Diagnosis of Fall Risk in Parkinson Disease: An Analysis of Individual and Collective Clinical Balance Test Interpretation

Facility and Participants

The site of data collection houses a Parkinsonism Exercise Program sponsored by the American Parkinson Disease Association. All people receiving care at this facility sign a University of Utah Health Sciences Center Institutional Review Board–approved consent form allowing the tracking of measures of physical performance. All people receiving traditional outpatient physical therapy or participating in a community-based risk reduction program from 2004 to 2006 represented the accessible population for this study. Inclusion criteria were a confirmed medical diagnosis of idiopathic PD,²¹ the physical ability to perform at least 3 of the 4 clinical balance tests (BBS, DGI, FRT, and Up and Go Test), and the cognitive ability to actively participate in the balance tests and report fall history. Exclusion criteria were an age of less than 60, an inability to follow 1- or 2-step commands via verbal or tactile cuing, and significant orthopedic (eg, fracture or moderate to severe osteoarthritis) or other neurologic (eg, stroke or traumatic brain injury) conditions.

Over the 2-year period during which data were collected, approximately 350 participants with varied neurologic and musculoskeletal diagnoses were examined at the facility. Seventy people with idiopathic PD (54 men and 16 women), with a mean age of 73.91 years (SD=6.45 years), and with a median modified Hoehn and Yahr staging level of 2.5 (range=1–4) met the inclusion and exclusion criteria and were the participants included in this analysis. Thirty-six people (51%) reported a history of at least 2 falls in the previous year (fall group), whereas the remaining 34 people reported fewer than 2 falls in the previous year (nonfall group) (Tab. 1). There were no significant differences between the fall and nonfall groups in any of the demographic variables (P>.05).

Procedure

All potential participants signed informed consent forms during their initial examinations. One physical therapist (DJB), who specializes in rehabilitation for PD, performed all of the physical therapy examinations. Participants were interviewed, and the following variables were recorded in their medical records: age, sex, fall and near-fall history (including the number of falls in the previous year), duration of PD (in years), past medical and surgical history, and current medication regimen. Following the clinical interview, each participant underwent a standard examination of vital signs, sensory and motor function, and physical performance capabilities. The physical performance examination included multiple clinical balance tests (at least 3 of the following 4 tests: BBS, DGI, FRT, and Up and Go Test). These tests were used because of their established reliability and validity and because we believed that they collectively examined several important components of balance abilities in people with PD. Recording of the Up and Go Test consisted of 3 individual trials, with the mean time for trials calculated and documented. Single trials of all other tests (BBS, FRT, and DGI) were performed primarily because of time limitations. All participants were given a 5-minute recovery period between individual balance tests to decrease fatigue. Each physical therapy examination took approximately 1 to 1.5 hours. All participants were instructed to be on the “on-dopamine replacement” phase of their medication cycle for their initial examinations.

For the purposes of this study, demographic as well as numerical clinical balance test data were extracted and recorded for each individual. Each participant's disease severity was determined using Hoehn and Yahr staging criteria.²² In 5 of the cases, the participants could not estimate their PD duration or the PD duration was not recorded in the medical records. In those cases, a value of 1 was assigned to acknowledge that the participants were diagnosed with PD. Participants were divided into groups based on fall history. Participants placed in the fall group were those who reported 2 or more falls within the previous year, and participants placed in the nonfall group were those with fewer than 2 self-reported falls in the previous year. A fall was classified as an unexpected contact of any part of the body with the ground.²³ This definition of a fall was consistent with previous studies of prediction of falls in people with PD and people without PD.^5,23,24

Data Analysis

Data analysis was performed with SPSS for Macintosh (version 11.0)* and Confidence Interval Analysis (version 2.0).^† Demographic and disease-related variables were summarized with descriptive statistics. Between-group comparisons were performed to determine whether people in the fall group and people in the nonfall group performed differently on the clinical balance tests of interest (BBS, DGI, FRT, and Up and Go Test). Comparisons of demographic and balance test performance differences between groups were accomplished with separate nonparametric tests for 2 independent groups (Mann-Whitney U tests) with a level of significance set at P<.05.²⁵

