American Journal of Epidemiology Copyright 2004 by the Johns Hopkins Bloomberg School of Public Health All rights reserved

Vol. 160, No. 8 Printed in U.S.A. DOI: 10.1093/aje/kwh288

PRACTICE OF EPIDEMIOLOGY Validity of the Clock-Drawing Test as a Screening Tool for Cognitive Impairment in the Elderly

Yuji Nishiwaki1, Elizabeth Breeze1, Liam Smeeth1, Christopher J. Bulpitt2, Ruth Peters2, and Astrid E. Fletcher1
Centre for Ageing and Public Health, London School of Hygiene and Tropical Medicine, University of London, London, United Kingdom. 2 Section of Care of the Elderly, Faculty of Medicine, Imperial College of Science, Technology, and Medicine, University of London, London, United Kingdom.
1

Downloaded from http://aje.oxfordjournals.org/ by guest on February 5, 2013

Received for publication February 3, 2004; accepted for publication May 14, 2004.

The authors studied the validity of the Clock-Drawing Test (CDT) in comparison with the Mini-Mental State Examination (MMSE) and the association between the CDT and mortality in a large general population sample of persons aged 75 years or more. Data were obtained from the MRC Trial of the Assessment and Management of Older People in the Community. Baseline assessments were conducted between 1995 and 1999 in the United Kingdom. A total of 13,557 subjects with both CDT and MMSE scores were included in the analysis. The median follow-up period was 4.7 years. The sensitivity and specificity of the CDT for detecting moderate-to-severe cognitive impairment (MMSE score 17) were 77% and 87%, respectively, for nurse administration and 40% and 91%, respectively, for postal administration. Sensitivity was higher among females and increased with age. In Cox proportional hazards regression, a worse CDT score was associated with increased risk of all-cause mortality and mortality from cerebrovascular and respiratory diseases, even after adjustment for potential confounders. The authors conclude that the CDT may have value as a brief face-to-face screening tool for moderate/severe cognitive impairment in an older community population but is relatively poor at detecting milder cognitive impairment. aged; cognition; mass screening; mortality; sensitivity and specificity

Abbreviations: CDT, Clock-Drawing Test; CI, confidence interval; HR, hazard ratio; MMSE, Mini-Mental State Examination.

As the population of the developed world ages, dementia is becoming an increasingly important public health concern. In developed countries, the prevalence of dementia is approximately 1.5 percent at age 65 years and doubles every 4 years, to reach approximately 30 percent by age 80 years (1). Patients with dementia require long-term care at home or in an institutional placement, and this entails significant societal costs. Several studies have shown that dementia or cognitive impairment predicts future survival (2). So far, screening for dementia has not been regarded as a priority, partly because of the lack of effective treatments. However,

as new medicines for early treatment or symptom management become available, screening for dementia in the older population will require consideration as a possible preventive strategy. The Mini-Mental State Examination (MMSE) is a wellvalidated measure of cognitive function (3) and has often been used to measure cognitive impairment in epidemiologic and clinical surveys. It consists of 20 items in seven categories. As a population screening tool for cognitive impairment in the community setting, the Clock-Drawing Test (CDT) warrants consideration as a quicker, simpler, and possibly

Correspondence to Professor Astrid E. Fletcher, Centre for Ageing and Public Health, London School of Hygiene and Tropical Medicine, Keppel Street, London WC1E 7HT, United Kingdom (e-mail: astrid.fletcher@lshtm.ac.uk).

797

Am J Epidemiol 2004;160:797807

798 Nishiwaki et al.

Downloaded from http://aje.oxfordjournals.org/ by guest on February 5, 2013

FIGURE 1. Profile of the MRC Trial of the Assessment and Management of Older People in the Community, United Kingdom, 19951999. CDT, Clock-Drawing Test; MMSE, Mini-Mental State Examination.

more cost-effective screening instrument for use in primary care. The reliability and validity of the CDT has been reported (412), but the majority of these studies were based on relatively small sample sizes in hospital- or clinic-based settings. Data on the validity of the CDT in the older general population are needed. Therefore, we conducted a study with the following objectives: 1) to assess the validity of the CDT in comparison with the MMSE as a population screening tool for cognitive impairment and to determine whether validity varies by gender, age group, or method of administration and 2) to examine the association between cognitive function assessed by the CDT and mortality.
MATERIALS AND METHODS The MRC Trial of the Assessment and Management of Older People in the Community

The MRC [Medical Research Council] Trial of the Assessment and Management of Older People in the Community was a cluster randomized trial of patients from general medical practices that compared different approaches to multidimensional screening for people aged 75 years or more, excluding anyone in long-term nursing care or with terminal disease (13, 14). The trial was designed to recruit patients from 108 practices, stratified to provide a representative

sample in terms of mortality (standardized mortality ratio) and deprivation (Jarman score) of general practices in the United Kingdom (England, Wales, and Scotland). The trial had two main arms, universal and targeted (figure 1). Each arm had further practice-level randomization to three methods of administration of a brief questionnaire: postal, interview by layperson, and interview by nurse. In the universal arm, all patients were invited to undergo a brief health assessment followed by a detailed assessment, which was done by the study nurse. In the targeted arm, only patients with predetermined triggers were offered the detailed assessments. The CDT and the MMSE were included in the brief assessment and the detailed assessment, respectively. Before starting the assessments, the nurses and lay interviewers attended a training session. Mortality follow-up was achieved by registering all eligible patients with the Office for National Statistics for notification of death and date and cause of death. Causes of death were coded according to the International Classification of Diseases, Ninth Revision. The major causes of death evaluated in the present analysis were coronary heart disease (codes 410414), cerebrovascular disease (codes 430438), respiratory disease (codes 460519), and malignant neoplasm (codes 140208). Ethical approval for the trial was obtained from relevant ethics committees.
Am J Epidemiol 2004;160:797807

