Introduction

The evaluation of the functionality is essential in studies of frailty and health-related quality of life (HRQoL) in the elderly. In fact, functional disability is one of the most important predictors of several outcomes, such as mortality, use of health services, or falls [1, 2], and, at times, it is considered to have more predictive capacity than other traditional variables [3].

When studying functional disability, different approaches and measurement strategies have been used. One of those strategies is the evaluation of activities of daily living (ADL) [4]. ADL refers to the everyday tasks that we perform when we are independent, and these are often divided into basic activities of daily living (BADL) and instrumental activities of daily living (IADL) [2].

Several questionnaires have been developed for the measurement of BADL [4], but the BI [5] is one of the most accepted and generally used in studies of functional capacity in the elderly [6].

The Barthel Index (BI) was initially developed to measure the level of functional independence of patients with neuromuscular and musculoskeletal pathologies [5], but it has been employed in other conditions [6, 7]. The review performed by Cabañero-Martínez et al. in 2009 [1] determined the scarcity of data on its psychometric properties, although it has been used in many studies of disability and it is still employed in current practice. In its Spanish version, only Baztán et al. [7] reviewed its characteristics, but they did not evaluate its validity and reliability with real data.

Any outcome measurement must comply with several characteristics before we can trust in its results. The aim of this paper was to evaluate the psychometric properties of the Spanish version of the BI and its functioning in four cohorts of elderly people.

Methods

The data employed in this paper came from four different cohorts. The first two were obtained from a study that included 2102 patients of 65 years or older who were recruited in the emergency departments (EDs) of six public hospitals in Spain [8, 9]. These patients were admitted with a hip (Cohort 1, 1008 patients) or wrist (Cohort 2, 1094 patients) fracture because of a fall. Patients were excluded if they had physical or psychological impairments that would prevent them from properly completing the questionnaires. Six months after the fracture, follow-up questionnaires were sent to all of the participants by mail. The participants who did not return the materials were sent a reminder letter at 21 days and again at 35 days, if necessary. After that, the participants who still had not returned the materials were telephoned to increase the response rate.

The third cohort (Cohort 3) came from a nationwide, cross-sectional study on HRQoL in 234 people 60 years and older who were living in 17 residential care settings across Spain [10]. Patients were excluded if they had moderate or severe cognitive impairment, which was operationally defined as four or more errors, adjusted by education level, in the Short Portable Mental State Questionnaire [11].

Cohort 4 included 1106 community-dwelling people of 65 years or older who were recruited from across Spain by a geodemographically based proportional multistage stratified sampling method [12]. These people were interviewed in their homes regarding health and HRQoL.

In all cases, the subjects were informed of the goals of each study and provided informed consent to participate. Approvals from the Ethics Committees and/or Review Boards of the participating institutions were obtained.

For the sake of the analysis of this paper, the following variables that were present in all of the studies were included in the analysis: age, gender, education level, comorbidities, and the questionnaires described below.

The BI [5] was used in all of the studies to assess functional status. It measures the ability to perform 10 BADL. The item scores are summed to yield a total score that range from 0 (totally dependent) to 100 (completely independent). A Spanish version of the BI was used [7].

Other questionnaires in Cohorts 1 and 2

The 12-Item Short-Form Health Survey (SF-12) [13] measures general HRQoL. The scores are transformed into the physical component score (PCS-12) and mental component score (MCS-12). Each of these subscales is scored from 0 to 100, with higher scores representing better function.

The Lawton and Brody Index (LBI) [14] was used to measure IADL. We used the dichotomous format provided by Lawton and Brody [14] because of its simplicity. Higher scores indicate greater ability.

The QuickDASH (only for Cohort 2) consists of 11 items derived from the Disabilities of the Arm, Shoulder, and Hand (DASH) questionnaire, which is an instrument designed specifically for upper-extremity disability [15]. Scores are converted to a range of 0 (no disability) to 100 (most severe disability). The QuickDASH has been shown to be equally reliable and valid as the full DASH [16].

The self-administered Western Ontario and McMaster Universities Osteoarthritis Index questionnaire (WOMAC) [17] (Cohort 2) is a health status instrument specific for patients with hip or knee osteoarthritis. We used a Spanish short-form of the questionnaire (WOMAC-SF) [18], which covers two dimensions: pain (PD-W) and function (FD-W). The scores for each dimension were standardized to range from 0 to 100, with 0 representing the best health status possible and 100 representing the worst.

