Sample

We retrospectively analysed 187 consecutive referrals to the Edinburgh Neuropsychological Assessment Service for Older People between the months of September 2004 and April 2006. Referrals were received at a tertiary level, stemming from consultants in old age psychiatry, geriatric medicine and neurology. All of these patients had undergone comprehensive psychiatric evaluation, relevant medical screening (including a standard battery of screening blood tests) and neuroimaging (computed tomography and/or magnetic resonance imaging or single-proton emission computed tomography) prior to being referred to our service. All but three patients were over the age of 50 years.

The original criteria for mild cognitive impairment set out by Petersen et al require that a person must present with a memory complaint, show evidence of objective memory decline in relation to age and education, demonstrate preservation of other areas of cognitive function and activities of daily life, and not fulfil criteria for dementia. ^{Reference Petersen, Stevens, Ganguli, Tangalos, Cummings and DeKosky8} Because it has since become apparent that not everyone who demonstrates cognitive impairment short of dementia has a ‘memory’ complaint, we used the recently expanded criteria that include people with non-memory complaints (single-domain non-memory MCI), as well as those exhibiting multiple domains of cognitive impairment who none the less fail to fulfil criteria for dementia (multiple domains slightly impaired). ^{Reference Petersen9,Reference Petersen and Morris10} The Mini Mental State Examination (MMSE) ^{Reference Folstein, Folstein and McHugh11} and Addenbrooke's Cognitive Examination ^{Reference Mathuranath, Nestor, Berrios, Rakowicz and Hodges12} were administered as a means of establishing the participants' general level of cognitive functioning. Level of everyday functioning was examined by means of the Clinical Dementia Rating scale ^{Reference Morris13} within the context of a clinical interview with the patient and the patient's primary carer (when available). A total of 112 patients fulfilled one or more of the following exclusion criteria and were therefore excluded from the analyses:

1. (a) dementia (MMSE score <24/30 or ACE score <80/100, plus fulfilling DSM–IV criteria; ¹⁴

2. (b) depression, assessed either by way of formal psychiatric consultation or, in a small proportion of cases, by a score greater than 10 on the Geriatric Depression Scale ^{Reference Yesavage, Brink, Rose, Lum, Huang, Adey and Leirer15} or clinical assessment by one of the authors (J.A.L.);

3. (c) one or more medical or psychiatric conditions that could conceivably account for the patient's cognitive impairment (head injury, schizophrenia, evidence of stroke or tumour on neuroimaging, alcoholism, epilepsy, cranial radiotherapy).

Of the remaining 75 patients, 15 showed cognitive impairments outside the domain of episodic memory, 15 returned a ‘normal’ cognitive profile and 45 showed memory function impaired for age (with or without additional areas of cognitive impairment). We present detailed neuropsychological baseline findings for 33 of these 45 patients as well as for 21 healthy individuals from the community who agreed to participate as a control group in a continuing longitudinal study examining neuropsychological markers of preclinical dementia. Control group participants were recruited through a local dementia support group or were spouses or carers of patients who had attended the neuropsychological assessment service.

Neuropsychological assessment measures

All the participants were given a comprehensive battery of neuropsychological tests. These tests were selected on the basis of their demonstrated validity for use within a population with MCI, and assessed the primary domains of verbal and visual episodic memory, semantic memory and language, processing speed, attention/executive function and visuospatial ability.

Comparison with other memory clinics

We searched the literature for studies employing neuropsychological test batteries similar to ours to examine the comparability of our sample with other published data.

Statistical analysis

We calculated z-scores to determine where scores fell below the tenth percentile of control performances. Visual inspection together with a one-sample Kolmogorov–Smirnov goodness-of-fit test indicated that the data were normally distributed. Group means were compared using independent sample t-tests. To determine whether there is an association between general level of cognitive function and the consistency of episodic memory impairment, we divided the participants with cognitive impairment into two groups: those who displayed episodic memory impairment on a single measure only and those who showed impairment on two or more episodic memory tests. Their Addenbrooke's Cognitive Examination and MMSE scores were compared using an independent samples t-test. Diagnostic categories between memory clinics were compared using chi-squared tests for diagnostic categories.

