Psychological Well-Being and Quality of Care: A Factor-Analytic Examination of the Palliative Care Outcome Scale

Transcription

1 Vol. 40 No. 1 July 2010 Journal of Pain and Symptom Management 67 Original Article Psychological Well-Being and Quality of Care: A Factor-Analytic Examination of the Palliative Care Outcome Scale Richard J. Siegert, PhD, Wei Gao, PhD, Frank H. Walkey, PhD, and Irene J. Higginson, BM BS, FRCP, PhD King s College London, Department of Palliative Care, Policy and Rehabilitation, (R.J.S., W.G., I.J.H.), School of Medicine at Guy s, King s College and St. Thomas Hospitals, King s College London, London, United Kingdom; and School of Psychology (F.H.W.), Victoria University of Wellington, Wellington, New Zealand Abstract Context. The Palliative Care Outcome Scale (POS) is a widely used outcome measure in palliative care research, and has good psychometric properties. It has been used for clinical or research purposes in specialist cancer centers, nursing homes, day hospice units, and hospice settings in a growing number of countries. However, the POS has not yet been examined using factor analysis. Objective. The aim of the present study was to examine the internal factor structure of the POS. Methods. Confirmatory and exploratory factor analyses were used for secondary analysis of two existing POS data sets of British patients, most of whom were cancer patients. Results. We began with a confirmatory factor analysis (CFA), which indicated that the POS is not a unidimensional scale. This was followed by an exploratory factor analysis that suggested two factorsdone reflecting a psychological wellbeing dimension and the other consisting of three items relating to the standard of professional care. A similar two-factor structure also was identified in the second sample using CFA. Conclusion. The POS appears to capture two factors, psychological status and quality of care, and to have three items that function independently (family anxiety, symptoms, and pain control). Our findings suggest that future evaluations of palliative care services should include assessment not only of symptoms and Financial support for Dr. Siegert in the preparation of this article was provided by the Dunhill Medical Trust and the Luff Foundation. Dr. Gao is supported by the COMPASS Collaborativeda National Cancer Research Institute supportive and palliative care research collaborative. Collection of data in the original studies was supported by the London Region NHS Research and Development program. Dr. Higginson is a National Institute for Health Research Senior Investigator. Ó 2010 U.S. Cancer Pain Relief Committee Published by Elsevier Inc. All rights reserved. Address correspondence to: Richard J. Siegert, PhD, Department of Palliative Care, Policy and Rehabilitation, School of Medicine at Guy s, King s College and St Thomas Hospitals, King s College London, Cicely Saunders Institute, Bessemer Road, London SE5 9PJ, United Kingdom. richard.siegert@ kcl.ac.uk Accepted for publication: November 25, /$esee front matter doi: /j.jpainsymman

2 68 Siegert et al. Vol. 40 No. 1 July 2010 well-being or quality of life, but also of quality of care, and that unidimensional measures will not capture all relevant aspects in palliative care. J Pain Symptom Manage 2010;40:67e74. Ó 2010 U.S. Cancer Pain Relief Committee. Published by Elsevier Inc. All rights reserved. Key Words Palliative Care Outcome Scale, POS, factor analysis, quality of care Introduction The Palliative Care Outcome Scale (POS) is a widely used measure for assessing outcomes in palliative care. 1 It has been used for clinical or research purposes in specialist cancer centers, nursing homes, and day hospice units in addition to hospice settings. 2e4 The POS has been used in many European countries, including the United Kingdom, the Netherlands, Germany, Italy, Portugal, and Spain and is also being used in Argentina and several countries in Africa. 2,3,5e7 The POS also has been successfully translated into several languages, including Spanish and German. 6,8 The POS is a 10-item questionnaire that has two formsdone for patients and one for staff. The patient questionnaire is the focus of the present article. This version of the POS has 10 items that are scored with a Likert scale and cover a range of important content areas, including physical symptoms, psychological symptoms, spiritual considerations, practical and emotional concerns, and psychosocial matters. 