Family Practice Vol. 19, No. 5, 489-495
© Oxford University Press 2002
Health Services Research |
What dimensions underlie patient responses to the General Practice Assessment Survey? A factor analytic study
National Primary Care Research and Development Centre, University of Manchester, Manchester, UK.
Dr Peter Bower, NPCRDC, 5th floor, Williamson Building, University of Manchester, Manchester M13 9PL, UK; E-mail: peter.bower{at}man.ac.uk
Bower P, Mead N and Roland M. What dimensions underlie patient responses to the General Practice Assessment Survey? A factor analytic study. Family Practice 2002; 19: 489–495.
Received 2 October 2001; Revised 13 February 2002; Accepted 13 May 2002.
 |
Abstract |
|---|
 Top Abstract IntroductionMethodsResultsDiscussionReferences |
|---|
Background. Patient self-report measures of primary care are being used increasingly for quality assessment and improvement. The General Practice Assessment Survey (GPAS) is a widely used measure. However, it is important that the measures used are valid and interpretable. Factor analysis is a useful method to assist in validation.
Objective. The aim of this study was to determine the underlying structure of responses to the GPAS.
Methods. Factor analysis of data from a number of patient surveys was carried out using the GPAS in primary care.
Results. Analysis indicated that three factors underlie responses to the GPAS. These were named ‘access’, ‘patient-centredness’ and ‘nursing’. These factors were replicated in a second sample of GPAS survey data.
Conclusion. Responses to the GPAS can best be summarized in terms of three underlying factors, which supports previous conceptual work. These factors may also have utility for reducing the overall length of the GPAS, and in reducing the need for multiple hypothesis testing associated with the use of the original scales.
Keywords. General practice, General Practice Assessment Survey, outcome measure, patients.
|
Introduction |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionReferences |
|---|
Patients’ views on the quality of primary care services are seen increasingly as crucial for the effective evaluation of services and their ongoing improvement.1 Although a number of methods have been used to ascertain such views, self-report measures remain popular because of their relative simplicity and efficiency. One such measure that is used increasingly by Primary Care Groups and Trusts is the General Practice Assessment Survey (GPAS).2
GPAS is a multi-item self-report questionnaire, and measures a number of separate dimensions relating to patients’ views of general practice, namely access, receptionists, continuity of care, communication, interpersonal care, GPs’ knowledge of the patient, specialist referral, nursing, general satisfaction and enablement.3 These constructs are hypothesized to be distinct facets of patients’ perceptions of the care they receive from their general practice.
Although preliminary data on the psychometric properties of the GPAS have been published (i.e. internal reliability, scale intercorrelations, associations with external variables),2,4 there are few validity data available. Correlating scale scores with external variables is considered the gold standard method for demonstrating validity, but factor analysis is also used in this context.5,6 Factor analysis of responses can indicate what dimensions (or ‘factors’) are actually represented in the pattern of correlations amongst responses. For example, factor analysis has indicated that responses to personality questionnaires can best be understood in terms of five broad factors.7 Factor analysis may have more pragmatic benefits as well, such as reducing redundancy in scales and assisting in the development of shorter measures.5
|
Methods |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionReferences |
|---|
Data collection
GPAS data were derived from a number of sources, including:
- a large-scale study of quality of care in general practice8
- a validation study of the GPAS2
- an ongoing evaluation of Personal Medical Services pilots9
- data gathered through the National Primary Care Research and Development Centre’s GPAS service (http://www.gpas.co.uk/)
Some of these data sets used version 1 of the GPAS, while others used version 2. The main difference between the two versions was the removal of scales of low validity (i.e. ‘technical competence’) or utility (‘trust’ and ‘co-ordination’) from version 2. The present analysis was restricted to those items common to both versions of the GPAS. Items measuring the concept of ‘enablement’3 were also added to version 2. Because these items were only available for a small proportion of the sample, they were excluded from the main analysis and used in the validation of the factors.
