Exploratory and confirmatory factor analysis of the questionnaire on Palliative Care for Advanced Dementia (qPAD) using a large sample of staff from Australian residential aged care homes

Abstract Background The Questionnaire on Palliative Care for Advanced Dementia (qPAD) is increasingly being used to assess residential aged care workers' knowledge and attitudes about palliative care for people with dementia. The qPAD developers performed an exploratory factor analysis and assessed the internal consistency using a small sample. Aim The aim of this study was to further assess the structural validity of the qPAD using a large sample of qPAD responses from staff who work in residential aged care homes in Australia. Methods Data from 727 care staff who participated in an Australian dementia palliative care training project were used for exploratory factor analyses, assessment of internal consistency, and confirmatory factor analysis of the knowledge test and attitude scale components of the qPAD. Results The exploratory factor analysis of the knowledge test produced a four‐factor solution. One item loaded weakly, and four items had cross‐loadings. Factor labels for the knowledge test were difficult to define. Factor analysis of the attitude scale produced a three‐factor structure with good internal consistency—Feeling valued and part of the care team (α = 0.88), Family and team engagement (α = 0.75) and Perceptions and beliefs (α = 0.83). Confirmatory factor analysis indicated improvements in model fit were needed for both the knowledge test and attitude scale. Conclusion The findings of this factor analysis differed from the original study. The attitude scale produced a three‐factor structure, but the knowledge test requires further development due to weak and cross‐loadings of several items, inadequate internal consistency of factors and poor model fit.


| INTRODUC TI ON
In Australia, over half the people in residential aged care homes (or nursing homes) are living with advanced dementia where they will spend the last months or years of life (Australian Institute of Health and Welfare, 2020). Care home staff therefore require knowledge, attitudes and skills that support the palliative and end- Several instruments have been developed to measure general knowledge and biomedical knowledge of dementia including the different causes of dementia, signs and symptoms, differential diagnoses, and risk factors (Spector, 2013). The Dementia Knowledge Assessment Scale (DKAS) also measures knowledge of psychosocial aspects of communication and care, and has been assessed among lay people, family carers and healthcare workers including people who work in residential aged care (Annear, 2017). Few instruments measure attitudes towards dementia, of which the Dementia Attitude Scale (DAS) is the most widely used (O'Connor, 2010) and has undergone psychometric assessment using samples of university students. However, the DAS validity and reliability have not been assessed among people who work in residential aged care. For our study, we wanted an instrument that had been tested among residential aged care workers and focussed on dementia-specific palliative care needs. The Questionnaire on Palliative Care for Advanced Dementia (qPAD) fulfilled these criteria (Long, 2012). The qPAD developers assessed the psychometric properties of the questionnaire in a sample of 85 caregiving staff from four care homes in Arizona, USA (Long, 2012 How could the findings be used to influence policy or practice or research or education?
• Our findings suggest more research is required to further develop and evaluate the knowledge test items of the qPAD before it can be used for education and research purposes.
• The qPAD attitude scale can be used to assess nursing home staff attitudes towards caring for people with advanced dementia, and to measure the impact of targeted interventions and improvement programs. attitude scale items using exploratory factor analysis, a method used to determine the number of distinct constructs (or factors) needed to account for the pattern of correlations among a set of measures (Fabrigar, 2011). In the original study, the exploratory factor analysis produced three factors from the knowledge test items-anticipating needs, preventing negative outcomes and insight and intuition, and three factors from the attitude scale-job satisfaction, perceptions and beliefs, and work setting support of families. Although the findings of this study suggested the two sections provided reliable and valid measures, the authors acknowledged the small sample size as a limitation and stated '…the instrument should be used with greater numbers of health caregivers before definitive statements can be made'.
Since its development, the qPAD has been used in several research studies and translated for use in Japan and Taiwan (Agar, 2017;Chen, 2018;Luckett, 2019;Nakanishi, 2016;Nakanishi, 2015). The Japanese study by Nakanishi et al. (2015) reported moderate to good Cronbach's alpha coefficients for the attitude scale items: 0.84 for the 12 items and between 0.55 and 0.79 for the three factors as identified in the original study. Similarly, the Taiwanese study by Chen and colleagues assessed the content validity index and internal consistency and reported good results. Agar et al. (2017) in a cluster randomised controlled trial of facilitated family case conferencing, used the qPAD to assess aged care workers' baseline and change in dementia palliative care knowledge and attitudes. They analysed baseline qPAD data to explore associations between qPAD scores and aged care facility/personal characteristics, and found being a nursing manager, registered nurse or enrolled nurse, and having a preferred language of English were associated with more favourable knowledge test scores; and having tertiary level education and greater experience in dementia care was associated with favourable attitudes (Luckett, 2019). None of these studies evaluated the structural validity of the qPAD using confirmatory factor analysis.