9 Each of these 10 items can be scored (0e4) and considered separately and they also can be summed to yield a total score ranging from 0 to 40, where 0 represents the best possible care and 40 the worst. The POS also includes two open questions on patients main concerns. Hearn and Higginson 1 developed the POS after a systematic review of measures in the field and in consultation with a multidisciplinary advisory group and input from patient and caregiver representatives. The 10 items were selected from existing measures of established reliability and validity, and were chosen to cover the physical, psychological, social, and spiritual domains of palliative care. They piloted the POS with 25 patients and, after some modifications, administered it to 168 patients for validation. The POS was validated against other relevant measures and found to have good concurrent validity, as well as being acceptable to both patients and staff. The internal consistency for patient data was 0.65, which is acceptable, although not especially high. This internal consistency probably reflects the diversity of topics covered in the POS, which was not originally intended to be a unidimensional scale. The authors of recent quality criteria for health status measures have noted that coefficient alpha is mostly relevant for questionnaires that concern a single underlying dimension, and is less relevant for those questionnaires in which the items are merely different aspects of a complex clinical phenomenon that do not have to be correlated. 10 (p. 36) Test-retest reliability was also good for 7 of 10 items but low for 3 (namely, pain, other symptoms, personal affairs). Factor analysis is a statistical approach that has traditionally played an important role in the psychometric development of multi-item questionnaires and rating scales. 11 The primary purpose of factor analysis in questionnaire development is to identify how specific items within a scale form subgroups or clusters of correlated itemsdcalled factors. The content or verbal meaning of the items comprising these factors can then be interpreted to give an indication of the constructs underpinning the full measure. For example, a 30-item depression questionnaire might have four separate factors concerned with cognitive, somatic, behavioral, and emotional symptoms of depression. Understanding these factors can help to clarify the different dimensions that a questionnaire seeks to assess, as well as indicating if there are ways to shorten the questionnaire, for example, if a lot of questions appear to tap into one factor. There are two different approaches to factor analysisd exploratory and confirmatory. 12 Exploratory factor analysis (EFA) is typically used for identifying the factor structure of

3 Vol. 40 No. 1 July 2010 POS Factor Structure 69 a scale when there is no obvious theoretical or empirical reason for expecting a specific structure. This might occur with a new scale. EFA typically involves two stages. The first stage identifies the factors accounting for the largest amounts of variance in the questionnaire data. However, the nature or meaning of these factors is often not very clear beyond the first factor extracted. A second stage, called factor rotation, is usually necessary to clarify the nature of these factors. Briefly, an item s correlation with a factor is called a loading and the items are plotted graphically using their factor loadings as their coordinates. Rotation means that the axes (i.e., factors) are then rotated in space to give a clearer picture of their relationship to the items. Part of the confusion surrounding EFA stems from the fact that numerous different methods have been developed for both the initial factor analysis (e.g., principal component analysis, principal factor analysis) and for the rotation stage (e.g., varimax rotation, oblique rotation), and these different methods sometimes give different solutions. Confirmatory factor analysis (CFA) is a more recent development that uses structural equation modeling. CFA involves specifying a hypothesized factor structure and then statistically testing how well that model actually fits the data. The chief advantage of CFA over EFA is that it allows for a statistical test of the goodness of fit of competing models. Higginson and Donaldson 13 included the POS with a generic quality-of-life measure (EuroQoL) and a measure of hope (Herth Hope Index) in a factor analysis to determine the relationships among the three scales. 