Literature search

We identified one other study reporting consecutive referrals to a memory clinic using similar diagnostic criteria and assessment measures. ^{Reference Alladi, Arnold, Mitchell, Nestor and Hodges2}

Sample characteristics

Our final sample for analysis consisted of 33 patients with aMCI (13 men and 20 women, with a mean age of 74.0 years, s.d.=6.4; social class I n=9, II n=12, IIIN n=9, social class not known n=3) and 21 healthy community-dwelling older adults without cognitive complaints (7 men and 14 women, with a mean age of 69.5 years, s.d.=7.4). These groups did not differ in terms of estimated premorbid level of intellectual function. The mean age of our control group was, however, significantly lower than that of our aMCI patient group, a finding similar to previous reports. ^{Reference Alladi, Arnold, Mitchell, Nestor and Hodges2,Reference Hodges, Erzinçlioglu and Patterson4}

Comparison of aMCI and control groups

Despite a mean ACE score that exceeded suggested cut-off points for dementia, ^{Reference Mathuranath, Nestor, Berrios, Rakowicz and Hodges12} the aMCI group had significantly lower mean scores than the control group on all neuropsychological measures, with the exception of the Rey Complex Figure Test copy task, the Trail Making Test part A and the letter (phonemic) fluency task. These findings were confirmed when the analysis was re-run with age as a covariate, with the exception of performance on the CANTAB PAL and the Trail Making Test part B, which just failed to reach significance. Demographic data and mean scores for the two groups on the individual neuropsychological measures are provided in Table 1. All aMCI group means are based on data from 33 participants, with the exception of the final three semantic memory measures, for which the participant numbers ranged between 13 and 31.

Table 1

Demographic data and performance of the sample on our neuropsychological test battery

	Control group (n=21) Mean (s.d.)	aMCI group (n=33) Mean (s.d.)	Control v. aMCIa P
Age, years	69.8 (7.4)	74.0 (6.4)	0.02*
NART score	118.3 (2.8)	116.4 (8.5)	0.331
MMSE score (30)b	29.1 (0.8)	28.0 (1.8)	<0.01**
ACE score
Total (100)	94.8 (3.3)	88.2 (5.9)	<0.001***
Delay (7)	6.4 (0.9)	3.9 (2.3)	<0.001***
Rey Complex Figure Test scores
Copy (36)	34.0 (2.4)	34.2 (2.5)	0.729
Immediate recall (36)	19.1 (6.3)	12.3 (6.0)	<0.001**
Delayed recall (36)	17.4 (7.4)	10.8 (6.9)	<0.01**
HVLT—R score
Total recall (12)	24.3 (5.0)	18.2 (4.6)	<0.001***
Delayed recall (12)	8.19 (2.8)	4.8 (3.2)	<0.001**
Discrimination (12)	10.3 (1.9)	8.5 (2.2)	<0.01**
PAL 6 errors	7.8 (6.9)	16.6 (14.8)	<0.01**
Trail Making Test score
Part A, s	40.3 (11.2)	48.4 (20.0)	0.095
Part B, s	87.5 (31.6)	131.7 (78.4)	<0.01**
Animal fluency score	21.1 (5.7)	15.1 (4.5)	<0.001***
P words score	15.7 (5.8)	16.0 (4.8)	0.860
Boston Naming Test score (60)	57.4 (3.1)	53.6 (5.5)	<0.01**
EENT score (50)	46.8 (3.0)	43.6 (4.5)	<0.01**
Graded Faces Naming Test score (30)	20.8 (3.1)	16.3 (4.9)	<0.001***
Graded Naming Test score (30)	23.4 (3.2)	19.5 (4.2)	0.015*

