RG, Salkeld G, Thomas M, Szony G. Prospective study of the impact of fear of falling on activities of daily living, SF-36 scores, and nursing home admission. J Gerontol A Biol Sci Med Sci. 2000;55: M299-M305.

(8) Shephard RJ. Aging, Physical Activity and Health. Champaign, Ill: Human Kinetics; 1997.

(9) Steinberg M, Cartwright C, Peel N, Williams G. A sustainable programme to prevent falls and near falls in community dwelling older people: results of a randomised trial. J Epidemiol Community Health. 2000;54:227-232.

(10) Yan J. Tai Chi practice improves senior citizens' balance and arm movement control. Journal of Aging and Physical Activity. 1998;6: 271-284.

(11) Shumway-Cook A, Gruber W, Baldwin M, Liao S. The effect of multidimensional exercises on balance, mobility, and fall risk in community-dwelling older adults. Phys Ther. 1997;77:46-57.

(12) Verfaillie D, Nichols J, Turkel E, Hovell M. Effects of resistance, balance, and gait training on reduction of risk factors leading to falls in elders. Journal of Aging and Physical Activity. 1997;5:215-228.

(13) Province MA, Hadley EC, Hornbrook MC, et al. The effects of exercise on falls in elderly patients: a preplanned meta-analysis of the FICSIT trials. JAMA. 1995;273:1341-1347.

(14) Rooks D, Ransil B, Hayes W. Self-paced exercise and neuromotor performance in community-dwelling older adults. Journal of Aging and Physical Activity. 1997;5:135-149.

(15) Bogle Thorbahn LD, Newton RA. Use of the Berg Balance Test to predict falls in elderly persons. Phys Ther. 1996;76:576-582.

(16) O'Brien K, Pickles B, Culham E. Clinical measures of balance in community-dwelling elderly female fallers and non-fallers. Physiotherapy Canada. Summer 1998:212-221.

(17) Berg KO, Wood-Dauphinee SL, Williams JI, Maki B. Measuring balance in the elderly: validation of an instrument. Can J Public Health. 1992;83(suppl 2):S7-S11.

(18) Berg K, Wood-Dauphinee S, Williams JI. The Balance Scale: reliability assessment with elderly residents and patients with an acute stroke. Scand J Rehabil Med. 1995;27:27-36.

(19) Tinetti ME, Williams TF, Mayewski R. Fall risk index for elderly patients based on number of chronic disabilities. Am J Med. 1986;80: 429-434.

(20) Podsiadlo D, Richardson S. The timed "Up & Go": a test of basic functional mobility for frail elderly persons. J Am Geriatr Soc. 1991;39: 142-148.

(21) Shumway-Cook A, Baldwin M, Polissar NL, Gruber W. Predicting the probability for falls in community-dwelling older adults. Phys Ther. 1997;77:812-819.

(22) Clark S, Rose DJ, Fujimoto K. Generalizability of the limits of stability test in the evaluation of dynamic balance among older adults. Arch Phys Med Rehabil. 1997;78:1078-1084.

(23) El-Kashlan HK, Shepard NT, Asher AM, et al. Evaluation of clinical measures of equilibrium. Larynngoscope. 1998;108:311-319.

(24) Trueblood PR, Hodson-Chennault N, McCubbin A, Youngclarke D. Performance and impairment-based assessments among community dwelling elderly: sensitivity and specificity. Issues on Aging. 2001;24(1): 2-6.

(25) 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;80:896-903.

(26) Manual for the Balance Master 6.1: The Objective Quantification of Daily Life Tasks. Clackamas, Ore: Neurocom International Inc; 1996.

(27) Hageman PA, Leibowitz JM, Blanke D. Age and gender effects on postural control measures. Arch Phys Med Rehabil. 1995;76:961-965.

(28) Brauer SG, Burns YR, Galley P. A prospective study of laboratory and clinical measures of postural stability to predict community-dwelling fallers. J Gerontol A Biol Med Sci. 2000;55:M469-M476.

(29) Hausdorff JM, Rios DA, Edelberg, HK. Gait variability and fall risk in community-living older adults: a 1-year prospective study. Arch Phys Med Rehabil. 2001;82:1050-1056.

(30) Cummings SR, Nevitt MC, Kidd S. Forgetting falls: the limited accuracy of recall of falls in the elderly. J Am Geriatr Soc. 1988;36: 613-616.

(31) Herndon J, Helmick C, Sattin R, et al. Chronic medical conditions and risk of fall injury events at home in older adults. J Am Geriatr Soc. 1997;45:739-743.

(32) Bulbulian R, Hargan ML. The effect of activity history and current activity on static and dynamic postural balance in older adults. Physiol Behav. 2000;70:319-325.

LK Boulgarides, PT, MS, is Lecturer, Kinesiology and Health Science Department, California State University, Sacramento, 6000J St, Sacramento, CA 95819-6073 (USA) ([e-mail]), and Faculty Director, CSUS LIFE Center for Senior Fitness and Wellness. Address all correspondence to Ms Boulgarides.

SM McGinty, PT, EdD, is Chair, Department of Physical Therapy, California State University, Sacramento.

JA Willett, PhD, ATC, is Associate Professor, Kinesiology and Health Science Department, California State University, Sacramento.

CW Barnes, PhD, is Professor, Department of Sociology, California State University, Sacramento, and Director, CSUS Institute for Social Research.

All authors provided concept/research design. Ms Boulgarides, Dr McGinty, and Dr Willett provided writing, data collection, project management, fund procurement, and subjects. Ms Boulgarides, Dr McGinty, and Dr Barnes provided data analysis. Dr McGinty and Dr Willett provided facilities/equipment and consultation (including review of manuscript before submission). Ms Boulgarides and Dr McGinty provided institutional liaisons. Dr McGinty provided clerical/secretarial support. The authors thank Edward Barakatt for assistance with statistical analysis and Heather Carling-Smith and Patricia A Harris for their contributions to data collection. The authors also thank the baccalaureate and post-baccalaureate students from the Department of Kinesiology and Health Science and the Department of Physical Therapy, California State University, Sacramento, for their assistance with subject recruitment and data collection.

This study was approved by the California State University, Sacramento, Committee for Protection of Human Subjects.

This study was supported by a grant from the California State University, Sacramento, Research and Creative Activity Committee.

This article was submitted May 8, 2002, and was accepted November 22, 2002.