Technical information on the development of the PAS
Technical information on the development of the PAS Users of the PAS do not need to know the details of the extensive psychometric work and validity studies which underpin the scales. However, some users have expressed interest in knowing more about this work. Below is a brief summary of the work which is now included as a Technical Appendix in the materials downloadable by users. Further psychometric and validity studies of the PAS have been completed and support the original findings. These have been published in the scientific literature.
JORM, A.F., MACKINNON, A.J., HENDERSON, A.S., SCOTT, R., CHRISTENSEN, H., KORTEN, A.E., CULLEN, J.S. & MULLIGAN, R. (1995) The Psychogeriatric Assessment Scales: A multi-dimensional alternative to categorical diagnoses of dementia and depression in the elderly. Psychological Medicine, 25, 447-460.
The PAS items were taken from the Canberra Interview for the Elderly (CIE). This is a standardised interview for the diagnosis for dementia, depression and related disorders by ICD-10 and DSM-III-R criteria. The CIE involves an interview with an informant as well as one with the subject. Responses are scored by a computer program to produce the diagnoses. The aim in developing the PAS was to produce a set of short scales made up of the best of the CIE items.
CIE data were available from an epidemiological study of 1045 persons aged 70+ living in Canberra or the adjacent town of Queanbeyan. This sample involved people living in the community as well as in institutions, with approximately equal numbers of males and females. There were 683 subjects who had relatively complete data on both the subject and informant sections of the CIE. Data from these subjects were analysed by principal components analysis followed by varimax rotation. A scree plot was used to determine that 5 factors should be rotated. These were labelled: Cognitive Impairment, Depression, Cognitive Decline, Behaviour Change and Stroke. The items loading highest on the Cognitive Impairment and Depression factors were all from the subject interview of the CIE, while the highest loading items for the Cognitive Decline and Behaviour Change factors were from the informant interview. The items loading on the Stroke factor were a mixture of both subject and informant items.
Items with loadings of 0.3 or greater on a factor were regarded as candidates for inclusion in a scale. Final selection of items was based on a two-parameter latent trait analysis. Items from each factor were analysed separately, confirming the unidimensional nature of the factor and giving separate slope and threshold parameters for each item. Items were selected to have steep slopes (i.e. to be highly discriminating items) and to have a range of thresholds (i.e. to cover a range of severity). The Stroke items were split into separate subject and informant scales, each having parallel items.
Reliability was assessed in the Canberra general population sample as well as with two clinic samples. The first clinic sample consisted of 76 geriatric and psychogeriatric patients from Sydney and the second consisted of 60 patients from Geneva (who had the questions administered in French). The patients in the two clinic samples had the items administered twice a few days apart, allowing the assessment of test-retest reliability. Internal consistency reliability was assessed in all three samples using Cronbach's alpha. Test-retest reliability was found to be high for all the scales, with alpha generally lower, reflecting the fact that alpha is an estimate of the lower bound of reliability. Reliability was generally higher for the informant scales than for the subject scales.
Validity was assessed against clinical diagnoses of dementia and depression using receiver operating characteristic (ROC) analysis. Diagnoses were available from the CIE computer program as well as from independent clinicians using the ICD-10 and DSM-III-R criteria. The Cognitive Impairment and Cognitive Decline scales were found to perform well as screening tests for dementia, while the Depression scale performed well as a screening test for depression. The Behaviour Change scale was non-specific, being affected by both dementia and depression. The Stroke scales performed well at discriminating vascular from non-vascular (mainly Alzheimer) types of dementia.
Further evidence of validity came from correlations with other commonly used scales. In the Canberra sample, the Cognitive Impairment scale correlated 0.80 with the MMSE and 0.45 with the IQCODE. The Cognitive Decline scale correlated 0.48 with the MMSE and 0.78 with the IQCODE. The Depression scale correlated 0.67 and 0.60 respectively with the Goldberg depression and anxiety scales. The Stroke scales correlated 0.71 and 0.65 with the Hachinski Ischemic Score.
Percentile rank norms were developed from the Canberra general population sample. Data on PAS scores were weighted by age group, sex and place of residence (community or institution) to match the structure of the population living in Canberra and Queanbeyan in 1990. The norms cover the whole population, including cases of dementia and depression. The main difference between the Canberra population and the rest of Australia is the higher level of education. However, only the Cognitive Impairment scale is affected by education, with a correlation of 0.18 in the Canberra sample.
The cutoffs on the PAS Summary Profile were set to detect around 80% of diagnosed cases of dementia and depression. To get sufficiently large groups for this analysis, cases were pooled from the Canberra, Sydney and Geneva samples. Individuals were included as a case if they satisfied either the ICD-10 or DSM-III-R criteria. The average profiles for cases of dementia and depression were also developed from these pooled data. The average profiles for cases of vascular and Alzheimer's dementia were based only on the Canberra data, as specific types of dementia were not diagnosed in the Sydney and Geneva studies.
This and other technical information about the PAS has been incorporated into the Online User's Guide and the printable version on the PAS Download page.
Top of Page