CERAD test performances in amnestic mildcognitive impairment and Alzheimer’sdisease

Patients with mild cognitive impairment (MCI)have been shown to be at risk for developingAlzheimer’s disease (AD) and other neurodegen-erative diseases (1, 2). In longitudinal studies ofAD and MCI, subjects showed decline in one orseveral cognitive domains 5–10 years before theclinical diagnosis of AD (3–6). There has beensome controversy in the definition and terminologyused to describe elderly subjects showing cognitivedecline in the absence of dementia. Due to differentdefinitions and classification criteria used [e.g. age-

associated memory impairment (AAMI) (7), cog-nitive impairment, no dementia (CIND) (8)] dif-ferent studies have incorporated varying patientpopulations, which has resulted in conflictingconversion rates and prognoses. In those studieswhere the MCI criteria have been more stringentlyused to refer to only those elderly subjects withsubjective and objective memory impairment butno dementia [amnestic MCI (9)], the results havebeen less controversial. Approximately half of thesubjects with isolated episodic memory impairment

Acta Neurol Scand 2005: 111: 172–179 DOI: 10.1111/j.1600-0404.2005.00380.x Copyright � Blackwell Munksgaard 2005

ACTA NEUROLOGICASCANDINAVICA

Karrasch M, Sinerva E, Gronholm P, Rinne J, Laine M. CERAD testperformances in amnestic mild cognitive impairment and Alzheimer’sdisease.Acta Neurol Scand 2005: 111: 172–179. � Blackwell Munksgaard 2005.

Objectives – The aim of the study was to examine the Consortium toEstablish a Registry for Alzheimer’s Disease (CERAD) testperformances cross-sectionally in patients suffering from amnestic mildcognitive impairment (MCI) and mild Alzheimer’s disease (AD).Moreover, we wanted to determine the sensitivity to amnestic MCI andmild AD, as well as the specificity of different CERAD subtests in ourstudy groups. Material and methods – Fifteen healthy elderlyindividuals, 15 amnestic MCI patients and 15 probable AD patientssuffering from mild dementia were tested with the CERADneurocognitive dementia screening test. Results – Significantdifferences were found in all CERAD tests except Constructionalpraxis (copy) and Clock drawing between the controls and the ADgroup. The MCI group was differentiated from the controls only in theWordlist learning test. In the language tests the sensitivity to MCI andAD was quite low and the specificity very high. In the savings scoresthe sensitivity to AD was high, but the specificity rather low. TheWordlist recognition test screened no false positives using the currentcut-off score and the sensitivity to AD was 0.6, but only one MCIpatient was detected using the current cut-off score. Raising the cut-offscore also raised the sensitivity to MCI without dramatic loss ofspecificity. Cut-off scores for the Wordlist learning test and Wordlistdelayed recall, which have been found to differentiate normal agingfrom dementia, are lacking in the Finnish CERAD. The current dataindicates that the Wordlist learning test might be relatively sensitive toMCI. Conclusions – The results indicate that the Finnish CERAD testbattery with its current cut-off scores has low sensitivity to MCI, andusing it as a sole cognitive screening instrument for MCI andpreclinical dementia might result in false negatives.

M. Karrasch1,2, E. Sinerv�2,3,P. Grçnholm1, J. Rinne4,M. Laine1,21Department of Psychology, �bo Akademi University,�bo, Finland; 2Centre for Cognitive Neuroscience,University of Turku, Turku, Finland; 3Department ofPsychology, University of Turku, Turku, Finland;4National PET-centre, University of Turku, Turku, Finland

Key words: aging; dementia; Alzheimer's disease; mildcognitive impairment; amnestic mild cognitiveimpairment; CERAD; dementia screening

Mira Karrasch, Department of Psychology, �bo AkademiUniversity, FIN-20500, �bo, FinlandTel.: +358-2-2154405Fax: +358-2-2154833e-mail: mira.karrasch@abo.fi

Accepted for publication November 16, 2004

172

developed AD in a 3–4-year follow-up time period(10–13). During a 9.5-year follow-up 60% of theinitially memory impaired subjects developed AD(6). The problem remains, however, that theamnestic MCI group seems to include healthyindividuals who might have long-standing poormemory functions and possibly patients havingother neurogenerative diseases.The diagnostic criteria for amnestic MCI include

subjective memory complaints, an objectivememory deficit compared with age-appropriatenorms, normal general cognitive function, normalactivities of daily living (ADL) and an absence ofdementia (9, 14, 15). Petersen et al. (16) found thatMCI patients differed from the control group inmemory measures, but other cognitive functionswere comparable. The performance of the MCIpatients was similar to that of the AD patients inmemory tests, but the AD patients were impairedin other areas as well. Especially delayed recallperformance and ability to benefit from cuesduring memory encoding beyond the immediatespan were compromised in the MCI group ascompared with normal elderly controls.Several studies have confirmed that impairment