13e15 They reported a separate principal component analysis for each of the three scales and then included all three scales in both exploratory and confirmatory factor analyses. They used a sample of 66 patients for the EFA (the historical group ). The principal component analysis of the POS showed all but three items (Given Information, Time Wasted, and Practical Matters Addressed) loaded above 0.30 on the first unrotated factor. Those three items showed a high loading on the second unrotated factor. The authors then included items from all the three questionnaires in a principal axis factor analysis with varimax rotation. They reported that the best solution in the exploratory analysis comprised four factors, then tested this with a CFA on a data set (the historical and concurrent groups, n ¼ 137). They again reported that the best solution comprised four factors, which they labeled as Symptoms, Self-Sufficiency, Positivity, and Spiritual. The present study aims to extend the previous work by focusing solely on the POS. Although informative about what factors or constructs the three measures have in common, the Higginson and Donaldson study 13 still does not fully clarify the internal structure of the POS itself, because of the influence of the items from the two other measures in the analysis. For example, in that analysis, only five of the POS items loaded high on any of the four factors (and four of these items loaded on just one factor). This is informative in telling us what these measures have in common and how they differdbut it does not show us the specific grouping of items within the factors (if any) that underpin the POS. What is needed is to examine the POS items alone and to follow the principal components analysis with a rotated factor solution. As the use of the POS is expanding in many countries in clinical care and in research, it is important to better understand the internal structure of the measure, so that it can be further developed and improved. The aim of the present study was to examine the internal structure of the POS using both EFA and CFA. 12 Methods Secondary analysis of two existing POS data sets was used to investigate the factorial structure of the POS. The first analysis was a CFA that tested the simplest model, one in which all 10 POS items load high on a single general factor. The second analysis was an EFA conducted in light of the results of the initial CFA. The results of the EFA were then tested using CFA on the same sample, to obtain an estimate of goodness of fit, and also tested on a different sample (Sample 2 below). There were two samples recruited for this study. In Sample 1, participants were 132 palliative care patients receiving community palliative care, home or hospice care, or day-care services, recruited as part of an evaluation of services in the south of England. They had a mean age of 70.5 years (standard deviation

4 70 Siegert et al. Vol. 40 No. 1 July 2010 [SD] ¼ 12.1), and 48% were male and 52% were female. All the participants were White. Their diagnoses included genitourinary/prostate cancer, 17%; gastrointestinal cancer, 15%; gynecological/breast cancer, 20%; lung cancer, 18%; other cancer, 13%; motor neuron disease, 6%; and unknown cancer primary, 11%. In Sample 2, participants were 99 palliative care patients receiving community palliative care, home or hospice care, or day-care services in the greater London region. They had a mean age of 64.8 years (SD ¼ 12.7), and 53% were male and 47% were female. Ethnicity comprised 94% White, 5% Black, and 1% other. Their diagnoses included genitourinary/prostate cancer, 13%; gastrointestinal cancer, 12%; gynecological/breast cancer, 19%; lung cancer, 25%; other cancer, 23%; motor neuron disease, 1%; and unknown cancer primary, 7%. An initial CFA was conducted using the AMOS program for structural equation modeling on the responses of the 132 Sample 1 participants. This tested the simplest model, in which all 10 items are thought to load highly on a single latent variable. The second analysis used principal components analysis with varimax rotation to examine the same data. This method was chosen because it is the most likely to produce a readily interpretable or simple structure, that is, where most of the items load high on only one of the rotated factors. 12 This makes it particularly suitable for EFA. The number of factors to retain was decided using Horn s method 16 of parallel analysis, incorporated into an online facility by Watkins. 17 In parallel analysis, the number of factors to be rotated is determined by the number with eigenvalues greater than the average value derived from random numbers drawn for the number of variables and the number of respondents in a large number of replications (100 replications were specified in the present study). For further details of the method, see Horn 16 and Watkins. 