Overall, 21 905 responses were available. Factor analysis requires complete data on all items, but there were missing data and ‘not applicable’ responses among a number of scales (e.g. ‘phoning the practice for advice’), and the total sample size available for analyses was considerably lower (n = 8025). Demographic details of the patients are shown in Table 1
.
|
Analysis
The GPAS measures some constructs with report– assessment pairs (e.g. report, "in general, how often do you see your usual doctor?"; assessment, "how do you rate this?"). Only assessment items are used in the calculation of scale scores, and thus only these items were included in the factor analysis. Some report items without assessment equivalents (e.g. "how long have you been a patient with the practice?") were also excluded, as were the items on referral to specialist care, which were in a report format. The item ‘general satisfaction’ and the enablement items were used to validate the factors and were also excluded (see below). The items included in the analysis are shown in Table 2
.
|
All assessment items use a six-point response format, ranging from ‘very poor’ to ‘excellent’.
Factor analysis is a statistical technique designed to simplify data sets (specifically correlation matrices). The observed correlations among multiple variables are described in terms of a smaller number of hypothetical ‘factors’. Examining the correlations (so-called ‘loadings’) between variables and factors indicates what dimension a factor may represent.
Factor analysis generates multiple factors, and a key decision concerns the criteria used to identify which factors are likely to be of substantive importance. Although maximum likelihood factor analysis has the advantage of providing statistical tests for the number of factors to extract, large sample sizes may mean that excessive numbers of factors are extracted.10,11 Therefore, a principal axis factor analysis was conducted in SPSS for Windows (release 9.0.0), and the scree test used to determine the number of factors. This test is a visual measure of the number of factors, independent of sample size, whereby multiple factors are plotted against a meas-ure of the proportion of variance explained by each factor. Any obvious discontinuity in the plot is used to distinguish substantive factors and those likely to be of less importance.
The scree test was performed on a principal components analysis, the number of factors determined, and the full principal axis factor analysis conducted with this number of factors.6 The solution was rotated using the varimax rotation procedure. Rotation is designed to improve the interpretability of factors, by finding the solution which maximizes the number of factors with a few high loadings, and the rest close to zero (so-called ‘simple structure’). Varimax rotation produces ‘orthogonal’ (i.e. uncorrelated) factors.
Because of the large sample available, the data were split randomly into two similarly sized groups, so that the factor solution generated in the first sample could be replicated in the second. No statistical comparisons were made between the solutions, but the number of factors extracted and the variables loading on each factor were compared to determine the stability of the solutions generated.6
|
Results |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionReferences |
|---|
Initial analysis
The initial data set comprised 3982 cases. Examination of the scree plot (Fig. 1
) suggested a three-factor solution, which also agreed with another conventional criterion for judging the number of factors (i.e. number of eigenvalues >1.0). Variables were judged to load significantly on a factor if the loading was >0.40. The principal axis factor analysis indicated that these three factors accounted for 64.1% of the variance. The rotated factor matrix is shown in Table 2
, showing the items and their loadings on each factor.
|
Based on the rotated factor matrix, the three factors were labelled as follows:
- Access—this factor included all the GPAS access items, as well as assessments of ‘receptionists’ and ‘continuity of care’.
- Patient-centredness—this factor included items from the ‘communication’, ‘interpersonal’ and ‘knowledge of the patient’ scales of the GPAS.
- Nursing—this factor included only the three GPAS items that relate specifically to nursing care.
Replication
The replication data set comprised 4043 cases. The second scree plot also identified three factors. This three-factor solution accounted for 65.3% of the variance, and the pattern of loadings of variables on factors was almost identical, although the item on ‘convenience of practice location’, which had loaded on the ‘access’ factor in the first analysis, did not meet the >0.4 criteria in the second analysis (see Table 2).