Use of the Clock-Drawing Test to Screen the Elderly 799

TABLE 1. Distribution of scores on the Clock-Drawing Test among 13,557 subjects, MRC Trial of the Assessment and Management of Older People in the Community, United Kingdom, 19951999
CDT* score No. of subjects Proportion (%) Design-adjusted 95% CI* Cumulative % Design-adjusted 95% CI

0 1 2 3 4

90 689 1,060 2,324 9,394

0.7 5.1 7.8 17.1 69.3

0.4, 1.1 4.3, 6.0 6.4, 9.5 15.0, 19.6 66.0, 72.4

0.7 5.8 13.6 30.7 100

0.4, 1.1 4.8, 6.8 11.6, 15.8 27.6, 34.0

* CDT, Clock-Drawing Test; CI, confidence interval.

The CDT and the MMSE

Patients interviewed face-to-face by a layperson or nurse for the CDT were given a blank sheet of paper, a pencil, and the following instruction: Draw a large clock face and put all the numbers in. After this had been done, patients were asked to draw in the clock hands to indicate a time of 10 past 11. For postal administration, patients were provided with a blank sheet of paper in the questionnaire and the instructions given above. A nurse scored all questionnaires regardless of whether questionnaire administration was postal or by layperson or nurse. Because nurses with no special psychological knowledge rated the CDT in this study, a simple scoring system was used (15). The nurse evaluated three items: correctly drawn clock shape, all numbers in the correct position, and hands of the clock set to the correct time. A score of 1 was assigned for each of these items if the nurse judged that the drawing, the assignment of numbers, and the time were correct; thus, the score could range from 0 (all items incorrect) to 3 (all items correct). If information on any of these three items was missing, the patient was excluded from the analysis. In addition, the nurses were trained to make a judgment as to whether there were abnormalities in the CDT that were not captured by the simple pass/fail scoring described above. For instance, if the drawing was a very disorganized, bizarre, or otherwise abnormal representation of a clock, the nurse recorded the presence of bizarre drawings, and these were included in the final score with a score of 0 if present and 1 otherwise. Therefore, final possible scores ranged between 0 (the worst) and 4 (the best). Since there were no predefined cutoff points to indicate cognitive impairment, we compared different threshold scores on the CDT with MMSE scores as part of this study. For the MMSE, two cutoff points were used: 23 or less for mild cognitive impairment (16) and 17 or less for moderateto-severe impairment (17).
Covariates

8084, and 8589 years) and a fourth group of persons aged 90 years or more. Home ownership as a marker of the socioeconomic status of subjects was categorized as owner occupier or renter. Persons living in sheltered or residential placements were regarded as having missing data, since there was no information on their previous residence. Lifestyle variables. Smoking status was categorized as nonsmoker, ex-smoker, and current smoker. Alcohol drinking was divided into four categories: never drinker, ex-drinker, 06 units per week, and 7 units per week. In the United Kingdom, a unit of alcohol is defined as the amount of ethanol contained in half a pint (284 ml) of beer, a single glass (125 ml) of table wine, or a single measure (25 ml) of spirits (18). A unit is equivalent to approximately 10 ml or 8 g of ethanol. Health status indicators. Reported current or recent history of heart attack, stroke, Parkinsons disease, diabetes mellitus, or cancer was categorized as yes or no. This information was determined by asking patients whether they had ever been told of these conditions by physicians. Selfassessed general health was categorized into poor/fair, good, and very good/excellent. The presence of depression was determined through the use of the Geriatric Depression Scale (19), and subjects were classified as either normal (score of 05) or depressed (score of 6). Number of falls was categorized as none, one, and two or more.
Statistical analysis

Downloaded from http://aje.oxfordjournals.org/ by guest on February 5, 2013

Through the brief and detailed assessments in the trial, a broad range of patient information was collected. To assess the association between cognition and mortality, we considered the following potential confounders. Demographic and social variables. Age at brief assessment was categorized into three 5-year age groups (7579,
Am J Epidemiol 2004;160:797807

In all analyses, we took account of the stratified cluster sampling by general practice in the estimation of standard errors. All statistical analyses were performed using Stata 7 software (Stata Corporation, College Station, Texas). Since the participants in the targeted arm with detailed assessments were not a representative sample, only people in the universal arm who responded to both the brief assessment (which included the CDT) and the detailed assessment (which included the MMSE) were entered into the analyses. The sensitivity, specificity, positive predictive value, and negative predictive value of the CDT in comparison with the MMSE, using different CDT cutoff points, were calculated for the whole sample and by gender, age group, and method of administration. Survival time was defined as time from the date of baseline assessment (between 1995 and 1999) to the date of

800 Nishiwaki et al.

TABLE 2. Validity of the Clock-Drawing Test in comparison with the Mini-Mental State Examination,* MRC Trial of the Assessment and Management of Older People in the Community, United Kingdom, 19951999
Variable and cutoff point for the MMSE Sensitivity 95% CI Specificity 95% CI Positive predictive value 95% CI Negative predictive value 95% CI

