Psychometric properties of four fear of falling rating scales in people with Parkinson’s disease

Background Fear of falling (FOF) is commonly experienced in people with Parkinson’s disease (PD). It is a predictor of recurrent falls, a barrier to physical exercise, and negatively associated with health-related quality of life. A variety of rating scales exist that assess different aspects of FOF but comprehensive head-to-head comparisons of their psychometric properties in people with PD are lacking. The aim of this study was to evaluate the psychometric properties of four FOF rating scales in people with PD. More specifically, we investigated and compared the scales’ data completeness, scaling assumptions, targeting, and reliability. Methods The FOF rating scales were: the Falls Efficacy Scale-International (FES-I), the Swedish FES (FES(S)), the Activities-specific Balance Confidence scale (ABC), and the modified Survey of Activities and Fear of Falling in the Elderly (mSAFFE). A postal survey was administered to 174 persons with PD. Responders received a second survey after two weeks. Results The mean (SD) age and PD duration of the 102 responders were 73 (8) and 7 (6) years, respectively. ABC had worse data completeness than the other scales (6.9 vs. 0.9–1.3% missing data). All scales had corrected item-total correlations exceeding 0.4 and showed acceptable reliabilities (Cronbach’s alpha and Intraclass Correlation Coefficient (ICC) >0.80) but only FES-I had ICC >0.90. The standard error of measurements ranged from 7% (FES-I) to 12% (FES(S)), and the smallest detectable differences ranged from 20% (FES-I) to 33% (FES(S)) of the total score ranges. ABC and FES(S) had substantially more outliers than mSAFFE and FES-I (10 and 15 vs. 3 and 4, respectively) when the two test occasions were compared. Conclusions When assessing FOF in people with PD, the findings in the present study favoured the choice of FES-I or mSAFFE. However, FES-I was the only scale with ICC >0.90 which has been suggested as a minimum when using a scale for individual comparisons.


Background
Parkinson's disease (PD) is a common neurodegenerative disorder that affects balance and people with PD fall more often than age-matched healthy controls [1,2]. Fear of falling (FOF) is commonly experienced [3,4] and is a predictor of recurrent falls [5], a barrier to physical exercise [6], and is negatively associated with health-related quality of life [4]. It is therefore important to detect and follow the progress of FOF in people with PD, and FOF should be considered a crucial endpoint for interventions [4,7].
High quality rating scales assessing FOF are important in both clinical practice and research. When choosing a rating scale, one has to consider which aspects the scale should cover as well as its psychometric properties (e.g., data completeness, scaling assumptions, targeting, and reliability). Increased knowledge of the psychometric properties of FOF rating scales will facilitate the interpretation of data obtained from the scales.
A variety of rating scales exist that are said to assess different aspects of FOF [8]. The Falls Efficacy Scale-International (FES-I) assesses concerns about falling and is recommended by the Prevention of Falls Network Europe (ProFaNE) [9]. FES-I was developed by combining and modifying items from three other scales: the original FES that assesses fall-related self-efficacy [10], the Activities-specific Balance Confidence scale (ABC) that assesses balance confidence [11], and the Survey of Activities and Fear of Falling in the Elderly (SAFE) that assesses both activity level, FOF and activity restriction [12]. SAFE has later been modified into a shorter version (mSAFFE) that taps activity avoidance due to the risk of falling [13].
In a recent study, we compared the content validity of FES-I, the Swedish FES (FES(S)), ABC and mSAFFE by linking them to the International Classification of Functioning, Disability and Health [14]. The linking process showed that all four scales mainly focus on FOF in relation to mobility. The ABC almost exclusively focuses on mobility, whereas the other rating scales cover a more diverse set of activities, such as self-care (FES-I, FES(S) and mSAFFE) and activities concerning community, social, and civic life (FES-I and mSAFFE) [14].
As psychometric properties, such as validity and reliability, are sample dependent [15], specific studies are needed to determine the psychometric properties of FOF rating scales in PD. One previous Swedish study has assessed the psychometric properties of FES(S) and mSAFFE in PD with satisfying results [3]. Four studies have assessed the psychometric properties of ABC in PD [16][17][18][19]. However, three of the ABC studies have limited PD samples (n = 19 to 37) [16][17][18] and three are based on a limited set of psychometric analyses [16,18,19]. To our knowledge, no study has assessed the psychometric properties of FES-I in people with PD. Thus, a comprehensive head-to-head comparison of psychometric properties of FOF rating scales in people with PD is warranted and will facilitate choosing a FOF rating scale for clinical practice and research in PD.
The aim of this study was to evaluate the psychometric properties of FES-I, FES(S), ABC and mSAFFE in people with PD. More specifically, we investigated and compared the scales' data completeness, scaling assumptions, targeting, and reliability.

