The Informant Questionnaire on Cognitive Decline in the Elderly (IQCODE) as a screening tool for dementia for a predominantly illiterate chinese population

There are brief cognitive tests such as the Mini-Mental State Examination (MMSE) ^[1] for screening for dementia. These instruments can provide objective measures of cognitive abilities, but performance on cognitive tests is strongly correlated with the subjects' educational backgrounds ^[2-4]. Because public education has become widely available only in the past several decades, older individuals generally have had less formal schooling. Consequently, screening for dementia based on cognitive testing alone may lead to oversampling of older and less-educated individuals ^[5].

The Informant Questionnaire on Cognitive Decline in the Elderly (IQCODE) ^[6,7] does not test for cognitive abilities but asks for ratings of an individual's changes in everyday cognitive functions during the previous 10 years. The original studies conducted in Australia showed that the IQCODE had high internal consistency (alpha = 0.95), based on a sample from the general population, and reasonably high test-retest reliability over 1 year (r = 0.75), based on a sample of dementing patients ^[7]. In addition, the IQCODE had little correlation with education or premorbid ability level, ^[6,7] could discriminate well between the general population and the dementing patients, and performed at least as well as the MMSE as a screening instrument for dementia ^[8].

One purpose of the present study was to evaluate the applicability of the IQCODE to a predominantly illiterate population of Chinese elderly, ie, to examine the influence of age, education, and gender on the IQCODE score among nondemented individuals and to compare its sensitivity and specificity with those of a cognitive test score for detecting dementia in this population. A low test score can reflect either low educational background, dementia, or both. Because the IQCODE is a measure of change in performance that is relatively independent of the premorbid level, it might be more suitable than a test score for populations with little or no formal education.

Another purpose of this study was to compare the relative sensitivity of the 26 items of the IQCODE in detecting dementia and to check whether some of the items can be omitted without appreciable reduction in its efficacy. A factor analysis showed that items of the IQCODE measured primarily one general factor of cognitive decline, ^[7] suggesting that some items may be redundant. In addition, items that ask about changes in reading and writing abilities do not apply to illiterate individuals. We felt that a shorter version of the IQCODE could be devised that would save time and effort in both clinical and epidemiologic studies.

Methods. Subjects. The subjects consisted of 399 nondemented community residents and 61 dementia patients. The community residents were sampled from the township of Kin-Hu on the islet of Kinmen situated west of Taiwan, just off the coast of mainland China. The sampling frame comprised all individuals aged 50 years and over according to 1990 household registration records. A random sample stratified by sex and age yielded approximately 100 men and 100 women in each of the four age groups 50 through 59, 60 through 69, 70 through 79, and 80 and older, for a total of 827 individuals. When actual field work began in the summer of 1992, only 683 individuals remained because 70 persons had died and 74 persons had moved away. Among the 683 individuals, 24 were hospitalized, 77 were not home during three visits, 102 declined or withdrew participation, and 58 did not have an appropriate informant at the time and place of the visit. In addition, 16 persons were demented and classified as patients, and seven persons were excluded because the IQCODE for them had six or more of the items not rated. These exclusions resulted in a final sample of 399 nondemented control subjects.

The dementia patients either were identified from the community survey (N = 16) or were outpatients at the Neurology Clinic of the Veterans General Hospital- Taipei (N = 48). Among the 64 patients, three were excluded from data analysis because, for each of them, six or more of the items on the IQCODE were not rated. The remaining 61 dementia patients consisted of 48 with probable Alzheimer's disease, 11 with vascular dementia, one with alcoholism, and one with Parkinson's disease.

Evaluation instruments. The Informant Questionnaire on Cognitive Decline in the Elderly. The IQCODE ^[6,7] has 26 items; it requires an informant to rate an individual's changes in memory and other cognitive abilities during the previous 10 years on a 5-point scale in which 1 indicates "much better," 3 indicates "not much change," and 5 indicates "much worse". The score is the average rating of the rated items.

The Cognitive Abilities Screening Instrument (CASI). The CASI ^[9,10] is a cross-cultural test that has a score range of 0 to 100 and provides quantitative assessment of attention, concentration, orientation, short-term memory, long-term memory, language abilities, visual construction, list-generating fluency, abstraction, and judgment. Some of its items are identical to or closely resemble those included in the MMSE; thus, an estimated MMSE score can be obtained from these items. This CASI-estimated MMSE score has been designated "MMSE-CE," in which CE represents "CASI-estimated" ^[10]. The MMSE-CE has been found to be nearly identical to the conventionally obtained MMSE score in one study: based on 88 normals and 57 dementia patients, the mean MMSE-CE score was 23.21, the mean MMSE score was 23.30, and the two scores had a correlation coefficient of 0.92 ^[11]. We chose the MMSE-CE for comparison with the IQCODE because the MMSE is probably the most frequently used screening test for dementia, and it was also the test chosen to be compared with the IQCODE in the original studies conducted in Australia ^[8]. (Our other findings with the CASI will be reported separately.)