of acquisition and consolidation of episodicmemory traces are the earliest signs of the under-lying degenerative process of AD (17, 18). Thismemory impairment is caused by the degenerativeprocesses in the medial temporal lobes, particularlyin the entorhinal cortex and the hippocampus(18–20). In the majority of both cross-sectional andlongitudinal studies, tests tapping verbal episodiclearning and recall (e.g. WMS Logical memory,WMS Verbal paired associates, Selective Remind-ing Test) have been shown to reveal early cognitivechanges in patients who will later develop AD(14, 21). Moreover, in a 6-year follow-up Backmanet al. (4) found that measures of verbal recall andrecognition were equally sensitive to preclinicalAD. The second most sensitive tests are thosemeasuring visual episodic memory (e.g. FuldObject Memory Evaluation) (14, 22). Semanticmemory impairment and naming problems (meas-ured, e.g. by the Boston Naming Test, wordfluency tests, WAIS Similarities subtest), on theother hand, usually appear at a later stage wherethe clinical criteria for dementia are fulfilled(14, 23, 24).The importance of neuropsychological evalua-

tion in the diagnosis and follow-up of patients withMCI and AD is well established (3, 25, 26). Studieshave confirmed that at the time clinical changes areevident and the diagnosis is made, patients havealready performed poorly in neuropsychologicaltests for many years (2, 27, 28). Sharper decline in

performance, although, seems to occur simulta-neously with the first clinical signs of dementia(29). Adequate diagnostic accuracy (85–90%) canbe reached using neuropsychological tests (2, 27,28, 30). Neuropsychological tests are sensitive forvery early and subtle cognitive changes (14, 21, 22)and are cost-effective.Neuropsychological assessments are, however,

time-consuming and may not be available to a vastnumber of patients. Efforts have been made todevelop cognitive screening tools that would beeasy to administer by health care personnel,minimally time-consuming, as well as having highsensitivity and specificity to AD. The Consortiumto Establish a Registry for Alzheimer’s Disease(CERAD) test battery (31, 32) includes measures inthe areas where impairments associated with ADfirst occur (verbal memory, naming). It has beenfound to have high re-test reliability, inter-rateragreement and longitudinal validity (33). A recentstudy (34) showed that CERAD test performancesin a sample of healthy elderly (58–85 years) sub-jects were unaffected by age. Delayed recall of awordlist and savings scores (i.e. delayed recalladjusted for acquisition) on the CERAD are wellpreserved in normal aging but impaired early on indementia, which is important for the detection ofearly impairment in cognitive function (31, 35).Greene et al. (35) have also found that bothcontrols and minimally impaired AD patientsbenefited from repeated trials on the wordlistlearning task, but the mild AD patients benefitedonly little. None of the CERAD memory measureshave been found to be very effective in staging theseverity of dementia beyond the mild stage (26, 36).The normative studies on CERAD have notincluded MCI patients and thus it is unknown asto which subtests show decline in MCI.As far as we are aware, no studies have been

reported on the cognitive performances of MCIpatients on the CERAD test battery although thetest has been recommended to be used as ascreening tool for people at risk of developingdementing diseases. It is thus of interest to deter-mine how well the CERAD subtests identify thecognitive features of amnestic MCI.With regard to clinical practice it is important to

note that in the Finnish CERAD cut-off scores forWordlist learning and Wordlist delayed recall arelacking. This might limit its use in detecting earlyphases of dementing processes. Thus we wanted toexplore its sensitivity to MCI and AD, as well asthe specificity of the CERAD subtests usingdifferent cut-off scores. Based on earlier findings(26, 35–37) it was hypothesized that the sensitivityto MCI would be high in Wordlist savings and

CERAD test performances in amnestic MCI and AD

173

Wordlist delayed recall. We also expected to find atleast moderate sensitivity to MCI in both Wordlistlearning and Wordlist recognition (4) once anoptimal cut-off score had been set. The sensitivityto MCI and mild AD, as well as the specificity ofthose tests where cut-off scores based on Finnishnormative data are available (verbal fluency,naming, MMSE, Wordlist learning savings, Word-list recognition, constructional praxis savings,clock drawing) was also studied. Earlier findingshave indicated that preclinical AD patients havelowered scores on recognition memory tests (4),which is why we also studied the effect of differentcut-off scores on the sensitivity and specificity ofthe Wordlist recognition test.

Materials and methods

Subjects

Fifty-four subjects were recruited in the study.They consisted of 22 controls, 17 MCI patients and15 probable AD patients. The control subjectswere recruited from various community sources(local organizations for retired people, patients�relatives). None of the volunteered control subjectsreported any neurological or psychiatric diseases.Subjects with head traumas, symptoms of depres-sion or subjective cognitive impairment were notincluded the study. A neuropsychological assess-ment objectively confirmed age-appropriate cogni-tive functioning in all controls. Tests used in theneuropsychological assessment were WMS-R,WAIS-R (Digit Span, Similarities, Block Design,Digit Symbol), Trail Making Test A + B, BentonVisual Retention Test, and Boston Naming Test.In order to match the controls for age and years ofeducation with the patients at the group level,seven normal subjects were retrospectively exclu-ded from the analysis. The excluded subjects wereeither younger or had more years of formaleducation compared with the respective meanvalues in the patient groups. The final number ofsubjects in the control group was 15 (11 females,four males). Seventeen subjects suffering fromsubjective MCI entered the study. Two subjectswere excluded from the study because of psychi-atric illness (depression, n ¼ 1; psychosis in theyouth, n ¼ 1). The remaining 15 subjects (ninefemales, six males) were referred to the study by aneurologist. Neurological findings for thesepatients did not meet with the NINCDS-ADRDA (38) criteria for probable AD. Therewere no other neurological or psychiatric disordersexplaining the subjective memory complaint inthese patients. Based on the neurological examina-