Methods
This postal survey study was sent to 174 persons with PD. It included socio-demographic and disease-related questions, as well as four FOF rating scales which were administered twice (hereafter referred to as t1 and t2), two weeks apart.

Participants and sample size
Participants were recruited from two outpatient hospital clinics in southern Sweden and included individuals with a clinically confirmed PD diagnosis (ICD-10: G 20.9) since at least one year. Exclusion criteria were difficulties reading and writing Swedish, clinically confirmed Alzheimer's disease, dementia, or cognitive or medical problems of a severity that were assumed to restrict giving informed consent or participating in the study. Moreover, individuals who were completely bedridden or wheelchair bound were excluded since most items in the FOF rating scales refer to walking ability. A PD specialized nurse at each of the outpatients clinics and one of the authors (SBJ) screened the medical records of all PD patients that had visited the two clinics during the past 14 months (n = 275). Fifty-nine persons (39% female) were excluded based on the exclusion criteria. Their mean (SD) age and PD duration were 76 (8) and 10 (6) years, respectively. In addition, 42 persons did not meet the inclusion criterion of a PD diagnosis of at least one year. A total of 174 possible participants remained, which was considered the final sample.
To reach a 'good sample size' according to recommendations for methodological quality and test-retest reliability analysis [20], we aimed at 50 to 99 participants with FOF total scores at both t1 and t2. Based on previous postal surveys in people with PD [21,22], we anticipated a response rate of approximately 65% at t1. Some additional drop outs were expected at t2, as well as some internal missing responses on the FOF rating scales.
All participants gave their written informed consent. The study was conducted in accordance with the Helsinki Declaration and was approved by the Regional Ethics Review Board in Lund, Sweden (Dnr 2013/118).

Procedure
All 174 possible participants were mailed the following: information about the study, an informed consent form, socio-demographic and disease-related questions, the four FOF rating scales (FES-I, FES(S), ABC, and mSAFFE), and a pre-stamped return envelope. A reminder was sent after two weeks to non-responders. Responders received a second survey after about two weeks, and a reminder was sent one week later to non-responders.
The internal order of the FOF rating scales was altered to minimize the risk that the ordering affected data completeness. Four different arrangements were used so that the scales appeared an equal number of times as the first, second, third, and fourth scale. Although the order of scales was altered, the original order of items within the scales remained unchanged.