Amnestic MCI group performance

Referral patterns

In this study we have shown that people who are neither depressed nor demented but who fulfil Petersen's expanded criteria ^{Reference Petersen9} for aMCI make up a significant proportion of referrals to our Old Age Clinical Neuropsychology service. Roughly a quarter of all patients referred during an 18-month period met Petersen's criteria, of whom a minority exhibited a memory deficit in isolation following comprehensive neuropsychological examination. This is an almost identical proportion of patients to that reported in the study by the Cambridge Memory Clinic (23%), ^{Reference Alladi, Arnold, Mitchell, Nestor and Hodges2} and is a very similar figure to the 20% of patients seeking help in Lehrner et al's clinic who met criteria for aMCI. ^{Reference Lehrner, Gufler, Guttmann, Maly, Gleiss, Auff and Dal-Bianco25} Furthermore, the observation that the highest numbers of people with MCI were those with more than one domain of cognitive deficit is in keeping with the findings from both population-based and other memory clinic studies. ^{Reference Loewenstein, Acevedo, Ownby, Agron, Barker, Isaacson, Strauman and Duara7,Reference Lopez, Jagust, DeKosky, Becker, Fitzpatrick, Dulberg, Breitner, Lyketsos, Jones, Kawas, Carlson and Kuller26}

It is possible that referral patterns might differ depending on whether the memory clinics are geriatrician- or psychiatrist-led. Surprisingly, the level at which referrals are received (whether primary, as in the Cambridge Memory Clinic, or tertiary, as in our clinic) appears not to influence the proportion of referrals of an MCI nature that are received.

Thus, it appears that both the concept and criteria are applicable and, indeed, a necessary adjunct to clinical practice.

Neuropsychological profile of aMCI

The applicability of Petersen's research criteria ^{Reference Grundman, Petersen, Ferris, Thomas, Aisen, Bennett, Foster, Jack, Galasko, Doody, Kaye, Sano, Mohs, Gauthier, Kim, Jin, Schultz, Schafer, Mulnard, van Dyck, Mintzer, Zamrini, Cahn-Weiner and Thal1} to clinical practice has recently been challenged on the grounds of exclusion of a significant number of patients who display episodic memory impairment of a visual nature only. ^{Reference Alladi, Arnold, Mitchell, Nestor and Hodges2} In contrast to the results from the Cambridge Memory Clinic reported by Alladi et al, ^{Reference Alladi, Arnold, Mitchell, Nestor and Hodges2} around half of the participants with aMCI in our study showed impairment of both verbal and visual memory, whereas all of the remaining participants with aMCI exhibited memory impairment of a verbal nature only. That is to say, we failed to uncover any case of isolated visual memory impairment.

A recent study ^{Reference Loewenstein, Acevedo, Ownby, Agron, Barker, Isaacson, Strauman and Duara7} supports the notion that the type of episodic memory measure used may affect whether or not impairments are detected – a point we will return to later in the discussion. However, the absence of participants showing episodic memory impairment of a solely visual nature in our sample cannot be readily explained in terms of differential test sensitivities, because nearidentical neuropsychological measures were used in previous studies (e.g. that of Alladi et al ^{Reference Alladi, Arnold, Mitchell, Nestor and Hodges2} ) to assess visual memory function. Only a small proportion of our patients with aMCI demonstrated impairment on the visual episodic memory tasks per se. Administrative procedures might go some way to explain this observation. Specifically, our inclusion of an immediate Rey Complex Figure Test recall trial might have resulted in higher delay scores, ^{Reference Lezak, Howieson, Loring, Hannay and Fischer27} thus serving to reduce the sensitivity of this measure in our aMCI group.

Findings of studies examining patients who are at risk of developing Alzheimer's disease suggest that measures of verbal episodic memory are most sensitive to changes early in the disease course, followed by measures of visual memory. ^{Reference Collie and Maruff28} It is therefore conceivable, taking into account our aMCI group's higher mean score on the Addenbrooke's Cognitive Examination, that our sample contained a greater number of patients who were at an earlier stage of their disease course. Longitudinal follow-up, in particular observation of annual performances on these visual episodic memory measures, will determine whether this is indeed the case.