With each participant's self-reported fall history as the gold standard for the measurement of fall risk, sensitivity, specificity, and LRs (positive and negative)^26,27 were calculated. In order to explore these validity indexes, we first had to interpret the test results in a dichotomous fashion in relation to an established cutoff score for fall risk. Because we had concerns that adjusting cutoff scores and performing collective interpretation of tests were overly complicated, for the purposes of this research, we chose to use cutoff scores that are currently accepted in the literature.^{13,14,18,23,28–30} Participants were determined to be positive for fall risk if they reached the threshold value for a specific clinical balance test (FRT value of ≤25.4 cm, BBS score of ≤46/56, DGI of ≤19/24, and Up and Go Test value of ≥8.5 s), whereas they were considered to be negative for fall risk if they did not reach the threshold value (FRT value of 25.4 cm, BBS score of 46/56, DGI of 19/24, and Up and Go Test value of <8.5 s).^{13,14,18,23,28–30}

Sensitivity in this context was defined as how often a clinical balance test detected fall risk for a participant in the fall group.^26,27,31 Sensitivity values close to 1.0 indicated that the majority of “true fallers” were identified.^26,27,31 Specificity was defined as how often a clinical balance test result was negative for a participant in the nonfall group. Specificity values close to 1.0 indicated that the majority of “true nonfallers” were identified.^26,27,31 Although sensitivity and specificity are the validity indices most recognizable to clinicians, what they actually tell clinicians is how likely a clinical balance test result is to be positive or negative given that a person with PD is at high risk for falls or is not at high risk for falls. There is a paradox of relying on sensitivity and specificity because knowledge of a person's degree of fall risk would eliminate the need for a diagnostic clinical balance test in the first place.

To overcome the limitations of sensitivity and specificity, positive and negative LRs were calculated. The advantage of LRs is that they allow a clinician to quantitatively estimate the posttest probability of an individual participant being in the fall group.^26,27,31 The positive LR was calculated as sensitivity/(1 − specificity) and used to answer the following question: How much more likely is a positive clinical balance test result to be found in a person in the fall group than in a person in the nonfall group? A larger positive LR amplified the probability of a person being in the fall group, given a positive clinical balance test result.^26,27,31 The negative LR was determined as (1 − sensitivity)/specificity and used to answer the following question: How much more likely is a negative clinical balance test result to be found in a person in the nonfall group than in a person in the fall group? A smaller negative LR reduced the probability of a person being in the fall group, given a negative clinical balance test result.^26,27,31

In addition to the analysis of each individual test, we calculated these same statistical measures (sensitivity, specificity, positive LR, and negative LR) for 3 differential groups of tests:

1. One or more positive tests: The criterion to meet this standard was interpretation of the results of at least one of the clinical balance tests performed as positive for fall risk on the basis of the operationally defined cutoff scores.

2. Two or more positive tests: The criterion to meet this standard was interpretation of the results of at least 2 of the clinical balance tests performed as positive for fall risk on the basis of the operationally defined cutoff scores.

3. Three or more positive tests: The criterion to meet this standard was interpretation of at least 3 of the clinical balance tests performed as positive for fall risk on the basis of the operationally defined cutoff scores.

Because sensitivity and a negative LR are both used to “rule out” a person from the fall group (minimizing false-negative results), they were grouped together in the presentation of our results. Because specificity and a positive LR are both used to “rule in” a person in the fall group (minimizing false-positive results), they were grouped together in the presentation of our results.

It is important to note that in the context of determining fall risk, we believe that misclassifying a participant as not being at risk for falls when that participant is at high risk for falls (a false-negative result) carries more significant consequences than misclassifying a participant as being at high risk for falls when that participant is not at high risk for falls (a false-positive result). For this reason, our goals for the interpretation of validity indexes were aimed at minimizing false-negative results, thus increasing sensitivity and reducing the negative LR.

Validity of Individual Clinical Balance Tests

For each test used in the present study, the mean clinical balance test values were significantly different between the fall and nonfall groups (P<.05). In each test, the performance of the nonfall group exceeded that of the fall group. In addition, in 3 of the 4 tests (all except the Up and Go Test), the 95% confidence intervals (CIs) of the fall and nonfall groups did not overlap (Tab. 2).