| AIM
The aim of this study was to assess the psychometric properties, including structural validity and internal consistency of the qPAD using a large sample of Australian residential aged care staff responses, and to compare these findings to the factor structure defined by the USA developers.

| Study design and subjects
The present study was part of a larger study-the IMproving The qPAD was used in the above-mentioned study as a secondary outcome measure of dementia knowledge and attitudes. It was chosen for the following two reasons: (i) it was developed specifically for direct care staff who work in residential aged care (also known as nursing homes and long-term care), and (ii) it measures staff knowledge of and attitudes about palliative and end-of-life care for people with dementia which was the focus of the IMPETUS-D project.
The study involved 24 residential aged care homes of a single private provider in Sydney, Melbourne and Adelaide, Australia. To be eligible for inclusion in the study, care homes had to have a minimum of 20 residents with a diagnosis of dementia and high care needs living permanently in the homes.
After recruitment of residential aged care homes, the qPAD data were collected to determine care home staff baseline level of knowledge and attitudes towards palliative dementia care. All direct care staff, defined as nurses and personal carers/attendants in nursing, who were working at the care homes, were invited to participate in the survey. Participation was voluntary, and staff were required to give informed consent prior to completion of the survey.

| Measurements
The survey instrument collected participant characteristics, including position, age, gender, education level, years worked in aged care, previous training in dementia and previous training palliative care.
This was followed by qPAD instrument measures. For each item of the knowledge test, respondents were asked to select from Agree, Disagree or Don't know. A score of 1 is allocated to each correct response and 0 for an incorrect or 'Don't know' response. Knowledge test scores range from 0 to 23 with higher scores indicating greater knowledge. For the attitude scale, respondents were asked to rate each item using a five-point scale of 1 Strongly disagree, 2 Disagree, 3 Neutral, 4 Agree and 5 Strongly agree. Attitude scale scores range from 12 to 60 with higher scores indicating more positive perceptions and attitudes. A total qPAD score is derived by adding the knowledge test and attitude scale scores. However, we did not use the total score for the purposes of the factor analysis.

| Data collection
Surveys were sent to all direct care staff (N = 1947) via email with a unique survey link generated by Research Electronic Data Capture (REDCap) (Harris, 2009). To increase participation at facilities with low response rates, personalised letters of invitation with hard copy surveys and reply-paid envelopes enclosed were also sent.
Responses were collected directly via REDCap, or hard copy responses were entered into REDCap by members of the research team. Data were then exported for analysis, and responses were randomly assigned into two datasets of approximately equal size for exploratory factor analysis sample (N = 364) and confirmatory factor analysis sample (N = 363). Only complete responses were included in the analysis. Statistical analysis was conducted using Stata 15.1 MP (StataCorp., 2017).

