Misclassification of dementia by the Mini-Mental State Examination - are education and social class the only factors?
One effect of the annual screening of the over-75s which general practitioners now have to undertake (Department of Health ) has been to create a need for short screening tests which detect, among other things, signs of physical and mental deterioration. These tests need to be concise, easy to administer and accurate. With respect to cognitive impairment, a number of short screens are available including the widely used Mini-Mental State Examination (MMSE) of Folstein et al. . Moreover, the MMSE, was recommended by the Medical Research Council Working Group on dementia in 1986 to be included in any future data set collected for research into Alzheimer's disease, to enable comparisons across studies.
However, no perfect short screening test for dementia yet exists. Although the MMSE has high sensitivity it can, under certain conditions, misclassify elderly persons as having signs of dementia. There is evidence from a number of community studies [3-7] that social factors and educational level are associated with poorer test scores on the MMSE, though this does not, in itself, show that these factors are the reason for any misclassification. Furthermore, attempts have been made to produce adjusted norms for the MMSE based on level of education  and age . Use of the MMSE in a community setting has further suggested to us that another factor, lack of manual dexterity, might also result in lower scores on the MMSE as items such as the writing of a sentence and the paper-folding exercise, in particular, are dependent on the level of this skill.
The aims of this paper are to investigate (i) the extent to which low social class and level of education result in elderly persons being miclassified as demented by the MMSE, and (ii) whether those persons with conditions which reduce their physical ability, particularly manual dexterity, or with hearing or visual impairment are at increased risk of misclassification.
The study population was taken from a large general practice which alone serves the town of Melton Mowbray, Leicestershire and its environs. All thone registered with the general practice and aged 75 years and over on 1 January 1988 were interviewed in their own homes or place of residence, including hospital, local authority and private residential home and nursing home. Trained field-workers conducted the interviews using a standard questionnaire which covered basic demographic data together with information on physical and mental health, social status and use of services. At this initial interview the MMSE was administered with the serial sevens option being used rather than the spelling of |world' backwards.
The second stage of the study, that of the confirmation of the dementia, proceeded from the results of the MMSE score, with missing or refused items scoring zero. All subjects scoring 21 or under, a one in two random sample of those scoring 22 or 23 and a one in ten sample of those scoring 24 and over were offered a full psychiatric assessment in the form of the Cambridge Mental Disorders of the Elderly Examination (CAMDEX)  administered by four experienced clinicians. This schedule contains a mental state examination, medical and psychiatric history, detailed cognitive testing, a physical examination and an interview with an informant. Using all the available information from the subject (although not the results of the cognitive tests per se) and informant and employing the detailed criteria from the CAMDEX, a clinical diagnosis of dementia was made together with a rating of its severity.
Social class was determined from the initial interview and coded to the 1980 Classification of occupations . Married and widowed women were classified according to their husbands' former occupations. For the present analysis, social class was dichotomized into non-manual (classes I, II and IIINM) and manual (classes IIIM, IV and V). Level of education was ascertained by two items from the CAMDEX schedule: the age at which subjects left school, categorized into (14 years and under/15 years and over) and number of years of further education after leaving school, categorized into (none/1 or more years).
A number of other factors which might result in misclassification by the MMSE were also considered. Visual and hearing impairments were assessed in the CAMDEX schedule and dichotomized into no impairment/some difficulty or totally impaired. Physical disability was classified (as none/some) using a scale of activities of daily living collected at the initial screening interview. The derivation of the scale has been described in detail elsewhere . Two other items assessing dexterity were included in the CAMDEX schedule: tremor (none/some) and physical difficulty with a manual task (yes/no).
Statistical analysis: Previous analysis of these data , confirming work of other researchers, had shown that a cut-point of 21/22 on the MMSE best detected moderate and severe dementia whilst 23/24 had the greatest sensitivity and specificity to detect mild, moderate or severe dementia. Because of this and the design of the study, misclassification was investigated by exploring the association between MMSE score in three categories (21 and under/22 or 23/24 and over) and the factors of interest (age, sex, social class, visual and hearing impairment, physical disability and manual dexterity etc.) in those subjects with no clinical diagnosis of dementia. Misclassification was only considered in this direction as the MMSE has high sensitivity . To enable conclusions to be drawn about the total population when the analysis was restricted to subjects selected for the CAMDEX assessment, entailed some form of weighting by the probability of selection, i.e. 1 for those with MMSE scored 21 or under, 0.5 for those scoring 22 or 23 and 0.1 for those scoring 24 and over.
