Enhancing precision of the 16-item Informant Questionnaire on Cognitive Decline in the Elderly (IQCODE-16) using Rasch methodology

Abstract

Objective: This study aimed to investigate psychometric properties and enhance precision of the 16-item Informant Questionnaire on Cognitive Decline in the Elderly (IQCODE-16) up to interval-level scale using Rasch methodology. Design: Partial Credit Rasch model was applied to the IQCODE-16 scores using longitudinal data spanning 10 years of biennial follow-up. Setting: Community-dwelling older adults aged 70–90 years and their informants, living in Sydney, Australia, participated in the longitudinal Sydney Memory and Ageing Study (MAS). Participants: The sample included 400 participants of the MAS aged 70 years and older, 109 out of those were diagnosed with dementia 10 years after the baseline assessment. Measurements: The IQCODE-16. Results: Initial analysis indicated excellent reliability of the IQCODE-16, Person Separation Index (PSI) = 0.92, but there were four misfitting items and local dependency issues. Combining locally dependent items into four super-items resulted in the best Rasch model fit with no misfitting or locally dependent items, strict unidimensionality, strong reliability, and invariance across person factors such as participants’ diagnosis and relationship to their informants, as well as informants’ age and sex. This permitted the generation of conversion algorithms to transform ordinal scores into interval data to enhance precision of measurement. Conclusions: The IQCODE-16 demonstrated strong reliability and satisfied expectations of the unidimensional Rasch model after minor modifications. Ordinal-to-interval transformation tables published here can be used to increase accuracy of the IQCODE-16 without altering its current format. These findings could contribute to enhancement of precision in assessing clinical conditions such as cognitive decline in older people.

Establishing conversion of the 16‐item Informant Questionnaire on Cognitive Decline in the Elderly scores into interval‐level data across multiple samples using Rasch methodology

“…This study applied Rasch analysis to investigate psychometric properties of a widely used informant-reported CC measure, the IQCODE-16, and re-evaluated the generalizability of the Rasch conversion algorithms proposed in a recent study by Truong and colleagues. 21 Good reliability of the IQCODE-16 was consistently obtained across analyses in this study, adding further empirical evidence supporting the robust psychometric properties of the scale. Interestingly, the Rasch solution suggested by Truong and colleagues 21 did not work equally well for all samples in the current study, potentially due to the previous study's biased sample.…”

“…21 Good reliability of the IQCODE-16 was consistently obtained across analyses in this study, adding further empirical evidence supporting the robust psychometric properties of the scale. Interestingly, the Rasch solution suggested by Truong and colleagues 21 did not work equally well for all samples in the current study, potentially due to the previous study's biased sample. Therefore, an alternative solution was found in the current study, which involved reorganizing super-items of the IQCODE-16 to achieve the best Rasch model fit across all samples.…”

“…It should be noted that the Rasch solution fit in Truong and colleagues' 21 study was not replicated in this study, possibly because we used different DIF factors in the current study. Also, the differences between the original conversion algorithm generated by Truong and colleagues 21 and the one generated in the current study are marginal, meaning both conversion tables can be used for ordinal-to-interval transformations of the data. However, the current conversion table has a higher degree of generalizability across different samples and is more robust with regard to personal factors such as participants' age and sex.…”

“…Descriptive statistics including mean, standard deviation (SD), Cronbach's alpha, and McDonald's omega for the IQCODE-16 across the four samples were computed using IBM SPSS v.27. Rasch analyses were conducted using the RUMM2030 software package 51 replicating the analytical solutions in Truong and colleagues, 21 which were based on the standardized criteria for the Rasch model fit as recommended elsewhere. 39,52 Replications involved the use of unrestricted Partial Credit models 24 and examined whether the Rasch model fit solution in Truong and colleagues' 21 study was appropriate for all samples in the current study.…”

“…CC levels). 21,25 Another issue related to the ordinal nature of the IQCODE-16 scores involves conducting parametric statistics, as they violate the arithmetic assumptions of such tests. 26,27 Rasch analysis 28 can help overcome these limitations.…”