Several studies have drawn attention to the substantial variability in MCI case definition as a function of the specific neuropsychological tests used. ^{Reference Alladi, Arnold, Mitchell, Nestor and Hodges2,Reference Loewenstein, Acevedo, Ownby, Agron, Barker, Isaacson, Strauman and Duara7} Consistent with this, in our study there was variability among the aMCI group as to which and how many episodic memory measures were impaired. This finding was previously demonstrated, ^{Reference Alladi, Arnold, Mitchell, Nestor and Hodges2} and highlights the inherent difficulty in specifying the use of any single measure as a means of establishing impaired episodic memory function in aMCI. Our aMCI sample could be roughly divided into thirds in terms of numbers of participants exhibiting impairment on one, two and three or more episodic memory measures. A similar breakdown in numbers has been previously reported. ^{Reference Alladi, Arnold, Mitchell, Nestor and Hodges2} It would appear entirely reasonable and indeed a matter of good clinical practice to seek to establish consistency in performance across a range of episodic memory measures in defining aMCI and it will be of interest to see whether this is a significant determinant of outcome.

The variability in case definition of aMCI as a function of the cognitive measures employed, coupled with the inherent difficulties in specifying the use of a single common measure in the evaluation of this condition, poses a major challenge for clinicians. Our findings suggest that employing Petersen's expanded criteria ^{Reference Petersen9} for MCI could conceivably lead to a patient's condition being classified as single-domain aMCI, multiple-domain aMCI or ‘worried well’, depending on the cognitive measures that were employed. If the MCI subdivisions prove useful in a prognostic sense, the means by which the cognitive aspects of the criteria are put into operation by clinicians will require further clarification.

Mean scores on cognitive screening measures were significantly lower for participants showing impairment on more than one episodic memory measure. This may reflect a more advanced disease course in this group. It is also possible that the single-measure impairment group will prove to be a less stable one over time, with a number of patients returning normal neuropsychological profiles when tested again at a later date. Alternatively, in cases in which participants show impairment on a single verbal memory measure only, this might have arisen secondarily to impairment in another cognitive domain, for example expressive language or attention/executive function (in which case the person's condition might be more accurately conceptualised as non-amnestic MCI). These possibilities and the prognostic implications of consistency and pervasiveness of impaired episodic memory performances remain to be examined by way of longitudinal follow-up.

Our study adds to the growing body of evidence supporting the rarity of a pure amnestic MCI syndrome, ^{Reference Alladi, Arnold, Mitchell, Nestor and Hodges2,Reference Tabert, Manly, Liu, Pelton, Rosenblum, Jacobs, Zamora, Goodkind, Bell, Stern and Devanand5,Reference Kramer, Nelson, Johnson, Yaffea, Glenna, Rosena and Millera6} and demonstrates that additional impairment often goes unnoticed unless participants undergo thorough neuropsychological assessment. Among 33 patients with aMCI, only 10 (30%) presented with isolated memory impairment. This figure is well within the range of previously reported rates. For example, Tabert and colleagues found that, following comprehensive neuropsychological assessment, only 19% of their aMCI cases were categorised as pure aMCI, ^{Reference Tabert, Manly, Liu, Pelton, Rosenblum, Jacobs, Zamora, Goodkind, Bell, Stern and Devanand5} whereas this figure reached 35% in the study by Alladi et al. ^{Reference Alladi, Arnold, Mitchell, Nestor and Hodges2} It should be borne in mind, however, that the rate of cases with purely amnestic MCI will vary in accordance with how impairment is defined. For example, Kramer and colleagues showed that the number of cases classified as pure aMCI was considerably higher (27%) when a cut-off of 1.5 s.d. below the mean, as opposed to 1 s.d. (resulting in a 5% rate), was used. ^{Reference Kramer, Nelson, Johnson, Yaffea, Glenna, Rosena and Millera6} It therefore remains a possibility that our less stringent definition of impairment (i.e. 1 s.d. below the mean performance of the healthy control group) might have resulted in an overestimation of the frequency of cases with non-pure aMCI. Identifying accompanying non-memory cognitive impairment none the less appears important in light of recent evidence indicating a higher risk of conversion to Alzheimer's disease in patients with aMCI who show additional areas of cognitive impairment compared with patients with pure aMCI. ^{Reference Tabert, Manly, Liu, Pelton, Rosenblum, Jacobs, Zamora, Goodkind, Bell, Stern and Devanand5}