Socio-demographic and disease-related questions
Current mobility when completing the survey at t1 and t2 was self-rated as: good (i.e., parkinsonian "on" state), good but hyperkinetic, or bad (i.e., parkinsonian "off" state). The survey at t1 included demographic questions (e.g., PD duration and living arrangements), as well as single-item questions targeting self-rated PD severity (response options: mild, moderate, or severe), self-rated general health (scored 1-5; higher = worse, inspired by the general health question in the Short Form-36) [23], activities of daily living (Parkinson's disease Activities of Daily Living Scale; PADLS) [24], and freezing of gait (item 3 of the selfadministered version [25] of the Freezing of Gait Questionnaire; FOGQsa) [26]. Both PADLS and FOGQsa have been shown to be valid and reliable in people with PD [24,25]. An open-ended question targeted the presence of diseases or health-related problems other than their PD. Dichotomous questions (Yes/No) targeted the following: dyskinesia; fluctuations with periods of increasing PD symptoms; FOF; activity avoidance due to the risk of falling; unsteadiness while walking; unsteadiness during turning in walking/standing; use of walking aid or personal support while walking indoors and outdoors, respectively; previous falls and/or near falls during the past six months. A fall was defined as "an event in which the respondent came to rest on the ground, floor, or lower level" (definition adopted from ProFaNE) [27]. A near fall was defined as "a fall initiated but arrested by support from a wall, railing, or other person, etc." [28]. Finally, participants were asked whether they had responded to the survey themselves (with or without assistance in reading and/or writing).

The four FOF rating scales
The FES-I assesses concerns about falling [9]. Respondents answer how concerned they are about the possibility of falling in relation to 16 different activities. Response categories are: not at all, somewhat, fairly, or very concerned (scored 1 to 4, respectively). The total score ranges from 16 to 64 (higher = worse) [9]. The Swedish translated FES-I was used in this study [29].
The FES(S) assesses fall-related self-efficacy [30]. Respondents answer how confident they are in performing 13 different activities without falling. Response categories range from 0 (not confident at all) to 10 (completely confident). The total score ranges from 0 to 130 (higher = better) [30].
The ABC assesses balance confidence [11]. Respondents answer how confident they are that they would not lose their balance or become unsteady when performing 16 different activities [11]. In this study, a Swedish translated and culturally adapted version of the ABC was used. The cultural adaptation implies that items related to stepping onto or off escalators are changed to traveling by bus (L. Lundin-Olsson, unpublished material, written personal communication, June 20, 2012). Response categories range from 0 (no confidence) to 10 (completely confident). The total score is the mean value of the 16 items, transformed into percentage, i.e., ranges from 0 to 100% (higher = better).
The mSAFFE assesses activity avoidance due to the risk of falling in relation to 17 different activities [13].

Analyses
The analyses were performed using the IBM SPSS Statistics 21.0 software and were based on four parts: i) data completeness, ii) scaling assumptions, iii) targeting, and iv) reliability. Data completeness and reliability (except Cronbach's alpha) were based on data from both t1 and t2. Scaling assumptions, targeting and Cronbach's alpha were based on t1 data only. The relationships between the rating scales were determined by calculating the Pearson's correlation coefficients (r) between the scales, based on t1 data.

Data completeness
Data completeness of the four rating scales was determined by calculating the percentage of missing data for items and total scores [15,31]. No imputation was done, i.e., a total score required absence of any missing item responses.

Scaling assumptions
Scaling assumptions were explored to examine the legitimacy of summing item scores to generate total scale scores, according to a series of criteria [15,31]. That is, mean scores, SDs, and distribution of item response option frequency should be roughly parallel across items. Also, corrected item-total correlations should exceed 0.4, indicating that items measure the same underlying construct and contain a similar proportion of information concerning FOF [15,31].

Targeting
Targeting refers to whether the rating scales' score distributions can adequately represent the true level of FOF in the sample [15]. This was evaluated by studying the rating scales' score distribution, skewness, and floor and ceiling effects. Mean total scores should be close to the scales' midpoint, total scores range the full span, skewness less than ±1 [15,32], and floor and ceiling effects (the percentage respondents receiving the minimum and maximum possible scores, respectively) should not exceed 15-20% [15,33].