| Exploratory factor analysis (EFA)
Descriptive statistics were calculated, including mean and standard deviation of knowledge test, attitude scale and total qPAD scores, the proportion correct for each knowledge test item and the mean for each attitude scale item. EFA were conducted to assess the structure of the knowledge test and attitudes scale. Initial exploration of the Australian data included replicating the principal factor analysis methods used in the USA study, retaining three factors, and employing an oblique rotation model for the knowledge test and orthogonal rotation model for the attitude scale. The findings are available as Tables S1 and S2.
However, the original qPAD factor analysis used traditional methods based on Pearson correlations and the assumption that the variables are continuous and follow a normal distribution. The EFA presented in this study differs from the original study. We have acknowledged the qPAD items are not continuous variables-the knowledge test uses a binary measure, and the attitude scale uses a five-point Likert scale which we consider ordinal. We used a more optimal approach to the factor analysis methods for non-continuous variables by using tetrachoric correlations for the knowledge test and polychoric correlations for the attitude scale to estimate the correlation had the measurement scales been continuous (Holgado-Tello, 2008). We used Promax oblique rotation to allow for correlated latent factors.
Prior to extraction of factors, the suitability of the data for factor analysis was assessed and met using Bartlett's chi-square test of sphericity (p < 0.05) and Kaiser-Meirer-Olgin (KMO > 0.5) measure of sampling adequacy. Scree plots were examined looking for an 'elbow' or distinct break in the slope of the scree plot (Cattell, 1966), and parallel analysis (Horn, 1965) was used to determine the number of factors to retain in the models. The exploratory analysis employed principal factor analysis with Promax oblique rotation. The internal consistency of the knowledge test, attitude scale and each of the scales' factors was assessed using Cronbach's alpha coefficient with an alpha equal to or greater than 0.70 considered satisfactory (Nunnally, 1978). Additional exploration was conducted retaining up to six factors for the knowledge test and up to five factors for the attitude scale.
The structure models were evaluated for the following: (i) number of items with salient loadings defined as loadings ≥0.40, (ii) number of items that have cross-loadings defined as loadings ≥0.40 onto more than one factor, (iii) internal consistency ≥0.70 and (iv) minimum of two items load onto a factor to allow meaningful interpretation for factors to be descriptively labelled.

| Confirmatory factor analysis (CFA)
Confirmatory factor analysis was performed using structural equation modelling with a maximum likelihood parameter estimation method on the confirmatory factor sample using the tetrachoric (knowledge test) and polychoric (attitude scale) correlations of the items identified from the best model in the EFA. Standardised factor loadings and corresponding p values were examined (p < 0.05).
The equation-level fit was assessed by examining R-squared values for each item, and overall model-level fit was evaluated using the  (Jackson, 2009). CFA using the factor structures from the original EFA was then performed, and goodness-of-fit statistics from both models were compared.

| Study participants
A total of 727 (37%) direct care staff from 24 care homes participated in the questionnaire. Table 1 summarises the participants' characteristics of the total sample, including the average qPAD scores. Most participants were personal carers (70%), female (85%), and 71% had worked less than 10 years in aged care homes. The mean (standard deviation) knowledge test, attitude scale and total qPAD scores were 14.9 (3.3), 44.7 (7.7) and 59.5 (8.7), respectively.

| Exploratory factor analysis-knowledge test
Examination of the scree plot of eigenvalues and parallel analysis of the knowledge test items suggested a three-factor and four-factor model, respectively (Supplementary Information S3). The three-and four-factor models were assessed against the stated criteria. 1 All items except item 11 (highest factor loading 0.391) loaded satisfactorily onto the four-factor structure. Item 12 (highest factor loading 0.285) and item 7 (highest factor loading 0.395) did not load well onto any factors of the three-factor model. Both models had an item cross-load onto more than one factor. The four-factor Promax rotated model was selected as the best EFA model and has been presented in Table 2. The four factors accounted for 55% of the total variance, and the internal consistency of the four-factor model was 0.75, 0.49, 0.59 and 0.40. Factor 1 was the only factor to meet the internal consistency criteria (≥0.70).
We were unable to label the four factors as items that loaded onto each factor covered a mix of themes and varied to the USA findings, for example the 'Factor 1-Anticipating needs' items loaded onto factors 1 and 4 of this study, 'Factor 2 -Preventing negative outcomes' items loaded across factors 1 and 4, and items from 'Factor 3 -Insight and intuition' loaded across factors 2, 3 and 4 of our study.