17 Two further CFAs were then completed to examine how well the solution from the EFA fitted the two data sets available (i.e., from Sample 1 and Sample 2). For each of the models tested using CFA, we obtained four indices of goodness of fit. The first was Chi-square. Here, we sought a low, nonsignificant value, which would indicate a close fit between the data and the model. However, as this index can be misleading with large samples, Ullman 18 has suggested as a rule of thumb that a Chi-square to degrees of freedom ratio (Chi-square/df) of less than 2.00 may be deemed to reflect a good fit to the model. This ratio was, therefore, used as our second index. For our third index, we used the goodnessof-fit index (GFI), where we sought a high value, approaching 1.00 and preferably >0.95, to indicate a good fit to the model. Finally, for our fourth index, we used the root mean square of approximation (RMSEA), which may be thought of as a measure of badness of fit, in which, therefore, we sought a very low value, approaching 0.00 and preferably <0.05, to indicate a good fit. For the fourth CFA (see Analysis 4 below), which tested a two-factor model on the responses from the 99 participants from Sample 2, we used item parceling to achieve a good model fit. Parceling involves combining small numbers of related items into a single score. Items were systematically allocated to the parcels to produce approximately equal correlations between each parcel and its associated factor. To make the item selection independent of the data analyzed (i.e., Centers 1e5), the selection of items for the four parcels was based on analyses of the other data set (Center 6). A one-factor, unrotated principal component analysis was undertaken for each of the item groups comprising a factor and the items were allocated to parcels to give the best possible relationship with the first principal component. On the first factor, we combined Items 6 and 7 (the highest and lowest loadings) and Items 3 and 8 (the two intermediate level loadings). On the second factor, we combined Items 9 and 10, and the highest loading item (Item 5) formed the other parcel. The parceling procedure was developed by Cattell 19 for EFA but has since become widely used for CFA to address a number of technical problems, including small sample size or nonnormal data. 19,20 Results Analysis 1: CFA of One-Factor Model (Sample 1) Results of the initial CFA testing a unifactorial model are presented in Table 1. Inspection of column one in Table 1 shows a GFI of 0.90,

5 Vol. 40 No. 1 July 2010 POS Factor Structure 71 Index of Fit to the Model Table 1 Results of Confirmatory Factor Analyses One Factor Two Factor a Two Factor Two Factor Sample 1 Sample 1 Sample 2 Sample 2 (n ¼ 132) (n ¼ 132) (n ¼ 99) (n ¼ 99) No. of items or parcels 10 items 7 items (3, 6, 7, 8/5, 9, 10) 7 items (3, 6, 7, 8/5, 9, 10) 7 items d4 parcels Chi-square df P-value Chi-square/df GFI RMSEA a Note: Factor 1 ¼ Item 3, Anxious/worried; Item 6, Share feelings; Item 7, Life worthwhile; Item 8, Feel good. Factor 2 ¼ Item 5, Given information; Item 9, Time wasted; Item 10, Practical matters. a RMSEA of 0.09, and a significant Chi-square value and a Chi-square/df ratio slightly above the desired value of 2.0. Taken together, the three fit indices suggest only a marginally adequate fit to the single factor model. In light of this result, the second analysis was undertaken using EFA to investigate for multidimensionality. Analysis 2: EFA (Sample 1) Parallel analysis suggested extracting two components for rotation and the results, both the unrotated principal components and the varimax rotation, are presented in Table 2. Table 2 shows two rotated factors, one comprising four items primarily about psychological well-being and the other consisting of three items relating to the standard of professional care received. The remaining three items did not load above 0.40 on either of the two factors. Table 2 Principal Component Analysis and Two-Factor Varimax Rotation of POS Items (Sample 1, n ¼ 132) a Item Principal Component Varimax Rotation C1 C2 F1 F2 POS7 Life worthwhile POS8 Feel good POS3 Anxious/worried POS6 Share feelings POS4 Family anxious 0.43 POS1 Pain control POS2 Symptom control POS5 Given information POS10 Practical matters POS9 Time wasted a Note: all factor loadings below 0.40 have been removed for clarity. Analysis 3: CFA of Two-Factor Model (Sample 1) A CFA of the same data from Sample 1 was then completed to obtain a measure of the goodness of fit of the EFA solution. In this case, we tested a two-factor model in which four items (3, 6, 7, and 8) load on one factor and three items (5, 9, and 10) load on the other. Results are presented in Table 1 and provide support for this model, with all three fit indices reflecting a very good fit. Analysis 4: CFA of Two-Factor Model (Sample 2) A CFA of the data for the 99 participants from Sample 2 was then completed. Initially, we tested the same two-factor model as for Sample 1 participants in Analysis 3, and the results are displayed in Table 1. The results suggested a promising but less than perfect fit as reflected in a GFI ¼ 0.92, RMSEA ¼ 0.11, and a significant Chi-square (P ¼ 0.005) and a Chi-square/df ratio above 2.0 (i.e., 2.17). Given the small sample size, we then repeated this CFA using item parceling. 19,20 This resulted in a very good fit, with a GFI approaching 1 (GFI ¼ 0.99), RMSEA ¼ 0.13, Chi-square/df close to 2.0, and a nonsignificant Chi-square value (P ¼ 0.11). Discussion In the present article, we used EFA and CFA to examine the structure of the POS using data from two separate palliative care samples. The results showed two orthogonal factorsdone representing a psychological well-being dimension and the other representing the perceived