tion and a neuropsychological assessment (thesame test battery as for the controls) they all metthe criteria for amnestic MCI (9). Fifteen patients(11 females, four males) meeting the NINCDS-ADRDA criteria for probable AD were referred tothe study by a neurologist. Five of the patients hadstarted on pharmacological treatment for ADbefore entering the study (two Reminyl, threeAricept). The MCI and AD patients were consecu-tive patients to a clinic. There were no statisticallysignificant differences between the three groups inage or years of education (means and standarddeviation are shown in Table 1). Retrospectivelythere were also no significant differences betweenthe medicated and unmedicated AD patients in anyCERAD variable (analyzed by Mann–WhitneyU-tests).

Procedure

The study was conducted at the Centre forCognitive Neuroscience, University of Turku.The joined ethical committee of the Turku Uni-versity and the Turku University Central Hospitalapproved the study. Informed consent for partici-pation was obtained from every subject. TheCERAD (31, 32) test procedure included a semi-structured interview of cognitive and affectivecomplaints with the patient and with a possibleinformant. It also included a short medical history.The MCI and AD patients were tested withCERAD within a month following the clinicalevaluation by a neurologist.

Data analyses

MANOVA was used to study overall differencesbetween the three groups on the CERAD meas-ures. Subsequent one-way ANOVAs were used tostudy differences between the three groups in thefollowing CERAD subtests: Verbal fluency,Naming, MMSE, Wordlist learning (sum of threetrials), Wordlist delayed recall, Wordlist savings,Wordlist recognition, Constructional praxis(copy), Constructional praxis (delayed recall),

Table 1 Age and years of education in the three groups studied

Age Years of education*

Mean SD Mean SD

Controls (n ¼ 15) 68.1 3.9 9.8 4.0MCI (n ¼ 15) 67.5 9.2 8.2 2.1AD (n ¼ 15) 71.9 5.8 8.9 3.3

* Due to the Finnish educational system most of the people going to school during1920s to 1950s have only completed a basic grammar school of 6–8 years and theratio of those having a higher education in these age cohorts is rather small.

Karrasch et al.

174

Constructional praxis (savings) and Clock draw-ing. Repeated measures ANOVAs was used tostudy differences between the groups in Wordlistlearning (trials 1, 2 and 3). Pairwise group differ-ences were analyzed by the Tukey post hoc test.Because several statistical comparisons were per-formed, a Bonferroni-corrected a level was usedboth in the one-way ANOVAs (P < 0.005) and inthe rmANOVA (P < 0.01), as well as in the posthoc tests. Sensitivity [correct positives/(correctpositives + false negatives)] to MCI and AD,and specificity [correct negatives/(correct negat-ives + false positives)] using the current cut-offscores based on Finnish normative data (32) werecalculated for the following tests: Verbal fluency,Naming, MMSE, Wordlist savings, Wordlistrecognition, Constructional praxis savings andClock drawing. The sensitivity and specificity ofthe Wordlist recognition test were also explored inthe cut-off range 81–95, and the optimal cut-offscores will be reported. In the Finnish CERAD nocut-off scores have been set in the Wordlistlearning and delayed recall tests, and thus thesensitivity and specificity of these tests wereexplored using different cut-off scores (16–20 forthe Wordlist learning test and 5–8 for the Wordlistdelayed recall test). The optimal cut-off scores forthese subtests will be reported.

Results

MANOVA revealed a significant main effect ofgroup on the combined CERAD variable including11 subtest measures (F(22,64) ¼ 2.790, P ¼ 0.001,Wilks� Lambda ¼ 0.261; partial eta squared ¼0.490). One-way ANOVA revealed significantmain effects for group in the following CERADsubtests: Verbal fluency (F(2,42) ¼ 7.343, P <0.005), Naming (F(2,42) ¼ 7.489, P < 0.005),MMSE (F(2,42) ¼ 13.221, P < 0.005), Wordlistlearning (sum of three trials) (F(2,42) ¼ 209.400,