| Confirmatory factor analysis-knowledge test
The CFA of knowledge test items assessed the fit of the above four-factor structure using the CFA sample (N = 363). Standardised factor loadings for twenty (83%) items reached statistical significance (p < 0.05), while factor loadings of items 3, 7 and 22 were weak and did not reach statistical significance (Table 3).

| Exploratory factor analysis-Attitude scale
The screen test and parallel analysis supported retaining three factors for the EFA of the attitude scale items (See Supplementary   Information S3). Table 5  Five items loaded onto factor 1 defined as 'Feeling valued and part of the care team', four items loaded onto factor 2 defined as 'Family engagement', and the remaining three items loaded onto factor 3 which we defined as 'Perceptions and beliefs'. In our study, items 10 and 12 loaded onto factor 2-Family engagement, whereas in the USA study these items loaded onto factor 1 which was labelled 'Job satisfaction'. Finally, mean values of items ranged from 3.2 (item 1 and item 2) to 4.2 (item 12).

| Confirmatory factor analysis-Attitude scale
The CFA of attitude scale items assessed the fit of the three-factor structure using the CFA sample (N = 363). All standardised factor loadings were statistically significant (p < 0.05) ( Table 6) (Table 7).

| DISCUSS ION
This study builds on the psychometric assessment of the qPAD instrument reported previously by Long et al. (2012). Using a large sample of qPAD responses from over 700 Australian residential aged care workers, our exploratory and confirmatory factor analyses showed some differences and similarities to the original USA study.

TA B L E 7
Goodness-of-fit statistics for the confirmatory factor analysis of the attitude scale (N = 363) The current study was limited to evaluating the psychometric properties of the qPAD using factor analysis and by assessing internal consistency. Further development of a knowledge measure may require some fundamental item and scale development work.
Evaluation of convergent and/or discriminant validity of the attitude scale would further our understanding.
Since the qPAD's development over 10 years ago, there has been With the ageing population and growing prevalence in dementia, a tool that comprehensively assesses nurses' and other aged care staff knowledge of and attitudes towards dementia-specific palliative care would be valuable, and could be used to understand the areas of knowledge and attitudes that need improvement, as well as evaluate the impact of targeted interventions. However, our findings indicated further development and evaluation of the qPAD is required before it can be used for education and research purposes.

| CON CLUSIONS AND IMPLIC ATIONS
In this study, exploratory factor analysis of the qPAD resulted in a four-factor structure of the knowledge test and a three-factor structure of the attitude scale. The factor structure and some of the factor labels differed to those proposed in the original study. This, together with confirmatory factor analyses suggested a need for redevelopment of the knowledge test items and further validation for the attitude scale. We therefore recommend revision of the qPAD, with a particular focus on knowledge test items.

ACK N OWLED G EM ENT
Thank you to Arlene Nunez, Gurleen Kaur, Sam Rojas-Ponce and Nahid Moradi for their assistance with survey data collection, and to the residential aged care home staff who participated in the survey and to Carol Long for sharing information from her original study. Scholarship. The funders do not have any input in the study design, analysis or manuscript development.

CO N FLI C T O F I NTE R E S T
The authors declare no conflicts of interest.

DATA AVA I L A B I L I T Y S TAT E M E N T
Data will be available after analyses is finalised and report / publication has been submitted and approved. Unidentifiable individual participant data and related data dictionaries will be available.

Access is subject to approval by the Principal Investigator Professor
Wen Kwang Lim.

E N D N OTE
1 Additional factor analyses retaining one, two, five and six factors for the knowledge test and one, two, four and five factors for the attitude scale did not achieve better results.