One solution would have been to reweight the data with weights inversely, related to the sampling probabilities. However, although this would produce unbiased estimates of parameters, the standard errors of the parameters would be artificially deflated and this would result in variables appearing to add significantly to a multiple regression model when, in fact, they did not.
Accurate estimation of the standard errors could have been achieved by fitting a multinomial logit model to the data, following Aitkin et al.  with, in addition, the logarithm of the sampling fraction as an offset to achieve reweighting. In this model, associations between the response variable (MMSE score) and the factor of interest would be shown by a significant interaction of these two variables in the model. However, this approach is very, unwieldy for contingency, tables formed by the cross-classification of a large number of explanatory variables as cell counts tend to be very small.
The actual analysis of these data used both methods. Owing to the problem of sparseness of data if the multinomial model had been fitted with all the factors of interest, the initial analyses involved fitting models to the two-way tables of MMSE score and the explanatory factor of interest to look at the effect of each of these factors separately. A multi-way table was then formed of all the significant factors of interest and the MMSE score, weighted according to the sampling procedure. Standard log-linear analysis was then performed to fit models to this table with MMSE score as the response variable. When the final selection of explanatory factors significantly associated with MMSE score had been found, the multinomial model, as described earlier, was fitted to the multi-way table formed by these factors and MMSE score to check that none of the factors had been erroneously included in the model.
On the age-sex register 1890 persons aged 75 years and over were identified; 7% died before interview, 4% could not be contacted and 5% refused with 1579 completing the first stage of the study. Further results from the initial screening stage of the study have already been published . From the initial MMSE score, 33% (518) were selected for the detailed assessment by the CAMDEX: 330 with a MMSE score of 21 or under, 81 who scored 22 or 23 and 107 with a score of 24 and above. One per cent (7) moved prior to interview, 7% (35) died before the CAMDEX interview, 7% (38) refused the CAMDEX interview and 85% (438) completed the second stage of the study.
Among those completing the CAMDEX, the very elderly, women, and those interviewed in an institution were over-represented compared with the total study population, reflecting the increase in prevalence of dementia in these subgroups. However, after reweighting (by the sampling procedure), the CAMDEX group and the total study population had very similar distributions of a number of the variables contained in both interviews, including age, sex, physical disability scale, place of interview and social class. A total of 155 persons had no clinical diagnosis of dementia and these formed the basis of the present analysis. Table I shows the distributions of MMSE score and socio-demographic characteristics such as age, sex, social class and level of education together with the distributions of physical disability, visual and hearing impairment, tremor and physical difficulty with a manual task for those persons not clinically demented both unweighted and reweighted by the sampling probabilities.
Table I. Distribution of the sociodemographic and physical ability measures in the CAMDEXed population free of dementia weighted according to the sampling probabilities and unweighted Unweighted Weighted[*] % No. % No. MMSE score < = 21 31 48 5 48 22, 23 16 24 5 48 24 + 53 83 90 830 Age group 75-79 43 67 55 513 80-84 34 53 33 307 85 + 23 35 11 106 Sex Male 36 55 40 369 Female 64 100 60 557 Social class Non-manual 30 46 38 355 Manual 64 100 56 517 Missing 6 9 6 54 Physical disability None 75 116 84 773 Some 24 38 16 152 Missing 1 1 0 1 Age left school 14 years and under 80 124 78 721 15 years and over 17 26 21 199 Missing 3 5 1 6 Years further education None 88 136 92 850 One or more 9 14 8 71 Missing 3 5 0 5 Hearing impaired None 70 109 74 684 Some 28 44 26 240 Missing 1 2 0 2 Visually impaired None 70 109 79 730 Some 28 44 21 194 Missing 1 2 0 2 Tremor None 93 144 97 894 Slight/severe 4 7 3 27 Missing 3 4 0 5 Physical difficulty with a manual task Yes 94 145 98 906 No 4 7 2 17 Missing 2 3 0 3 [*] Weighted according to sampling procedures.