P < 0.005), Wordlist learning (delayed recall)(F(2,42) ¼ 19.362, P < 0.005), Wordlist savings(F(2,42) ¼ 15.262, P < 0.005), Wordlist recogni-tion (F(2,42) ¼ 22.628, P < 0.005), Construc-tional praxis (delayed recall) (F(2,42) ¼ 10.585,P < 0.005) and Constructional praxis (savings)(F(2,42) ¼ 6.869, P < 0.005). Means, standarddeviations and ranges of the test performances ofthe three groups, as well as statistically significantdifferences between the groups are shown inTable 2. Statistically significant pairwise differ-ences as examined by Tukey’s post hoc tests werefound between the control group and the ADgroup, with the AD patients performing signifi-cantly worse in all tests except Constructionalpraxis (copy) and Clock drawing. In MMSE,Wordlist learning (delayed recall), Wordlist learn-ing (savings) and Wordlist learning (recognition)there were also statistically significant differencesbetween the MCI and the AD group, with the ADpatients performing worse than the MCI patients.A significant main effect of Wordlist learning

trial (F(2,84) ¼ 78.641, P ¼ 0.000, partial etasquared ¼ 0.652) was found in the repeated meas-ures ANOVA, which reflected the fact that thenumber of remembered items increased in allsubjects during the three trials. There was asignificant main effect for the factor group(F(2,42) ¼ 13.489, P ¼ 0.000, partial etasquared ¼ 0.391) in the Wordlist learning test,indicating that there were significant differences inthe average amount of items learned between thethree groups. Post hoc analyses showed significantpairwise differences between the controls and theAD group (P < 0.01), as well as between thecontrols and the MCI group (P < 0.01), but notbetween the MCI and the AD group, indicatingthat the Wordlist learning of the MCI patients wasalmost as poor as that of the AD group. Theinteraction term was non-significant (F(4,84) ¼0.994, P ¼ 0.443, partial eta squared ¼ 0.043)

Table 2 Means, standard deviations and ranges on CERAD tests in the three groups

Control MCI AD Post hoc (Tukey)

Verbal fluency 21.2 (3.8, 16–30) 18.9 (5.9, 10–30) 14.3 (5.2, 6–26) C > ADNaming 13.5 (1.2, 12–15) 12.8 (2.4, 7–15) 10.1 (3.5, 4–15) C > ADMMSE 27.4 (1.7, 24–30) 26.5 (2.3, 22–30) 22.8 (3.4, 18–29) C > AD, MCI > ADWLL sum score (three trials) 21.6 (3.2, 15–26) 17 (3.4, 10–21) 14.2 (4.9, 6–24) C > ADWLL (delay) 7.5 (1.7, 6–10) 5.7 (2.1, 1–8) 2.5 (2.7, 0–8) C > AD, MCI > ADWLL (savings) 92.2 (15.6, 75–120) 82 (26.7, 17–120) 39.8 (36.4, 0–88) C > AD, MCI > ADWLL (recognition) 98.7 (3, 90–100) 91.3 (7.9, 75–100) 81.3 (11.5, 65–100) C > AD, MCI > ADConstr. praxis (copy) 10.3 (1.3, 7–11) 9.6 (1.6, 7–11) 9.5 (1.8, 6–11) nsConstr. praxis (delayed recall) 7.9 (2.8, 2–11) 6.3 (3.9, 0–11) 2.27 (3.5, 0–11) C > ADConstr. praxis (savings) 75 (32, 18–120) 67.3 (43.5, 0–114) 27.5 (36.9, 0–100) C > ADClock drawing 5.2 (0.9, 3–6) 5.4 (0.6, 4–6) 4.3 (1.4, 1–6) ns

The between-group differences are shown in the last column, based on Bonferroni-adjusted (P < 0.005) a levels. WLL, wordlist learning.

CERAD test performances in amnestic MCI and AD

175

indicating that the shape of the learning curve wassimilar in all three groups (see Table 3).The sensitivity and specificity of those CERAD

subtests having cut-off scores based on Finnishnorms were calculated (Table 4). The subtestsVerbal fluency, Naming and MMSE had highspecificity, but the sensitivity to MCI was poor.The Wordlist learning (savings) had a high sensi-tivity to AD, but the sensitivity to MCI and thespecificity was low. In Wordlist learning (recogni-tion) the specificity was perfect, and the sensitivityto AD mediocre, but almost no MCI patients wereidentified. Cut-off scores of 86 and 92% yielded ahigher sensitivity and only a slightly loweredspecificity (Table 5).As the Wordlist learning test was the only

discriminator between the controls and the MCIgroup, the sensitivity and specificity of cut-offscores 16–20 were calculated for the Wordlistlearning test (sum score of three learning trials).The three optimal cut-off scores are presented inTable 5. A cut-off score of 16 yielded a specificityof 0.93. The sensitivity to MCI was 0.33 and overhalf of the mild AD patients were identified ascorrect positives. Increasing the cut-off score low-ered the specificity. Previous studies have found thedelayed recall of wordlist to be very sensitive toAD and thus the sensitivity and specificity of thismeasure using the cut-off scores 5–8 were calcula-ted. The optimal cut-off score was 6: it yielded nofalse positives and 80% of the mild AD patients

performed below this level. The sensitivity to MCIwas rather low (0.26).

