Predicting Which Older Adults Will or Will Not Fall Using the Fullerton Advanced Balance Scale

Methods 

This study formed part of a larger research project conducted through the Center for Successful Aging and the Fall Prevention Center of Excellence at California State University, Fullerton. The Center for Successful Aging Fall Prevention Dataset was comprised of 307 community-residing older adults who had previously participated in a single community-based fall risk screening session conducted by trained personnel from the Fall Prevention Center of Excellence at California State University, Fullerton. The dataset was examined and sorted according to predetermined eligibility criteria. To be eligible for inclusion in the present study, participants had to be 65 years or older, independently residing in the community, of normative cognition, able to ambulate independently for a distance of 60m, not diagnosed with a progressive neurologic condition or severe musculoskeletal condition, and to have sustained 2 or more falls in the past year or not to have sustained a fall in the past year. For the purpose of this study, a fall was defined as “an event resulting in an individual unintentionally coming to rest on the ground, floor, or other lower level, not as the result of a major intrinsic event or overwhelming hazard.”22(p1702) Normative cognition was determined by the participant's ability to complete all screening forms, questionnaires, and assessments with no or minimal assistance. Older adults were excluded from the study if they were currently participating in a balance-training program. All participants read, understood, and signed an informed consent form approved by the Institutional Review Board at California State University, Fullerton, prior to participating in the larger research study. A total of 192 community-residing older adults aged 65 years and older who were included in the original dataset met the inclusion criteria associated with the present study.

During the larger research study, participants attended a single, 90-minute fall-risk screening session being conducted at a community-based facility. During registration, participants were asked to read and complete an informed consent form and all required questionnaires. Once all forms were completed, participants were randomly assigned to 1 of 4 testing stations. Trained personnel from the Fall Prevention Center of Excellence administered all questionnaires and physical performance assessments. At the end of the screening session, participants received feedback concerning their test results. Participants were also provided with a packet of informational materials on preventing falls and an exercise brochure and resistance band.

During the fall-risk screening session, we administered 7 physical performance tests (ie, FAB scale, 30-foot walk at preferred and maximum speed,23 8-foot Up-and-Go test, 30-second arm curl and chair-stand tests,24 Limits of Stability, and Modified Clinical Sensory Integration Test). Although 7 performance tests were administered for the purposes of the larger research project, the only performance test data used as secondary data for the present study was the FAB scale. The HAQ was also obtained for secondary data use.

Results 

A dataset composed of 133 older adults who had reported no history of a fall in the previous 12 months (nonfallers) and 59 who reported falling 2 or more times (recurrent fallers) was used to examine the predictive properties of the FAB scale relative to faller status. Important demographic characteristics for the faller and nonfaller groups are presented in table 1. There was no statistical group difference between fallers and nonfallers based on the significance values that were calculated using independent t tests for continuous variables and Pearson chi-square tests for dichotomous variables for each of the demographic variables. Significance values (P values) for analyses are also included in table 1. The frequency distribution table for the total scores on the FAB scale is presented in appendix 1. None of the subjects in the present study scored 40 out of 40 on the FAB scale.

Table 1. Demographic Characteristics of Study Sample

	Characteristics	Nonfallers (n=132)	Recurrent Fallers (n=59)	P⁎
	Age	76.5±6.08(65–91)	78±7.70(65–91)	.061
	Falls	0±0(0)	3±1.20(2–9)	<.001†
	Number of medications	3.6±2.94(0–16)	4.6±3.31(0–14)	.161
	Days active	3±2.40(0–7)	2±2.33(0–7)	.785
	CPF scale	20±4.12(8–24)	18±4.38(5–24)	.356
	Total FAB score (out of 40)	25±6.71(6–40)	20±7.34(4–34)	.192

NOTE. Values are mean ± SD (range). Abbreviation: CPF scale, composite physical function scale.

⁎ Significance was calculated using independent t test for continuous variables and Pearson chi-square tests for dichotomous variables for each of the demographic variables. † Significant values.

The results of the binary logistic regression analysis indicated that a test of the full model was significantly reliable (N=192, χ12=21.038, P<.001) in that it differed from the constant model (intercept only model). It also indicated that the total FAB scale score was predictive of faller status. The Hosmer-Lemeshow test,28 a deciles-of-risk statistic, was used to formally evaluate goodness-of-fit. The test was nonsignificant, indicating that the model assessed was a good model. The majority of the fallers were represented in the higher deciles of risk when compared with the nonfallers. The overall prediction success rate was 71.4% with 90.2% (n=120 of 133) of nonfallers being correctly classified but only 28.8% (n=17 of 59) of recurrent fallers correctly classified.

The logistic regression also produced an OR of .902 for the likelihood of sustaining a fall. An OR of .902 indicates that for every 1-point increase in the total FAB score there is an 8% decrease in the older adult's likelihood of sustaining a fall. The probability of falling was calculated for every possible score on the FAB scale (fig 1). An inverse linear relationship between the total FAB scale score and the probability of falling was evident. As the total score on the FAB scale increased, the probability of sustaining a fall decreased. For example, a total score of 5 on the FAB scale was associated with a 72.4% probability of falling, whereas a total score of 35 was associated with a 10.7% probability of falling.