Reliability
Reliability is a measure of the random error associated with scale scores and the reproducibility of scores [15]. This was assessed in several ways. The internal consistency was examined by means of Cronbach's alpha [15].
The test-retest reliability was studied in terms of oneway random, single measures Intraclass Correlation Coefficient (ICC 1,1 ) with absolute agreement definition of concordance [34]. Cronbach's alpha and ICC >0.75 or >0.80 are considered acceptable for group level [35,36], while ICC >0.90 has been suggested as a minimum when using scales for individual comparisons [36,37]. The standard error of measurement (SEM) was calculated using the formula SD baseline Â ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ffi 1−reliability p [38]. The smallest detectable difference (SDD) was calculated using the formula SEM Â 1:96 Â ffiffi ffi 2 p [39]. Due to differences in scoring ranges between the scales, SEM and SDD values were also expressed as percentages of the possible scoring ranges, to facilitate comparisons.
The mean difference (đ) in scale scores between t1 and t2 and the 95% CI around đ were calculated. If the 95% CI includes 0, there are no systematic differences between t1 and t2 [40]. The number of outliers for each rating scale was calculated (an outliers was defined as a participant with differences between t1 and t2 outside the first or third quartile ± 1.5 × interquartile range) [41]. Finally, test-retest data were plotted and visually inspected in the form of Bland-Altman graphs (the individual differences between t1 and t2 were plotted against the individual mean of t1 and t2) [40]. Since these graphs did not contribute any additional information than the numerical analyses, they are not presented here.

Results
Of the 174 possible participants, 63 persons did not respond and 6 explicitly expressed that they did not want to participate; they (n = 69; 54% women) had a mean (SD) age of 77 (9) years. One hundred and five persons returned the first postal survey, but three surveys were not answered by the person with PD and were therefore excluded. This resulted in 102 included participants and a conservative response rate of 59%. Ninety-seven persons responded to the second survey (t2). Basic demographic data and participants characteristics are presented in Table 1. A majority (n = 60) of the participants stated that they had one or more disease or health-related problem, apart from their PD. The most common problems were cardiovascular (n = 22) and musculoskeletal (n = 22). Current mobility at t1 was rated as good (i.e., parkinsonian "on" state) by 48 participants, good but hyperkinetic by 17, and bad (i.e., parkinsonian "off" state) by 35 participants (2 missing responses). Corresponding mobility ratings at t2 were: good ("on") 52 participants, good but hyperkinetic 15, and bad ("off") 23 participants (7 missing responses).

Relationship between the scales
The correlations (r) between the four FOF rating scales ranged from 0.80 to 0.93 (P < 0.001); the weakest correlation was found between mSAFFE and ABC and the strongest between mSAFFE and FES-I.

Data completeness
One of the 102 participants left FES-I completely blank and another person left both FES(S) and ABC blank. Four additional persons misunderstood the ABC: three persons responded by writing "X" instead of specifying a digit after the items, and the fourth person supplied double digits on each item, resulting in uninterpretable responses. The number of participants that obtained a total score was: ABC, n = 82; mSAFFE, n = 86; FES(S), n = 90; and FES-I, n = 92. The overall mean of missing responses were: FES-I, 0.9%; FES(S), 1.0%; mSAFFE, 1.3%; and ABC, 6.9% (those that left the scales completely blank are not included in these numbers). The number of participants that obtained a total score at t2 Parkinson's disease Activities of Daily Living Scale. Dichotomized: "no difficulties" and "mild difficulties" are counted as "no". "Moderate difficulties", "high levels of difficulties" and "extreme difficulties" are counted as "yes". c Item 3 of the self-administered Freezing of Gait Questionnaire. Dichotomized: all response options but "never" are counted as "yes".

Scaling assumptions
Item means and SDs, respectively, were roughly parallel for most items in each of the FOF scales. Some items of FES-I, ABC and mSAFFE had a larger proportion of participants that chose the worse response options, resulting in worse mean scores (i.e., more difficult items). These were: FES-I items 11 (Walk on slippery surface), 14 (Walk on uneven surface) and 15 (Walk up/down a slope), ABC items 6 (Stand on chair to reach) and 16 (Walk on icy sidewalks), and mSAFFE item 8 (Go out when it is slippery) (Tables 2, 3, 4, 5). A larger proportion of responders chose the best response option for FES-I item 3 (Preparing simple meals) and mSAFFE items 4 (Go to the doctor/ dentist), 6 (Take a shower) and 12 (Walk around indoors) (data available on request). All four scales had corrected item-total correlations exceeding 0.4 (Tables 2, 3, 4, 5).

