Predicting Falls In Individuals with Parkinson Disease

A Reconsideration of Clinical Balance Measures

Abstract

Background and Purpose 

Postural control problems, falls, and fall-related injuries are a common source of morbidity in elderly individuals, especially those with Parkinson disease (PD). Clinical balance tests such as the Berg balance scale and the functional reach have been reported to be useful in assessing fall risk in elderly individuals. However, the utility of commonly used clinical balance tests as accurate screens for fall risk has not been sufficiently examined in persons with PD. The purposes of this study were to identify which commonly used clinical balance tests of persons with PD were predictive of falls, to re-examine the cut-of scores for these tests with the goal of maximizing sensitivity as well as minimizing the negative likelihood ratio, and to determine which of the clinical balance tests had the most value in predicting falls.

Participants 

Forty-five persons with a diagnosis of idiopathic PD, aged 39–90 years (mean [sd] =69.94 [11.28]; mean [sd] Hoehn and Yahr level = 2.60 [.66]) participated.

Methods 

Upon initial contact with participants, demographics and fall history were gathered and baseline physical examinations were performed. Each individual underwent balance testing with the functional reach test, the Berg balance scale, the dynamic gait index, timed up and go, and the cognitive timed up and go. Fallers and nonfallers were divided based on fall history and groups were compared on balance test performance. Sensitivity, specificity, likelihood ratios, and receiver operator characteristic curves were calculated for all balance tests.

Results 

Twenty-five (55%) participants had a history of falls. Using cut-of scores reported in previous studies, the sensitivity of all tests was low (less than 0.60) and the specificity was high (greater than 0.85). Reconsideration of the cut-of scores resulted in increased sensitivity for all tests (greater than 0.75) and low negative likelihood ratios (less then .30).

Discussion and Conclusion 

Given the large financial, psychological, and physical complications that are associated with a fall and relatively little harmful effects of fall prevention interventions, we propose consideration of cut-of scores that maximize sensitivity for individuals with PD. Regardless of the performance on any individual clinical balance test, the multifactorial nature of postural instability in PD may necessitate a battery of tests to provide the most accurate identification of fall risk.

INTRODUCTION

Parkinson disease (PD) is a chronic disabling condition that is characterized by rigidity, bradykinesia, tremor, and postural instability.¹ As PD severity worsens, profound clinical deficits in postural control, including instability during centrally initiated postural adjustments (ie, gait initiation, transition to single limb stance) and perhaps most devastating, life threatening falls occur with increasing frequency. A recent review of prospective studies examining falls in PD concluded that almost 70% of patients fall at least once each year and nearly 50% fall more then twice each year.^2,3 Falls resulting from postural instability have an enormous impact on the health of persons with PD as they are a source of considerable morbidity and mortality. For example, researchers have reported that PD increases risk for fractures, with a recent prospective study reporting that PD more than doubled hip fracture risk.^4,5 Consistently, studies examining this topic have concluded that there is a critical need for better treatment of postural control deficits.^6–8

In individuals without PD, it has been suggested that 30% to 40% of all falls can be prevented if individuals who are at risk for falling are identified and are given appropriate therapeutic interventions to help reduce their risk of falling.⁹ Unfortunately, the impact of accurate identification of fall risk and the effects of balance training on individuals with PD is unclear at this time and much research is needed. However, several researchers have found that individuals with PD can have improvements in strength, aerobic capacity, muscle reaction time, and functional characteristics of gait in response to rehabilitation.^10–12 These responses are similar to those observed in individuals without PD who are participating in fall prevention programs.^13–17 If therapeutic interventions can improve these areas, future fall occurrence may be minimized. Postural instability and falls have been implicated as significant contributors to depression and decreased quality of life in persons with PD.^18,19 If successful, fall reduction interventions may decrease the consequences of falls on individuals and society as well as improve the quality of life of those with PD. In order to do this, the ability to identify individuals with PD who are at risk for falling must be improved.