View Large Image Download to PowerPoint

Fig 1. Predicted probability of falling as a function of total FAB scale score.

Sensitivity and specificity values were calculated for every possible 5-point increment in total FAB scale score. The obtained values were then plotted against each other (sensitivity on the y-axis and specificity on the x-axis) to produce an ROC curve (fig 2). The graph was then visually analyzed to find the cut-off point at which the total FAB scale score produced the highest sensitivity and specificity values. The ROC curve analysis showed that a cut-off score of 25 of 40 on the FAB scale produced the highest sensitivity (74.6%) and specificity (52.6%) in predicting faller status (table 2).

View Large Image Download to PowerPoint

Fig 2. ROC analysis. Sensitivity and specificity of the total FAB scale score as a function of faller status.

Table 2. Sensitivity and Specificity of Possible Cut-Off Scores (in 5-Point Increments) for the Total FAB Scale Score

	Possible Cut-Off Score	Specificity (%)	Sensitivity (%)
	5	100	3.4
	10	97.0	11.9
	15	88.7	30.5
	20	75.2	50.8
	25⁎	52.6⁎	74.6⁎
	30	18.8	94.9
	35	2.3	100

⁎ Indicates the optimal cut-off score that produces the best specificity and sensitivity values.

One significant discriminant function was obtained (Wilks' λ=0.865; χ2=26.723, P<.003), which accounted for 100% of the between-group variability. The functions at group centroids further indicated that the discriminant function maximally discerned recurrent fallers (−.588) from nonfallers (.263). The structure matrix of the pooled within-groups correlation between predictor variables and standardized canonical discriminants was used to determine which predictor variables were included in the discriminant function. Although the discriminant function was comprised of the variables with loading values of 0.3 or higher (a commonly used cut-off), a more stringent cut-off value of 0.5 or higher was used in the present study. The strongest predictor variables in discerning faller status were item 4 (up and over the bench, .862), item 7 (standing on foam with eyes closed, .593), item 8 (2-footed jump, .580), item 5 (tandem walk, .562), and item 6 (single leg stance, .529).

In order to determine the item scores associated with fallers and nonfallers, the group statistics were used. The group statistics, which indicate the mean score associated with each of the strongest predictor variables, were used to report the mean scores for recurrent fallers and nonfallers for each significant individual FAB scale test item. Recall that each individual test item is scored using a 5-point ordinal scale (0–4) with higher numbers on each test item indicating better performance.

The classification results indicated that the discriminant functions comprised of the 5 predictive items of the FAB scale correctly classified 72.3% of the original cases. Because there were unequal sample sizes for the faller and recurrent faller groups, the a priori probabilities were used to adjust for the groups being influenced by sample size, because the larger group (nonfallers) may have a higher probability than the smaller group (recurrent fallers) of being classified correctly.28 The discriminant function correctly classified 88.6% (n=117 of 132) of the nonfallers, but only 36% (n=21 of 59) of the recurrent fallers.

Discussion 

The objective of the present study was to investigate the predictive properties of the FAB scale relative to faller status. As the results of the binary logistic regression analysis indicated, the FAB scale is a predictive measure of faller status among independently functioning older adults (P<.0001). In fact, an inverse relationship was evident between the total FAB scale score and the probability of falling. Specifically, every 1-point decrease in the total FAB scale score was associated with an 8% increase in the older adult's probability of sustaining a fall. At the optimal cut-off score of 25 out of 40 on the FAB scale, a moderately high level of sensitivity (74.6%) but lower level of specificity (52.6%) was evident. Although a little more than 7 out of 10 actual recurrent fallers were correctly classified, only about 5 out of 10 older adults with no history of falls were correctly classified in the present study sample.

Using a similar analysis to evaluate the predictive properties of the BBS, Shumway-Cook et al17 reported a sensitivity of 77% and specificity of 86% using a cut-off score of 49 out of 56. Although the sensitivity and specificity values obtained using the BBS were higher when compared with those in the present study, the sample used in the previous study was quite small (n=44) and was comprised of older adults with higher levels of functional impairment. For example, the sample used in the present study included 192 older adults, of which only 20% (39 of 200) used some form of assistive device as compared with 46% in the sample tested by Shumway-Cook.17 The faller and nonfaller groups in the previous study were also equal in size, which is not truly representative of the proportion of fallers to nonfallers (ie, 35%) that exist in the actual older adult population.1, 2

More recently, Brauer et al5 investigated the predictive properties of 4 clinical balance tests (ie, BBS, Functional Reach test, Lateral Reach test, and the Step-Up test) and 3 laboratory tasks (ie, postural stability in quiet stance, limits of stability, and a reaction-time step task) in a sample of 100 female adults (65–86y), with similar demographic characteristics to the present study sample. The investigators found that none of the clinical balance assessments, either alone or in combination, was predictive of faller status in the higher functioning group of older adults studied. A model that included the BBS, anterior reach distance, lateral reach distance, and number of steps was not significantly predictive of faller status (P=.76).5 Boulgarides et al7 also studied the predictive properties of 5 clinical tests of balance and gait, including the BBS, in a sample of 99 relatively healthy, high functioning older adults. Their results indicated that neither the BBS nor any of the 4 other clinical balance tests selected (eg, Modified Clinical Test of Sensory Interaction for Balance, Limits of Stability, Timed Up & Go test, and the Dynamic Gait Index) could predict faller status. In fact, the authors of both studies just described concluded that current clinical balance tests lack the ability to detect the subtle changes in balance abilities that are likely occurring in independently functioning older adults.

