Diagnostic accuracy of the Preclinical AD Scale (PAS) in cognitively mildly impaired subjects

J Neurol (2002) 249 : 312 319 Steinkopff Verlag 2002 ORIGINAL COMMUNICATION Pieter Jelle Visser Frans R. J. Verhey Philip Scheltens Marc Cruts Rudolf W. H. M. Ponds Christine L. Van Broeckhoven Jellemer Jolles Diagnostic accuracy of the Preclinical AD Scale (PAS) in cognitively mildly impaired subjects Received: 21 February 2001 Received in revised form: 20 July 2001 Accepted: 25 July 2001 P. J. Visser ( ) F. R. J. Verhey R. W. H. M. Ponds J. Jolles Department of Psychiatry and Neuropsychology Institute of Brain and Behaviour University of Maastricht P. O. Box 616 6200 MD Maastricht, The Netherlands Tel.: +31-20/6 17 17 21 Fax:+31-43/3875444 E-Mail: pj.visser@np.unimaas.nl P. Scheltens Department of Neurology VU Medisch Centrum P. O. Box 7057 1007 MB Amsterdam, The Netherlands M. Cruts C. L. Van Broeckhoven Department of Molecular Genetics Flanders Interuniversity Institute for Biotechnology (VIB) Laboratory of Neurogenetics Born-Bunge Foundation (BBS) University of Antwerp (UIA) Universiteitsplein 1 2610 Antwerp, Belgium Presented, in part, at the World Alzheimer Congress 2000 Pivotal Research, Washington, USA, 9 13 July, 2000 Abstract The Preclinical AD Scale (PAS) is a newly developed scale for the diagnosis of preclinical Alzheimer s disease (AD). The PAS combines six markers of preclinical AD, namely age, MMSE score, functional impairment, cognitive test performance, medial temporal lobe atrophy, and the apolipoprotein E (APOE) genotype. The aim of the study was to investigate whether the PAS can accurately identify subjects with preclinical AD who become demented during a 2 or 5 year follow-up from among subjects with mild cognitive impairment for other reasons. We also investigated whether a stepwise scoring of the PAS could reduce the number of elaborate or expensive diagnostic procedures. The PAS was scored retrospectively in two independent samples of non-demented subjects with mild cognitive impairment older than 55 years (average age 65.6 years), who were selected from a memory clinic population. In the first sample, the follow-up was 5 years (5-year follow-up sample; n=69). In the second sample, the follow-up was 2 years (2-year follow-up sample; n=23). The PAS item medial temporal lobe atrophy was not scored in the 5-year follow-up sample. A PAS cut-off score of 4/5 could best identify subjects with AD-type dementia at follow-up (n=25) in the 5-year follow-up sample with a sensitivity of 80 % and a positive predictive value of 77 %. A PAS cutoff score of 5/6 could best identify subjects with AD-type dementia at follow-up (n=8) in the 2-year follow-up sample with a sensitivity of 88 % and a positive predictive value of 70 %. The positive predictive value could be increased to 94 % in the 5-year follow-up sample and to 80 % in the 2-year follow-up sample by using higher cut-off scores, but this reduced the sensitivity. Step-wise scoring of the PAS had the same diagnostic accuracy as the total PAS score and reduced the number of cognitive assessments by 22 to 38 %, the number of assessments of medial temporal lobe atrophy by 57 to 74 %, and the number of APOE genotypings by 74 %. It is concluded that the PAS is a useful scale to identify subjects with preclinical AD who will become demented during the next 2 or 5 years. Step-wise scoring of the PAS can reduce the number of elaborate or expensive diagnostic procedures considerably. Key words Alzheimer s disease Diagnosis MCI Follow-up studies MRI JON 623

313 Introduction Before subjects with Alzheimer s disease (AD) become demented there is a long preclinical phase in which they experience mild cognitive impairments [13]. The Preclinical Alzheimer s disease Scale (PAS) was developed recently as a means to identify subjects who are in the preclinical phase of AD (Table 1) [27, 30]. The PAS combines six markers of preclinical AD, i. e. age, the score on the Mini-Mental State Examination (MMSE), the degree of functional impairment, cognitive test performance, medial temporal lobe atrophy, and the apolipoprotein E (APOE) genotype. These markers were selected on the basis of a meta-analysis of 21 prospective studies of subjects with mild cognitive impairment [27, 30]. Each marker is scored on a 3 or 4-point scale and the sum of the scores for each marker indicates the risk for preclinical AD. The aim of the present study was to investigate whether the PAS can accurately identify subjects with preclinical AD who become demented during a 2 or 5 year follow-up from among subjects