The highest level of sensitivity (95% CI) of any individual clinical balance test was 0.77 (0.61–0.88), for the FRT, whereas the smallest negative LRs (95% CIs) for any individual balance test were calculated for the FRT and the DGI: 0.42 (0.21–0.83) and 0.42 (0.24–0.73), respectively. In the context of a negative balance test result, such sensitivity and negative LR values indicate a small number of false-negative results and a moderate ability to reduce the posttest probability of being in the fall group.

The highest level of specificity (95% CI) of any individual clinical balance test was 0.85 (0.68–0.94), for the DGI, whereas the largest positive LR value (95% CI) for any individual balance test was calculated for the DGI: 4.26 (1.67–11.18) (Tab. 3). In the context of a positive clinical balance test result, such specificity and positive LR values indicate a relatively small number of false-positive results and clinical significance through amplification of the pretest probability of being in the fall group.

Validity of Collective Interpretation of Multiple Clinical Balance Tests

The combination of one or more positive tests resulted in the highest levels of sensitivity (95% CI) and negative LR (95% CI): 0.97 (0.83–0.99) and 0.14 (0.012–0.65), respectively; however, the levels of specificity and positive LR were relatively low. In the context of one or more positive clinical balance tests, such sensitivity and negative LR values indicate very few false-negative results and clinical significance through reduction of the posttest probability of being in the fall group to a larger extent than that realized with any individual test.

The combination of 3 or more positive tests resulted in the highest levels of specificity (95% CI) and positive LR (95% CI): 0.80 (0.57–0.85) and 2.48 (1.35–4.57), respectively; however, the levels of sensitivity and negative LR were relatively low. In the context of 3 or more positive clinical balance tests, such specificity and positive LR values indicate a moderate number of false-positive results. However, such positive LR values support clinical significance by helping to “rule in” fall risk through amplification of the pretest probability of being in the fall group.

The combination of 2 or more positive tests resulted in validity indexes between the extremes seen with the other 2 combinations (sensitivity [95% CI]: 0.82 [0.67–0.92]; negative LR [95% CI]: 0.34 [0.16–0.76]; specificity [95% CI]: 0.51 [0.36–0.67]; and positive LR [95% CI]: 1.70 [1.17–2.47]) (Tab. 4).

Testing the Difference Between People With a History of Falls and People Without a History of Falls

The first research question analyzed pertained to the validity of each individual clinical balance test as a measure of fall risk. We hypothesized that individual clinical balance tests could accurately distinguish people with PD and a history of falls from those without a history of falls. In agreement with other studies, all of the clinical balance tests examined in the present study could discriminate people with PD and a history of falls from those without a history of falls.^16,18,32 These findings strengthen the evidence that all of these measures have a basic level of content validity in that they have the ability to detect a difference in balance test performance in people with PD when one exists.

In contrast to previous studies of the Timed “Up & Go” Test,^16,33,34 2 factors question the diagnostic utility of the Up and Go Test in our sample. First, the overlap of the 95% CIs of the fall and nonfall groups on the Up and Go Test suggests that this test was less useful than the other tests at distinguishing people in the fall group from those in the nonfall group (Tab. 2). In addition, the 95% CIs of the LRs for the Up and Go Test encompassed 1, indicating that the Up and Go Test results contributed little ability to modify the posttest probability of falling.

Are Multiple Tests More Clinically Useful Than an Individual Test?

Once we felt confident that the clinical balance tests were valid, the next diagnostic decision analyzed was the relative benefit of individual test interpretation versus collective interpretation of multiple tests. This analysis was done with a priority placed on ruling out a diagnosis of fall risk (maximizing sensitivity and minimizing the negative LR). The results of the present study indicated that the collective interpretation of multiple clinical balance tests (when more than 1 or 2 tests were positive) reduced the false-negative rate to a greater degree than did the interpretation of any individual test alone. To our knowledge, although this type of analysis was previously suggested,^16,18,19 such a study had not been previously reported.

Despite the improved sensitivity of collective interpretation, the results of the present study in isolation are of limited clinical utility. As stated in the Method section, sensitivity and specificity alone are limited in their diagnostic abilities because they require a clinician to know the fall diagnosis (falls versus no falls). The more clinically useful tools derived from our data are the LRs.

To emphasize the clinical utility of our calculated LRs, a clinical case example is provided. Clinical case 1 illustrates the utility of the negative LR in answering the following question: How much more likely is a negative result on multiple clinical balance tests to be found in a person who has PD but no fall history than in a person who has PD and a fall history?