It could be argued that if the majority of persons with PD are at risk for falls, then all individuals with PD should be referred for fall risk reduction interventions. Realistically, this is not feasible from a time or cost effectiveness standpoint. In addition, there are subpopulations of individuals with PD who appear to have less problems with postural instability (ie, those early in the disease process or those with tremor as their predominate motor sign).²⁰ In order to distinguish those individuals with PD not at risk from those at risk, clinical balance tests that provide accurate screening are needed.

The motor subsection of the Unified Parkinson Disease Rating Scale (UPDRS) has been traditionally used by neurologists as a standardized clinical rating to assess the motor signs and symptoms of PD.²¹ However with respect to balance and postural control, there are significant limitations to this scale. For example, only one item in the scale specifically exams balance (the retropulsion test). Although the test for retropulsion examines reactive balance responses, an important component of postural instability, it does not examine many other components and may not predict falls in an accurate fashion.²

To account for the shortcomings of the UPDRS, several research groups have examined the ability of common clinical balance tests (the functional reach test [FRT], components of the Berg balance scale [BBS], the timed up and go test [TUG], and the cognitive timed up and go test [CTUG]) to screen for and predict fall risk in individuals with PD.^2,22–32 These efforts have emphasized that the validity of the measures in predicting falls in individuals with PD could be improved. The majority of these studies demonstrate a balance test performance difference between those persons with PD with a history of falls and those without a history of falls. However, only one study has examined sensitivity and specificity in a sample of individuals with PD. Behrman et al²² reported that the use of the cut-of score for the FRT reported in previous research (25.40 cm) resulted in poor sensitivity (0.30) in prediction of true fallers (when self reported fall history was used as a marker for the condition of interest).

The cut-of scores reported in previous research have been determined using general samples of community dwelling elderly individuals. Studies similar to Behrman et al are needed that examine the sensitivity and specificity of the BBS, DGI, TUG, and CTUG in PD specific samples. We have reported, in abstract form, that application of the cut-of scores reported in previous research to a PD specific sample resulted in high levels of specificity (> .85) and lower levels of sensitivity (< .60).³³ In clinical practice, interpretation of balance test performance using these cut-of scores would result in persons who are at risk may not be correctly identified as being at risk (false negatives). Such persons would likely not be referred for falls risk intervention and may subsequently experience a fall and injury. In contrast, if a clinician were to consider balance test performance relative to a criterion value that was set to maximize sensitivity, more individuals with PD who are not truly at risk for falls (false positives) may be referred for falls risk intervention and treatment. The risk of ‘over-treating’ some individuals likely has fewer adverse consequences than missing the opportunity to treat someone who is at risk for falling. In order to decrease the number of false negatives, and accurately screen more persons with PD at risk for falls, we believed that the cut-of scores for these tests should be reconsidered. We were also interested in the utility of additional measures of test performance in order to clarify the post-test probability of falls in our sample.

The specific objectives of this study were: (1) to examine which commonly used clinical balance tests accurately distinguished between persons with PD with a history of falls and those without a history of falls, (2) to re-examine the cut-of scores for these tests with the goal of maximizing sensitivity and consideration of likelihood ratios, and (3) to examine which of the clinical balance tests had the most value in predicting falls (using measures of test performance) in a sample of persons with PD. To address objective 1, we hypothesized that there would be a difference between the mean clinical balance scores of persons with PD with a history of falls from those persons with PD without a history of falls. Because previously reported cut-of scores were established in populations other than from individuals with PD, to address objective 2, we hypothesized that sensitivity and likelihood ratios of these tests could be improved through a re-examination of the cut-of scores for the FRT, BBS, DGI, TUG, and CTUG. Lastly, to address objective 3, we hypothesized that there would be a difference between the clinical balance scores in their fall prediction ability.

METHODS

Participants

Prior to recruitment of participants for this study, approval for the study protocol was received from the University of Utah Health Sciences Institutional Review Board. All individuals with idiopathic PD referred for outpatient physical therapy care in the University of Utah Rehabilitation and Wellness Clinic during a 1-year period were eligible to participate. The following inclusion criteria were used: a medically confirmed diagnosis of idiopathic PD,³⁴ ability to ambulate at least household distances (50 feet) with no greater than minimal assistance, and a willingness and the ability to participate in a clinical balance examination and accurately report fall incidence, situation, and consequences. Patients were excluded from the study if they had a history of any other neurologic or an orthopedic disorder that would affect their ambulatory or balance ability (eg, stroke, fracture), or any cognitive deficits that precluded cooperation with the procedures of this study. Forty-five persons met the criteria and were eligible to participate.