with mild cognitive impairment for other reasons. We also investigated whether a step-wise scoring of the PAS could reduce the number of elaborate or expensive diagnostic procedures without loss of diagnostic accuracy. The stepwise scoring of markers of preclinical AD has received little attention in the literature.yet it is important to make a rational use of diagnostic measures in order to reduce costs and the burden of diagnostic procedures to the patient. For the stepwise scoring we determined a priori four steps. The first step included the PAS items that could be most easily scored, i. e. age, MMSE score, and degree of functional impairment. The second step consisted of the PAS item cognitive test performance, the third step of the PAS item medial temporal lobe atrophy, and the fourth step of the PAS item APOE genotype. The order of these steps was motivated as follows. Cognitive test performance was chosen as the second step because it is a stronger marker of preclinical AD than the APOE genotype [24] and it is more widely available and less expensive than assessment of medial temporal lobe atrophy. Medial temporal lobe atrophy was chosen as the third step because it is a stronger marker of preclinical AD than the APOE genotype is [11]. Finally, we investi- Table 1 The Preclinical AD scale (PAS) 1 0 1 2 Score A. Age (years) 59 60 64 65 74 75... B. MMSE Age <75years Edu 8 years 27 25,26 24 Edu 8 14 years 28 26,27 25 Edu 14 years 29 27,28 26... Edu 8 years 26 24,25 23 Age 75 years Edu 8 14 years 27 25,26 24 Edu 14 years 28 26,27 25 C. Functional impairment 1 C.1 GDS GDS 1 GDS 2 GDS 3 C.2 CDR C.2.1. Total Box score 0 0.5 1 1.5 C.2.2. Final score CDR 0 CDR 0.5... C.3 CAMDEX Min Dem D. Cognitive tests 2 Memory Other 1 impaired score 2 impaired scores... 50th perc E. MTL atrophy 1 E.1 Qualitative rating 3 Age < 75 years 0 1 2 Age 75 years 0 1 2 3... E.2 Volumetry 4 66th perc 33 66th perc 10 33th perc 10th perc F. APOE genotype Other e2e4/e3e4 e4e4... TOTAL SCORE... MMSE=Mini-Mental State Examination [7]; Edu=Education; GDS=Global Deterioration Scale [18]; CDR=Clinical Dementia Rating scale [10, 16]; CAMDEX=Cambridge Mental Disorders of the Elderly Examination [20]; Min Dem= minimal dementia; perc=percentile; MTL=Medial Temporal Lobe; APOE=Apolipoprotein E. 1 Only one of the scoring options of the PAS item should be used. 2 Including 1 test that assesses memory (delayed recall or learning) and 1 to 3 tests for other cognitive domains (e. g., language, executive functions, abstract reasoning, visuoconstruction). Impairment is defined as a score below the 10th percentile or above the 90th percentile (speed related tasks). Percentile scores are corrected for age and, if possible, for sex and education as well. 3 Qualitative rating according to [21] or [6]. 4 Volumetry of hippocampus, parahippocampal gyrus, or entorhinal cortex; percentiles scores are corrected for age and intracranial volume and, if possible, for sex as well.

314 gated the predictive accuracy of individual PAS items for outcome. Materials and methods Subjects Subjects were selected from the population of an ongoing follow-up study of subjects who were referred to the Maastricht Memory Clinic because of cognitive impairment but who were not demented at the time of the first visit [26, 31, 33]. In this follow-up study, subjects were reassessed on average 2 and 5 years after the baseline assessment [31, 33]. From the follow-up study we selected two independent samples. No subjects were added to either sample. Inclusion criteria for both samples were age higher than 55 years, a MMSE score higher than 23, and a score on the Global Deterioration Scale [18] of 2 or 3 at baseline. Subjects were excluded if they had cognitive impairment at baseline that was related to any neurological disorder, any somatic disorder, any vascular disorder, or any major psychiatric disorder other than mild affective disorders. The first sample was selected from among subjects who were eligible for the 5-year follow-up assessment at the time of the present analysis (n=81). The outcome with respect to the presence of dementia at follow-up was not available for 12 of these subjects (15%) (2 subjects had died before follow-up, 3 subjects had refused the follow-up, 6 subjects were untraceable, and one subject had experienced cerebral bleeding and could not be tested). The remaining subjects (n=69, 85%) were included in the study and will be referred to as the 5-year follow-up (FU) sample. The PAS item medial temporal lobe atrophy was not assessed in this sample. The average follow-up of these