Discussion

Almost all CERAD subtests in this study werefound to differentiate controls from AD patients.These differences were systematic and statisticallyhighly significant although the sample sizes weresmall. The MCI group was also differentiated fromthe AD group in four tests, indicating that the MCIgroup performed better than probable AD patientson one-third of the CERAD tests.None of the CERAD measures showed statisti-

cally significant differences between all threegroups. Moreover, the performances of the con-trols and the MCI patients were significantlydifferent only in the Wordlist learning test. In thistest the MCI group performed nearly as poorly asthe AD group. This finding is partly in line withprevious findings that the encoding of material inverbal episodic memory is deficient in MCI (14, 21,39). Previous studies have, however, also shownthat episodic learning is compromised in MCI. Inour study the MCI patients (and the AD patients)were, contrary to expectation, able to gain materialacross trials. Petersen et al. (39) have suggestedthat a sensitive measure of preclinical AD is theinability to acquire information beyond the short-term memory span, and that in order for thisphenomenon to be apparent several learning trialsare needed. It is thus possible that three trials areinadequate to show this �plateau effect�.In order to examine the reliability of the

CERAD as a screening instrument for MCI, wecalculated the sensitivity and specificity of the dif-ferent CERAD measures. Delayed recall perform-ance in the Wordlist learning test has previouslybeen found to be the best discriminator betweennormal aging and early AD (33). A cut-off scorefor the delayed recall of the wordlist is lacking inthe Finnish CERAD and our calculations showed

Table 4 Sensitivity and specificity of the CERAD tests where cut-off scoresbased on Finnish norms are available

Sensitivity Specificity

MCI vscontrols

AD vscontrols

Controls vsMCI and AD

Verbal fluency (15) 0.26 0.53 1Naming (11) 0.13 0.4 1MMSE (25) 0.13 0.73 0.93WLL (savings) (80) 0.33 0.8 0.66WLL recognition (80) 0.07 0.6 1Constr. praxis savings (60) 0.33 0.8 0.66Clock drawing (5) 0.06 0.33 0.86

Cut-off scores in parentheses. WLL, wordlist learning.

Table 5 Sensitivity and specificity for the wordlist learning test, wordlist delayedrecall and wordlist recognition

Sensitivity Specificity

MCI vscontrols

AD vscontrols

Controls vsMCI and AD

WLL sum score of three trials (16) 0.33 0.6 0.93WLL sum score of three trials (18) 0.33 0.8 0.86WLL sum score of three trials (20) 0.73 0.86 0.8Wordlist delayed recall (6) 0.26 0.86 1Wordlist recognition (86) 0.33 0.73 1Wordlist recognition (92) 0.46 0.8 0.93

Cut-off scores in parentheses. WLL, wordlist learning.

Table 3 Means, standard deviations and ranges on CERAD tests in the threegroups in the Wordlist learning trials

Control MCI AD

Wordlist learning (trial 1) 5.9 (1.2, 4–8) 3.9 (1.2, 1–6) 3.4 (1.9, 0–6)Wordlist learning (trial 2) 7.8 (1.2, 5–9) 6.2 (1.5, 3–8) 5.3 (1.7, 3–8)Wordlist learning (trial 3) 7.9 (1.4, 5–10) 6.9 (1.5, 4–9) 5.5 (1.8, 3–10)

Tukey's post hoc revealed significant differences between the control group andthe AD group, as well as between the control group and the MCI group.

Karrasch et al.

176

that a cut-off score of 6 seemed to be very specificand sensitive to clinically diagnosed AD, but only26% of the MCI patients performed at this level.One explanation for this result might be that thedelays in the CERAD memory tests are too short(usually ca. 5 min) to tap the episodic memoryconsolidation problems in MCI. Follow-up studieshave found that delayed recall of the WMS-RLogical memory test is a good predictor as towhich patients will later develop AD (5, 28), butone should note that in the standard administra-tion of the WMS-R, the delay usually is 20–30 min.The effectiveness of CERAD in tapping memoryconsolidation problems in MCI might thus beenhanced by prolonging the delays, e.g. by chan-ging the order of subtest presentation so thatWordlist learning and Constructional Praxis wouldbe the first tests administered and that the delayedrecalls would be the last subtests. This could, onthe other hand, result in floor effects in diagnosedAD patients.Another unexpected finding was that the Savings

scores (delayed recall adjusted for acquisition) inboth Wordlist learning and Constructional praxiswere relatively unspecific, i.e. several healthy con-trols were identified as false positives using thismeasure. This suggests that the current cut-offscores in the Finnish CERAD Savings scores mightbe too high.In the Finnish CERAD, no cut-off score has

been assigned to the Wordlist learning test. In lightof the current and previous findings (3, 14, 39) thisposes problems in the clinical use of the test. Cut-off scores are needed to study the sensitivity andspecificity of this test. We explored the sensitivityand specificity for the sum score of the three trialsof the Wordlist learning test using a range of cut-off scores. A cut-off score of 16 yielded a sensitivityof 0.33 to MCI and 0.6 to AD with a specificity of0.93. Raising the cut-off score also raised thesensitivity of the test, but the number of falsepositives also increased (Table 5).Wordlist recognition was found to be lower in

the MCI patients than in the controls, although thedifference did not reach the level of statisticalsignificance. Backman et al. (4) studied episodicmemory using recall of 12 words and recognitionof 24 words (12 targets, 12 distracters), and foundthat both measures were equally sensitive topreclinical AD 6 years before diagnosis. The cur-rent cut-off score for the Finnish CERAD Word-list recognition is 80% correct, which seems ratherlow as according to the findings of Karrasch andLaine (34), healthy elderly people almost reachedthe ceiling in this test (mean score in subjects withless than 10 years of formal education 97% correct