Validation
Items loading on a factor provide evidence only for face validity, and demonstrations of construct validity require examining associations with other variables.6 The scope for such validation in the present data set is limited because of the lack of relevant variables external to the GPAS itself. However, for a preliminary analysis, it was hypothesized that scores on the ‘access’, ‘patient-centredness’ and ‘nursing’ factors would all be related to ‘general satisfaction’, as this item attempts to capture a global perception of the quality of general practice services and will thus be influenced by all aspects of that service. In contrast, it was hypothesized that the correlation between ‘enablement’ and ‘patient-centredness’ would be of greater magnitude than the correlations with ‘general satisfaction’. Enablement relates to the degree to which consultation processes provide patients with the resources they need to manage their illness.3 Enablement is thus most likely to be predicted by items that relate specifically to the doctor–patient relationship and the effectiveness of care delivery, such as those in the ‘patient-centredness’ factor, rather than items relating to access, which deal with the issue of whether patients find it easy to get the care they need in the first place.12 A recent observational study in general practice found that aspects of patient-centredness (i.e. patients’ perceptions of the doctor’s interest in the effect of the problem on patients’ lives, health promotion and a positive approach) all predicted enablement scores.13 Since the ‘enablement’ items in the GPAS refer specifically to the doctor, examining associations with the nursing factor was considered inappropriate.
Factor scores were computed based on the factor analysis of all cases (n = 8025), using the regression method, although the analysis was based on 1035 cases where both factor scores and enablement scores were available. The Pearson correlations can be found in Table 3. The analysis indicated that the relationship between ‘enablement’ and ‘patient-centredness’ was higher than that between the other variables.
|
Multiple regression was used to examine the degree to which the factor scores predicted variance in ‘global satisfaction’ and ‘enablement’. All three factors were entered simultaneously into the equation. The regression results are shown in Table 4
. Overall, the factors predicted a reasonable amount of variance in ‘enablement scores’, but only a small proportion of the variance in the measure of ‘global satisfaction’.
|
|
Discussion |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionReferences |
|---|
Overall, the analyses indicated that responses to the GPAS can be summarized most efficiently in terms of three orthogonal factors, measuring dimensions of ‘access’, ‘patient-centredness’ and ‘nursing’. These factors accounted for nearly two-thirds of the variance, were replicable across two different respondent samples and were based on large samples of data.
The identified factors fit broadly with a recent theoretical model concerning quality in primary care,12 which hypothesizes two dimensions, ‘access’ and ‘effectiveness’. Although ‘effectiveness’ is conceptualized as concerning both clinical and interpersonal care, items relating to patient perceptions of the quality of clinical care were removed from version 2 of the GPAS as qualitative interviews with patients indicated that they do not readily distinguish good technical care (e.g. GPs’ diagnostic or management skills) from good interpersonal care.14 Although two items on the ‘nursing’ factor have equivalents in the ‘patient-centredness’ factor (i.e. "how well they listen to what you say" and "how well they explain your health problems or any treatment that you need"), the fact that these items load on another factor suggests that patients do make distinct judgements as to the interpersonal aspects of their consultations with different professional groups. This may also reflect the relative lack of contact that some patients may have with nurses in their practice.
The factor analysis of scales such as the GPAS differs from the methods used in the development of many psychological scales, where a wide selection of items are generated, rated and then factor analysed to derive meaningful dimensions from a diffuse selection of variables. Rather, items used in the GPAS were developed with a specific structure in mind, and are presented in a structured fashion within the questionnaire itself, with the items grouped in related sections. The identified factor structure may relate as much to the design of the scale i.e. the common complaint about factor analysis that "you only get out what you put in".6 Therefore, the solution might be viewed as relatively pre-determined and uninformative, in that visual inspection of the item content may lead one to the same conclusions as formal statistical analysis. Nevertheless, the analysis could have provided alternative structures (e.g. a separate continuity factor, conflation of the ‘patient-centredness’ and nursing factors), and the analysis does provide some evidence that patient ratings agree with the suggested model.