subjects was 5.1 years (range 4.7 5.5). The second sample was selected from among subjects in whom medial temporal lobe atrophy was assessed at baseline (n=25). Outcome with respect to the presence of dementia at follow-up was not available in one subject who refused the follow-up assessment. We also excluded one subject with frontal lobe dementia at follow-up. The remaining subjects (n=23, 92 %) were included in the study and will be referred to as the 2-year FU sample. The average follow-up in these subjects was 2.0 years (range 1.1 3.5). In the 5-year FU sample, 44 subjects were not demented at follow-up and 25 subjects had probable (n=24) or possible AD (n=1) at follow-up. In the 2-year FU sample, 15 subjects were not demented at follow-up and 8 subjects had probable AD at follow-up. Subjects from the 5-year FU sample had similar baseline characteristics as the subjects from the 2-year FU sample (P > 0.10 for all comparisons) (Table 2).Subjects with no follow-up or subjects who were excluded had similar baseline characteristics as the subjects from the 5-year and 2-year FU sample (Table 2).All subjects gave their informed consent prior to their inclusion in the study. Clinical assessment and clinical diagnosis at baseline and follow-up Table 2 At baseline, all subjects underwent a standardized assessment, which included a detailed history provided by the patient and a significant other, a psychiatric, neurological, and physical examination, clinical rating scales (i. e. the MMSE [7], the Global Deterioration Scale (GDS) [18], the Hamilton Depression Rating Scale 17 items [9], and the Blessed Dementia Rating Scale part I [2]), appropriate laboratory tests, a neuropsychological assessment, and neuroimaging as described elsewhere [26]. The diagnosis of dementia and AD was made according to the DSM-IV and NINCDS-ADRDA criteria [1, 15]. The follow-up assessment consisted of a standardized questionnaire about medical history and cognitive complaints, the MMSE, the GDS, the Hamilton Depression Rating Scale, the Blessed Dementia Rating Scale, and a neuropsychological test protocol [31, 33]. If a subject refused to come for the follow-up assessment,a telephone interview was conducted which included a standardized questionnaire about medical history and cognitive complaints, and the Telephone Interview for Cognitive Status [5]. The diagnosis of dementia and AD at follow-up was made by a neuropsychiatrist and a neuropsychologist, who were unaware of the results of the baseline assessment and the PAS score at baseline and who made their diagnosis independently of each other. If there was disagreement about the clinical diagnosis, a consensus meeting was held and if no agreement was reached the subject was considered not demented. PAS scoring Baseline characteristics 5-year FU 2-year FU No Follow-up sample sample or excluded (n=69) (n=23) (n=14)* Age (years) 65.0 (7.2) 67.3 (6.7) 68.7 (7.9) range 55 81 55 79 57 84 Sex (%male) 51% 65% 36% Education (years) 10.3 (3.1) 11.0 (3.3) 10.2 (3.8) MMSE score 28.0 (1.8) 27.6 (1.7) 28.1 (1.5) HDRS 9.6 (5.7) 9.6 (6.7) 12.8 (6.3) GDS 2 30 12 7 3 39 11 7 Delayed recall (z-score) 0.99 (1.4) 0.91 (1.1) 0.39 (1.1) SCWT card 3 (z-score) 0.65 (1.2) 1.24 (1.5) 1.38 (2.0) MST-L1 (z-score) 0.94 (1.3) 0.67 (0.93) 0.36 (1.0) Fluency (z-score) 0.50 (1.0) 0.85 (1.1) 0.47 (0.9) IQ 116.6 (14.2) 113.1 (11.7) 107.0 (12.3) Hippocampal volume 0.31 (1.3) (z-score) APOE genotype 1; 5; 19; 24; 3 0; 1; 8; 8; 2 (22; 23; 33; 34; 44) MMSE=Mini-Mental State Examination; HDRS=Hamilton Depression Rating Scale; GDS=Global Deterioration Scale; SCWT=Stroop Color Word Test; MST-L1=Memory Scanning Task letter 1; IQ= Intelligence Quotient according to [14] or [34]; = no data available. Data are means (SD) except for sex (%), GDS score, and APOE genotype (number of subjects). * Including 12 subjects who were eligible for the 5-year follow-up and 2 subjects who were eligible for the 2-year follow-up. The PAS was scored in both samples in the same way on the basis of the data collected at baseline. The PAS item functional impairment was scored with the GDS [18]. The PAS item cognitive tests was scored using the delayed recall measure of the Auditory Verbal Learning Test [3, 12], the time to complete the Memory Scanning Task Letter 1 [4], the time to complete card 3 from the Stroop Color Word Test (SCWT) [23], and verbal fluency (the ability to name as many professions/trades (n=90) or animals (n=1) as possible within 1 minute). In four subjects the SCWT had not been done and the time to complete Trailmaking B [19] was used instead. If