Other questionnaires in Cohorts 3 and 4

The EuroQol-5D (EQ-5D) gives an overall value of the HRQoL based on five questions regarding the state of health [19]. Each item is rated on a three-level scale from 1 (no problem) to 3 (inability to perform or extreme problem).

The Functional Independence Scale (FIS) (Cohort 4) measures functional independence in several activities of daily living [20]. Higher scores indicate greater functional independence.

The depression subscale of the Hospital Anxiety and Depression Scale (HADS-D) is a self-report measure developed to screen depression in non-psychiatric patients [21] (Cohort 4). It consists of seven items rated on a 0–3 Likert-type scale. Thus, the total possible score ranges from 0 to 21.

The Duke-UNC Functional Social Support (DUFSS) measures an individual’s perception of personal social support [22] (Cohort 4). It is composed of 11 items, answered on a 5-point scale. The score ranges between 11 and 55, with higher scores indicating greater perceived social support.

Statistical analysis

All the analyses were performed separately in the four cohorts, except responsiveness, which was analyzed only in Cohorts 1 and 2.

The statistical description of the sample was performed using means and standard deviations (SDs) or frequencies and percentages. Sociodemographic data and BI were compared among the four cohorts of patients. The analysis of variance (ANOVA) with the Scheffe’s test for multiple comparisons or the non-parametric Kruskal–Wallis test was used to compare continuous variables, and the Chi-square or Fisher’s exact tests for categorical variables.

Reliability was assessed with Cronbach’s alpha coefficient [23]. A coefficient >0.70 was considered acceptable [24].

The structure validity was evaluated by means of a confirmatory factor analysis (CFA) for categorical variables to confirm the structure proposed by the original authors [5]. Various indexes of fit were evaluated [25]: (a) the root mean squared error of approximation (RMSEA), in which a value <0.08 was considered acceptable; and (b) the comparative fit index (CFI) and Tucker–Lewis Index (TLI), both of which had to be >0.90 to be satisfactory. Factor loadings were also examined, and those ≥0.40 were considered suitable.

The convergent and discriminant validity of the BI were explored by its correlations with the questionnaires described above. Regarding convergent validity, we expected high correlations with PCS-12, LBI, QuickDASH, the items of self-care, and mobility for the EQ-5D, FIS, and HADS-D. For discriminant validity, we sought to find lower correlations with MCS-12, PD-W, and DUFFS.

For known-groups validity, we expected that the BI would capture the differences between participants divided by different criteria: (a) age (≤75, 75–85 and >85 years); (b) number of comorbidities (0, 1–2 and ≥3); and (c) gender. Following the results found in other studies and/or our experience, we hypothesized that the BI score would be lower (greater dependency) for older people [26], those with more comorbidities [27] and for women. Therefore, the BI scores were compared among the different subgroups using the t test or the ANOVA with Scheffe’s test for multiple comparisons or using the non-parametric Wilcoxon or Kruskal–Wallis tests.

The predictive validity was analyzed by studying the ability of the Barthel Index to predict mortality at 6 months by means of the area under the ROC curve (AUC). Further, to establish the optimal cut-off value of the Barthel Index for this prediction, the ROC analysis was used considering the one that maximized the sum of sensitivity and specificity. This analysis was performed in patients with hip or wrist fracture.

Regarding responsiveness, we first compared sociodemographic characteristics and baseline BI between patients who responded to the follow-up and those who did not. We used the Chi-square test or Fisher’s exact test for the comparison of categorical variables, and the t test or the non-parametric Wilcoxon test for quantitative variables. Then, among responders, the responsiveness study was performed separately according to groups defined by the change score obtained in the LBI as follows: those patients with LBI at baseline equal or lower than at follow-up were classified as “unchanged or improved” and those with LBI at baseline higher than at follow-up were considered “worsened.” We decided to combine the improved and the unchanged patients because the first group was small in both cohorts, and the change scores were similar to the unchanged group.

The means and SDs were calculated for the BI at baseline and at the 6-months, and a paired t test was used for the comparison of scores at the two time points. Ceiling and floor effects at baseline and at 6 months were examined to evaluate the acceptability of the scales. We used 15% as the critical value [28]. To measure the responsiveness, we used the standardized effect size (SES) and the standardized response mean (SRM) [29]. Cohen’s benchmarks were used to classify the magnitude of the effect sizes [30]: not significant, <0.20; small, 0.20‒0.50; moderate, 0.50‒0.80; and large, >0.80. We expected a larger SES or SRM in patients who were classified as “worsened” than among those classified as “unchanged or improved.”