In clinical case 1, a 65-year-old man with PD, mild hypometria, unilateral resting tremor, and no prior history of falling attends a clinic. Using the incidence of falling in our sample as the pretest probability of falling, a 51% pretest probability of falling is assumed. During the physical examination, the man is able to reach only 20 cm on the FRT but does not reach the fall risk threshold for any of the other clinical balance tests. Because the man has only 1 positive test (and 3 negative clinical balance tests), a negative LR of 0.09 is applied (Tab. 4). By using an LR nomograph^27,31,35 (Fig. 1), a clinician can calculate that the man has an 8% posttest probability of falling. On the basis of severity and symptoms,^36,37 the clinician may determine that the man is not currently at high risk and that intensive one-on-one physical therapy may be deferred. The man can be counseled on alternative therapeutic options, such as community- or home-based exercises designed to reduce fall risk or improve postural stability.^38–40 It is critical that serial examinations over time be scheduled and performed because it is likely that postural instability and fall risk will worsen over time as the severity of the PD progresses (path A in Fig. 2).

Figure 1.

Nomograph for interpreting diagnostic test results. The bold line indicates the application of a negative likelihood ratio (LR) of greater than one positive clinical balance test (0.09) to calculate the posttest probability of falling in a person with Parkinson disease. Reprinted with permission from: Centre for Evidence-Based Medicine (www.cebm.net/downloads.asp).

Figure 2.

Proposed clinical decision-making algorithm that involves the serial use of clinical balance tests. See text for details. PD=Parkinson disease.

Clinical Decision-Making Algorithm

Although our analysis strongly supports the collective interpretation of clinical balance tests as a means of minimizing false-negative results and reducing the posttest probability of falling, it does not provide guidance on the order in which the tests should be administered. To gain insight, we used our individual test data to construct a clinical decision-making algorithm based on the relative validity indices of the individual tests. Because the DGI possesses the best overall positive LR compared with the other clinical balance tests (BBS, FRT, and Up and Go Test), we recommend that the DGI be used first, followed by the BBS (which contains the FRT).

The following clinical case illustrates the utility of this clinical decision-making algorithm. In clinical case 2, a 72-year-old woman with PD, bradykinesia, dementia, and a history of 2 falls in the previous year is referred to a physical therapist for examination and treatment. During the examination, the woman scores 15 of 24 on the DGI. Because the woman has a positive DGI result, she can be classified as having a significant fall risk, and physical therapy intervention may be justified (path B in Fig. 2). If the woman had a negative score on the DGI, then the physical therapist could perform the BBS (because it possesses the second largest positive LR and contains the FRT). If the woman had a positive score on the BBS or the FRT, then she would be classified as having a significant fall risk, and physical therapy would be justified (path C in Fig. 2). Given her report of previous falls, if the woman had negative scores on both the BBS and the FRT, then the physical therapist should consider other potential causes for the previous falls (path D in Fig. 2).

Limitations and Directions for Future Research

Although our findings appear to have clinical relevance, they should be interpreted cautiously. From a research design standpoint, our sample was relatively small and was derived from only one outpatient facility. For this reason, people with mild PD and people with severe PD were not represented in large numbers. In addition, although the research plan and data gathering were prospective, we relied on each participant's self-reported fall history as the diagnostic gold standard for fall risk. A similar method should be applied to a sample of people with PD, with prospective observation of fall history as the diagnostic gold standard. Finally, we used the cutoff scores most commonly referenced in the literature, as opposed to altering the cutoff scores, as has been suggested in other research.^16,18 The rationales for the recalculation of cutoff scores and the collective interpretation of multiple tests are similar, that is, minimizing false-negative results. In the present study, we chose to constrain our analysis and examine only the effect of collective interpretation. Future research should compare these methods directly.

Although the accurate identification of fall risk is important so that people at risk and their caregivers are aware of the risk for injury, we have made the assumption that accurate identification of fall risk will lead to effective treatment. However, there are relatively few studies documenting improved postural stability, reduced fall risk, or reduced fall occurrence as a result of physical therapy intervention. Additional research should be performed to add to the few studies that have directly examined the effects of rehabilitation interventions on fall risk and fall occurrence.^39–42