subjects could not complete the test for reasons other than sensory impairment, the test performance was considered impaired. Seven subjects had missing data for verbal fluency (n=1) or for the Memory Scanning Task (n=7). The PAS cognitive tests item score in these subjects was calculated on the basis of cognitive data that were available. Four of them had maximum PAS scores for the cognitive tests item. All cognitive scores were corrected for age, sex, and education and expressed as z-scores on the basis of a reference population of 1870 cognitively normal subjects [33]. The sign of the z-scores of the SCWT, Trailmaking B, and the Memory Scanning Task was inverted such that a z-score below zero indicated a below average performance. Medial temporal lobe atrophy was assessed by volumetry of the hippocampus on MRI [28]. MRI scans were not available for volumetry in two

315 subjects and in these subjects a qualitative rating of medial temporal lobe atrophy was performed [21]. APOE genotyping was performed by a polymerase chain reaction [22]. Blood samples for genotyping were taken at the follow-up assessment from 1995 onwards. Therefore, no samples were available for subjects seen before 1995, for subjects in whom the follow-up assessment was done by telephone, and for subjects without follow-up. The APOE genotype was available for 52 subjects of the 5-year FU sample (75 %) and for 20 subjects of the 2-year FU sample (83 %). Compared with subjects who were genotyped, subjects in the 5-year FU sample who were not APOE genotyped had a lower MMSE score (26.8 vs 28.4, P < 0.01) and delayed recall score ( 1.69 (z-score) vs 0.76, P < 0.01) at baseline, and more frequently had AD-type dementia at follow-up (65% vs 27 %, P=0.01). There were no differences in baseline characteristics between subjects with and without APOE genotyping in the 2- year FU sample. Subjects in which the APOE genotype was not available were given a score of zero on that item. Statistical Analysis The data were analysed using SPSS for Macintosh 6.1.1 (SPSS Inc., Chigaco, IL, USA). In groups with more than 10 subjects, continuous variables were compared by means of a t-test. In groups with 10 or fewer subjects, continuous variables were analysed with the Mann- Whitney test corrected for ties. Categorical data were analysed with a chi square test with continuity correction. The differences between the groups in PAS scores and scores on individual PAS items were analysed with the Mantel-Haenszel test for linear association. Cox regression with time to follow-up as the censoring time was used to investigate the predictive accuracy of individual PAS items in the combined sample of subjects with a 5- and 2-year follow-up. Results In both samples, the total PAS score was higher in subjects with AD-type dementia at follow-up than in subjects without AD-type dementia at follow-up (5.7 vs 2.3 in the 5-year FU sample and 6.4 vs 3.3 in the 2-year FU sample, P 0.01 for both comparisons) (Table 3). In the 5-year FU sample, all PAS items were scored significantly higher in subjects with AD-type dementia at follow-up than in subjects without dementia at follow-up (Table 4). In the 2-year FU sample, the PAS items age and Table 3 PAS score according to outcome 5-year FU sample 2-year FU sample No dementia AD-type No dementia AD-type at FU dementia at FU at FU dementia at FU (n=44) (n=25) (n=15) (n=8) 0 9 0 1 0 1 7 1 2 0 2 7 1 3 0 3 7 0 3 1 4 8 3 0 0 5 5 5 3 0 6 1 5 2 3 7 0 8 1 3 8 0 1 0 0 9 0 1 0 1 10 0 0 0 0 Average PAS 2.3 (1.9) 5.7 (1.8)* 3.3 (2.3) 6.4 (1.7)* score (SD) AD = Alzheimer s disease; FU = Follow-up. * Significantly different from subjects with no dementia at follow-up (P < 0.01) MTL atrophy were scored significantly higher in subjects with AD-type dementia at follow-up than in subjects without dementia at follow-up (Table 4). Univariate cox regression analyses with time to follow-up as the censoring time showed that all PAS items were significantly associated with outcome in the combined sample, except for the APOE item (P=0.38). After excluding subjects in which the APOE genotyping was not performed, the APOE item was significantly associated with outcome (P=0.04). The receiver operating characteristic (ROC) curve for the total PAS score is shown in Fig. 1. A cut-off between 4 and 5 could best differentiate between subjects with and without AD-type dementia at follow-up in the 5-year FU sample. A cut-off between 5 and 6 could best differentiate between subjects with and without ADtype dementia at follow-up in the 2-year FU sample. The sensitivity, specificity, positive