6 72 Siegert et al. Vol. 40 No. 1 July 2010 quality of palliative care received. The psychological well-being factor also tapped some elements of spirituality with one item (Have you felt that life was worthwhile?). However, three POS items, namely, symptom control, pain, and family anxiety, did not load on either of these two factors. There was little evidence from either the principal component analysis or the CFA of a single general factor underpinning the POS. This is perhaps not surprising given that the POS was originally developed as 10 single items, each one reflecting an important aspect of palliative care, rather than as a unidimensional measure. These two dimensions emerged in the analyses from both samples, and this suggests that these two dimensions, psychological well-being and perceived quality of palliative care, are fairly robust features of the POS. Interestingly, the dimension of perceived quality of palliative care was not evident in any of the four rotated factors identified by Higginson and Donaldson, when the 10 POS items were factor analyzed together with items from other questionnaires. 13 However, it was present when those authors performed a principal components analysis on the 10 POS items alone and was the second largest component (in terms of percentage of variance accounted for). In that article, the authors examined three QoL measures used in palliative care, and the best rotated factor solution composed of four factors, which they labeled as Symptoms, Self-Sufficiency, Positivity, and Spiritual. However, only five POS items had high loadings on any of these four factors, with Pain Control loading on the Symptoms factor and four POS items loading on the Positivity factor (Anxious/Worried, Share Feelings, Life Worthwhile, and Feel Good). Thus, the factor we have called Well-Being seems identical to Higginson and Donaldson s Positivity factor. Thus, the POS seems to capture an important outcome variable, namely Quality of Care, which is missing from some quality-of-life measures commonly used to measure outcome in palliative care research and measuring psychological well-being in the dying person. In our analysis, we found three POS items that did not correlate closely with any factor (Pain Control, Symptom Control, and Family Anxious). However, the item Family Anxious did load moderately highly on the first principal component (i.e., the psychological wellbeing factor), although less so after rotation. Thus, it appears that the Pain and Symptom items are functioning as individual items and perhaps best considered on their own rather than in conjunction with other items. One possible explanation for this is that these two items, while they reflect information that is essential for the clinician involved, are best regarded as causal indicators not indicator variables, and hence are not fully suitable for the traditional factor-analytic paradigm. 21,22 Causal indicators are variables external to the person that might be expected to directly affect their quality of life, and indicator variables are those unobservable variables (e.g., anxiety, depression, QoL) that are inferred from scores on multiple self-report items. It has been argued that including indicator variables such as physical symptoms and side effects in factor analysis will likely result in misleading or uninterpretable results. This question of whether or not these two items are better conceptualized as causal indicators rather than indicator variables is beyond the scope of the present study. However, it could readily be tested with a different and much larger POS sample using structural equation modeling. The identification of two virtually orthogonal factors and a small group of unrelated items accounts well for the relatively low reliability estimates found by Hearn and Higginson 1 for the POS as a whole. Put simply, the measure is not in any sense unidimensional, and any evaluation using the reliability estimate as a criterion is almost certainly inappropriate. This would accord with the development of the POS, which deliberately aimed to be brief and yet to capture the broad elements of palliative care, so that it could be used in a wide range of clinical and research settings. Limitations The present study has a number of limitations that should be considered. Both of the samples used were relatively small, with ns of 99 and 132, respectively, and both comprised palliative care patients from British centers. Moreover, both samples were largely composed of cancer patients, with a very small