All statistical analyses were performed with SAS for Windows statistical software, version 9.2 (SAS Institute, Inc., Cary, NC) and Mplus version 6.1 software (Muthen & Muthen, Los Angeles).

Results

The descriptive data of the four cohorts are presented in Table 1. The mean age ranged between 72.07 years for Cohort 4 and 83.66 years for Cohort 1. The majority of the participants were women, and most of the participants had comorbidities, particularly in Cohort 3, in which almost 80% of the participants had 3 or more comorbidities.

Table 1 Descriptive data of cohorts
Full size table

Reliability

The Cronbach’s alpha coefficients were all greater than 0.70: 0.92 for Cohorts 1 and 3, 0.91 for Cohort 2, and 0.92 and 0.88 for Cohort 4.

Confirmatory factor analysis

The fit indexes of the CFA satisfied the thresholds in all of the cohorts. The RMSEA was below 0.08 in all cases, and the CFI and TLI values were all above 0.90.

All of the factor loadings were above 0.40 and were statistically significant (p < 0.05) (Table 2).

Table 2 Confirmatory factor analysis: factor loadings and fit indexes
Full size table

Convergent and discriminant validity

The discriminant validity was confirmed. For Cohort 1, the correlation between the BI and MCS-12 was 0.20 and that between the BI and PD-W was −0.25. For Cohort 2, the correlation between the BI and MCS-12 was also low (r = 0.20). Finally, in Cohort 4, the BI and the DUFFS showed a correlation of r = 0.10.

Regarding the convergent validity, the highest correlations were found between the BI and the LBI in Cohorts 1 and 2 (r = 0.76 and r = 0.58, respectively), the FD-W and PCS-12 in Cohort 1 (r = 0.73 and r = 0.57, respectively), and with EQ-5D, in the self-care item, in Cohort 3 (−0.66) (Table 3).

Table 3 Convergent and discriminant validity
Full size table

Known-groups validity

More dependency was observed in the older people in all of the cohorts, except the third one (p = 0.7098). As age increased, the BI scores decreased, and the differences were significant between all age groups (Table 4).

Table 4 Known-groups validity
Full size table

Similar results were observed with comorbidities: as the number of comorbidities increased, the BI decreased. The differences were statistically significant in Cohorts 1, 2, and 4.

The hypothesis for gender was only confirmed in Cohort 3, in which women scored significantly lower than men.

Predictive validity

The AUC of the Barthel Index for the prediction of mortality at 6 months in patients with hip fracture was 0.674, and the optimal cut-off value was 85, obtaining a sensitivity of 63.64% and a specificity of 64.77%. In patients with wrist fracture, the AUC was 0.739, with an optimal cut-off value of 90 resulting in a sensitivity of 58.82% and specificity of 83.98%. Considering both hip and wrist fracture patients together, the AUC was 0.742, with a sensitivity and specificity of 79.05 and 61.93%, respectively, corresponding to a cut-off value of 95.

Responsiveness

Differences between the respondents and non-respondents were only found in Cohort 1. In this group, respondents had a higher education level and higher scores in the PCS-12 and lower scores in the PD-W and the FD-W than non-respondents.

Although the floor effects were low, the ceiling effects were high in both cohorts and in the worsened and unchanged groups (Table 5). These ceiling effects decreased in the post-fracture evaluation but were still beyond 15% in almost all of the groups.

Table 5 Responsiveness parameters after 6 months follow-up, according to the worsened and unchanged groups, in Cohorts 1 (n = 707) and 2 (n = 857)
Full size table

The differences between baseline and follow-up BI were statistically significant. The change was always negative because all of the participants perceived less functionality and thus more dependency after the facture. In the worsened groups for both cohorts, the SES and SRM indexes were above the Cohen’s threshold of 0.20 for considering that a change has been found (SES: 1.81 for Cohort 1 and 0.93 for Cohort 2; SRM: 1.13 for Cohort 1 and 0.68 for Cohort 2). For the unchanged groups, although the differences between baseline and follow-up were also significant, these indexes were lower in both cohorts (SES: 0.41 for Cohort 1 and 0.11 for Cohort 2; SRM: 0.57 for Cohort 1 and 0.14 for Cohort 2), as expected.