Reliability
The mean time between responses to the first and the second survey was 16.7 (SD 3.8, min-max 13-38) days. Reliability coefficients, SEM and SDD values for the four FOF scales are presented in Table 6. All scales had Cronbach's alpha >0.90 and ICC >0.80, and one (FES-I) had ICC >0.90. The đ was close to 0 with CI including 0 for all four scales. There were 3 outliers in mSAFFE, 4 in FES-I, 10 in ABC, and 15 in FES(S).

Discussion
This is the first comprehensive comparison of the psychometric properties of four commonly used FOF scales in people with PD. Our main findings were: ABC had markedly worse data completeness than the other scales, all scales showed acceptable reliability (Cronbach's alpha and ICC >0.80) but only FES-I had ICC >0.90, and FES(S) and ABC had substantially more outliers than mSAFFE and FES-I when comparing t1 and t2.
Our sample consisted of more males than females, which is in agreement with prevalence studies of PD Nine persons did not answer all 16 items and did therefore not receive a total score. Possible item score range 1-4, possible total score range 16-64, higher = worse.
[42]. The mean age and PD duration were 73 and 7 years, respectively, which correspond well with a previously reported mean age at symptom onset of 62 to 70 years [43]. Our sample contained fewer fallers than previous studies [3,44], whereas the prevalence of FOF in our sample (55%) was within a previously reported range (38-59%) [3,4]. Self-reported PD severity ranged from mild to severe. The present sample thus seems fairly representative, although it needs to be noted that those with severe cognitive or medical problems were excluded.

Relationship between the scales
The four FOF rating scales correlated ≥0.80, which is not surprising since the content is similar [14]. However, the scales are said to assess different aspects of FOF, i.e., concerns about falling, fall-related self-efficacy, balance confidence and activity avoidance due to the risk of falling [9,11,13,30]. Previous studies have stated that these constructs are not interchangeable and that scale selection should be based on the specific construct of interest [3,8]. Thus, more studies are needed to confirm the relationships between the different FOF scales.

Data completeness
ABC had a substantially higher proportion of missing data than the other scales (6.9 vs. 0.9-1.3%). Four persons completely misunderstood the ABC, implying that the instructions need to be clarified. It should, however, be noted that the Swedish version of the ABC was used in this study and the instructions might be perceived as more clear in the original ABC. To our knowledge, no previous study has presented data completeness for ABC in people with PD or other samples. The percentage of missing data was highest (12.9%) for ABC items 14 and 15. These are the items that are culturally adapted in the Swedish version (changed from stepping on/off escalators into traveling by bus, L. Lundin-Olsson, written personal communication, June 20, 2012). The high number of missing data suggests that these items are difficult to understand or irrelevant to the participants [31]. In fact, three participants had written supplementary comments, stating that they did not travel by bus. An additional 19 participants stated that they always avoided traveling by public transport according to item 15 of mSAFFE. While the original ABC includes instructions on how to respond to activities that the respondent does not engage in, the Swedish translated ABC does not. This might explain the high number of missing responses in these items. However, even if these two items are removed, missing data remains higher for ABC than for the other scales (6.0% vs. 0.9-1.3%).

12.
Hurrying up to answer the telephone 6.5 (3. Eleven persons did not answer all 13 items and did therefore not receive a total score. Possible item score range 0-10, possible total score range 0-130, higher = better.

