Persistence of Neuropsychological Testing Deficits in Mild Cognitive Impairment

Introduction

Mild cognitive impairment (MCI) often represents a transitional stage between normal aging and Alzheimer's disease (AD) [1]. Subjects meeting diagnostic criteria for MCI progress to AD at significantly higher rates than normal elderly controls (10% vs. 1 to 2% per year) [2, 3]. These findings have led to increased efforts to identify subjects with MCI, particularly since therapeutic interventions initiated in this population may potentially prevent or delay progression to clinical AD [1].

These initiatives have been complicated by the variability in reported rates of progression from MCI to AD, which range from 2 to 32% per year [4, 5]. Of equal concern is the variability in rates of reversion from MCI to normal cognition, which range from 4 to 53% [6, 7]. High reversion rates in MCI indicate that many subjects may not have incipient AD and its utility for predicting progression to dementia reduce.

A number of factors are likely to contribute to the variance in reported rates of progression and reversion. Population-based samples [4, 7] often have lower rates of progression to AD than memory clinic-based samples [5, 8], which may include a higher percentage of subjects with preclinical AD [9]. Rates of progression may also depend on the length of follow-up, which can range from 1 to 10 years [10, 11]. Calculations of annual rates of progression typically assume linear rates of progression over time [2], but the risk of progression may be greatest over the first 18 months of follow-up [12].

Other potentially significant sources of variability are differences in the operational definition of MCI between individual studies [9]. The Petersen criteria establish general guidelines for the diagnosis of MCI: (1) subjective cognitive complaint, (2) essentially intact activities of daily living, (3) objective cognitive impairment, and (4) not demented [1]. Unfortunately, the operationalization of these criteria, particularly for objective cognitive impairment, differs widely across research groups.

Individual investigators may rely on clinician judgment [13], global assessment tools [14, 15], and/or formal neuropsychological testing [7, 16] to diagnose MCI [1]. Petersen's original cohort of MCI subjects averaged 1.5 standard deviations (SD) below age-adjusted normative means on memory testing [17]. Most subsequent studies have specified performance ≥1.5 SD below the mean as a threshold for impairment [7, 16, 18], but others have used cutoffs of ≥1.0 SD [19, 20]. Individuals diagnosed with MCI may need to demonstrate neuropsychological deficits in single tests [7, 21, 22], multiple tests [16], or averaged across cognitive domains [18]. Subjects subsequently diagnosed with AD are most frequently characterized by deficits in episodic memory, executive function, and perceptual speed [23]. However, the specific tests that best predict progression to AD vary across studies [21, 22,24,25,26,27,28], and subjects demonstrate differential impairments on tests putatively assessing the same cognitive domains [29].

Since MCI subjects with incipient AD exhibit consistent cognitive impairments [30, 31], another approach to optimizing the operationalization of the criterion for objective cognitive impairment is to focus on deficits that remain persistent across longitudinal assessments. The exploratory analyses presented here sought to determine whether commonly used neuropsychological tests effectively identify persistent cognitive impairments, and whether increasing the stringency of the operational definition of MCI improves the persistence of identified impairments. We were specifically interested in whether subjects fulfilling different arbitrary criteria for impaired performance on cognitive testing at baseline continued to exhibit similar impairments at follow-up assessment.

Methods

Results

Discussion

The data presented here demonstrate that the use of different neuropsychological tests and thresholds for impairment in the diagnosis of MCI identifies subsets of subjects with different rates of persistent cognitive impairment. The prevalence of impaired performance varied widely across individual tests. Amnestic deficits were more persistent than non-amnestic deficits. Baseline deficits in VR-II and the CVLT were the best predictors of persistent memory impairment. Subjects who at baseline were impaired on multiple memory tests or had poorer overall memory performance were more likely to exhibit persistent memory deficits. Our work confirms and extends previous findings indicating that different methods of operationalizing ‘objective cognitive impairment for age and education’ for the Petersen criteria result in the identification of heterogeneous and incompletely overlapping subject populations [29].

The tests included in our neuropsychological battery exhibited significant variability in their ability to detect deficits, implying that the different assessments used to define MCI impact the composition of individual study populations. In particular, inclusion criteria for several previous clinical trials for MCI required impaired performance on delayed story recall tests similar to LM-II [50]. Although deficits in LM-II are likely to reflect persistent memory impairment, relatively few AMN subjects performed poorly on this test, which may reflect the more structured context in which information to be remembered is presented and/or the broad normative ranges reported for older subjects, particularly on the WMS-III, which was administered to the majority of our subjects [34, 35].