Given that the FAB scale consists of more challenging test items than those included on the BBS and many other clinical tests currently used to evaluate balance abilities in community-residing older adults, it is perhaps not surprising that lower total FAB scores are associated with higher fall-risk (ie, 25 of 40 on FAB scale vs 49 of 56 on BBS). The inability to perform some of the more challenging tasks included on the FAB scale may not necessarily equate to an immediate high probability of falling but rather to subtle declines in balance abilities that act as a precursor to an increased probability of falling. As previously noted, the primary reasons for development of the FAB scale were to (1) provide a functional balance measure that was less prone to the ceiling effects evident with the BBS when used with higher functioning older adults and (2) provide an assessment tool that measured more dimensions of balance and thereby provide the clinician with more guidance in the design of an individual treatment plan. To achieve these objectives, the FAB scale includes individual test items that are more difficult to perform without imbalance and are therefore more sensitive to subtle changes occurring in the multiple dimensions of balance. The inclusion of more individual test items that specifically assess sensory reception and integration abilities, which are integral to good motor planning and execution, also increases the FAB scale's sensitivity in detecting subtle changes in balance and mobility.

In order to obtain a high score on the FAB scale, a higher level of balance skill is required than is needed to successfully perform the majority of test items included on the BBS. Thus, some items on the FAB, although difficult to perform, may not be critical items in discerning faller status. The fact that an individual who scores a 0 on the FAB scale only has an 82.2% probability of falling as opposed to a 100% probability provides support for this argument.

The results of the DFA indicated that 5 of the 10 individual FAB items were found to be significant predictor variables (based on the more stringent cut-off value of 0.5) in discerning faller status. The mean scores for the fallers indicated that these individuals were unable to step up and over a bench without contacting it in 1 or both directions (item 4), experienced difficulty walking along a line with feet in a tandem position (item 5), were unable to stand on 1 leg for longer than 5 seconds (item 6), were unable or unwilling to close their eyes while standing on a foam surface (item 7), and were unable to successfully complete a 2-footed jump without initiating or landing with one foot before the other foot (item 8). This latter test item requires lower-body muscle strength and power as well as motor coordination. Because individuals with lower-body weakness are 4.4 times as likely to sustain a fall,1, 2 and reduced muscle power has been associated with higher levels of disability among older adults,29, 30 lacking sufficient muscle strength and power to perform the 2-footed jump may explain why this test item proved to be a good predictor of faller status in this sample.

Previous studies have shown that the ability to stand on 1 leg is a strong predictor of who will and will not fall.8, 10 Interestingly, item 4 (step up and over the bench) was a stronger predictor of faller status than item 6 (standing on 1 leg) in the present study. Standing on foam with eyes closed (item 7) was also predictive of falls in the present study. The primary goal of this test item is to evaluate sensory reception and integration abilities in an altered sensory environment. The inability to maintain standing balance in this deprived sensory condition has previously been shown to discriminate between older adult fallers and nonfallers.31

There are also similarities between certain BBS items and FAB scale test items that have previously been found to be predictive of faller status. For example, Chiu et al8 found the following individual items on the BBS to be the most predictive of faller status in a sample of 78 community-residing older adults: test items 9 (pick up an object from the floor), 14 (stand on 1 leg), and 12 (place alternate feet on a stool). This latter test item on the BBS requires the individual to perform a total of 8 alternating toe touches on a 6-inch stool within 20 seconds. Although less difficult to perform than item 4 (step up and over) on the FAB scale, dynamic weight shift, and single leg stance abilities are still being evaluated. Thus, it is understandable that these items would also be predictive of faller status.

Conclusions 

The present study has shown that the FAB scale is a predictive measure of faller status when used with independently functioning older adults. A practitioner can be confident in more than 7 out of 10 cases that an older adult who scores 25 or lower on the FAB scale is at high risk for falls and in need of immediate intervention. Finally, unlike other tests of balance, the FAB scale appears less prone to ceiling effects when used with independently functioning older adults. As the frequency distribution of the total FAB scale scores showed, none of the 192 participants in the present study achieved the maximum score of 40 out of 40 and only 1 participant achieved the next highest score of 38 out of 40. In fact, the total FAB scale scores were generally well distributed across the group of participants screened. As such, this tool is likely to serve as both a useful screening tool for identifying subtle changes in balance abilities as well a responsive outcome measure for showing change after a treatment intervention.