7 Vol. 40 No. 1 July 2010 POS Factor Structure 73 proportion of motor neuron disease patients. Consequently, it would be desirable to replicate the present findings with a larger sample, with samples from other countries, and with noncancer samples. Another limitation of the study was that in Analysis 3, we completed a CFA on the data from Sample 1 and these were the same data used in the EFA in Analysis 2. The advantage of doing this is that it gives a quantitative fit index for that solution rather than just relying upon a subjective judgment based on a visual inspection of the rotated factor loadings. However, CFA should be conducted on a sample that is quite separate from the sample used in the EFA. This was indeed the next step in the analyses, although to achieve a very good fit with the Sample 2 data, it was necessary to use item parceling. Moreover, the use of only two indicators (i.e., parcels) per factor is considered less than ideal for CFA. 20 At the same time, parceling is a well-accepted technique in CFA for situations where the sample size is small or item responses are not normally distributed. 19,20,23 Clinical Implications The POS was originally developed using clinical and patient/family involvement to capture the core important outcomes that palliative care seeks to address. The independent items and two factors found in the present study fit neatly with Cicely Saunders s model of total pain having physical, psychological, social, and spiritual/existential components. However, psychological (e.g., anxious/worried) and spiritual items (i.e., felt that life was worthwhile) within the POS appear to be loading together, suggesting that, for most of the patients at least, these aspects are closely related. This close relationship may mean that assessment of psychological concerns is at least a potential proxy screening for spiritual concerns as well, which may be valuable for staff who often find it difficult to assess spiritual concerns especially when they do not know the patient well. Spirituality is a newer concept than religion and is typically thought to represent a search for existential meaning, 24 which when low, is a risk factor for hopelessness and a desire for hastened death and is correlated with depression. 25 Our findings also suggest that when a patient appears to have psychological distress, it is important to also assess spiritual concerns, as it may be these concerns that underpin the psychological distress. Our findings also identified a factor representing the patient s perception of the quality of palliative care received. This is in accord with the trend toward the measurement of patient-oriented outcomes in palliative care beyond simply the measurement of quality of life and including measurement of the way in which services are provided. When patients have advanced or progressive illness and their function declines, the way that services are provided may be important in determining their quality of life. Our findings suggest that future evaluations of palliative care and similar holistic services for people with advanced illness and/or high levels of physical impairment should include assessment of symptoms, wellbeing, and also quality of care. The most important clinical message from the present study is that clinicians must look at the patient s score on each of the 10 POS items and also consider clusters of related items, as well as considering the total score. More importantly, a high score on even a single item could reflect a high level of patient distress or discomfort for one factor even if the person s total score is low. Items 3, 6, 7, and 8 reflect aspects of a patient s psychological well-being or quality of life and should be considered together. Certainly, where a patient has high scores on several of these items, it is important to explore for possible depression, anxiety, or existential/spiritual distress. Similarly, Items 5, 9, and 10 each represent an aspect of the quality of palliative care received and these items should be considered together. High scores on these three items suggest dissatisfaction with the quality of care. Items 1 and 2 (pain, symptoms) should always be considered as individual items regardless of scores on the other items, and a high score on either of these two items suggests that a more in-depth assessment is required promptly. Item 3 represents the family anxiety and equally needs to be considered independently, although this item is closely related to psychological distress. Taken together, our results show that the 10 items in the POS cover five important domains: pain (single item), symptoms (single item), well-being (four items), family anxiety (single item), and quality of

