Artificially low mild cognitive impairment to normal reversion rate in the Alzheimer's Disease Neuroimaging Initiative

Abstract

INTRODUCTION:

We examined reasons for low mild cognitive impairment (MCI)-to-cognitively normal (CN) reversion rates in the Alzheimer’s Disease Neuroimaging Initiative (ADNI).

METHODS:

CN and MCI participants were identified as remaining stable, progressing, or reverting at one-year follow-up (Year 1). Application of ADNI’s MCI criteria at Year 1 as well as Alzheimer’s disease (AD) biomarkers by group were examined.

RESULTS:

The MCI-to-CN reversion rate was 3.0%. When specific components were examined, 22.5% of stable MCI participants had normal memory performance at Year 1 and their AD biomarkers were consistent with the stable CN group. At Year 1, when all MCI criteria were not met, the more subjective CDR rather than objective memory measure appeared to drive continuation of the MCI diagnosis.

DISCUSSION:

Results demonstrate an artificially low one-year MCI-to-CN reversion rate in ADNI-diagnosed participants. If the LM cutoffs had been consistently applied, the reversion rate would have been at least 21.8%.

1. Introduction

Mild cognitive impairment (MCI) represents a transitional stage between normal cognitive aging and the onset of dementia, including Alzheimer’s disease (AD) [1,2]. However, not all who are diagnosed with MCI exhibit progressive decline–many remain at the level of MCI and a significant portion (up to 30-50%) revert to a cognitively normal (CN) state, depending on diagnostic criteria [3,4]. Recent meta-analyses found reversion rates of 18% [5] and 24% [6], with community-based studies containing a greater proportion of “reverters” (31%) than clinically-based studies (14%) [6]. In stark contrast, one-year reversion rates within the Alzheimer’s Disease Neuroimaging Initiative (ADNI) were previously estimated at 2.2% [7].

ADNI’s MCI diagnostic criteria are relatively consistent with the conventional Petersen/Winblad criteria [7–9]. Specifically, ADNI relies upon subjective cognitive complaint, global cognitive screening, a self- and study-partner-informed global functioning interview, and one memory measure [7]. Alternative neuropsychological actuarial MCI criteria [10,11] have reliably identified a large proportion of individuals with “false positive” MCI diagnoses who, despite receiving a diagnosis of MCI in ADNI, exhibit biomarker profiles, neuropsychological performance, and functional trajectories most consistent with CN participants [11–16]. The susceptibility of these diagnostic errors and the lack of clear guidelines for the application of the individual components of ADNI’s MCI criteria at follow-up occasions suggest that the basis of the very low one-year MCI-to-CN reversion rate warrants further investigation.

Therefore, we comprehensively examined 1) one-year diagnostic stability and reversion rates in ADNI, 2) possible explanations for ADNI’s very low reversion rate, and 3) how stable MCI participants who would have been better classified as “reverters” differed from other stable MCI participants in terms of biomarkers, cognition, depressive symptoms, and rates of progression to dementia. We hypothesized, based on previous work in ADNI [7], that 1) we would replicate the low MCI-to-CN reversion rates with a larger ADNI sample, 2) diagnoses at the one-year follow-up exam (Year 1) would be driven more by clinical judgment (i.e., Clinical Dementia Rating) than objective memory performance, and 3) that those participants who were diagnosed as stable MCI despite having normal objective memory performance at Year 1 would have biomarker profiles and progression rates that are more similar to stable CN participants than stable MCI participants.

2. Methods

Data used in the preparation of this article were obtained from the ADNI database (adni.loni.usc.edu). ADNI was launched in 2003 as a public-private partnership. The primary goal of ADNI has been to test whether serial magnetic resonance imaging (MRI), positron emission tomography (PET), other biological markers, and clinical and neuropsychological assessment can be combined to measure the progression of MCI and early AD. For up-to-date information on ADNI, see www.adni-info.org. This study was approved by the Institutional Review Boards at each of the participating institutions, and written informed consent was obtained from all participants or authorized representatives at each site.

2.1. Participants

Specific enrollment criteria for ADNI have been described elsewhere [7]. All non-demented ADNI participants from ADNI 1, ADNI GO, and ADNI 2 cohorts who completed a baseline and one-year neuropsychological assessment were considered for analyses (total N=1,208). Table 1 shows the specific ADNI diagnostic criteria for distinguishing CN, MCI, and AD participants [7,17]. At baseline, all MCI participants were considered “amnestic” (single or multi-domain) except two participants who were considered “non-amnestic” MCI (one progressed to AD at Year 1, one remained MCI at Year 1).