Cognitive deficits were more persistent in the AMN group than in the NON group. AMN subjects with deficits in ≥2 memory tests or average memory z-scores ≤−1.5 at baseline demonstrated higher frequencies of stable memory deficits. More stringent criteria for memory impairment may more specifically identify amnestic MCI subjects most likely to have incipient AD. However, increasing the number of impaired tests necessary to constitute a non-amnestic deficit did not appreciably improve the persistence of non-memory impairments. Our data are consistent with the lower rates of reversion to normal cognition [31, 51, 52] and higher rates of progression to AD [6, 15, 51, 53, 54] seen in amnestic MCI relative to non-amnestic MCI. Neuropathological studies indicate that neurofibrillary tangle deposition in medial temporal lobe structures suggestive of incipient AD is seen more frequently in amnestic MCI [55,56,57] than in non-amnestic MCI [58]. Non-memory testing in elderly subjects may be more susceptible to cognitive phenomena associated with normal aging, such as increased intra-individual variability in task performance [59].

There are several possible alternative explanations for the poorer persistence of non-amnestic deficits. Increasing the number of tests in a battery increases the likelihood that performance on at least 1 test will be in the impaired range [49]. The composition of our battery (4 memory tests vs. 10 non-memory tests) may have predisposed the NON group to include a greater proportion of false-positive MCI subjects than the AMN group, and the poorer persistence of deficits in the NON group may reflect regression to the mean. Nevertheless, with the notable exception of the RCFT-R, the persistence of individual test-specific deficits was consistently higher on memory tests than non-memory tests.

Subjects received essentially identical neuropsychological batteries at consecutive assessments; alternate versions were not used. Poorer persistence of deficits in the NON group might have been due to more substantial test-retest improvements in the non-memory tests. However, analyses of test-retest effects in the NC group using paired t tests revealed significant improvement in composite memory z-scores [baseline: mean = 1.18, SD = 0.60; follow-up: mean = 1.41, SD = 0.64; t(39) = 2.68; p = 0.011] but not on composite non-memory z-scores [baseline: mean = 0.87, SD = 0.37; follow-up: mean = 0.90, SD = 0.41; t(39) = 0.66; p = 0.51].

Although the importance of isolated non-amnestic deficits remains uncertain, the combined presence of both amnestic and non-amnestic deficits appears to be a marker of more advanced incipient AD. AMN-MD subjects were more likely to have persistent memory impairments than AMN subjects with isolated memory deficits, and more likely to have persistent non-memory deficits than NON subjects. These results are consistent with prior work demonstrating that progression rates to AD are highest in multiple-domain amnestic MCI [18, 27, 53, 54, 60, 61]. Alternatively, it remains possible that our baseline assessment of functional decline was insufficiently sensitive and some AMN-MD subjects were already mildly demented.

VR-II and the CVLT were the best predictors of persistent memory deficits. Our findings are consistent with previous work in MCI demonstrating that poorer performance on Visual Reproduction [21, 26] or delayed recall of word lists [22, 25] is associated with subsequent AD. Conversely, although some investigators have advocated the use of the RCFT-R for the identification of amnestic MCI [29, 62], and a significant proportion of AMN subjects exhibited baseline deficits in this task, the persistence of these deficits was poorer than that of the other memory measures, suggesting that the RCFT-R may be less useful for identifying incipient AD.

AMN subjects who performed poorly at baseline on the RCFT-R also performed poorly at baseline on the copy portion of the Rey Complex Figure Test [RCFT-C; χ²(1,62) = 5.06, p = 0.025], but exhibited similar improvements across assessments on both tasks. Poor performance on the RCFT-R may be due in part to deficits in visuospatial function, and poor persistence of RCFT-R deficits may mirror the poor persistence of visuospatial deficits in our MCI population. However, baseline visual memory deficits in both the RCFT-R and VR-II remained robust after correction for visuospatial performance (RCFT-C, Visual Reproduction I) using encoding scores [63].