Procedure

Upon entry into the physical therapy clinic, participants signed an informed consent form and then underwent a physical therapy examination. During the personal interview portion of the examination the following information was gathered: age, duration of disease, Hoehn and Yahr level, medications taken and their schedule, and fall history. For the purposes of this study, a fall was considered an event in which the individual unintentionally came to rest on the ground or other level. Individuals were classified as a faller if they self-reported 2 or more falls in the previous 12 months. This definition of a faller was consistent with previous studies of fall prediction in persons with and without PD.^35,36

Prior to any rehabilitation interventions, a physical examination was performed while the patients were in the ‘on’ phase of their medication cycle (generally 1–3 hours after taking their anti-Parkinson's medications). Immediately following the history and past medical history portion of the examination, each individual underwent the following standardized clinical assessments of balance (in order) in the manner described in previous literature: the functional reach test (FRT),³⁷ the Berg balance scale (BBS),^23,29 the dynamic gait index (DGI),³⁸ timed up and go (TUG), and the cognitive timed up and go (CTUG).^26,35 Two trials of the FR, TUG, and CTUG were performed and the mean of the trials was taken as the test measure. Because of the length of the BBS and DGI, only one complete trial of each of these tests was performed. Testing was performed by 1 of 2 of physical therapists who specialized in treating persons with PD. Since these results represented data from clinical practice, the physical therapists were not masked to fall history and inter-rater reliability was not tested. Approximately 5 minutes of rest were given between tests and the overall physical therapy examination took 1 to 1.5 hours.

Data analysis

Statistical examination of the data was performed using SPSS for Windows, Version 11.0. Descriptive statistics were calculated for demographic and disease-specific variables, while comparison of fallers and nonfallers for each clinical measure of balance was performed using separate independent t-tests or their nonparametric alternative (Mann Whitney U tests).

In order to determine appropriate cut-of scores and the relative predictive values of the tests, multiple statistical procedures were performed. First, sensitivity and specificity values were determined for each clinical balance measure, using the cut-of scores reported in previous research.^{22,23,26,35,38} A self reported history of greater than 2 falls in the past year was considered the target condition for these calculations (ie, a person with PD who truly falls [a faller]). Sensitivity was calculated as the proportion of participants with a history of falls that had a positive test score (FRT ≤ 25.40 cm, BBS < 46/56, DGI ≤ 19/24, TUG ≥ 13.5 sec, CTUG ≥ 15.00 sec). Specificity was calculated as the proportion of subjects without a history of falls with a negative test score (FRT > 25.40 cm, BBS > 46/56, DGI > 19/24, TUG < 13.5 sec, CTUG < 15.00 sec).^{22,23,26,35,38}

Second, for each balance test, we determined sensitivity, specificity, negative likelihood ratio (LR), and positive likelihood ratio (PR) values for each value found in our sample. These values were analyzed as potential cut-of scores of these clinical balance tests (ie, the score above which a score would be considered ‘positive’). We then used likelihood ratio (LR) values, which combined sensitivity and specificity to describe the odds of being a faller given a certain test result.³⁹ The negative LR was calculated as (1-sensitivity)/specificity and indicated the change in odds of being a faller given a negative balance test result. A negative LR of 1 indicated that the balance test result did nothing to change the odds of being a faller, whereas an negative LR less than 1 diminished the odds of being a faller.⁴⁰ The positive LR was calculated as sensitivity (1-specificity) and indicated the change in odds of being a faller given a positive balance test result. A positive LR of greater than 1 increased the odds of being a faller.