Scaling assumptions
While FES(S) items were roughly parallel, this was not the case for the other three scales. These findings are not unexpected since the various items within the scales are of different difficulty level. Moreover, items that were rated as more difficult by our sample have been rated as more difficult in previous PD studies as well as in older healthy populations [3,9,11,17]. One could argue that some variation in item difficulty levels is preferable, since this result in a scale that is able to assess FOF in individuals with both low and high levels of FOF. Although classic test theory states that items within a scale should be "roughly parallel" to allow for a summed total score [15,31], no guidelines exist that describe how rigid this judgement should be. Previous studies using the FOF scales studied here have, in fact, all used regular total scores [3,9,11,13,17,18,45].

Targeting
All four scales seem fairly well targeted and met the criterion of floor and ceiling effects below 20% [15]. FES(S) had 17.8% ceiling effect, which is higher than the other scales (4.9-10.5% floor/ceiling effect). A previous PD study found a lower ceiling effect of FES(S) (10.1-10.6%), but a higher floor effect of mSAFFE (18.3-19.4% vs. 10.5% in our study) [3]. No previous study has presented data on floor and ceiling effects on FES-I and ABC in people with PD.

Reliability
All four scales had high internal consistency (Cronbach's alpha >0.90) and acceptable test-retest reliability (ICC >0.80) [35,36]. However, only FES-I had an ICC >0.90, which has been suggested as a minimum when using scales for individual comparisons [36,37]. In comparison with previous reliability studies in PD, our results of internal consistency were consistent with previous studies [3,[16][17][18][19]. Test-retest reliability of FES(S) was lower than previously reported (ICC = 0.82 vs. ICC = 0.87) [3]. The situation was similar for mSAFFE (0.85 vs. 0.92) [3]. The ICC of ABC in the current study was in between the results of the two previous ABC studies that assessed test-retest reliability (0.86 vs. 0.79 and 0.94) [17,18]. These differences are likely to appear as psychometric properties are sample dependent [15]. Missing data includes the four persons who misunderstood the scale, but not the person who left it blank. c Fifteen persons did not answer all 16 items and did therefore not receive a total score. Possible item score range 0-10, possible total score range 0-100, higher = better.
SEM% for the four rating scales varied from 7 to 12%. This implies that a change in a mean score greater than 7 to 12% of the possible scoring range indicates a "real" change (above measurement error), when assessing FOF for a group of people with PD [38]. SDD% were 20 to 33%, indicating that the smallest change in an individual's FOF score that can be interpreted as a "real" change (above measurement error) should exceed 20 to 33% of the possible scoring ranges [39]. FES-I had the lowest SEM% and SDD%, where a difference of at least 4 and 10 points indicated a "real" change on a group and individual level, respectively.

Study limitations
There is a variety of FOF scales [8], and it needs to be acknowledged that this psychometric comparison is not fully comprehensive since only four Swedish translated scales were included. We selected FES-I because it is Sixteen persons did not answer all 17 items and did therefore not receive a total score. Possible item score range 1-3, possible total score range 17-51, higher = worse. recommended by ProFaNE [9], and its forerunners (or adaptations of them) since they are commonly used. More studies of other FOF rating scales are needed, as well as cross-national comparisons, to establish which rating scale that is the best of all available FOF scales. The postal survey study design means that all scales were self-administered, and it needs to be underlined that the present findings may not apply if the scales are administered as an interview. Furthermore, the crosssectional design does not enable us to determine either the responsiveness of the FOF scales, nor the minimal important differences. However, it has been argued that SEM is a reasonable approximation of the minimal important difference [46].

Conclusions
All four FOF scales showed acceptable internal consistency and test-retest reliability. ABC revealed insufficiencies in terms of data completeness, and ABC and FES(S) had many outliers when comparing t1 and t2. When assessing FOF in people with PD, the findings in the present study favoured the choice of FES-I or mSAFFE. However, FES-I was the only scale with ICC >0.9, which has been suggested when using a scale for individual comparisons.