Our administration of the RCFT-R incorporated a 3-min delay, which is shorter than the delays used in the LM-II, VR-II, and CVLT tasks. Although other studies of the RCFT-R in MCI populations have used longer delays [29, 62], performance on this task appears to be stable across delays ranging up to 60 min [45]. The RCFT-R also differs from the other memory tests in our battery, at least during subjects’ initial exposure, because it represents an incidental rather than intentional memory task. In subsequent administrations, the RCFT-R likely becomes an intentional memory task given subjects’ prior experience [64]. This may result in improved performance, particularly in older populations [65]. Impaired performance on the initial administration of this test may therefore be more indicative of poor attention and/or motivation than poor memory [64].

Since participants with MCI in this study were recruited from patients initially seen in the UCLA Memory Disorders Clinic, the base rate of underlying AD is likely to be higher in our sample than in the general elderly population [66]. Therefore, both the prevalence and persistence of deficits in our neuropsychological battery are also likely to be elevated in our MCI cohort relative to an unselected group of elderly individuals. Conversely, setting the threshold for impaired performance on individual tests at ≥1.5 SD below normative means results in approximately 7% of subjects with normal cognition appearing to exhibit deficits. The inclusion of such subjects in our MCI group may underestimate the true persistence of deficits in specific measures due to regression to the mean.

In evaluating for MCI, it is important to weigh the possible outcomes attached to false positive versus false negative errors. False positive diagnoses carry the potentially negative consequences associated with being ‘labeled’ with a cognitive disorder. Alternatively, if the criteria are excessively stringent, they may fail to identify subjects with incipient AD. Our data suggest that if a single memory test is used to diagnose amnestic MCI, VR-II, delayed recall of the CVLT, or similar visual or verbal memory paradigms achieve the optimal balance between these undesirable outcomes. The accuracy of clinical evaluations based on neuropsychological assessments may be further improved through the addition of biochemical and neuroimaging biomarkers [67]. The poor persistence of non-memory deficits in this study indicates that the optimal operationalization of non-amnestic MCI remains uncertain and further investigations are necessary to clarify the clinical significance of such impairments.

A number of factors may limit the interpretation of our results. Clinical diagnoses of MCI were not determined independently of subjects’ neuropsychological performance. Additionally, the cutoffs used to define impaired performance on each test were obtained from different published normative samples. Therefore, the relative prevalence of impairments on individual tests in our battery may not consistently reflect the utility of each test for diagnosing MCI. The interval between neuropsychological assessments was relatively short (16.4 months) and varied somewhat between individual subjects. However, the annualized progression rate from MCI to AD in the AMN group was 8.9%, comparable to that reported across other longitudinal studies [2], and reversion rates from MCI to normal cognition appeared to be relatively insensitive to length of follow-up both in the present study and in previous reports [68, 69]. Relatively few subjects were impaired on specific non-memory tests, raising the possibility that this study may be insufficiently powered to evaluate the persistence of deficits on these assessments. Nevertheless, persistence of non-memory deficits remained poor even when impairment in any test was considered. Practice effects and test-retest improvement may also have affected performance in our MCI subjects, potentially reducing the observed persistence of baseline cognitive deficits. However, previous work suggests that although normal aged control subjects demonstrate significant improvement when administered an identical verbal learning test over an interval of 2 to 3 years, subjects with MCI or AD do not [70]. Our study population was primarily Caucasian. There was a relatively higher proportion of non-Caucasian subjects in the NON group, consistent with previous work demonstrating poorer performance on non-memory neuropsychological testing in non-Caucasian samples [71, 72]. Thus, our results may not be generalizable to ethnically diverse populations, which may require assessment with specific community-based norms [73]. The NC group consisted of individuals who consistently demonstrated normal cognition to minimize the inclusion of subjects with incipient cognitive impairment. These criteria are likely to have selected for subjects with higher premorbid level of function and greater cognitive reserve, which may protect against cognitive decline and limit the generalizability of their results to a broader elderly population.

Our current findings add to the existing literature on the neuropsychological assessment of MCI by assessing the clinical utility of several common neuropsychological measures for the identification of persistent cognitive deficits. Ideally, these data will contribute to improving the operational definition of this heterogeneous condition. Additional work in larger populations of MCI subjects should focus on developing a consensus neuropsychological battery that yields predictable rates of progression to AD and low rates of reversion to normal cognition.