In addition, there are a number of other issues that are suggested by the factor analysis that may be of relevance to further development of the GPAS. The item for practice location did not load on the ‘access’ factor in the replication analysis, although the loading was relatively high and its failure to load may reflect the arbitrary nature of the criterion used to identify variables with factors (>0.40). The receptionist scale is generally scored separately in GPAS from the other ‘access’ items, although the present analysis would suggest that it is viewed as an aspect of access by patients. Finally, the analysis indicated that continuity of care seems related to ‘access’ rather than ‘patient-centredness’, although it had a relatively high loading on the latter factor as well (0.36 in both analyses).
The ‘patient-centredness’ factor is made up from three different scales within the GPAS: ‘communication’, ‘interpersonal skills’ and ‘knowledge of the patient’. The present analysis would suggest that these scales are not highly distinct, and it would seem reasonable to sug-gest that patients are making a relatively global judgement about the ability of their GP to relate to them in an effective manner, rather than highly specific judgements about each facet. However, patient ratings of ‘patient-centredness’ may be better predictors of patient outcomes than external ratings made by experts,15,16 so even this global judgement may be of utility.
The associations between the factor scores and measures of ‘general satisfaction’ and ‘enablement’ provide some evidence of construct validity, although the fact that all measures are derived from the same scale means that this demonstration of validity is relatively weak, and would be much improved by the use of external measures. The fact that the factors only predict a small amount of variance in ‘global satisfaction’ suggests that the latter cannot be considered an effective proxy for scores on the specific scales.
A major problem in the administration of any questionnaire (especially in the busy context of primary care) is length. The present analysis might suggest ways of reducing the questionnaire, by eliminating redundancy. For example, the use of ten items (four in the ‘communication’ and three each in the ‘interpersonal’ and ‘knowledge of the patient’ scales of the GPAS) to measure the ‘patient-centredness’ factor may be excessive. Although scale scores are more reliable when they involve more items (assuming similar correlations between items), it may be possible to reduce the length of a scale by selecting those items loading most highly on that factor. For example, the items "how often do you leave your doctor’s surgery with unanswered questions?" and "your doctor’s knowledge of your responsibilities at home, work or school?" have lower loadings than the other ‘patient-centredness’ items and might be candidates for exclusion.
An additional issue relates to the scoring of the GPAS. The conventional scoring mechanism involves summing items into scales and computing scale scores. This provides multiple, correlated dimensions (e.g. those measuring ‘communication’, ‘interpersonal skills’ and ‘knowledge of the patient’). Factor analysis allows the calculation of factor scores that normally are uncorrelated, which may have several advantages. For example, uncorrelated factors may be useful in relation to intervention studies, where two ‘pure’ dimensions of patients’ perceptions may be superior to multiple correlated dimensions in terms of simplicity and avoidance of multiple hypothesis testing.
However, the use of factor scores does require that the results of the factor analysis are stable. The large amount of data available for the present analyses suggests that the estimates may be relatively stable, but there are legitimate concerns about the representativeness of the samples, e.g. one study providing data for the present analysis reported a response rate of only 38%.8 Calculation of factor scores based on biased samples may not generalize to the wider population.
In conclusion, the present analysis does suggest that patient responses to the GPAS are interpretable in terms of a model involving two dimensions of ‘access’ and ‘effectiveness’, together with an additional dimension concerning the quality of nursing care.
|
Acknowledgments |
|---|
The authors would like to thank Stephen Campbell and Sophie Jerrim for assistance with the database of GPAS questionnaires, and all those who provided data, especially John Campbell, Department of General Practice and Primary Care, GKT School of Medicine.
|
References |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionReferences |
|---|
1 Secretary of State for Health. The NHS Plan. London: HMSO, 2000.
2 Ramsay J, Campbell J, Schroter S, Green J, Roland M. The General Practice Assessment Survey (GPAS): tests of data quality and measurement properties. Fam Pract 2001; 17: 372–379.
3 Howie J, Heaney D, Maxwell M, Walker J. A comparison of Patient Enablement Instrument (PEI) against two established satisfaction scales as an outcome measure of primary care consultations. Fam Pract 1998; 15: 165–171.