Total (n = 13,557) 23 17 Gender Male (n = 5,277) 23 17 Female (n = 8,280) 23 17 Age group (years) 7579 (n = 6,596) 23 17 8084 (n = 4,249) 23 17 8589 (n = 2,061) 23 17 90 (n = 651) 23 17 Method of administration Postal questionnaire (n = 5,934) 23 17 Lay interview (n = 4,254) 23 17 Nurse interview (n = 3,369) 23 17 44.5 76.5 40.1, 49.0 68.0, 83.3 91.2 87.1 89.6, 92.6 85.1, 88.9 53.7 20.0 49.3, 58.2 15.5, 25.3 87.8 98.9 84.6, 90.3 98.2, 99.3 42.5 72.9 34.4, 51.0 60.2, 82.7 88.2 84.7 82.9, 92.0 79.4, 88.8 40.7 10.9 33.5, 48.3 7.4, 15.8 88.9 99.2 86.4, 91.0 98.8, 99.5 23.2 39.9 19.6, 27.2 29.9, 50.7 92.5 90.6 90.1, 94.4 88.3, 92.5 39.3 10.1 30.3, 49.2 6.3, 15.9 85.2 98.3 80.1, 89.2 97.2, 98.9 48.0 68.3 39.7, 56.5 54.6, 79.4 80.6 72.6 74.8, 85.4 66.5, 78.0 67.9 26.4 61.0, 74.2 20.2, 33.7 64.5 94.1 58.7, 69.8 91.2, 96.1 42.5 68.4 36.6, 48.7 56.1, 78.6 86.1 80.8 83.0, 88.6 77.4, 83.8 54.2 17.3 48.7, 59.7 13.2, 22.2 79.4 97.8 75.4, 83.0 96.6, 98.6 33.6 63.1 28.3, 39.4 50.6, 74.1 89.0 86.4 86.9, 90.8 84.2, 88.4 39.7 12.1 33.8, 45.9 8.9, 16.3 86.2 98.8 83.4, 88.5 98.2, 99.1 23.4 41.0 19.7, 27.4 29.2, 54.0 93.7 92.3 92.0, 95.1 90.4, 93.8 30.9 6.0 26.0, 36.3 4.3, 8.3 91.0 99.2 89.2, 92.6 98.9, 99.5 37.0 63.2 31.7, 42.7 52.7, 72.5 88.9 85.5 86.6, 90.8 83.0, 87.7 44.7 13.5 39.9, 49.6 10.7, 16.8 85.4 98.5 82.5, 87.9 97.9, 98.9 29.1 56.8 24.5, 34.1 44.8, 68.0 93.6 91.5 92.0, 94.9 89.8, 93.0 42.0 12.6 35.7, 48.6 9.1, 17.0 89.2 99.0 87.4, 90.8 98.6, 99.3 34.5 61.4 29.8, 39.6 51.9, 70.0 90.8 87.9 88.9, 92.4 85.8, 89.7 43.9 13.2 39.2, 48.8 10.7, 16.2 86.9 98.7 84.6, 89.0 98.3, 99.0

Downloaded from http://aje.oxfordjournals.org/ by guest on February 5, 2013

* The cutoff point for the Clock-Drawing Test was a score of 2 or less. MMSE, Mini-Mental State Examination; CI, confidence interval.

death, the date of emigration, or December 31, 2001, whichever was earlier. The Kaplan-Meier method was used to plot the unadjusted survivor function by level of baseline cognitive function as assessed by the CDT. To quantify the impact of cognitive function on mortality, we used Cox proportional hazards regression. Adjustment for potential confounders was accomplished via a series of nested models. The initial step included only age group, gender, and home ownership. The next step added lifestyle variables (model 1). The health status indicators described above were then added (model 2). Wald tests were used to determine the statistical significance

of variables, to check linearity for categorical variables (by giving consecutive integers to each category), and to evaluate interactions. It was confirmed that all variables included in model 2 were significant predictors of mortality and that there was no evidence of interaction between variables. On the basis of the test for linearity, age group, smoking category, self-assessed general health, and number of falls were included in the model in a linear manner. Because subjects with cognitive impairment might have had preceding diseases that caused both impairment and death, we repeated the analyses after excluding deaths occurring within the first
Am J Epidemiol 2004;160:797807

Use of the Clock-Drawing Test to Screen the Elderly 801

Downloaded from http://aje.oxfordjournals.org/ by guest on February 5, 2013

FIGURE 2. Receiver operating characteristic curves for the Clock-Drawing Test in comparison with the Mini-Mental State Examination (MMSE), by method of administration, MRC Trial of the Assessment and Management of Older People in the Community, United Kingdom, 19951999. The cutoff point for the MMSE score was set at 17 or less.

year of follow-up. The proportional hazards assumption was graphically confirmed using an Aalen-Nelson plot of cumulative rate on a log scale (20).
RESULTS Study population