Table II shows the results of the analyses of each of the explanatory factors of interest taken singly. Using a significance level of 0.005 to compensate for the increased risk of Type I error with multiple tests, clearly significant associations were found between MMSE score and age, social class, visual impairment, physical disability and physical difficulty with a manual task.
[TABULAR DATA OMITTED]
For the multivariate analysis, models were fitted to the multi-way table formed by the cross-classification of MMSE score and age, social class, visual impairment, physical disability and physical difficulty. with a manual task. From this analysis using a stepwise procedure, age, social class and visual impairment showed highly significant associations with MMSE score.
Finally, a multinomial logit model was fitted to the table formed by these three, explanatory variables and MMSE score. The interactions between MMSE score and all three factors remained significant, the changes in deviance when each of the interactions was dropped from the model being age (25.02, df = 4, P < 0.0001), social class (15.29, df = 2, p = 0.0005) and visual impairment (8.09, df = 2, p = 0.018). The final model provided a good fit to the data with a residual deviance of 15.94 on 14 df.
The Figure shows the fitted probabilities, from the final model, by age group, of falling into each of the MMSE categories for each of the combinations of non-manual/manual social class and with/without visual impairment. The strong fitted relationship between lower MMSE score and increasing age is clearly visible with increasing age resulting in a higher probability of scoring 21 and under and a correspondingly lower probability of scoring 24 or above. The association with visual problems and manual social class is also strong. Even in the youngest age group, those of manual social class and with some visual impairment had a 42% chance of scoring 21 or under on the MMSE despite having no clinical signs of dementia.
The analysis reported here has demonstrated that certain groups of subjects with no clinical diagnosis of dementia have an increased likelihood of obtaining low scores on the MMSE, and therefore of being misclassified as having features of dementia. At the highest risk of misclassification are those persons aged 85 years and over, of manual social class and with some visual impairment. The response rates for both phases of the screening were very high and as Melton Mowbray is similar to England and Wales in terms of age, sex and social class distribution it is unlikely that the results are unrepresentative.
There have been two previous studies based in UK general practices using similar designs, which have investigated the association between MMSE score and both social class and educational status. Brayne and Calloway  surveyed women between the ages of 70 and 79 years in rural Cambridgeshire. All the women had a full psychiatric assessment using the CAMDEX and increasing age, manual social class and a low level of education were associated with lower scores on the MMSE.
O'connor et al. , in the Hughes Hall Project for Later Life conducted in the city of Cambridge, studied both sexes aged 75 years and over. A two-stage sampling procedure was employed with all those scoring 23 and below and a one-in-three sample of those scoring 24 or 25 going forward for a CAMDEX assessment. Analysis of data from the screening interview showed associations between MMSE score and age, education and social class. The present study had similar age, sex, and social class profiles to the Hughes Hall data although Melton Mowbray appeared to have a greater percentage who left school at or before 14 years of age (75% vs. 68%).
The problems with both these previous analyses and with USA studies [3, 4, 7] were that they were studies of association of factors with MMSE rather than misclassification. If the prevalence of dementia increases with lower education, lower social class and increasing age, an analysis of the total population, including the true cases of dementia will show a relationship between these variables and MMSE score. It does not follow that these variables are resulting in falsely positive cases of dementia. Age- and education-specific norms have been derived despite the equivocal evidence so far but both have used USA populations and selected case and control groups rather than community populations. The ability to generalize these norms is therefore still open to question.
Another approach to addressing the problem of level of education and the MMSE was made by Jorm et al. [1 6] who compared the predictive and construct validity of the MMSE between two groups defined as more and less educated elderly persons. Jorm found no evidence that the MMSE was biased towards the better educated, though the two groups were of small size and differences may have been missed. None of these papers properly addresses the issue, of whether low social class and low educational level are causing misclassification. The present study shows that increased age and low social class are associated with an increased likelihood of misclassification as false positive. Visual impairment can also result in low MMSE scores in those without clinical signs of dementia. This study showed no evidence of misclassification due to the sex of the subject. Educational level, physical disability and manual dexterity when considered singly were significantly associated with low MMSE score but these effects were subsumed by the effects of age, social class and visual impairment. It is important that the MMSE score is used in its proper context, as a screening tool, and not as a definitive diagnosis and that factors such as manual social class and visual impairments be taken into account.