predictive value, and neg- Table 4 PAS score on individual items according to outcome 5-year FU sample 2-year FU sample No dementia AD-type No dementia AD-type at FU dementia at FU at FU dementia at FU (n=44) (n=25) (n=15) (n=8) A. Age ( 1, 0, 1, 2) 18, 15, 10, 1 2, 3, 12, 8 1 4, 2, 8, 1 0, 1, 4, 3 2 B. MMSE (0, 1, 2) 36, 7, 1 11, 10, 4 1 9, 4, 2 6, 1, 1 C. Functional impairment (0, 1, 2) 0, 25, 19 0, 5, 20 1 0, 9, 6 0, 3, 5 D. Cognitive tests ( 1, 0, 1, 2) 13, 8, 14, 9 1, 1, 4, 19 1 5, 2, 2, 6 0, 0, 3, 5 E. MTL atrophy ( 1, 0, 1, 2) 5, 8, 1, 1 0, 2, 4, 2 1 F. APOE genotype (0, 1, 2) 23, 13, 2 2, 11, 1 1 7, 4, 2 2, 4, 0 Indicated are the number of subjects with each PAS item score. Subjects with missing data for the items E. and F. were excluded. AD = Alzheimer s disease; FU = Follow-up; MMSE = Mini-Mental State Examination; MTL = Medial temporal lobe; APOE = Apolipoprotein E; = no data available. 1 Significantly different from subjects with no dementia at follow-up (analysis of trend) (P 0.01) 2 Significantly different from subjects with no dementia at follow-up (analysis of trend) (P < 0.05)

316 Table 5 Sensitivity, specificity, positive predictive value, and negative predictive value for AD-type dementia at follow-up at several PAS cut-off scores Sample Cut-off Sensitivity Specificity PPV NPV 5-year FU 4/5 80% 86% 77% 88% 5-year FU 5/6 60% 98% 94% 81% 5-year FU 6/7 40% 100% 100% 75% 2-year FU 4/5 88% 60% 54% 90% 2-year FU 5/6 88% 80% 70% 92% 2-year FU 6/7 50% 93% 80% 78% AD = Alzheimer s disease; PPV = positive predictive value; NPV = negative predictive value. ative predictive value of the cut-off scores 4/5, 5/6, and 6/7 are shown in Table 5. The decision rules for the step-wise scoring were determined on the basis of the total PAS score in the combined sample of subjects with a 5-year and 2-year follow-up. First, we determined which total PAS score was associated with a low risk for AD-type dementia at follow-up (< 30 %) or a high risk for AD-type dementia at follow-up (> 70 %). Based on the combined 5-year and 2-year FU sample, the low-risk total PAS score was 4 and the high-risk total PAS score was 6 (Table 3). We then determined which score at each step identified with 100 % accuracy subjects with a low-risk or a high-risk total PAS score. These subjects did not proceed to the next step. For example, all subjects with a score 1 after step 1 had a low-risk total PAS score and all subjects with a score 6 after step 1 had a high-risk total PAS score. Only subjects with a score between 1 and 6 proceeded to the next step. The decision rules for the stepwise scoring are shown in Table 6. With these decision rules, stepwise scoring reduced the number of cognitive assessments by 38 % in the 5-year FU sample and by 22 % in the 2-year FU sample, the number of assessments of medial temporal lobe atrophy by 74 % in the 5-year FU sample and by 57 % in the 2-year FU sample, and the number of APOE genotypings by 74 % in both samples (Table 7). Discussion Fig. 1 ROC curve of PAS. The italic cut-off scores refer to the 5-year FU sample and the bold cut-off scores to the 2-year FU sample. The main conclusions are that the PAS is useful for identifying subjects with preclinical AD who will become demented during the next 2 or 5 years, and that step-wise scoring of the PAS can reduce the number of elaborate or expensive diagnostic procedures considerably. The 5-year and 2-year FU sample differed with respect to the best cut-off scores to identify subjects with AD-type dementia at follow-up and the sensitivity, specificity, and positive predictive value. The fact that the best cut-off score was higher in the 2-year FU sample than it was in the 5-year FU sample may have resulted from the fact that medial temporal lobe atrophy was scored in the 2-year FU sample but not in the 5-year FU sample. In addition, it is possible that subjects who develop AD-type dementia during a 2-year follow-up have higher PAS scores at baseline than subjects who develop it during a 5-year follow-up. The sensitivity for detecting subjects with AD-type dementia at follow-up Table 6 Decision rules stepwise scoring of the PAS Low risk AD-type Borderline risk High risk AD-type dementia at FU AD-type dementia dementia at FU at FU Step 1 Age, MMSE, functional impairment 1 2 5 6 Step 2 Cognitive performance 3 4 5 6 Step 3 MTL atrophy 3 4 5 6 Step 4 APOE genotype 4 5 6 Only subjects with a borderline risk proceed to the next step. AD = Alzheimer s disease; FU = follow-up; MMSE = Mini-Mental State Examination; MTL = medial temporal lobe; APOE = Apolipoprotein E.