In order to maximize the clinical utility of the cut-of score, we sought to minimize false negatives (saying a true faller was not at risk), by selecting single cut-of scores for each clinical balance test that maximized sensitivity and minimized the negative LR. As a component of the sensitivity and specificity calculations, receiver operating characteristic (ROC) curves were generated for each of the clinical balance tests.⁴¹ Separate ROC curves were generated by plotting the sensitivity (true positive rate) along the Y-axis and 1-specificity (false positive rate) on the X-axis for all values of each clinical balance measure represented in our sample. Gallagher describes ROC curves as a ‘signal to noise’ comparison of a specific test. Tests that have ROC curves closer to the upper left corner of the plot maximize the signal to noise ratio and have greater clinical utility.⁴¹

Lastly, the area under the ROC curve (AUC) was calculated to quantify the clinical utility of each test. The nonparametric method described by Hanley and McNeil⁴² and Bamber⁴³ was used. The nonparametric method was chosen because it did not require a binomial data distribution. The AUC was interpreted as the probability of correctly identifying the person with PD who will fall from a randomly selected pair of patients and may range from 0.50 (no predictive value) and 1.00 (perfect predictive ability). The clinical balance test with the largest AUC represented the clinical balance test with the greatest predictive ability. For this reason, the AUC calculations were used to assist in comparison of the relative predictive value of the tests.

RESULTS

The overall sample had a mean (standard deviation [sd]) of 69.94 (11.28) years of age, had a mean PD duration of 7.43 (5.62) years and a mean Hoehn and Yahr level of 2.60 (.66). Sample demographics subdivided into faller and nonfaller groups are given in Table 2. Twenty-five of the 45 participants (54.9%) were classified as fallers and the remaining 20 were classified as nonfallers. Comparison of these groups determined that they differed significantly with respect to age, years with PD, and Hoehn and Yahr score (p < .05) (Table 1).

Table 1: Participant Characteristics

Table 2: Sensitivity, Specificity, and Likelihood Ratios for Each Clinical Balance Test

Sensitivity and specificity were determined for the 5 clinical balance tests, using previously reported cut of scores. Sensitivity ranged from 0.35–0.57, and specificity ranged from 0.87–1.0 (Table 2).

FRT

The mean score (sd) of the sample on the FRT was 27.43 (8.38) cm. The FRT scores of the fallers and nonfallers were significantly different (faller mean (sd) = 23.11 (8.12) cm; nonfallers mean (sd) = 31.70 (5.61) cm; p < .05). A cut-of score of 31.75 cm resulted in a sensitivity of 0.86, a specificity of 0.52, and a negative LR of .30 (Tables 2 and 3). The positive LR was 1.79. The ROC curve for the FRT generated from the data in this investigation is displayed in Figure 1 found on page 63. The AUC for the FRT was 0.80.

Figure 1: ROC Curves for FRT, DGI, and BBS. Test curves above the reference line indicate a screening test that has better predictive ability then a random choice. Label number 1 indicates a cut-off score of 31.75 cm on the FRT, label 2 indicates a cut-off score of 22/24 on the DGI, and label 3 indicates a cutoff score of 54/56 on the BBS. The area under the ROC curves (95% CI) for the FRT, DGI, and BBS were 0.80 (0.69–0.92), 0.84 (0.73–0.95), and 0.83 (0.71–0.95), respectively.

Table 3: Between Group Comparisons of Clinical Balance Test Performance

BBS

The mean score (sd) of the sample on the BBS was 50.20 (7.90) out of 56. Te BBS scores of the fallers and nonfallers were significantly different (faller mean [sd] = 46.4 0 [8.79] points; nonfallers mean [sd] = 54.69 (1.69) points; p < .05). A cut-off score of 54 resulted in a sensitivity of 0.79, a specificity of 0.74, and a negative LR of 0.29 (Tables 2 and 3). The positive LR was 3.07. The ROC curve for the BBS generated from the data in this investigation is displayed in Figure 1. The AUC for the BBS was 0.83.

DGI

The mean score (sd) of the sample on the DGI was 19.92 (3.94) out of 24. The DGI scores of the fallers and nonfallers were significantly different (faller mean [sd] = 17.92 [4.36] points; nonfallers mean [sd] = 21.82 [3.42] points; p < .05). A cut-off score of 22 resulted in a sensitivity of 0.89, a specificity of 0.48, and a negative LR of .27 (Tables 2 and 3). The positive LR was 1.86. The ROC curve for the DGI generated from the data in this investigation is displayed in Figure 1. The AUC for the DGI was 0.84.