Validity of the CDT

Of 21,241 participants in the universal arm, 123 (0.6 percent) were excluded because they were under 75 years of age or resided in a nursing home. Of the remaining 21,118 subjects, 14,707 (69.6 percent) who completed both the brief assessment and the detailed assessment were included. Males were more likely to respond than females (73.1 percent vs. 67.6 percent; p < 0.001). Responders were slightly younger than nonresponders (median age, 80.3 years vs. 81.1 years; p < 0.001). There were 869 and 430 subjects with missing data for the CDT and the MMSE, respectively (149 subjects had missing data for both tests); therefore, 13,557 subjects with complete data from both tests (64.2 percent of the 21,118) were eligible for analysis. Females, older people, nonhomeowners, and depressed subjects were more likely to be excluded. Brief assessments were done by postal questionnaire (19 practices), lay interview (17 practices), and nurse interview (17 practices) for 5,934 subjects (44 percent of the 13,557), 4,254 subjects (31 percent), and 3,369 subjects (25 percent), respectively.
Am J Epidemiol 2004;160:797807

CDT scores were distributed as shown in table 1. If the cutoff point was set at a score of 2 or less, the prevalence of cognitive impairment was 13.6 percent. The prevalence of moderate/severe cognitive impairment by the MMSE was 2.9 percent, and the prevalence of at least mild cognitive impairment was 17.3 percent. Sensitivity, specificity, positive predictive value, and negative predictive value according to different MMSE cutoff points are shown in table 2, where a cutoff point of 2 or less was used for the CDT score. The CDT scores showed higher sensitivity and lower specificity for more severe cognitive impairment than for milder scores. The positive predictive value was lower and the negative predictive value was higher for a cutoff point of 17 or less, reflecting the lower prevalence of moderate/severe cognitive impairment in comparison with milder impairment. Sensitivity tended to be higher in females than in males, and specificity was lower. Among subjects aged 8589 years and 90 years, higher sensitivities and lower specificities were observed when subjects were compared with those in other age groups. Regardless of cutoff points for the MMSE, the postal questionnaire had a lower sensitivity than face-to-face methods. In contrast, lay interviews and nurse interviews provided acceptable sensitivity and specificity if the cutoff point for the MMSE was set at 17 or less, as shown by the receiver operating characteristic curves graphed by administration method (figure 2). For nurse administration, sensitivity and specificity versus

802 Nishiwaki et al.

TABLE 3. Characteristics of the study population according to score on the Clock-Drawing Test, MRC Trial of the Assessment and Management of Older People in the Community, United Kingdom, 19951999
CDT score 2 (n = 1,839) No. Age group (years) at baseline* 7579 8084 8589 90 Gender* Male Female Home ownership* Owner occupier Renter Smoking* Nonsmoker Ex-smoker Current smoker Alcohol drinking* Never drinker Ex-drinker Current drinker (units/week) 06 7 Health status indicators Reported heart attack No Yes 1,630 198 89.2 10.8 10,401 1,266 89.2 10.9 1,016 215 57.1 12.1 7,200 1,970 62.5 17.1 412 136 23.2 7.6 1,686 664 14.6 5.8 845 797 195 46.0 43.4 10.6 4,446 6,124 1,130 38.0 52.3 9.7 915 622 59.5 40.5 7,768 3,095 71.5 28.5 502 1,337 27.3 72.7 4,775 6,943 40.8 59.3 537 638 452 212 29.2 34.7 24.6 11.5 6,059 3,611 1,609 439 51.7 30.8 13.7 3.8 % 3 or 4 (n = 11,718) No. %

TABLE 3. Continued
CDT score 2 (n = 1,839) No. Reported stroke* No Yes Reported diabetes mellitus* No Yes Reported Parkinsons disease* No Yes Reported cancer** No Yes Self-assessed general health* Poor/fair Good Very good/excellent Depression* Normal Depressed No. of falls* 0 1 2 1,279 272 284 69.7 14.8 15.5 9,490 1,388 805 81.2 11.9 6.9 1,538 234 86.8 13.2 10,690 792 93.1 6.9 423 685 717 23.2 37.5 39.3 1,635 4,269 5,765 14.0 36.6 49.4 1,663 163 91.1 8.9 10,416 1,221 89.5 10.5 1,796 36 98.0 2.0 11,530 115 99.0 1.0 1,686 152 91.7 8.3 10,977 730 93.8 6.2 1,617 214 88.3 11.7 10,746 930 92.0 8.0 % 3 or 4 (n = 11,718) No. %

Downloaded from http://aje.oxfordjournals.org/ by guest on February 5, 2013

Table continues

* p < 0.001 for design-adjusted chi-squared test; ** p = 0.06 for designadjusted chi-squared test. CDT, Clock-Drawing Test. Because of missing values, numbers of subjects do not always add up to 1,839. Because of missing values, numbers of subjects do not always add up to 11,718. A unit of alcohol is defined as the amount of ethanol contained in half a pint (284 ml) of beer, a single glass (125 ml) of table wine, or a single measure (25 ml) of spirits (18).

the MMSE using a cutoff point of 17 or less were 76.5 percent and 87.1 percent, respectively. If a cutoff point of 1 or less for the CDT was used, corresponding sensitivity and specificity were 46.3 percent and 96.2 percent, showing much lower sensitivity. For a CDT cutoff point of 3 or less, corresponding sensitivity and specificity were 92.7 percent and 68.1 percent, which means that lower specificity would produce many more false-positive cases.
Predictive validity of the CDT for mortality