We would like to thank Drs H. Hollis, P. W. E. Johnston, P. J. C. Howe, M. N. G. Halford, G. E. Martin, B. C. McG. Williamson, R. J. Thew, B. Kirkup, D. J. Corvin, T. D. W. Smith, D. A. Barrow, A. D. Firkin, D. M. Lovett, C. O'shea, B. E. Holt and Mr J. Bishop for their co-operation; the community nursing services, paramedical services, social services department and voluntary, agencies for their help; Finian Kelly and Morag Campbell Stern for their help with the CAMDEX interviews; John Woods and Neil Raymond for computer programming; Linda Bingham, Alison Hipkin and Lesley Harris for office management; Carolyn Douglas for help with the manuscript; and David Clayton and David Jones for helpful discussion.
We are grateful to Nuffield Provincial Hospitals Trust for funding the main 1988 survey and for continued support.
[1.] Department of Health and Welsh office. General practice in the National Health Service: a new contract. London: Department of Health, 1989. [2.] Folstein MF, Folstein SE, McHugh PR. |Mini-Mental State': a practical method for grading the cognitive state of patients for the clinician. F Psychiatr Res 1975;12:189-98. [3.] Escobar JI, Burnam A, Karno M, et al. Use of the Mini-Mental State Examination in a community population of mixed ethnicity". F Nerv Ment Dis 1986; 174:607-14. [4.] Fillenbaum GG, Hughes DC, Heyman A, et al, Relationship of health and demographic characteristics to Mini-Mental State Examination score among community residents. Psychol Med 1988;18:719-26. [5.] O'Connor DW, Pollitt PA, Treasure FP, Brook CPB, Reiss BB. The influence of education, social class and sex on Mini-Mental State scores. Psychol Med 1989;19:771-6. [6.] Brayne C, Calloway P. The association of education and socioeconomic status with the Mini Mental State Examination and the clinical diagnosis of dementia in elderly people. Age Ageing 1990;19:91-6. [7.] Ganguli M, Ratcliff G, Huff J, et al. Effects of age, gender and education on cognitive tests in a rural elderly community sample: norms from the Monongahela Valley Independent Elders Survey. Neuroepidemiology 1991;10:42-52. [8.] Uhlmann RF, Larson EB. Effect of education on the Mini-Mental State Examination as a screening test for dementia. F Am Geriatr Soc 1991; 39:876-80. [9.] Bleeker ML, Bolla-Wilson K, Kawas C, Agnew J. Age-specific norms for the Mini-Mental State Exam. Neurology 1988;38:1565-8. [10.] Roth M, Huppert FA, Tym E, Mountjoy CQ. The Cambridge examination for mental disorders of the elderly. Cambridge: Cambridge University Press, 1988. [11.] Office of Population Censuses and Surveys. Classification of occupations. London: HMSO, 1980. [12.] Jagger C, Clarke M, Davis RA. The elderly at home: indices of disability. F Epidemiol Commun Health 1986;40:139-42. [13.] Clarke M, Jagger C, Anderson J, Battcock T, Kelly F, Campbell Stern M. The prevalence of dementia in a total population: - a comparison of two screening instruments. Age Ageing 1991; 20:396-403. [14.] Aitkin M, Anderson D, Francis B, Hinde J. Statistical modelling in GLIM. Oxford: Oxford University Press, 1989;235-6. [15.] Jagger C, Clarke M, Clarke SJ. Getting older - feeling younger: the changing health profile of the elderly. Int F Epidemiol 1991;20:234-8. [16.] Jorm AF, Scott R, Henderson AS, Kay DWK. Educational level differences on the Mini-Mental State: the role of test bias. Psychol Med 1988;18:727-31.
|Printer friendly Cite/link Email Feedback|
|Author:||Jagger, C.; Clarke, M.; Anderson, J.; Battcock, T.|
|Publication:||Age and Ageing|
|Date:||Nov 1, 1992|
|Previous Article:||Longitudinal diagnosis of memory disorders.|
|Next Article:||Spinal dural arteriovenous malformations - a treatable cause of progressive paraparesis in elderly people.|