317 Table 7 Outcome of stepwise scoring of the PAS Low-risk score Borderline-risk score High-risk score No dementia AD-type No dementia AD-type No dementia AD-type at FU dementia at FU dementia at FU dementia at FU at FU at FU 5-Year FU sample Step 1 24 2 20 23 0 0 Step 2 11 (35) 1 (3) 8 10 1 (1) 12 (12) Step 3* Step 4 3 (38) 2 (5) 5 5 0 (1) 3 (15) 2-Year FU sample Step 1 4 1 11 7 0 0 Step 2 4 (8) 0 (1) 6 4 1 (1) 3 (3) Step 3 1 (9) 0 (1) 5 1 0 (1) 3 (6) Step 4 0 (9) 0 (1) 3 0 2 (3) 1 (7) Indicated are number of subjects with a low-, borderline-, or high-risk score after each step. The cumulative number of subjects with low-risk or high-risk scores is given in parentheses. AD=Alzheimer s Disease; FU = follow-up; = no data available. * Medial temporal lobe atrophy (step 3) was not determined in this sample. was at all cut-off scores higher in the 2-year FU sample than it was in the 5-year FU sample. This may be due to the fact that medial temporal lobe atrophy was scored in the 2-year FU sample but not in the 5-year FU sample, because previous studies have demonstrated that assessment of medial temporal lobe atrophy can increase the sensitivity for detecting subjects who will become demented compared with that of age and cognitive measures [6, 29]. The higher sensitivity in the 2-year FU sample may also have resulted from the short follow-up period in this sample since, as indicated above, subjects who develop AD-type dementia during a 2-year followup may have higher PAS scores at baseline than subjects who develop it during a 5-year follow-up. The specificity and positive predictive value were at all cut-off scores lower in the 2-year FU sample than they were in the 5- year FU sample. This may be due to the short follow-up in the 2-year FU sample because some subjects with high PAS scores and no dementia at the 2-year follow-up may develop dementia after a longer follow-up interval. The step-wise scoring of the PAS indicated that the number of assessments of cognitive function, medial temporal lobe atrophy, and APOE genotype can be reduced substantially without loss of diagnostic accuracy. It should be noted that the step-wise scoring decision rules in the present study were chosen post-hoc. The decision rules may be different in other settings, as may be the best order of steps. The fact that all individual PAS items were predictors of outcome in the Cox regression analysis supported the selection of variables for the PAS. Cox regression with backward step selection selected the items age,cognitive tests, and medial temporal lobe atrophy (data not shown), which indicates that the PAS may be shortened for use in the present sample. The difference in average annual conversion rate between the 5-year FU sample (7.2 %) and 2-year FU sample (17 %) may suggest that these samples are not comparable. There was, however, no significant difference in average annual conversion rate if we compared the conversion rate after 2 years of follow-up in the 5-year FU sample with the conversion rate in the 2-year FU sample (10 % versus 17 %, P= 0.26). This indicates that the conversion rate levels off at longer follow-up intervals and suggests that the lower annual conversion rate in the 5- year FU sample is probably due to the longer follow-up in this sample. The scores on the single PAS items of subjects who had AD-type dementia at follow-up demonstrated great heterogeneity. This is consistent with the observation that there is no simple profile of preclinical AD [11, 25, 29, 32, 33]. One important advantage of the PAS is that it does not require subjects with preclinical AD to have a certain degree of cognitive impairment, or functional impairment, or brain atrophy in order to qualify for the diagnosis of preclinical AD. Therefore, the PAS will probably have a better sensitivity for detecting subjects with preclinical AD than criteria for preclinical AD that require a certain degree of cognitive impairment, or functional impairment, or brain atrophy. In the present study, mild cognitive impairment was defined as a score of 2 or 3 on the GDS. Another definition of mild cognitive impairment that has recently received much attention is the definition according to the criteria of Mild Cognitive Impairment [17]. These criteria are more stringent than the criteria of mild cognitive impairment in the present study. In order to investigate the diagnostic accuracy of the PAS in subjects with Mild Cognitive Impairment, we scored the PAS in a subgroup of patients from the 5-year FU sample who met these criteria (n=21 (including 13 subjects with AD-type dementia at follow-up), average age 68.5 years). The average PAS score in these subjects was 5.6 (range 3 8). A cut-off between 5 and 6 could best discriminate between subjects with or without AD-type dementia at follow-up. The sensitivity for Alzheimer-type dementia at this cut-

