Zarit Caregiver Burden Interview: Development, reliability and validity of the Chinese version
*Lie Wang, MS, Department of Social Medicine, School of Public Health, China Medical University, no. 92 Beier Road, Heping District, Shenyang, 110001, China. Email: email@example.com
Aims: To develop a specific scale used in measuring caregiver burden in China, and to evaluate its reliability and validity.
Methods: Participants from the First and Second Affiliated Hospital of China Medical University and the Hospital of Tiefa Coal Industry Group completed the Chinese version of the Zarit Caregiver Burden Interview.
Results: A total of 523 caregivers were included. The internal consistency of the Zarit Caregiver Burden Interview was high (Cronbach's α 0.875). The item–total correlations were all statistically significant (P < 0.01). Confirmatory factor analysis confirmed the five factors of the Zarit Caregiver Burden Interview in this study, and the goodness-of-fit indices reported for this 5-factor model all fell within the acceptable range.
Conclusions: The Chinese version of the Zarit Caregiver Burden Interview is reliable and valid for use. This study has important implications for burden measurement in Chinese caregivers.
CAREGIVER BURDEN IS defined as a multidimensional biopsychosocial reaction resulting from an imbalance of care demands relative to caregivers' personal time, social roles, physical and emotional states, financial resources, and formal care resources given the other multiple roles they fulfill.1 Research has indicated that caregiving over a long period can have a negative impact on caregivers both mentally and physically.2–4 Given the magnitude of services provided and the sacrifices made by caregivers, caregiver burden has been recognized as a serious public health concern.
The caregiving problem has become a prominent problem in China due to aging and the only child family policy. It is estimated that there will be 400 million elderly aged ≥60 in China by the year 2050, who will account for 25% of the total population.5 The demands for and utilizations of the health service are higher in the elderly compared with younger citizens, and the number of people who pay for the service will reduce, so care insurance has been emphasized by many researchers. This research aimed to develop a specific scale used in measuring caregiver burden in China, and to evaluate the reliability and validity.
This study was carried out at the First and Second Affiliated Hospital of China Medical University in Shenyang and the Hospital of Tiefa Coal Industry Group in Tieling of Liaoning province. The study included 523 caregivers who cared for inpatients examined between January 2008 and March 2008 in the three hospitals. After providing written informed consent for the conduct of the survey, caregivers were interviewed by nurses, and the total number who completed the survey was 523, with a valid response rate of 98.3%. The protocol of this study was approved by the Research Ethics Committee, China Medical University.
The Zarit Caregiver Burden Interview
It was Zarit who first proposed an operational definition of caregiver burden and developed an assessment tool for feelings of caregiver burden, the Zarit Caregiver Burden Interview (ZBI).6 The ZBI is now the instrument most widely used in North America and Europe for assessing the burden experienced by family caregivers who look after the community-residing impaired elderly.7 It comprises 22 questions graded on a scale from 0 to 4, according to the presence or intensity of an affirmative response, and measures the caregiver's health, psychological well-being, social life, finances, and the relationship between the caregiver and patient. The ZBI was adapted to several languages, and the internal consistency ranged from 0.85 to 0.94.8 However, until now there has not been a Chinese version.
The adaptation process
The English and Japanese version of the ZBI was translated into Chinese by two bilingual translators, both of whom have a medical background. This version was viewed and discussed by the two translators and experts specializing in social medicine. Discrepancies were resolved by consensus to achieve conceptional equivalence. For an accuracy check, this version was translated back to English and Japanese by two other bilingual translators. The back-translated version was compared with the original English and Japanese versions, and they were found to be comparable.
Statistical analysis was conducted using spss 11.5 for Windows and Amos 6.0. The internal consistency reliability was assessed using Cronbach's α coefficient and item–total correlation. A Cronbach's α coefficient above 0.80 is indicative of a good internal consistency. Factor analysis was used to assess the construct validity of the first 21 items. Item 22, a global measure of the burden of caregiving was deleted from the factor analysis because of the global nature of the item and because it is known to correlate highly with all other items.9 A principal component exploratory factor analysis was performed on the 21 items using a varimax rotation, and confirmatory factor analysis was performed using Amos 6.0. The appropriateness of the theoretical models to the empirical data was evaluated through maximum likelihood goodness-of-fit χ2-test statistics and empirical goodness-of-fit indices. Although a nonsignificant P-value indicates that the model fits the actual data, the χ2 statistic and resulting P-value are sensitive to sample size as well as to structural and distributional assumptions. The fit indices in general represent the proportion of the variation among measured variables, captured by the model. Although there is no consensus on the best measures of fit indices, the criteria that are widely accepted include root mean square error of approximation (RMSEA) (values of 0.08 or less indicating good fit), normed fit index (NFI) (values of 0.80 or higher indicate good fit), goodness-of-fit index (GFI) (values of 0.85 or higher indicate good fit), and comparative fit index (CFI) (values greater than 0.80 indicate good fit).10
Demographics of caregivers and patients
Caregiver ages ranged from 16 to 79 (44.2 ± 12.4 years), and there were 213 (40.7%) men and 310 (59.3%) women. In these subjects, the majority of the caregivers were spouses (32.9%) and children (43.2%). Patients in this study came from orthopedics (14.7%), neurosurgery (7.6%), neurology (9.0%), cardiovasology (8.6%), medical oncology (49.1%) and surgical oncology (11.0%). The median of patients' duration of illness was 3 months (lower quartile = 1 and upper quartile = 8.3, respectively).
