Presence of and correction for interviewer error on an instrument assessing cognitive function of older adults.

Abstract

To determine the extent of interviewer error on the Short Portable Mental Status Questionnaire (SPMSQ), the association between interviewer-level variables and SPMSQ score, compare the prevalence of SPMSQ cognitive status categories before and after correction for interviewer error, and identify SPMSQ items prone to interviewer error. Data of 4542 older adults from a national survey in Singapore were utilized. Multilevel models, which adjusted for respondent-level variables known to be associated with cognitive status, were utilized to determine the intraclass correlation on the SPMSQ, association of interviewer-level variables (age, ethnicity, education and number of interviews carried out) with SPMSQ and identify SPMSQ items prone to interviewer error. The intraclass correlation, after adjusting for respondent-level variables, was 0.265. Interviewer educational status was associated with SPMSQ score. Correction for interviewer error resulted in an "improvement" in cognitive status of a substantial proportion of those initially classified as moderately or severely impaired. Two of the SPMSQ items, "Please count backward from 20 by 3" and "When were you born?" were particularly prone to interviewer error. Investigators using the SPMSQ, and potentially, other instruments, for assessing cognitive function in surveys of older adults and using face-to-face interviews for data collection should be cognizant of the sensitivity of such instruments to interviewer error. Intensive training of interviewers to standardize instrument administration as well as assessment for interviewer error and its correction, using appropriate statistical models, such as multilevel models, is warranted.