The results of our study are also consistent with evidence indicating accompanying semantic memory impairment in aMCI, ^{Reference Grundman, Petersen, Ferris, Thomas, Aisen, Bennett, Foster, Jack, Galasko, Doody, Kaye, Sano, Mohs, Gauthier, Kim, Jin, Schultz, Schafer, Mulnard, van Dyck, Mintzer, Zamrini, Cahn-Weiner and Thal1–Reference Tabert, Manly, Liu, Pelton, Rosenblum, Jacobs, Zamora, Goodkind, Bell, Stern and Devanand5} with just 10 patients of 33 exhibiting episodic memory impairment in isolation, and 22 of the remaining 23 displaying evidence of semantic memory compromise. This finding may reflect an increased risk of conversion to Alzheimer's disease from aMCI, although early semantic memory failure is by no means specific to the former disease, ^{Reference Clague, Dudas, Thompson, Graham and Hodges29,Reference Graham, Emery and Hodges30} and although some studies report prognostic significance of performance on semantic memory measures, ^{Reference Blackwell, Sahakian, Vesey, Semple, Robbins and Hodges18,Reference Artero, Tierney, Touchon and Ritchie31,Reference Estevez-Gonzàlez, Garcia-Sànchez, Boltes, Otermín, Pascual-Sedano, Gironell and Kulisevsky32} others have failed to do so. ^{Reference Beatty, Salmon, Troster and Tivis33} The stage at which impairments in this domain become apparent does appear to vary in accordance with the sensitivity of the measure employed. ^{Reference Hodges, Erzinçlioglu and Patterson4} The intact performance of participants with aMCI on measures of lexical (letter) fluency, also previously reported, ^{Reference Alladi, Arnold, Mitchell, Nestor and Hodges2,Reference Hodges, Erzinçlioglu and Patterson4} suggests that the ‘initiation’ aspects of semantic fluency tasks do not pose any difficulty to patients with this impairment subtype.

In view of the sound mean performances of our participants with aMCI on cognitive screening measures (MMSE score 28/30, Addenbrooke's Cognitive Examination score 88/100), it seems unlikely that consideration of such scores will be of any value in ruling out the presence of additional domains of cognitive impairment. Reliance on clinical judgement to determine the presence or absence of additional domains of subtly impaired cognition is similarly likely to prove difficult when dealing with patients with above-average premorbid IQ scores who are performing at sound levels on cognitive screens. Taken together, the above observations raise the question of whether global screening measures coupled with clinical judgement are a sufficient means of investigating MCI, and if not, whether additional resources or an expanded skill base will be required to handle this population clinically.