The internal consistency of the ZBI was high (Cronbach's α 0.875). The item–total correlations ranged from 0.138 to 0.708 and were all statistically significant (P < 0.01). Cronbach's α coefficient increased to 0.877 if item 20 was deleted from the scale, and increased to 0.881 if item 21 was deleted (Table 1).
Table 1. Item–total correlation and alpha if item deleted from Zarit Caregiver Burden Interview
Exploratory factor analysis
Factor analysis was appropriate as the Kaiser–Meyer–Olkin measure of sampling adequacy was 0.867 (the items measure a common factor), and the Bartlett test of sphericity was 3487.085 (P < 0.001), indicating adequate sampling and suitable correlation matrix for the analysis. Principal component analysis (PCA) with varimax rotation was conducted. Item 7 was removed because it failed to have adequate loadings (<0.5) within important factors. Finally, five factors had an eigenvalue greater than 1 and accounted for 56.415% of total variance (Table 2). The first factor had high loadings on such items as ‘social life suffering’, ‘having less privacy’ and ‘being depended on’. These items are regarded as the sacrifice for caregiving. Therefore, factor 1 was termed ‘Sacrifice’. Factor 2 mainly contained items representing financial and time adequacy because of caregiving. Therefore, factor 2 was interpreted as representing ‘Loss of control’. The items in factor 3 included embarrassment, anger and other negative feelings. Therefore, factor 3 was named ‘Embarrassment/Anger’. Factor 4 was composed of items 20 and 21, which were evaluations on caregiving. Therefore, factor 4 was named ‘Self-criticism’. Factor 5 had high loadings on such items as ‘asking for more help than needed’ and ‘having not enough time for oneself’, which are regarded as patient's dependency. Therefore, factor 5 was termed ‘Dependency’.
Table 2. Exploratory factor analysis
|Explained variance (%)||28.691||9.064||7.504||6.154||5.001|
Confirmatory factor analysis
The 5-factor model of this study was tested for goodness of fit. Although the χ2 statistic indicated that the model did not provide a good fit to the data, χ2(160) = 660.9, P < 0.01, other goodness-of-fit indices reported for this 5-factor model (RMSEA = 0.077; CFI = 0.841; NFI = 0.802; GFI = 0.886) all fell within the acceptable range. Table 3 showed Embarrassment/Anger and Self-criticism were not significantly correlated (r = 0.025, P > 0.05), and so did Self-criticism and Dependency (r = 0.053, P > 0.05). There were significant correlations between other factors (P < 0.01).
Table 3. Standardized estimate correlations between factors
|Loss of control||–||1||0.465*||0.255*||0.676*|
A high Cronbach's α coefficient indicated that the internal consistency of ZBI was good. The item–total correlations that were all statistically significant (P < 0.01) indicated the representation of measurement theme. Cronbach's α coefficient was used for adjusting if the item was deleted. If the item was not correlated enough with the measurement objective, Cronbach's α coefficient would increase after deleting the item. In this study, Cronbach's α coefficient increased to 0.877 if item 20 was deleted from the scale, and increased to 0.881 if item 21 was deleted. Therefore, items 20 and 21 should be considered for deletion.
It was suggested that burden is a multidimensional construct and a global score may not provide a complete and accurate assessment. Caregivers with an identical score may be affected by different aspects of burden.11 Whitlatch et al. examined the ZBI in a sample of caregivers of family members with dementia, and proposed a two-factor (Personal Strain and Role Strain) structure for the ZBI that contained 18 items.12 However, there was no further information on the model fit or variance explained regarding this factor analysis. Knight et al. proposed a three-factor (Embarrassment/Anger, Patient's Dependency and Self-criticism) structure for ZBI that contained 14 items.13 According to his explanations, the remaining seven items did not load on the three factors nor coalesce to form a meaningful additional factor that seemed to tap specific aspects of burden.
However, data in this study formed a new five-factor structure that yielded a good fit. This may bring a new understanding of the fact that the Chinese caregivers were probably faced with more complicated feelings associated with caregiving. In Chinese culture, due to the Confucian thought of filial piety and the Chinese law, it is regarded as an obligation for family members to provide care to patients even if they suffer from financial strain or lack of time. Moreover, it is immoral to express feelings such as ‘boring’ or ‘embarrassed’. Therefore, complicated feelings may exist during the caregiving period. However, this might also have resulted from the variety of demographic factors in patients and caregivers of this study. Although there were differences of sampling, the ‘Self-criticism’ factor reported in this study consists of two items that are identical to those of the ‘Self-criticism’ factor reported by Knight et al. Item 7 did not load on the five factors, and nor did the studies by Whitlatch and Knight. This may indicate a specific aspect of caregiver burden in future research. As for the relationships between factors in this study, all the factors correlated with each other significantly except Self-criticism with Embarrassment/Anger and Dependency, which indicated a weaker relation between Self-criticism and other factors.
The study population only comprised inpatients with consideration to the practicability of sampling. Data about outpatients should also be collected. The other limitation was that we did not study the durations of patients' hospitalization.
Our special thanks to the participants and nurses in the study.