Table 1.

Specific components of ADNI classification criteria to distinguish CN, Early MCI, Late MCI, and AD.

	CN	Early MCI	Late MCI	AD
Subjective complaint	None, aside from those common to other normal subjects of that age range.	Yes, by subject (verified by study partner) or study partner or clinician	Yes, by subject (verified by study partner) or study partner or clinician	Yes, by subject (verified by study partner) or study partner or clinician
MMSE*	≥24	≥24	≥24	20-26 (Inclusive)
Logical Memory	≥9 for 16+ years of education	9-11 for 16+ years of education	≤8 for 16+ years of education	≤8 for 16+ years of education
	≥5 for 8-15 years of education	5-9 for 8-15 years of education	≤4 for 8-15 years of education	≤4 for 8-15 years of education
	≥3 for 0-7 years of Education	3-6 for 0-7 years of education	≤2 for 0-7 years of education	≤2 for 0-7 years of education
CDR	CDR=0	CDR=0.5	CDR=0.5	CDR=0.5 or 1.0
	Memory Box score must be 0	Memory Box score of at least 0.5	Memory Box score of at least 0.5
General cognition and functional status	Cognitively normal, based on an absence of significant impairment in cognitive functions or activities of daily living.	General cognition and functional performance sufficiently preserved such that a diagnosis of Alzheimer’s disease cannot be made.	General cognition and functional performance sufficiently preserved such that a diagnosis of Alzheimer’s disease cannot be made.	NINCDS/ADRDA criteria for probable AD

Note. CN=Cognitively Normal; MCI=Mild Cognitive Impairment; AD=Alzheimer’s Disease; MMSE=Mini Mental State Exam; CDR=Clinical Dementia Rating; NINCDS/ADRDA=National Institute of Neurological and Communication Disorders and Stroke/Alzheimer’s Disease and Related Disorders Association.

^*MMSE exceptions may be made for subjects with less than 8 years of education at the discretion of the project director. Logical Memory (LM) impairment was based on education-adjusted cutoffs on one paragraph of the Logical Memory II subscale of the Wechsler Memory Scale—Revised (maximum score of 25). Late MCI criteria were the only MCI criteria for ADNI 1 phase; Early MCI criteria were only included in ADNI GO and ADNI 2 phases. This table was adapted from the procedure manuals for ADNI 1, ADNI GO, and ADNI 2 available at http://adni.loni.usc.edu/methods/documents/.

2.2. Procedure

2.2.1. Components of diagnostic criteria

For CN and MCI participants, cutoffs for the Mini Mental State Exam (MMSE; general cognition), Logical Memory II (LM; memory), and Clinical Dementia Rating (CDR; global function) were examined in more detail to determine whether these individual components were applied at Year 1 and which component(s) of the diagnostic criteria were prioritized when cut-offs for all three components were not met. For these analyses, the diagnostic criteria components/cutoffs were based on those shown in Table 1. For MCI participants, the Early MCI (E-MCI) LM cutoffs (rather than the Late MCI [L-MCI] cutoffs) were used because, depending on the phase of ADNI, the MCI LM cutoffs varied, and the E-MCI cutoffs are the most diagnostically inclusive (i.e., more MCI participants will meet the less-strict cutoffs than if the L-MCI cutoffs were used). Importantly, only the CDR component had fully independent cutoffs for CN and MCI participants since there was overlap between the CN and E-MCI LM component and complete overlap between CN and MCI on other criteria (e.g., MMSE) except for the subjective complaint.

2.2.2. Additional measures

A large subset of participants underwent a baseline lumbar puncture (CN n=289, MCI n=581). AD cerebrospinal fluid (CSF) marker positivity was examined by diagnostic group, and biomarkers were measured using Elecsys® immunoassays. Cut-off scores proposed by Hansson and colleagues [18] and optimized for ADNI were used to determine biomarker positivity: <976.6pg/ml for β-amyloid_1-42 (Aβ), >0.0251pg/ml for the hyperphosphorylated-tau (p-tau)/Aβ ratio, and >0.27pg/ml for the total-tau (t-tau)/Aβ ratio. Participants with at least one apolipoprotein E (APOE) ε4 allele were also considered at-risk (i.e., ε4-positive). The MMSE was used to measure global cognition, and the Geriatric Depression Scale (GDS) was used to measure depressive symptoms. The Rey Auditory Verbal Learning Test (AVLT) delayed recall score was used as a second measure of memory, in addition to LM, in post-hoc analyses.