Our results reveal an absence of any significant difference in performance between the aMCI and control groups on measures of visuospatial function and processing speed. In-depth longitudinal evaluation of neuropsychological performance in MCI and questionable dementia suggests that visuospatial functions tend to fail secondarily to episodic memory and category fluency performances, ^{Reference Hodges, Erzinçlioglu and Patterson4,Reference Perry and Hodges34} although some heterogeneity is known to exist. ^{Reference Caine and Hodges35} It is therefore once again possible that our failure to demonstrate group differences on a visuospatial copying task reflects an earlier disease stage of our aMCI sample. Alternatively, it is conceivable that the varied and somewhat subjective scoring methods for the Rey Complex Figure Test copy task across different studies might be responsible for this finding. Cross-sectional findings pertaining to visuomotor processing speed in MCI vary, with some studies reporting significant differences between MCI and control groups ^{Reference Grundman, Petersen, Ferris, Thomas, Aisen, Bennett, Foster, Jack, Galasko, Doody, Kaye, Sano, Mohs, Gauthier, Kim, Jin, Schultz, Schafer, Mulnard, van Dyck, Mintzer, Zamrini, Cahn-Weiner and Thal1,Reference Arnaiz, Almkvist, Ivnik, Tangalos, Wahlund, Winblad and Petersen36,Reference Nordlund, Rolstad, Hellstrom, Sjogren, Hansen and Wallin37} and others, like ourselves, failing to do so. ^{Reference Albert, Moss, Tanzi and Jones38–Reference Fox, Warrington, Seiffer, Agnew and Rossor40} The disparity in findings may simply reflect the heterogeneity of aMCI or alternatively the disease stage. Group differences in processing speed might be more likely to exist where samples contain significant numbers of patients in the preclinical stages of a subcortical dementia of a cerebrovascular nature. For example, there is some evidence to suggest a disproportionately strong association between perceptual speed and parkinsonian signs in MCI. ^{Reference Boyle, Wilson, Aggarwal, Arvanitakis, Kelly, Bienias and Bennett41}

Limitations

Several study limitations should be noted. The significantly higher mean age of our patient group opens up the possibility that some of their performance deficits were explicable in terms of age-related cognitive decline. Ideally, control for age should have been better. However, aMCI group participants were identified on the basis of their performance on age-standardised tests; therefore, the discrepancy would not have influenced patient group membership. Longitudinal follow-up of these patients will help to clarify the relevance of this difference. Furthermore, our aMCI sample was characterised by a high average level of estimated premorbid general intellectual function, which introduces problems of applicability. Similar issues were present in a recent comparable study, ^{Reference Alladi, Arnold, Mitchell, Nestor and Hodges2} although other socio-demographic characteristics (i.e. gender, ethnicity, education and occupation) were not reported, preventing further comparison between that and our study. There may therefore be a need to replicate these findings employing greater numbers of age- and IQ-matched healthy controls and aMCI patients with average premorbid IQs, together with other socio-demographic markers more closely resembling the population mean.

Implications

Patients with MCI make up a significant number of referrals to older adult memory assessment services, with the most common referral subtype in our sample being that of aMCI, followed by equal numbers of non-amnestic MCI and worried well. Relatively few people with aMCI exhibit episodic memory compromise in isolation and fewer still show a visual but not verbal episodic memory deficit. Both the concept and criteria for MCI therefore appear to be relevant and indeed necessary adjuncts to clinical practice.

Our findings highlight the inherent difficulties of specifying a single measure in the assessment of memory and other cognitive functions in MCI, while at the same time emphasising the need for clarification of the means by which MCI criteria can be put into operation clinically. Initial attempts at better defining neuropsychological aspects of the aMCI criteria have been made, ^{Reference Grundman, Petersen, Ferris, Thomas, Aisen, Bennett, Foster, Jack, Galasko, Doody, Kaye, Sano, Mohs, Gauthier, Kim, Jin, Schultz, Schafer, Mulnard, van Dyck, Mintzer, Zamrini, Cahn-Weiner and Thal1} but their application in a clinical sense remains inconsistent and their poor definition has not gone unnoticed. ^{Reference Portet, Ousset, Visser, Frisoni, Nobili, Scheltens, Vellas and Touchon42} The existence of a number of neuropsychological measures of well-documented sensitivity in aMCI and the strikingly similar mean performances of different clinic aMCI groups on such measures suggest that this need not be the case. Although the importance of exercising clinical judgement in arriving at a diagnosis of MCI cannot be ignored, it would none the less seem inevitable that further definition of the neuropsychological aspects of MCI criteria will be needed to facilitate identification of the subtypes of impairment and to further our understanding of their respective prognoses.