For the present study, the IQCODE and the CASI were translated and adapted from English into Chinese. Translation and adaptation were accomplished through group discussions among those coauthors who are fluent in both languages, as well as knowledgeable about the Chinese culture and experienced in clinical and research studies of dementia. After pilot testing with individuals who were not subjects of the present study, the draft versions of the IQCODE and the CASI in Chinese were further polished and finalized before their use in the present study. The full code names of these two instruments are IQCODE version C-1.0 and CASI version C-2.0; for brevity, the version codes will be omitted in the rest of this article.

Procedure. The IQCODE was administered to informants and the CASI was administered to the subjects by trained research assistants who were medical or nursing students and who did not know the subjects' diagnostic status.

All subjects were also evaluated by one of a team of six neurologists or by a psychiatrist who did not know the scores on the IQCODE and the CASI. The physicians' assessments included the following: (1) a semistructured interview comprising inquiries about daily activities and testing of memory, orientation, abstract thinking, and judgment that were different from those used in the IQCODE and the CASI; (2) a complete neurologic examination; (3) the Center for Epidemiological Studies Depression scale ^[12,13]; and (4) standardized assessments of Parkinson's disease, stroke, and transient ischemic attacks. When dementia was suspected, a family member was also interviewed for history, daily activities, and social functioning. These cases were presented for group discussions among the physicians, and diagnosis was reached by consensus. The diagnosis of dementia was made according to the DSM-III-R criteria ^[14]. The severity of dementia was rated from 1 (mild) to 3 (severe) according to the Clinical Dementia Rating (CDR) scale ^[15]. Neither the IQCODE nor the CASI scores were used in diagnosis or classification.

Statistical analysis. Descriptive and analytic statistics were performed with the SAS and BMDP computer programs. Two-tailed tests were used in pairwise comparisons. A p value of <0.01 was adopted for statistical significance.

Results. Demographic characteristics of the subjects. The 61 patients in the dementia group included 39 men and 22 women. The 399 subjects in the control group included 196 men and 203 women. The male preponderance in the patient group reflects the fact that most of the patients were from a veterans' hospital.

The ages and education levels of the subjects are presented in Table 1. The patient group was older and better educated than the control group; the mean differences in age and education were significant at the 0.01 level by the t test. However, the educational backgrounds of both groups were low. Sixty-three percent of the 460 subjects had 0 years of formal schooling.

Relationship of the informants to the subjects. The relationship of the informants to the subjects is also shown in Table 1. The length of time the informant had lived with or had frequent contacts with the subject ranged from 10 to 61 years (mean +\- SD = 40.5 +\- 12.3) in the patient group and from 10 to 65 years (mean +\- SD = 29.5 +\- 15.3) in the control group. Separate data analyses for the patients and for the controls showed that the IQCODE score had no significant association with either the length of the relationship (according to Pearson's r) or the type of the relationship (by the ANOVA test).

Effect of age, education, and sex on the IQCODE and the MMSE-CE scores in the control group. Forward stepwise regressional analysis showed that the IQCODE score was somewhat influenced by age (R² = 0.0449, p < 0.0001), but the rate of decline increased only 0.04 points per 10 years of age. Education and gender had no significant additional effect on the IQCODE when the influence of age was controlled. In contrast, the MMSE-CE score was more strongly associated with age, education, and gender. The values of the R² were 0.1881, 0.3182, and 0.3938 after the successive entering of the independent variables of age, education, and gender.

Score on the IQCODE in relation to dementia status. According to the CDR scale, dementia was mild (CDR = 1) in 34 patients, moderate (CDR = 2) in 10 patients, and severe (CDR = 3) in 17 patients. The mean (+\- SD) and the 10th to the 90th percentile range of the IQCODE scores were 3.12 (0.23), 1.92 to 4.92 for the control group; 3.65 (0.39), 3.04 to 4.52 for the mild dementia group; 4.37 (0.39), 3.27 to 4.88 for the moderate dementia group; and 4.71 (0.33), 3.76 to 5.00 for the severe dementia group. A one-way ANOVA showed that the differences among the four means were highly significant (F = 203.87, df1 = 3, df2 = 456, p < 0.0001). A priori planned comparisons of the means between adjacent groups showed that the mean differences were significant at the 0.01 level between control and mild dementia and between mild and moderate dementia. Although the mean difference between moderate and severe dementia was nonsignificant, the trend was in the expected direction.