Of the 13,557 subjects who completed both the CDT and the MMSE, 4,521 died and four were censored because of emigration during a median follow-up period of 4.7 years. Characteristics of the study subjects by CDT score are shown in table 3. All covariates except for reported heart attack were associated with cognitive impairment. The Kaplan-Meier survival curve showed a consistent reduction

in survival probability with a lower baseline level of cognitive function (figure 3). Cox regression revealed that the age-, gender-, and home ownership-adjusted hazard ratio for all-cause mortality was 1.77 (95 percent confidence interval (CI): 1.64, 1.91) among subjects who scored 2 or less on the CDT as compared with those with scores of 3 or more (table 4). Adjustment for lifestyle variables did not alter this hazard ratio (hazard ratio (HR) = 1.76, 95 percent CI: 1.63, 1.90). Further adjustment by health indicator variables reduced this association, but it still remained statistically significant (HR = 1.60, 95 percent CI: 1.48, 1.74). This relation was observed across all levels of cognitive function as determined by the CDT. Even mild cognitive impairment was associated with higher all-cause mortality. Exclusion of deaths occurring in the first year of follow-up altered the parameters little. Cognitive impairment assessed by the CDT was associated with mortality from cerebrovascular disease, respiratory
Am J Epidemiol 2004;160:797807

Use of the Clock-Drawing Test to Screen the Elderly 803

FIGURE 3. Kaplan-Meier unadjusted survival estimates by baseline level of cognitive function, as measured by score on the Clock-Drawing Test, MRC Trial of the Assessment and Management of Older People in the Community, United Kingdom, 19952001. The Clock-Drawing Test had a five-point score scale with 4 as the best score and 0 as the worst.

Downloaded from http://aje.oxfordjournals.org/ by guest on February 5, 2013

disease, and malignant neoplasms but not with coronary heart disease mortality (table 5). Adjusted hazard ratios from model 2 are shown in the table. Subjects who scored 2 or less were 1.7 times and 1.9 times more likely to die from cerebrovascular disease and respiratory disease, respectively, than those who scored 3 or 4. These relations were observed across all levels of cognitive function but not for malignant neoplasms (data not shown). Exclusion of deaths occurring in the first year of follow-up produced similar results.
DISCUSSION

To our knowledge, this is the largest study to have assessed the validity of the CDT in a representative community sample. For nurse-administered interviews, sensitivity (76.5 percent) and specificity (87.1 percent) for moderate/severe cognitive impairment were fairly acceptable. This is comparable to other currently accepted population screening tools, such as mammography for breast cancer, with sensitivities in the range of 7590 percent and specificities of 9095 percent (21). However, on the whole, the sensitivity and specificity of the CDT for mild impairment were low, suggesting that the CDT was better at detecting moderate/severe cognitive impairment. If we apply these figures to a hypothetical population of 1,000 people aged 75 years or older, assuming that the prevalence of moderate/severe cognitive impairment is the same as in this study population (2.9 percent), a sensitivity of 77 percent would result in 22 cases being detected and seven people being missed. Clearly, some people with MMSE scores of 17 or less may well have already come to the attention of health-care providers, so this is an upper estimate of new cases. It should also be noted that even this relatively high specificity (87 percent) would yield many false-positive cases. In the above example, 125 falsepositive cases would be produced. Since mild cognitive impairment may be a sign of early dementia, screening for mild impairment would become
Am J Epidemiol 2004;160:797807

more important as new medicines for early treatment became available. However, the CDT alone did not appear to be a good screening tool for mild cognitive impairment, and a two-step strategy (the CDT followed by the MMSE in persons scoring positive) might lead to improved validity. Since there was no clinical measure of dementia for use as a gold standard in this study, it was impossible to assess the validity of the combination of the CDT and the MMSE for detecting early dementia. However, we were able to investigate the effects of both scores on mortality. Hazard ratios adjusted for demographic, social, and lifestyle variables and health status indicators were higher for the combined MMSE and CDT (HR = 2.22, 95 percent CI: 1.85, 2.66) than for the CDT alone (HR = 1.55, 95 percent CI: 1.42, 1.70) and the MMSE alone (HR = 1.74, 95 percent CI: 1.26, 2.39). The postal questionnaire showed lower sensitivity than the nurse interview, though gender and age distributions were comparable among administration groups reflecting random allocation. It is unlikely that this reflected any bias due to the fact that the nurse both scored the CDT and administered the test, because similar results were observed for administration by the lay interviewer, which was also scored by the nurse. Response bias might also have affected the estimates of validity by method of administration. Of subjects who participated in the brief assessments, 9.6 percent of the postal administration group did not respond to the CDT, as compared with 4.5 percent for nurse administration and 3.9 percent for lay administration. Women, older people, nonhomeowners, and depressed subjects were less likely to complete the CDT. These factors were associated with worse MMSE scores. If cognitively impaired patients, who could be easily detected by both the CDT and the MMSE, were more likely to refuse the CDT, sensitivity would be underestimated. Indeed, the prevalence of subjects being assessed as cognitively impaired by the CDT was lower for postal administration (10.2 percent, 95 percent CI: 8.5, 12.2) than for lay administration (16.8 percent, 95 percent CI: 12.8, 21.7) and nurse

804 Nishiwaki et al.

TABLE 4. Association between baseline cognitive function as assessed by the Clock-Drawing Test and all-cause mortality, MRC Trial of the Assessment and Management of Older People in the Community, United Kingdom, 19952001*
No. of subjects No. of deaths Adjustment for age, gender, and home ownership HR 95% CI HR Model 1 95% CI HR Model 2 95% CI