TUG

The mean score (sd) of the sample on the TUG was 11.67 (5.51) seconds. The TUG times of the fallers and nonfallers were significantly different (faller mean (sd) = 13.71 (6.02) seconds; nonfallers mean (sd) = 9.66 (3.18) seconds; p < .05). A cut-off score of 7.95 seconds resulted in a sensitivity of 0.93, a specificity of 0.30, and a negative LR of .27 (Tables 2 and 3). The positive LR was 1.31. The ROC curve for the TUG generated from the data in this investigation is displayed in Figure 2. The AUC for the TUG was 0.77.

Figure 2: ROC Curve for TUG and CTUG. Test curves above the reference line indicate a screening test that has better predictive ability then a random choice. Label number 1 indicates a cutoff score of 7.95 sec on the TUG and label 2 indicates a cut-off score of 8.50 sec on the CTUG. The area under the ROC curves (95% CI) for the TUG and CTUG were 0.77 (0.58–0.86) and 0.80 (0.67–0.93), respectively.

CTUG

The mean score (sd) of the sample on the CTUG was 16.48 (11.63) seconds. The CTUG scores of the fallers and nonfallers were significantly different (faller mean [sd] = 21.45 (13.79) seconds; nonfallers mean [sd] = 11.29 [3.92] seconds; p < .05). A cut-off score of 8.5 seconds resulted in a sensitivity of 0.92, a specificity of 0.40, and a negative LR of .23 (Tables 2 and 3). The positive LR was 1.42. The ROC curve for the CTUG generated from the data in this investigation is displayed in Figure 2. The AUC for the CTUG was 0.80.

DISCUSSION

Accurate identification of individuals with PD at risk for falls may assist clinicians to appropriately prescribe therapeutic interventions for fall prevention that will ideally reduce fall risk and fall related injury.⁴⁴ This study operated on the assumption that accurate screening of fall risk in individuals with PD can assist clinicians to appropriately prescribe therapy for reduction of fall risk and fall related injuries. To meet this goal, we sought to (1) determine if these 5 common clinical balance tests accurately discriminated between those persons with PD with and without a history of falls, (2) to determine the sensitivity, specificity, and likelihood ratios of the FRT, TUG, CTUG, DGI, and BBS and recalculate the cut-off scores to maximize sensitivity and minimize the negative LR, and (3) determine which of the tests possessed the best predictive validity, based on the area under the ROC curves of each of the clinical balance tests.

Our results demonstrated that the calculated sensitivity of the FRT, TUG, CTUG, DGI, and the BBS using cut-off scores reported in previous studies is poor (< .60). These results are similar to previous studies which found that clinical balance tests (such as the FRT, TUG, CTUG, DGI, and BBS) may be able to discriminate between fallers and nonfallers, but lack the ability to minimize false negative results.^2,22,27

However, as hypothesized, the sensitivity of these assessments could be improved by using cut-off scores more targeted to individuals with PD. Our results agree with the results of Behrman et al²² with regards to the FRT, and extend their results by suggesting that the previously reported cut-of scores for other clinical balance tests must be reconsidered for a pathology such as PD which affects the postural control system.

Likelihood ratio values are a useful statistic for examining screening tests and risk levels in individual patients are because they allow the determination of the probability of a condition given a certain test result. In this study, we sought to determine the negative LRs of the clinical balance tests in order to determine the post-test probability of being a faller. Using the calculations described by Deeks and Altman,⁴⁵ we multiplied the pretest probability of being a faller in our sample (55%) by the negative LR to determine the post-test probability of being a faller. For example, multiplying 55% by the smallest negative LR in our sample (0.22 for the DGI with a test score of 22 points or greater) equals .12. This suggests that if a person with PD in our sample scored greater than 22 on the DGI, the clinician could revise their post-test probability of this person being a faller from 55% to 12%.