with SD 6, for subjects with more education thevalues were 99% correct with SD 2). A cut-offscore of 86% in our data yielded a sensitivity of0.33 for MCI and 0.73 for AD. No false positiveswere identified using this cut-off. When the cut-offscore was raised to 92% correct, the specificitydropped slightly (0.93), but the sensitivity rose to0.46 for MCI and 0.8 for AD. These findingsindicate that the cut-off in this subtest couldpossibly be adjusted to increase sensitivity withoutlosing specificity.The present study clearly shows that when

comparing normal elderly controls and clinicallydiagnosed probable AD patients, almost allCERAD tests are sensitive enough to revealsignificant differences. These differences werefound in verbal fluency, naming, learning, forget-ting and recognition memory even if the clinicaldementia severity in our AD patients was mild(mean MMSE score 22.8). Previous clinical studieson CERAD test performances have not includedMCI patients and our findings indicate that theonly discriminator between healthy aging and MCIon the CERAD is the Wordlist learning test. Manystudies have also indicated that the first cognitivesymptoms of AD would encompass the acquisitionand consolidation of episodic memories beyond theimmediate span (35, 37). This view is also suppor-ted by the cognitive roles of the medial temporallobes and the hippocampus, which are the firstareas to suffer from the pathological changes inAD.There were no significant differences between the

three groups in the performance of Constructionalpraxis (copy), indicating that this subtest is of littleuse in differentiating between normal aging, MCIand mild dementia. Yuspeh et al. (40) foundsignificant differences in this test between controlsand AD patients, but based on the mean MMSEscore their patient group was closer to moderatethan mild dementia. This indicates that problemsin basic visuospatial functions as measured by theConstructional praxis test do not appear at themild stage of the disease. The AD patients in ourstudy did, however, perform significantly worse onthe delayed recall of this test than the controls, butthe performance of the MCI group was close to thenormal range. In the clock drawing test there wereno significant differences between the groups.These results are in line with Powlishta et al. (41)and indicate that the clock drawing test is a poorinstrument for detecting MCI and mild AD.There are limitations to the generalization of the

results of this study. Small sample sizes may notprovide enough statistical power to detect allsignificant group differences. Larger samples

CERAD test performances in amnestic MCI and AD

177

should be used to validate the statistical differencesbetween the three groups found in this study. Inorder to assign cut-off scores for Wordlist learningand Wordlist delayed recall in the Finnish CERADlarger samples should be studied. As we notedearlier, in a previous study, where a verbal learningtest including 16 items and six learning trials wereused, learning curves in early AD patients reachedan early plateau and they were unable to acquirematerial beyond the immediate span (39). Thus it ispossible that the sensitivity of the Wordlist learn-ing test to MCI might be increased by increasingboth the number of items to be remembered andthe number of learning trials. Our findings alsoindicate that the current cut-off score for Wordlistrecognition could be raised without losing specif-icity. Background factors, such as educationalhistory, have been shown to affect performanceson the CERAD in healthy elderly controls (34) andthe effect of background variables on CERAD testperformances in MCI and AD patients should thusalso be studied. Our subjects were matched by ageand level of education in order to control for theseconfounds. Another problem is the unknown causeof memory impairment in the MCI subjects (42).The MCI group is most probably heterogeneous;some subjects may progress to AD, while othersmay remain stable or develop other dementingdisorders. Moreover, some tests may have highpositive predictive value (PPV) even if the sensi-tivity is low (43). Once reliable incidence andprevalence rates of MCI or preclinical AD becomeavailable in Finland, it could be useful in futurestudies to calculate PPVs.The results of this study show that by the time

clinical diagnosis of probable AD is reached,performances on all CERAD tests (except Con-structional praxis copy) are very poor. This mightbe due to the fact that diagnoses in primary healthcare are in part based on short screening tests likethe MMSE. The MMSE has been shown to havean excellent specificity, but its sensitivity to clinic-ally diagnosed probable AD has been found to berelatively low (�0.65) (44–46). Its sensitivity topreclinical/incipient AD has in recent studiesshown to be even lower (�0.35) (47, 48). In ourstudy there were no significant differences inMMSE scores between the MCI patients and thecontrols. The sensitivity of the MMSE to MCI inour data was 0.13, which underlines the import-ance of developing new screening tools.In sum, it can be concluded that at the group

level MCI patients performed significantly worsethan the controls on the Wordlist learning test.When examining the ability of CERAD to identifyamnestic MCI, we found that the sensitivity of the

Wordlist recognition test to amnestic MCI wasmoderate and approximately 75% of the ADpatients were correctly identified once the cut-offscore was adjusted to perfect specificity. We alsoexplored the sensitivity and specificity of theWordlist learning test (sum of three trials) usingdifferent cut-off scores. A cut-off score of 16yielded moderate sensitivity to amnestic MCI andAD (0.33 and 0.6, respectively) with a reasonablespecificity (0.93). Savings scores had a specificity of0.66, indicating that several false positives will beidentified using the current Finnish cut-off scores.Our results suggest that the CERAD does not havehigh sensitivity for amnestic MCI. Thus potentialpreclinical AD patients quite likely would receive afalse-negative result if tested only with theCERAD. Based on our data some CERAD meas-ures (verbal learning, delayed recall and recogni-tion) could prove to be more sensitive to amnesticMCI than the widely used MMSE if cut-off scoresare modified, but larger samples need to be studiedlongitudinally to validate this assumption.