CDT score

All subjects 3 or 4 0, 1, or 2 4 3 2 1 0 Exclusion of subjects who died within 1 year of baseline assessment 3 or 4 0, 1, or 2 4 3 2 1 0 9,821 1,299 7,950 1,871 765 477 57 2,543 588 1,964 579 320 239 29 1.00 1.80 1.00 1.21 1.81 1.98 2.09 1.10, 1.33 1.62, 2.03 1.76, 2.23 1.55, 2.82 1.66, 1.96 1.00 1.79 1.00 1.21 1.81 1.96 2.07 1.10, 1.33 1.61, 2.02 1.72, 2.22 1.52, 2.81 1.64, 1.95 1.00 1.65 1.00 1.19 1.68 1.77 1.92 1.08, 1.31 1.49, 1.89 1.57, 2.00 1.33, 2.76 1.52, 1.80 10,278 1,416 8,294 1,984 835 518 63 3,000 705 2,308 692 390 280 35 1.00 1.77 1.00 1.23 1.83 1.90 2.00 1.13, 1.34 1.64, 2.03 1.70, 2.14 1.49, 2.69 1.64, 1.91 1.00 1.76 1.00 1.22 1.81 1.88 1.98 1.12, 1.33 1.63, 2.01 1.67, 2.13 1.45, 2.70 1.63, 1.90 1.00 1.60 1.00 1.20 1.67 1.68 1.80 1.09, 1.31 1.49, 1.86 1.49, 1.89 1.25, 2.58 1.48, 1.74

Downloaded from http://aje.oxfordjournals.org/ by guest on February 5, 2013

* Figures and analyses were restricted to the subset of 11,694 subjects who had complete data for all covariates. CDT, Clock-Drawing Test; HR, hazard ratio; CI, confidence interval. Model 1 included smoking and alcohol drinking in addition to age, gender, and home ownership as covariates. Model 2 included depression, reported current/past diseases (heart attack, stroke, diabetes mellitus, Parkinsons disease, and cancer), selfassessed general health, and number of falls in addition to the covariates included in model 1.

administration (15.5 percent, 95 percent CI: 13.3, 17.9), which suggests that differences in response rates might partly account for the lower sensitivity for postal administration. Additionally, for postal administration, family

members might have assisted the elderly person, and this would have resulted in more false-negatives. This suggests that, for the CDT as a population screening tool, face-toface administration may be preferable to postal administra-

TABLE 5. Association between cognitive impairment as assessed by the Clock-Drawing Test and causespecific mortality, MRC Trial of the Assessment and Management of Older People in the Community, United Kingdom, 19952001*,
All subjects Cause of death Adjusted HR, 95% CI Exclusion of subjects who died within 1 year of baseline assessment Adjusted HR 95% CI

Coronary heart disease (n = 601) Cerebrovascular disease (n = 511) Respiratory disease (n = 624) Malignant neoplasm (n = 791)

1.18 1.73 1.91 1.33

0.92, 1.52 1.39, 2.15 1.58, 2.31 1.05, 1.67

1.25 1.59 1.98 1.43

0.98, 1.59 1.22, 2.08 1.64, 2.40 1.13, 1.81

* Figures and analyses were restricted to the subset of 11,694 subjects who had complete data for all covariates. Cognitive impairment was assessed using a cutoff point of 2 or less for the Clock-Drawing Test score. HR, hazard ratio; CI, confidence interval. Adjusted for age, gender, home ownership, smoking, alcohol drinking, depression, reported current/past diseases (heart attack, stroke, diabetes mellitus, Parkinsons disease, and cancer), self-assessed general health, and number of falls.

Am J Epidemiol 2004;160:797807

Use of the Clock-Drawing Test to Screen the Elderly 805

tion. The validity also varied by age group; this variation was independent of gender or administration methods (data not shown). The reason for this interesting variation is unclear, and additional studies using the same scoring and cutoff points in other populations are needed. In contrast to almost all previous studies, the standard used in this study was the MMSE rather than clinical diagnosis of dementia, because an extensive clinical evaluation of over 13,000 elderly patients was impractical. Although the MMSE is a tool that is commonly used to detect cognitive impairment, it is not a diagnostic test. Cognitive function detected by the CDT might be different from that detected by the MMSE (5). Some studies found only moderate correlations between the CDT and the MMSE in clinical settings (7, 22). Evaluation of the validity of the CDT against a clinical diagnosis of dementia as a gold standard is still needed. The educational level and cultural background of participants may influence the validity of the CDT (9). There were very few older participants from minority ethnic groups in our study, reflecting the very low proportion of older people of minority ethnicity currently in the British population. We did not ask about educational level. Housing tenure, which might be considered a weak proxy measure of educational attainment, was associated with the CDT score in unadjusted analyses and also after adjustment for other confounders. However, the validity of the CDT varied little by housing tenure. For nurse administration, sensitivity and specificity were 75 percent and 90 percent, respectively, for people who owned their homes and 71 percent and 84 percent, respectively, for those in rental accommodations. We did not consider the visual acuity of the participants in evaluating CDT performance. Exclusion of those with binocular presenting acuity less than 6/18 (the conventional definition of vision impairment) reduced sensitivity and increased specificity slightly; however, this was entirely due to the effect of age. People with poor visual acuity were older, and the CDT was less sensitive and specific among older people. The role of the CDT as a screening tool has been evaluated in previous studies (6, 9, 2325). In a review article, Shulman (9) reported that most studies achieved sensitivities and specificities of approximately 85 percent and concluded that the CDT, in conjunction with other widely used tests such as the MMSE, could provide a significant advance in the early detection of dementia. In contrast, Powlishta et al. (24) concluded from their study that the CDT did not appear to be a useful screening tool for detecting very mild dementia. Other authors have concluded that the CDT should not be used alone as a dementia screening test because of its overall inadequate performance (25, 26). However, most previous studies were based on relatively small sample sizes or were undertaken in a clinical setting, and their results may not be applicable to a larger community population. Paganini-Hill et al. (27) conducted a large postal questionnaire survey including the CDT among 5,000 community-dwelling elderly persons and concluded that the CDT might help predict cognitive frailty and future disability. However, the validity of the CDT was not examined in their study.
Am J Epidemiol 2004;160:797807