4 Campbell J, Ramsay J, Green J. Age, gender, socioeconomic, and ethnic differences in patients’ assessments of primary health care. Qual Health Care 2001; 10: 90–95.
5 Fitzpatrick R, Davey C, Buxton M, Jones D. Evaluating patient-based outcome measures for use in clinical trials. Health Technol Assess 1998; 2: 14.
6 Kline P. An Easy Guide to Factor Analysis. London: Routledge, 1994.
7 Costa P, McRae R. Solid ground in the wetlands of personality: a reply to Block. Psychol Bull 1995; 117: 216–220.[ISI][Medline]
8 Campbell S, Hann M, Hacker J et al. Identifying predictors of high quality care in English general practice: observational study. Br Med J 2001; 323: 1–6.
9 The PMS National Evaluation Team. National Evaluation of First Wave NHS Personal Medical Services Pilots. University of Manchester: NPCRDC, 2000.
10 Everitt B. Making Sense of Statistics in Psychology. Oxford: Oxford University Press, 1996.
11 Kim J, Mueller C. Factor Analysis: Statistical Methods and Practical Issues. London: Sage Publications, 1978.
12 Campbell S, Roland M, Buetow S. Defining quality of care. Soc Sci Med 2000; 51: 1611–1625.[ISI][Medline]
13 Little P, Everitt H, Williamson I et al. Observational study of effect of patient centredness and positive approach on outcomes of general practice consultations. Br Med J 2001; 323: 908–911.
14 Chapple A, Rogers A, Roland M. Users’ understanding of medical knowledge in general practice. Soc Sci Med 2002; 54: 1215–1224.[ISI][Medline]
15 Stewart M, Brown J, Weston W, McWhinney I, McWilliam C, Freeman T. Patient-centred Medicine: Transforming the Clinical Method. London: Sage, 1995.
16 Stewart M. Towards a global definition of patient centred care. Br Med J 2001; 322: 444–445.
GPAS questionnaires, manuals and analysis software can be found on the GPAS website (www.gpas.co.uk)
CiteULike 
Connotea 
Del.icio.us What's this?
This article has been cited by other articles:
|
|
 |
|
  U. W. Jayasinghe, J. Proudfoot, C. Holton, G. P. Davies, C. Amoroso, T. Bubner, J. Beilby, and M. F. Harris Chronically ill Australians' satisfaction with accessibility and patient-centredness Int. J. Qual. Health Care, April 1, 2008; 20(2): 105 - 114. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 M. Hann, P. Bower, S. Campbell, M. Marshall, and D. Reeves The association between culture, climate and quality of care in primary health care teams Fam. Pract., August 1, 2007; 24(4): 323 - 329. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 J. Haggerty, F. Burge, J.-F. Levesque, D. Gass, R. Pineault, M.-D. Beaulieu, and D. Santor Operational Definitions of Attributes of Primary Health Care: Consensus Among Canadian Experts Ann. Fam. Med, July 1, 2007; 5(4): 336 - 344. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
R. G Evans, A. Edwards, S. Evans, B. Elwyn, and G. Elwyn Assessing the practising physician using patient surveys: a systematic review of instruments and feedback methods Fam. Pract., April 1, 2007; 24(2): 117 - 127. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
J. Howie, D Heaney, and M Maxwell Quality, core values and the general practice consultation: issues of definition, measurement and delivery Fam. Pract., August 1, 2004; 21(4): 458 - 468. [Full Text] [PDF]
|
|
|
|
 |
|
 P Bower, S Campbell, C Bojke, and B Sibbald Team structure, team climate and the quality of care in primary care: an observational study Qual. Saf. Health Care, August 1, 2003; 12(4): 273 - 279. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 P. Bower, M. Roland, J. Campbell, and N. Mead Setting standards based on patients' views on access and continuity: secondary analysis of data from the general practice assessment survey BMJ, February 1, 2003; 326(7383): 258 - 258. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||