8 74 Siegert et al. Vol. 40 No. 1 July 2010 care (three items), which support the holistic and multiprofessional approach of palliative care and appear to function independently. Acknowledgments The authors are very grateful to all the participants who completed the POS in the present study and also the clinical staff who supported this research and assisted with participant recruitment. The authors are grateful to two anonymous reviewers whose comments and suggestions helped to substantially improve this paper. References 1. Hearn J, Higginson IJ. Development and validation of a core outcome measure for palliative care: the palliative care outcome scale. Qual Health Care 1999;8:219e Stevens AM, Gwilliam B, A Hern R, Broadley K, Hardy J. Experience in the use of the palliative care outcome scale. Support Care Cancer 2005;13(12): 1027e Brandt HE, Deliens L, van der Steen JT, et al. The last days of life of nursing home patients with and without dementia assessed with the Palliative care Outcome Scale. Palliat Med 2005;19(4): 334e Slater A, Freeman E. Is the Palliative Care Outcome Scale useful to staff in a day hospice unit? Int J Palliat Nurs 2005;11(7):346e Eisenchlas JH, Harding R, Daud ML, et al. Use of the Palliative Outcome Scale in Argentina: a cross-cultural adaptation and validation study. J Pain Symptom Manage 2008;35(2):188e Serra-Prat M, Nabal M, Santacruz V, Picaza JM, Trellis J. Validation of the Spanish version of the Palliative Care Outcome Scale. Med Clin (Barc) 2004; 123(11):406e Powell RA, Downing J, Harding R, Mwangi-Powell F, Connor S. Development of the APCA African Palliative Outcome Scale. J Pain Symptom Manage 2007;33(2):229e Bausewein C, Fegg M, Radbruch L, et al. Validation and clinical application of the German version of the palliative care outcome scale. J Pain Symptom Manage 2005;30(1):51e Aspinall F, Hughes R, Higginson I, et al. A user s guide to the Palliative Care Outcome Scale. London: Department of Palliative Care, Policy and Rehabilitation, King s College London, Terwee CB, Bot SDM, de Boer MR, et al. Quality criteria were proposed for measurement properties of health status questionnaires. J Clin Epidemiol 2007;60:34e Nunnally JC, Bernstein IH. Psychometric theory. New York: McGraw Hill Inc., Thompson B. Exploratory and confirmatory factor analysis: Understanding concepts and applications. Washington, DC: American Psychological Association, Higginson IJ, Donaldson N. Relationship between three palliative care outcome scales. Health Qual Life Outcomes 2004;2(68). 14. Herth K. Hope in the family caregiver of terminally ill people. J Adv Nurs 1993;18:538e Brazier J, Jones N, Kind P. Testing the validity of the EuroQoL and comparing it with the SF-36 health survey questionnaire. Qual Life Res 1993;2(3):169e Horn JL. A rationale and test for the number of factors in factor analysis. Psychometrika 1965;32: 179e Watkins MW. Monte Carlo PCA for parallel analysis [Computer freeware]. 18. Ullman JB. Structural equation modeling. In: Tabachnick BG, Fidell LS, eds. Using multivariate statistics. Boston: Allyn & Bacon, 2001: 653e Cattell RB. Radial parcel factoring vs. item factoring in defining personality structure in questionnaires: theory and experimental checks. Aust J Psychol 1974;2:103e Hau KT, Marsh HW. The use of item parcels in structural equation modeling: non-normal data and small sample sizes. J R Stat Soc Series A 2004;57: 327e Fayers PM, Hand DJ. Factor analysis, causal indicators and quality of life. Qual Life Res 1997;6:139e Fayers PM, Hand DJ. Causal variables, indicator variables and measurement scales: an example from quality of life. J R Stat Soc 2002;165:233e Bandalos DL, Finney SJ. Item parcelling issues in structural equation modelling. In: Marcoulides GA, Schumacker RE, eds. New developments and techniques in structural equation modelling. Mahwah, NJ: Lawrence Erlbaum Associates, 2001: 269e Koffman J, Morgan M, Edmonds P, Speck P, Higginson IJ. I know he controls cancer : the meanings of religion among Black Caribbean and White British patients with advanced cancer. Soc Sci Med 2008;67:780e Rodin G, Lo C, Mikulincer M, et al. Pathways to distress: the multiple determinants of depression, hopelessness, and the desire for hastened death in metastatic cancer patients. Soc Sci Med 2009;68: 562e569.