Comparison between the IQCODE and the MMSE-CE on their sensitivity and specificity for dementia. The Pearson's correlation coefficient between the IQCODE score and the MMSE-CE score was -0.69 (p < 0.0001). To compare the sensitivity and specificity of various IQCODE and MMSE-CE cutoff scores for detecting dementia, receiver operating characteristic (ROC) curves ^[16,17] were plotted and are shown in the Figure 1. The area under the ROC curve, an index of the overall performance of a screening test, ^[18,19] was 91.3% (SE = 2.4%) for the IQCODE and 84.0% (SE = 3.1%) for the MMSE-CE; the difference between the two areas was significant at the 0.01 level (z = 2.57). For the IQCODE, the optimal cutoff score of >=3.4 for dementia yielded an 89% sensitivity and an 88% specificity. For the MMSE-CE, the optimal cutoff score of <=20 for dementia yielded a 72% sensitivity and an 81% specificity. The positive and negative predictive values, calculated based on a 10% disease prevalence, ^[20] were 45% (positive) and 99% (negative) for the IQCODE and 30% (positive) and 96% (negative) for the MMSE-CE.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1. The receiver operating characteristic curves of the IQCODE and the MMSE-CE

IQCODE item analysis. Based on the control subjects, the IQCODE items had a Cronbach's alpha of 0.97, indicating a high level of internal consistency. A principal components factor analysis showed that the first factor accounted for 60.5% of the common variance and that the next three factors respectively accounted for 6.6%, 4.6%, and 4.0%. The loadings of the 26 items on the first factor ranged from 0.60 to 0.85.

Stepwise discriminant analysis followed by the jackknife validation procedure was performed to select a subset of IQCODE items that would perform nearly as well as using all 26 items. Seventeen items were selected by the computer program, and these yielded 97.1% classification accuracy. This subset of 17 items is shown in Table 2. Area under the ROC curve based on the 17 items was 91.1% (SE = 1.8%), and that based on the MMSE-CE was 84.0% (SE = 3.1%). The difference between the two areas was significant at the 0.01 level.

It is noteworthy, however, that using the first two selected items of "recalling conversations a few days later" and "handling financial matters, eg, the pension, dealing with the bank" alone gave 96% classification accuracy. Based on the average rating on these two items, a cutoff score of >=4.0 for dementia provided 87% sensitivity and 95% specificity, and a cutoff score of 3.5 provided 92% sensitivity and 80% specificity.

Discussion. The IQCODE was originally written in English and used in Australia. For the present study, we translated and adapted it for use in a population of predominantly illiterate Chinese elders. Despite the differences in language and educational background of the subjects between the Australian and the Chinese studies, we corroborated the original findings of the IQCODE: the items in the questionnaire have very high internal consistency and they measure primarily a general factor of cognitive decline; its score has little or no association with the subjects' educational backgrounds or gender; and its score has a low association with the subjects' ages and a moderately high association with tested cognitive abilities. Both the Australian and the present studies showed that the IQCODE could effectively discriminate between normals and dementia patients. We further demonstrated that the IQCODE could also discriminate between normals and patients in the early stage of dementia and between different stages of dementia.

We had expected that, for populations with little or no formal schooling, the IQCODE would be more sensitive than cognitive tests such as the MMSE for detecting dementia. This expectation was confirmed by the finding that the area under the ROC curve was significantly larger for the IQCODE than for the MMSE-CE. However, our sample was characterized not only by the low average level of educational background but also by a large variation in educational background: 63% of the 460 subjects, including 271 controls and 20 dementia patients, had 0 years of education (the "no-education" group); the educational background of the remaining 128 controls and 41 dementia patients (the "low-education" group) ranged from 1 to 19 years (mean +\- SD = 6.21 +\- 4.36). To examine further the circumstance or circumstances for the advantage of the IQCODE over the MMSE-CE, we repeated separately for the low-education group and the no-education group the comparison between the area under the ROC curve for the IQCODE and that for the MMSE-CE. A significant (p < 0.01) advantage of the IQCODE over the MMSE-CE was obtained only for the low-education group. For the no-education group, a nonsignificant trend in the opposite direction was obtained. Because the 20 patients in the no-education group was a very small number for ROC analysis, the nonsignificant finding should be regarded as tentative only. Taken at their face value, the findings suggest that ratings of within-subject changes have an advantage over objective test scores when the subjects have large variations in their educational backgrounds but not when the subjects are illiterate. In the low-education group, the average level of education was higher among the dementia patients (mean = 9.7 years) than among the controls (mean = 5.1 years). The higher premorbid cognitive abilities of the patient group might also have reduced the sensitivity and specificity of the MMSE-CE. At the same time, the advantage of the IQCODE in its relative insensitivity to premorbid ability level is again highlighted.

The present study showed that the IQCODE could be shortened from 26 to 17 items without appreciable reduction in its usefulness. This finding is based on a particular sample of predominantly illiterate Chinese elders, and the omitted items included those concerning reading and writing abilities. To what extent the same 17 items can be used effectively in different populations remains to be studied.

Although the present study also showed that the use of only two items on the IQCODE was almost as good as the use of 17 items for detecting dementia, this finding should be regarded with caution. In general, a measure based on a smaller number of items will be less reliable. In addition, the average rating based on two items will show fewer gradations within the score range of 1 to 5 than one based on more items; the latter can provide finer differentiations among individuals and can follow within-individual changes with greater sensitivity.

We conclude that an informant questionnaire about the change in cognitive function may be more effective in population studies than is the traditional screening test of cognitive function, especially when there is large variation in the subjects' educational backgrounds. It should be cautioned, however, that the use of the IQCODE assumes that informants are objective and honest in their ratings, although this assumption may not be true in some cases.