There are various scoring systems for the CDT and no universally accepted standards. More complex scoring systems require greater rating skill and increase training time. In our study, because nurses with no special training in the detailed interpretation of the test were to be used, we adopted a simple scoring method previously used in the CAMCOG battery (the cognitive portion of the Cambridge Examination for Mental Disorders of the Elderly) (15, 28) that we considered suitable for mass screening of the elderly. Nurses were trained to recognize drawings as bizarre/disorganized using examples or a general impression, as was also done in other studies (4, 7, 8). Recently, Scanlan et al. (29) found that a simple binary rating (normal or abnormal) made by naive raters was surprisingly effective in detecting dementia. The association of MMSE scores with survival has previously been reported (3033). This study provided the first opportunity to assess the predictive validity of the CDT with regard to mortality. The result revealed that cognitive impairment assessed by the CDT was associated with higher mortality across all levels of cognitive function. Though the mechanism for the association between cognitive impairment and mortality has not been clear, there is a hypothesis that this relation could be explained by the preceding condition of patients. Although inclusion of health indicators such as current/past diseases and selfassessed general health in the model did reduce this association, it still remained significant. Moreover, exclusion of subjects who died within 1 year of the baseline assessment did not alter the result. Cognitive impairment seemed to be an independent risk factor for mortality in this study population, though residual confounding or confounding by other factors not included in this study should be taken into consideration. Cognitive impairment was associated with risk of death from cerebrovascular disease, respiratory disease, and malignant neoplasms. The increase in cerebrovascular disease mortality suggests that a proportion of cognitive impairment in this study population was vascular in origin. Increased mortality from respiratory disease was consistent with the findings of another study (34). Although the pathway is unclear, death due to respiratory disease has already been found to be a common cause of death among persons with Alzheimers disease (35). In contrast to the case with cerebrovascular and respiratory diseases, the association with cancer death was not observed across all levels of cognitive function. Therefore, this could be a chance finding. This analysis was based on 64 percent of eligible subjects in the universal arm. Because there were missing values for some covariates, multivariate analyses for the association between the CDT and mortality were restricted to a subset of 11,694 subjects. Unless the association between cognition and mortality was different among subjects who were excluded from the analysis, this would not have introduced selection bias. One drawback of the present study was that data on all possible confounders except age and gender were self-reported. Since cognitively impaired subjects could have memory disturbances, recall among those subjects may have been less reliable. If subjects with cognitive impair-

Downloaded from http://aje.oxfordjournals.org/ by guest on February 5, 2013

806 Nishiwaki et al.

ment were less likely to recall a history of stroke, the prevalence of reported stroke would have been underestimated among subjects with cognitive impairment; this would have led to overestimation of the hazard ratio. We found that the CDT is a quick and easily completed face-to-face screening tool for detecting moderate/severe cognitive impairment in a community elderly population. The CDT had acceptable validity in comparison with the MMSE using simple scoring by a nurse without special knowledge. However, the standard used in this study was not clinical diagnosis. Considering the many false-positive cases generated by the CDT, we caution against using the CDT as a stand-alone instrument to screen for cognitive impairment.

ACKNOWLEDGMENTS

The MRC Trial of the Assessment and Management of Older People in the Community was funded by the Medical Research Council, the United Kingdom Department of Health, and the Scotland Office. Dr. Yuji Nishiwaki was supported by the Uehara Memorial Foundation (Tokyo, Japan). The authors thank the nurses, general practitioners, and other staff in the participating medical practices and Judith Nickson at the University of Cambridge for training nurses in the administration of the Clock-Drawing Test and the Mini-Mental State Examination.

REFERENCES 1. Ritchie K, Lovestone S. The dementias. Lancet 2002;360: 175966. 2. Dewey ME, Saz P. Dementia, cognitive impairment and mortality in persons aged 65 and over living in the community: a systematic review of the literature. Int J Geriatr Psychiatr 2001;16:75161. 3. Folstein MF, Folstein SE, McHugh PR. Mini-mental state: a practical method for grading the cognitive state of patients for the clinician. J Psychiatr Res 1975;12:18998. 4. Death J, Douglas A, Kenny RA. Comparison of clock drawing with Mini-Mental State Examination as a screening test in elderly acute hospital admissions. Postgrad Med J 1993;69: 696700. 5. Juby A, Tench S, Baker V. The value of clock drawing in identifying executive cognitive dysfunction in people with a normal Mini-Mental State Examination score. CMAJ 2002; 167:85964. 6. Kirby M, Denihan A, Bruce I, et al. The Clock Drawing Test in primary care: sensitivity in dementia detection and specificity against normal and depressed elderly. Int J Geriatr Psychiatr 2001;16:93540. 7. Mendez MF, Ala T, Underwood KL. Development of scoring criteria for the clock drawing task in Alzheimers disease. J Am Geriatr Soc 1992;40:10959. 8. Shulman KI, Gold DP, Cohen CA, et al. Clock-drawing and dementia in the community: a longitudinal study. Int J Geriatr Psychiatr 1993;8:48796. 9. Shulman KI. Clock-drawing: is it the ideal cognitive screen-