Acknowledgements

This study was supported, in part, by grants from the Academyof Finland, Finnish Alzheimer Foundation, The Miina Silla-npaa Foundation, The Paulo Foundation, The Jenny and AnttiWihuri Foundation and Oskar Oflund Foundation.

References

1. Flicker C, Ferris SH, Reisberg B. Mild cognitive impair-ment in the elderly: predictors of dementia. Neurology1991;41:1006–9.

2. Small BJ, Fratiglioni L, Viitanen M, Winblad B, Backman L.

The course of cognitive impairment in preclinical Alzhei-mer’s disease. Arch Neurol 2000;57:839–44.

3. Linn RT, Wolf PA, Bachman DL et al. The �preclinicalphase� of probable Alzheimer’s disease. A 13-year pros-pective study of the Framingham cohort. Arch Neurol1995;52:485–90.

4. Backman L, Small BJ, Fratiglioni L. Stability of the pre-clinical episodic memory deficit in Alzheimer’s disease.Brain 2001;124:96–102.

5. Marquis S, Moore MM, Howieson DB et al. Independentpredictors of cognitive decline in healthy elderly persons.Arch Neurol 2002;59:601–6.

6. Morris JC, Storandt M, Miller JP et al. Mild cognitiveimpairment represents early-stage Alzheimer’s disease.Arch Neurol 2001;58:397–405.

7. Koivisto K, Reinikainen KJ, Hanninen T et al. Prevalence ofage-associated memory impairment in a randomly selectedpopulation from eastern Finland. Neurology 1995;45:741–7.

8. Palmer K, Wang H-X, Backman L, Winblad B, Fratiglioni L.

Differential evolution of cognitive impairment in non-demented older persons: results from the Kungsholmenproject. Am J Psychiatry 2002;159:436–42.

9. Petersen RC, Doody R, Kurz A et al. Current concepts inmild cognitive impairment. Arch Neurol 2001;58:1985–92.

Karrasch et al.

178

10. Bowen J, Teri L, Kukull W, McCormick W, McCurry SM,

Larson EB. Progression to dementia in patients with iso-lated memory loss. Lancet 1997;349:763–5.

11. Luis CA, Barker WW, Loewenstein DA et al. Conversion todementia among two groups with cognitive impairment.Dement Geriatr Cogn Disord 2004;18:307–13.

12. Meyer JS, Xu G, Thornby J, Chowdhury M, Quach M.

Longitudinal analysis of abnormal domains comprisingmild cognitive impairment (MCI) during aging. J NeurolSci 2002;201:19–25.

13. Bozoki A, Giordani B, Heidebrink JL, Berent S, Foster NL.

Mild cognitive impairments predict dementia in non-demented elderly patients with memory loss. Arch Neurol2001;58:411–6.

14. Collie A, Maru P. The neuropsychology of preclinicalAlzheimer’s disease and mild cognitive impairment. Neu-rosci Biobehav Rev 2000;24:365–74.

15. Kantarci K, Jack CR, Xu C et al. Regional metabolicpatterns in mild cognitive impairment and Alzheimer’sdisease. Neurology 2000;55:210–6.

16. Petersen RC, Smith GE, Waring SC, Ivnik RT, Tangalos

EG, Kokmen E. Mild cognitive impairment: clinicalcharacterisation and outcome. Arch Neurol 1999;56:303–8.

17. Perry RJ, Watson P, Hodges JR. The nature and stagingof attention dysfunction in early (minimal and mild)Alzheimer’s disease: relationship to episodic and seman-tic memory impairment. Neuropsychologia 2000;38:252–71.

18. Fox NC, Warrington EK, Freeborough PA et al. Presymp-tomatic hippocampal atrophy in Alzheimer’s disease. Alongitudinal MRI study. Brain 1996;119:2001–7.

19. Jack CR, Petersen RC, Xu YC et al. Prediction of AD withMRI-based hippocampal volume in mild cognitiveimpairment. Neurology 1999;52:1397–403.

20. Pennanen C, Kivipelto M, Tuomainen S et al. Hippocampusand entorhinal cortex in mild cognitive impairment andearly AD. Neurobiol Aging 2004;25:303–10.

21. Morris JC, McKeel DW, Storandt M et al. Very mildAlzheimer’s disease: informant-based clinical, psychomet-ric, and pathologic distinction from normal aging. Neur-ology 1991;41:469–78.

22. Masur DM, Sliwinski M, Lipton RB, Blau AD, Crystal HA.

Neuropsychological prediction of dementia and theabsence of dementia in healthy elderly persons. Neurology1994;44:1427–32.

23. Laatu S, Portin R, Revonsuo A, Tuisku S, Rinne J. Know-ledge of concept meanings in Alzheimer’s disease. Cortex1997;33:27–45.

