The ASCQ-Me
research team used exploratory and confirmatory factor analysis including structural
equation modeling (PROC CALIS and MPlus) followed by item response theory modeling
(Multilog and IRT-PRO) to analyze the ASCQ-Me field test data to construct item
banks. Our purpose in using these methods was to understand the dimensional
structure underlying the responses of patients to ASCQ-Me questions, including where
the responses fell on the underlying health dimension tapped by each item bank
and the amount of information each response and each question contributed to
the dimension. The resulting statistics were used to create computerized adaptive tests (CATs). The most useful and informative items in each ASCQ-Me item
bank were identified to create five-item, short forms which could be employed
in place of the CATs for those users without access to computers.
Analysis of Subdomains and Question Scoring
The ASCQ-Me SCD severity questions and the pain episode
questions were not analyzed using IRT. Exploratory and confirmatory factor
analyses were used to identify two subdomains within the pain episode question
responses – frequency and severity. Then classical psychometric analyses were used to evaluate the construct validity and reliability of the two dimensions including internal consistency reliability (Cronbach, 1951) for the subscales,[1] and the magnitude of the item total correlations. We also compared the correlation of an item with its own scale (corrected for overlap[2]) with its correlation to the other scale.[3] We did not use structural methods to analyze responses to the
SCD severity questions because we did not hypothesize that there should be any
relationship between the responses to these questions. These were simply a
checklist of medical conditions and treatments that are often associated with
SCD but which could be independent of each other. Thus, we scored the ASCQ-Me
SCD severity questions as an index using a method commonly employed for
co-morbidity indices which is to create a score from the simple sum of the
number of questions endorsed.[4,5,6] We realize that this treats each question with
equal weight and encourage others to conduct research using alternative
weighting schemes for each question or, indeed, to use alternative questions.
These analyses, including the evaluation of unidimensionality
and differential item functioning, the development of calibrated item banks,
and the construction of short forms is detailed here.
[1]
Nunnally, JC (1978). Psychometric theory. Second edition. New York: McGraw-Hill Book Company.
[2]
Howard, K.I. and Forehand, G.G.(1962).: A Method for Correcting Item-Total Correlations for the Effect of Relevant Item Inclusion. Educational and Psychological Measurement 22:731-735
[3] Campbell, D.T. & Fiske, D.W. (1959). Convergent and discriminant validation by the multitrait-multimethod matrix. Psychological Bulletin, 56, 81-105.
[4] Michelson H, Bolund C, Brandberg Y. Multiple chronic health problems are negatively associated with health related quality of life (HRQoL) irrespective of age. Qual Life Res 2000;9:1093-1104.
[5] Wensing M, Vingerhoets E, Grol R. Functional status, health problems, age and comorbidity in primary care patients. Qual Life Res 2001;10:141-148
[6] Meyer HH. Methods for scoring a check-list type rating scale. J Appl Psychol 1951; 35(1):46-49