318 off was substantially higher (77 %) in subjects who fulfilled the criteria of Mild Cognitive Impairment than it was at the same cut-off in subjects from the 5-year FU sample, while the specificity, positive predictive value, and negative predictive value were comparable.the sensitivity, however, was much lower (40 %) if it was taken into account that 11 subjects with AD-type dementia at follow-up were excluded because they did not meet the inclusion criteria of Mild Cognitive Impairment. The decision rules for stepwise scoring after step 2 indicated that the PAS can be used to increase the predictive accuracy for Alzheimer-type dementia in subjects who meet the criteria of Mild Cognitive Impairment since, according to these criteria, subjects can have a low-risk score of 3 (age below 65 years, high MMSE score, Clinical Dementia Rating scale (CDR) score of 0.5, one impairment on cognitive tests) up to a high-risk score of 8 (age above 75 years, low MMSE score, CDR score of 0.5, two impairments on cognitive tests) (Table 6). A limitation of the study was that information about the APOE genotype was not available for all subjects. This may have led to underestimation of the diagnostic accuracy in the 5-year FU sample because the APOE genotype was tested significantly less often in subjects with AD-type dementia at follow-up than in subjects without dementia at follow-up, while subjects with ADtype dementia had significantly higher scores on the APOE item than subjects without dementia at follow-up. Another limitation was that medial temporal lobe atrophy was not assessed in the 5-year FU sample. In addition, the follow-up period in the 2-year FU sample was short and the number of subjects in the 2-year FU sample was small. Furthermore, some of the subjects that were included in the present study were also included in 3 of the 21 studies that were used for the meta-analysis on the basis of which the PAS was constructed. This may have led to overestimation of the diagnostic accuracy in the present study. Finally, we did not investigate the PAS scores of subjects who were in the preclinical stage of non-ad type dementia because these other types of dementia were infrequently diagnosed in the study (in less than 2 % of the subjects). One subject who was eligible for the 2-year FU sample had frontal lobe dementia at follow-up. The PAS score in this subject was 3 but on the basis of this single case no conclusions can be made. The reason for the low incidence of non-ad type dementia in the present study may have been the exclusion criteria at baseline that may have excluded subjects at high risk for non-ad type dementia, or the relatively young age of the study population which has made, for example, vascular disorders less common. The study population had some specific characteristics which may limit the ability to generalize from the findings. First, the study was performed in the setting of a Memory Clinic which is a secondary and tertiary referral center. The non-demented subjects referred to it may be a specific subset of the population of the subjects with mild cognitive impairment and the results may therefore not apply to other settings because of possible referral and spectrum bias [8]. Second, the diagnostic accuracy may be different if subjects are selected according to other criteria of mild cognitive impairment (see above). Third, the positive predictive value of the PAS may be different in settings where the conversion rate to AD-type dementia is different. Fourth, the predictive accuracy may be less in samples where the average age is higher because, in general, predictors of conversion to AD-type dementia are less powerful in older subjects. We found a similar ROC curve of the PAS scores in a subsample of subjects from the 5- year FU sample older than 65 years (n=31, average age 71.9 years) compared with the ROC curve based on the whole 5-year FU sample (data not shown). This suggests that the PAS also has a good accuracy in older subjects, but the predictive accuracy in the oldest old remains to be investigated. The clinical applicability of the PAS may depend on the setting in which it is used. In the setting of a specialized Memory Clinic, the PAS may be helpful for the clinician to select subjects who should remain under clinical supervision (for example, if the PAS score is 4 (5 PAS items) or 6 (full PAS)) or subjects who may benefit from pharmacological treatment (for example, if the PAS score is 6 (5 PAS items) or 7 (full PAS)), once drugs for subjects with preclinical AD become available. The decision rules for stepwise scoring may be helpful to guide the diagnostic investigations in order to reduce costs and the burden of diagnostic procedures to the patient, although some physicians in the setting of a specialized Memory Clinic will possibly not be inclined to abandon cognitive testing or neuroimaging on the basis of these decision rules. In less specialized clinical settings, where some of the PAS items can not be scored, a shortened version of the PAS may be useful to identify subjects with preclinical AD, although the diagnostic accuracy will then be less. In general practice or in an epidemiological setting, the first step of the PAS may be used to select subjects with cognitive complaints who should be referred for further evaluation. This retrospective study showed that the PAS may be a useful tool in a setting of a memory clinic for the assessment of relatively young subjects with mild cognitive impairment in order to identify subjects with preclinical AD. Future studies should investigate the diagnostic accuracy of the PAS in other settings. In addition, it should be investigated whether the PAS can detect subjects with preclinical AD who become demented at longer follow-up intervals and what is the PAS score of subjects who are in the preclinical stage of other types of dementia.