Relative Predictive Validity

As stated in the data analysis section, the AUC can be interpreted as the probability of correctly identifying the person with PD who will fall from a randomly selected pair of patients and may range from 0.50 (no predictive value) and 1.00 (perfect predictive ability). The variable with the largest AUC represents the variable with the greatest predictive ability. In this study, the DGI demonstrated the highest AUC at 0.84. However the 95% CI for the AUC for all the balance test measures overlapped substantially (Figures 1 and 2). Considered alone, the AUC values suggest that all of the tests examined in this research possess similar strong predictive validity. However, the BBS did show a higher level of specificity and a modestly higher positive LR (3.07) than the other tests. These values suggest that the BBS may overall have stronger predictive validity.

Individuals with PD demonstrate a variety of motor deficits (ie, muscle co-contraction that decreases their ability to generate torque around a joint).^46–48 In addition, they may have difficulty adapting their motor temporal sequencing to new/varied situations.^46–48 These fundamental postural control differences between individuals with PD and non-neurologically impaired individuals may cause individuals with PD to fall when placed in a situation where a non-neurologically impaired individual would not fall. Assuming that persons with PD fall more easily, the clinical balance tests used to screen for fall risk must reflect this difference. This study attempted to address the likelihood that people with PD fall more easily by using alternate methods for analyzing clinical balance test performance.

Assuming that our sample is representative of a broader population of community dwelling individuals with mild to moderate PD, performance of any of the balance tests of interest in this research and interpretation of these results using the cut-of scores reported in previous research would lead to a false sense of security among both clinicians and individuals with PD. The low sensitivity might lead clinicians to inaccurately believe that an individual does not need therapeutic interventions because they are not at risk of falling.

Given the large financial, psychological, and physical complications that are associated with a fall and relatively little harmful effects of fall prevention interventions, we propose a reconsideration of the cut-off scores of these tests for individuals with PD. The values shown in Table 2 are, in our opinion, values that provide an acceptable trade in specificity for sensitivity. However these values are associated with a lower specificity and therefore might misclassify a nonfaller as a faller (notice that these values correctly classify at least 4/5 true fallers and in most cases 3/5 true nonfallers). Making this sort of false positive judgment is not associated with the severe consequences of a false negative result. For this reason, this trade of appears warranted.

However, it should be stressed that even at these stricter cut-off scores these tests are still not 100% sensitive and may reflect the particular characteristics of our sample. Based on these results, we advocate that clinicians avoid classification into high or low fall risk based on performance on one test relative to one cut-off score. We also advocate the use of LR as a means of making clinical judgments of balance test results for persons with PD. Through the use of the proposed cut-off scores and LRs, clinicians may be able to make more accurate clinical management decisions.

Further research is necessary on multiple issues surrounding falls in persons with PD. Future studies using similar methodology to this study should include larger and more diverse sample of persons with PD. These results appear to indicate that the development of new tests or a battery of existing tests may be needed for more accurate predictive ability for falls in persons with PD. Although the sample size of this study did not support the use of multiple regression analyses, a prospective examination of the predictive value of a battery of clinical balance tests in individuals with PD would be a valuable contribution to the literature and to clinical practice. In addition, in focusing on accurate fall prediction, we have made the assumption that fall risk can be reduced in persons with PD. Studies documenting the effects of fall risk reduction programs in PD, using number of falls as an outcome, are critically needed.

SUMMARY

This study demonstrated that while commonly used clinical balance tests have the ability to distinguish between fallers and nonfallers, the previously reported cut-off scores of these tests have poor sensitivity for use when screening for fall risk in individuals with PD. Based on our data and because of the potential consequences associated with a fall, we advocate consideration of cut-off scores on clinical balance tests that maximize sensitivity and therefore minimize false negatives. Due to the nature of their postural control deficits, individuals with PD may be at higher risk for falls than those without PD. Such differences may require alteration in the clinical interpretation of balance test results. Additional research is needed to develop more accurate tests and/or to determine which battery of existing tests might best predict fall risk in persons with PD.

ACKNOWLEDGEMENTS

We thank the persons with PD that participated in this research. We also thank Julie Fritz, PT, PhD, ATC for her statistical consultations and review of earlier versions of this manuscript. This project was supported in part by a grant from the Utah Chapter of the American Parkinson Disease Association.