ing test? Int J Geriatr Psychiatr 2000;15:54861. 10. Sunderland T, Hill JL, Mellow AM, et al. Clock drawing in Alzheimers disease: a novel measure of dementia severity. J Am Geriatr Soc 1989;37:7259. 11. Tuokko H, Hadjistavropoulos T, Miller JA, et al. The Clock Test: a sensitive measure to differentiate normal elderly from those with Alzheimer disease. J Am Geriatr Soc 1992;40:579 84. 12. Wolf-Klein GP, Silverstone FA, Levy AP, et al. Screening for Alzheimers disease by clock drawing. J Am Geriatr Soc 1989;37:7304. 13. Fletcher AE, Jones DA, Bulpitt CJ, et al. The MRC Trial of Assessment and Management of Older People in the Community: objectives, design and interventions [ISRCTN23494848]. (Electronic article). BMC Health Serv Res 2002;2:21. 14. Smeeth L, Fletcher AE, Stirling S, et al. Randomised comparison of three methods of administering a screening questionnaire to elderly people: findings from the MRC Trial of the Assessment and Management of Older People in the Community. BMJ 2001;323:14037. 15. Roth M, Huppert FA, Tym E, et al. CAMDEX: The Cambridge Examination for Mental Disorders of the Elderly. Cambridge, United Kingdom: Cambridge University Press, 1988. 16. OConnor DW, Pollitt PA, Hyde JB, et al. Do general practitioners miss dementia in elderly patients? BMJ 1988;297: 110710. 17. Hofman A, Rocca WA, Brayne C, et al. The prevalence of dementia in Europe: a collaborative study of 19801990 findings. Eurodem Prevalence Research Group. Int J Epidemiol 1991;20:73648. 18. Health Development Agency, National Health Service. Prevention and reduction of alcohol misuse. London, United Kingdom: Health Development Agency, 2002. 19. Almeida OP, Almeida SA. Short versions of the Geriatric Depression Scale: a study of their validity for the diagnosis of a major depressive episode according to ICD-10 and DSMIV. Int J Geriatr Psychiatr 1999;14:85865. 20. Clayton D, Hills M. Statistical models in epidemiology. Oxford, United Kingdom: Oxford University Press, 1993. 21. Ferrini R, Mannino E, Ramsdell E, et al. Screening mammography for breast cancer: American College of Preventive Medicine practice policy statement. Am J Prev Med 1996;12: 3401. 22. Manos PJ, Wu R. The ten point clock test: a quick screen and grading method for cognitive impairment in medical and surgical patients. Int J Psychiatry Med 1994;24:22944. 23. Siu AL. Screening for dementia and investigating its causes. Ann Intern Med 1991;115:12232. 24. Powlishta KK, Von Dras DD, Stanford A, et al. The Clock Drawing Test is a poor screen for very mild dementia. Neurology 2002;59:898903. 25. Borson S, Brush M, Gil E, et al. The Clock Drawing Test: utility for dementia detection in multiethnic elders. J Gerontol A Biol Sci Med Sci 1999;54:M53440. 26. Storey JE, Rowland JT, Basic D, et al. A comparison of five clock scoring methods using ROC (receiver operating characteristic) curve analysis. Int J Geriatr Psychiatr 2001;16:3949. 27. Paganini-Hill A, Clark LJ, Henderson VW, et al. Clock drawing: analysis in a retirement community. J Am Geriatr Soc 2001;49:9417. 28. Roth M, Tym E, Mountjoy CQ, et al. CAMDEX: a standardised instrument for the diagnosis of mental disorder in the elderly with special reference to the early detection of dementia. Br J Psychiatry 1986;149:698709. 29. Scanlan JM, Brush M, Quijano C, et al. Comparing clock tests

Downloaded from http://aje.oxfordjournals.org/ by guest on February 5, 2013

Am J Epidemiol 2004;160:797807

Use of the Clock-Drawing Test to Screen the Elderly 807

for dementia screening: naive judgments vs formal systems what is optimal? Int J Geriatr Psychiatr 2002;17:1421. 30. Neale R, Brayne C, Johnson AL. Cognition and survival: an exploration in a large multicentre study of the population aged 65 years and over. Int J Epidemiol 2001;30:13838. 31. Smits CH, Deeg DJ, Kriegsman DM, et al. Cognitive functioning and health as determinants of mortality in an older population. Am J Epidemiol 1999;150:97886. 32. Bassuk SS, Wypij D, Berkman LF. Cognitive impairment and mortality in the community-dwelling elderly. Am J Epidemiol

2000;151:67688. 33. Gussekloo J, Westendorp RG, Remarque EJ, et al. Impact of mild cognitive impairment on survival in very elderly people: cohort study. BMJ 1997;315:10534. 34. Helmer C, Joly P, Letenneur L, et al. Mortality with dementia: results from a French prospective community-based cohort. Am J Epidemiol 2001;154:6428. 35. Burns A, Jacoby R, Luthert P, et al. Cause of death in Alzheimers disease. Age Ageing 1990;19:3414.

Downloaded from http://aje.oxfordjournals.org/ by guest on February 5, 2013

Am J Epidemiol 2004;160:797807