Discussion

The BI is a very widely used questionnaire for the assessment of the functionality of the elderly and has been used in several conditions [6, 7].

Although the BI has been translated into Spanish [7], and there is some information on its psychometric properties [1], no papers have been found regarding its validation process. We employed this questionnaire in four different cohorts for the assessment of functionality in older adults in Spain, and we used these data to evaluate the psychometric properties of the Spanish version of the questionnaire.

Our results show that this version of the BI has good psychometric properties. It has good reliability, we have confirmed its structure via CFA, and it can discriminate between groups and detect differences between follow-up points.

The internal consistency complies with the standards [23]: the Cronbach’s alpha coefficient was greater than 0.70, indicating that all the items are measuring the same concept. The internal consistency of the BI was also good in other psychometric analyses [31] or cultural validations [32]. Other reliability analyses could be evaluated in future studies to provide more evidence of this psychometric characteristic of the BI. In fact, a review published by Sainsbury et al. in 2005 stated that there is still a lack of data on test–retest reliability and inter-observer disagreement in chronic older people [6].

The results of the CFA supported the hypothesis of the unidimensionality of the BI, with good fit indexes, as other authors have also found [33]. Nevertheless, other researchers have identified two different factors [34]. Laake et al. [34] found differences in the number of factors depending on the type of patients who are studied. They proposed a two-factor structure in a group of patients with hip fracture, but in our study the factor analyses indicated a single-factor structure. It is possible that the different sample sizes (102 in the Laake et al. study and 1008 in ours) could have influenced the results of the factor analysis, resulting in different structures.

With regards to convergent and discriminant validity, almost all our hypotheses were confirmed. Although all of the correlations for convergent validity were statistically significant, some of them were weak, in accordance with those found in other studies [31, 33], and they were lower than the correlations observed by Cabañero-Martínez et al. [1]. Regarding discriminant validity, it has been difficult to identify other studies that evaluate this type of validity. Cabañero-Martínez et al. [1] did not find any discriminant analyses in their review of the Spanish versions of the BI.

The hypotheses for evaluating known-group validity were confirmed in all cases, although the differences in the BI baseline scores for gender were statistically significant only in one of the cohorts. In fact, some other researchers did not find differences in BI scores between men and women [26]. Other researchers did find differences in the BI scores for age [26] and comorbidities [27], showing that older people and those with more comorbidities had lower BI scores.

Finally, the responsiveness indexes found moderate to large changes in the worsened patients, indicating that the BI can detect differences between two follow-up points. Other studies have also evaluated the responsiveness of the BI by employing different indexes and have reported similar [6] or weaker results [35].

The strengths of our study include the comparison of the psychometric properties of the Spanish version of the BI in four different cohorts. Despite the fact that the samples contain some differences, the results are similar, which gives confidence in the validity and reliability of the Spanish version of the BI. Another strength of this study is the large size of the samples, which complies with the minimum standards required for these types of analyses.

This study has also some limitations that must be addressed. First, these analyses were performed with four different cohorts obtained from different studies. This could introduce some bias because each study was performed under different conditions and with different patients. However, this limitation could be considered irrelevant because the psychometric results were similarly good in the four samples. Second, we evaluated the reliability of the questionnaire by means of the internal consistency with the Cronbach’s alpha coefficient, but better evidence on the reliability should be added with a test–retest analysis. The third limitation is related to the missing data, which particularly affects the responsiveness results. We found some differences between the respondents and non-respondents in Cohort 1; thus, the responsiveness data should be interpreted with these facts in mind.

For future studies, we recommend performing some new type of analysis to increase the evidence of the reliability and validity of this version of the BI. In addition, other measurement properties could be analyzed according to the item response theory.

We have employed the BI in these cohorts mainly as a measure of functional status, but as frailty is a complex and multicomponent concept, where the new concept of cognitive frailty plays a crucial role [36, 37], future studies could explore the use of the BI as a measure of the association between physical and cognitive deterioration, not only in the field of an specific pathology, but also in studies of general health in the elderly.

Because the increasing age of the population is currently influencing the incidence of chronic conditions, it is essential to have access to reliable, valid, and easy-to-use instruments for the evaluation of the health status of patients. The Spanish version of the BI fulfills these characteristics because it has good psychometric properties; thus, it can be used in the evaluation of functionality and basic activities of daily living in elderly people with different conditions.