319 Acknowledgements The authors thank Nico Roozendaal for help in the data collection and data management, Sally Serneels for performing the APOE genotyping, and Dr Jane Sykes for her linguistic advice. The APOE genotyping was supported by a grant of the Fund of Scientific Research Flanders (FWO) and a special research grant of the University of Antwerp (UIA). Marc Cruts is a postdoctoral fellow of the FWO. References 1. American Psychiatric Association (1994) Diagnostic and statistical manual of mental disorders. American Psychiatric Association, Washington DC 2. Blessed G, Tomlinson BE, Roth M (1968) The association between quantitative measures of dementia and of senile changes in the cerebral grey matter of elderly subjects. Br J Psychiatry 114:797 811 3. Brand N, Jolles J (1985) Learning and retrieval rates of words presented auditorily and visually. J Gen Psychol 112:201 210 4. Brand N, Jolles J (1987) Information processing in depression and anxiety. Psychol Med 17:145 153 5. Brandt J, Spencer M, Folstein M (1988) The Telephone Interview for Cognitive Status. Neuropsychiat Neuropsychol Behav Neurol 1:111 117 6. de Leon MJ, George AE, Convit A, et al. (1993) The radiologic prediction of Alzheimer disease: the atrophic hippocampal formation. AJNR 14:897 906 7. Folstein MF, Folstein SE, McHugh PR (1975) Mini-Mental State, a practical method for grading the cognitive state of patients for the clinician. J Psychiatr Res 12:189 198 8. Gifford DR, Cummings JL (1999) Evaluating dementia screening tests. Methodological standards to rate their performance. Neurology 52:224 227 9. Hamilton M (1960) A rating scale for depression. J Neurol Neurosurg Psychiatry 23:56 62 10. Hughes CP, Berg L, Danziger WL, Coben LA, Martin RL (1982) A new scale for the staging of dementia. Br J Psychiatry 140:566 572 11. Jack CR, Petersen RC, Xu YC, et al. (1999) Prediction of AD with MRIbased hippocampal volume in mild cognitive impairment. Neurology 52:1397 1403 12. Lezak M (1995) Neuropsychological Assessment. Oxford University Press, New York 13. Linn R, Wolf P, Bachman D, et al. (1995) The preclinical phase of probable Alzheimer s disease. Arch Neurol 52:485 490 14. Luteyn F, van der Ploeg FAE (1983) Groninger Intelligentie Test. Swets en Zeitlinger, Lisse, The Netherlands 15. McKhann G, Drachman D, Folstein M, Katzman R, Price D, Stadlan EM (1984) Clinical diagnosis of Alzheimer s disease: report of the NINCDS-ADRDA Work-Group under the auspices of the Department of Health and Human Services Task Force on Alzheimer s disease. Neurology 34:939 944 16. Morris J (1993) The Clinical Dementia Rating (CDR): Current version and scoring rules. Neurology 43:2412 2414 17. Petersen R, Smith G, Waring S, Ivnik R, Tangalos E, Kokmen E (1999) Mild cognitive impairment. Clinical characterization and outcome. Arch Neurol 56:303 308 18. Reisberg B, Ferris SH, De Leon MJ, Crook T (1982) The global deterioration scale for assessment of primary degenerative dementia. Am J Psychiatry 139:1136 1139 19. Reitan R (1958) Validity of the Trail Making Test as an indication of organic brain damage. Percept Motor Skill 8:271 276 20. Roth M, Tym E, Mountjoy CQ, et al. (1986) A standardized instrument for the diagnoses of mental disorder in the elderly with special reference to the early detection of dementia. Br J Psychiatry 149:698 709 21. Scheltens P, Leys D, Barkhof F, et al. (1992) Atrophy of medial temporal lobes on MRI in probable Alzheimer s disease and normal ageing: diagnostic value and neuropsychological correlates. J Neurol Neurosurg Psychiatry 55:967 972 22. Slooter AJC, Cruts M, Kalmijn S, et al. (1998) Risk estimates of dementia by apolipoprotein E genotypes from a population-based incidence study: The Rotterdam study. Arch Neurol 55:964 968 23. Stroop J (1935) Studies of interference in serial verbal reactions. J Exp Psychol 18:643 662 24. Tierney MC, Szalai JP, Snow WG, et al. (1996) A prospective study of the clinical utility of ApoE genotype in the prediction of outcome in patients with memory impairment. Neurology 46:149 154 25. Tuokko H, Vernon-Wilkinson R, Weir J, Beattie B (1991) Cued recall and early identification of dementia. J Clin Exp Neuropsychol 13:871 879 26. Verhey FRJ, Jolles J, Ponds RWHM, et al. (1993) Diagnosing dementia: a comparison between a monodisciplinary and multidisciplinary approach. J Neuropsychiatry Clin Neurosci 5:78 85 27. Visser PJ (2000) Predictors of Alzheimer type dementia in subjects with mild cognitive impairments. Neuropsych Publishers, Maastricht 28. Visser PJ, Krabbendam L, Verhey FRJ, et al. (1999) Brain correlates of memory dysfunction in alcoholic Korsakoff s syndrome. J Neurol Neurosurg Psychiatry 67:774 778 29. Visser PJ, Scheltens P, Verhey FRJ, et al. (1999) Medial temporal lobe atrophy and memory dysfunction as predictors for dementia in subjects with mild cognitive impairment. J Neurol 246:477 485 30. Visser PJ, Verhey FRJ, Kester A, Jolles J (2001) Predictors of AD-type dementia in subjects with mild cognitive impairment A meta-analysis [abstract]. Int Psychogeriatr 13 (Suppl. 2):S140 31. Visser PJ, Verhey FRJ, Ponds RWHM, Cruts M, Van Broeckhoven CM, Jolles J (2000) Course of objective memory impairment in non-demented subjects attending a memory clinic and predictors of outcome. Int J Geriatr Psychiatry 15:363 372 32. Visser PJ, Verhey FRJ, Ponds RWHM, Jolles J (in press) Characteristics of preclinical Alzheimer s disease. Int J Geriatr Psychiatry 33. Visser PJ, Verhey FRJ, Ponds RWHM, Kester A, Jolles J (2000) Distinction between preclinical dementia and depression. J Am Geriatr Soc 48:479 484 34. Wechsler D (1955) Wechsler Adult Intelligence Scale. Manual. Psychological Corporation, New York