24. Henry JD, Crawford JR, Phillips LH. Verbal fluency per-formance in dementia of the Alzheimer’s type: a meta-analysis. Neuropsychologia 2004;42:1212–22.

25. Salmon DP, Thomas RG, Pay MM et al. Alzheimer’s diseasecan be accurately diagnosed in very mildly impaired indi-viduals. Neurology 2002;59:1022–8.

26. Welsh KA, Butters N, Hughes J, Mohs R, Heyman A.

Detection and staging of dementia in Alzheimer’s disease.Arch Neurol 1992;49:448–52.

27. Rubin EH, Storandt M, Miller JP et al. A prospective studyof cognitive function and onset of dementia in cognitivelyhealthy elders. Arch Neurol 1998;55:395–401.

28. Elias MF, Beiser A, Wolf PA, Au R, White RF, D’Agostino

RB. The preclinical phase of Alzheimer disease. ArchNeurol 2000;57:808–13.

29. Storandt M, Grant EA, Miller JP, Morris JC. Rates ofprogression in mild cognitive impairment and earlyAlzheimer’s disease. Neurology 2002;59:1034–41.

30. Becker J, Boller F, Lopez O, Saxton J, McGonicle K. Thenatural history of Alzheimer’s disease. Arch Neurol1994;51:585–94.

31. Welsh KA, Butters N, Mohs RC et al. The Consortium toEstablish a Registry for Alzheimer’s Disease (CERAD).Part V. A normative study of the neuropsychologicalbattery. Neurology 1994;44:609–14.

32. Pulliainen V, Hokkanen L, Salo J, Hanninen T: CERAD-kognitiivinen tehtavasarja. Kasikirja. [Manual of the Fin-nish version of CERAD]. Kuopio: Suomen Alzheimer-tutkimusseura ry, 1999.

33. Welsh-Bohmer KA, Mohs RC. Neuropsychological assess-ment of Alzheimer’s disease. Neurology 1997;49:S11–S13.

34. Karrasch M, Laine M. Age, education and test performanceon the Finnish CERAD. Acta Neurol Scand 2003;108:97–101.

35. Greene JDW, Baddeley AD, Hodges JR. Analysis of theepisodic deficit in early Alzheimer’s disease: evidence fromthe doors and people test. Neuropsychologia 1996;34:527–51.

36. Welsh KA, Butters N, Hughes J, Mohs R, Heyman A.

Detection of abnormal memory decline in mild cases ofAlzheimer’s disease using CERAD neuropsychologicalmeasures. Arch Neurol 1991;48:278–81.

37. Wang Q-S, Zhou J-N. Retrieval and encoding of episodicmemory in normal aging and patients with mild cognitiveimpairment. Brain Res 2002;924:113–5.

38. McKhann G, Drachman D, Folstein M, Katzman R, Price D,

Stadlan E-M. Clinical diagnosis of Alzheimer’s disease:report of the NINCDS-ADRDA Work Group under theauspices of Department of Health and Human ServicesTask Force on Alzheimer’s Disease. Neurology1984;34:939–44.

39. Petersen RC, Smith GE, Ivnik RT, Kokmen E, Tangalos EG.

Memory function in very early Alzheimer’s disease.Neurology 1994;44:867–72.

40. Yuspeh RL, Vanderploeg RD, Kershaw DAJ. CERAD praxismemory and recognition in relation to other measures ofmemory. Clin Neuropsychol 1998;12:468–74.

41. Powlishta KK, Von Dras DD, Stanford A et al. The clockdrawing test is a poor screen for very mild dementia.Neurology 2002;59:898–903.

42. Petersen RC. Mild cognitive impairment or questionabledementia? Editorial. Arch Neurol 2000;57:650–1.

43. Smith GE, Ivnik RJ. Normative neuropsychology. In:Petersen RC, ed. Mild cognitive impairment: aging toAlzheimer’s disease. Oxford: OxfordUniversity Press, 2003.

44. Wind AW, Schellevis FG, Van Staveren G, Scholten RJPM,

Jonker C, Van Eijk JTM. Limitations of the Mini-MentalState Examination in diagnosing dementia in the generalpractice. Int J Geriatr Psychiatry 1997;12:101–8.

45. Tangalos EG, Smith GE, Ivnik RJ et al. The Mini-MentalState Examination in general medical practice: clinicalutility and acceptance. Mayo Clin Proc 1996;71:829–37.

46. Gallassi R, Morreale A, Di Sarro R, Lorusso S. Value ofclinical data and neuropsychological measures in probableAlzheimer’s disease. Arch Gerontol Geriatr 2002;34:123–34.

47. Tang-Wai D, Knopman D, Geda Y et al. Comparison of theshort test of mental status and the Mini-Mental StateExamination in mild cognitive impairment. Arch Neurol2003;60:1777–81.

48. Tierney MC, Szalai JP, Dunn E, Geslani D, McDowell I.

Prediction of probable Alzheimer disease in patients withsymptoms suggestive of memory impairment: value of theMini-Mental State Examination. Arch Fam Med 2000;9:527–32.

CERAD test performances in amnestic MCI and AD

179