2.3. Statistical Analyses

Baseline demographic and clinical characteristics by group (CN and MCI) were compared using independent t-tests or chi-squared tests. Group differences in AD biomarker positivity were examined using chi-squared tests. Group differences in LM were examined using a one-way analysis of variance (ANOVA), and MMSE and depressive symptoms were examined using Kruskal-Wallis tests; Bonferroni corrections were used for multiple comparisons. A Cox regression adjusting for age, sex, and education was used to determine the hazard ratio (HR) and 95% confidence interval (CI) of progression to dementia. In these analyses, time-to-dementia was the number of months from the baseline assessment to the assessment when the participant first met criteria for dementia. Participants who did not progress to dementia during their follow-up period (range of follow-up: 12-60 months) were censored at their last visit. Kaplan-Meier curves were used to show the rate of progression to dementia by group. A post-hoc mixed between-within ANOVA examining the change in memory scores (LM and AVLT) from baseline-to-Year 1 by group, with Bonferroni-corrected post-hocs, assessed for differential practice effects/regression to the mean.

3. Results

3.1. Baseline Characteristics

At baseline, there were 420 CN and 788 MCI participants based on ADNI diagnoses. CN and MCI diagnostic groups significantly differed from one another on all demographic, functional, neuropsychological, and biomarker data examined (Table 2).

Table 2.

Baseline demographic and clinical characteristics [mean (SD) or %] by group.

	CN (N=420)	MCI (N=788)	p-value
Age	74.70 (5.76)	73.18 (7.54)	<.001
Education	16.34 (2.67)	15.99 (2.81)	.036
Female, %	48.9%	40.8%	.007
GDS	0.76 (1.08)	1.68 (1.43)	<.001
MMSE	29.07 (1.14)	27.61 (1.80)	<.001
CDR=0.5, %	0.5%	99.7%	<.001
CDR Memory=0.5, %	0.6%	96.9%	<.001
FAQ	0.22 (0.94)	3.14 (4.06)	<.001
Logical Memory II (raw)	13.16 (3.30)	5.68 (3.42)	<.001
AVLT delayed recall (raw)	7.56 (3.89)	3.90 (3.86)	<.001
Aβ	1352.57 (659.56)	1013.88 (553.69)	<.001
p-tau/Aβ	0.02 (0.02)	0.04 (0.03)	<.001
t-tau/Aβ	0.23 (0.18)	0.36 (0.27)	<.001
p-tau	21.70 (8.91)	27.81 (14.19)	<.001
t-tau	236.39 (87.09)	287.19 (126.30)	<.001
APOE ε4+, %	27.6%	51.1%	<.001

Note. CN=Cognitively Normal; MCI=Mild Cognitive Impairment; GDS=Geriatric Depression Scale; MMSE=Mini-Mental State Exam; CDR=Clinical Dementia Rating; FAQ=Functional Activities Questionnaire; AVLT=Rey Auditory Verbal Learning Test; Aβ= β-amyloid; p-tau=hyperphosphorylated tau; t-tau=total tau APOE=Apolipoprotein E. For the CSF markers, the CN group had n=289 and MCI group had n=581.

3.2. Classification at Year 1

Examination of stability, progression, and reversion using ADNI’s reported diagnoses showed that among participants classified as CN at baseline (N=420), 403 (96.0%) remained CN, 17 (4.0%) progressed to MCI, and 0 (0%) progressed to dementia at Year 1. Of those who were classified as MCI at baseline (N=788), 661 (83.9%) remained MCI, 24 (3.0%) reverted to CN, and 103 (13.1%) progressed to dementia at Year 1. Stability, progression, and reversion rates for longer follow-up durations (e.g., Baseline-to-Year 2, Baseline-to-Year 3, etc.) for CN and MCI participants are included in Supplementary Tables 1 and 2.

3.3. Follow-up on ADNI criteria

3.3.1. Components of criteria

At baseline, 99.5% of CN participants met all three required components of the CN criteria that were examined. At Year 1, however, 14.5% of stable CN participants did not meet all three components of the criteria (1 component n=4, 1.1%; 2 components n=53, 13.5%). Table 3 shows the decomposition of the specific ADNI criteria components (MMSE, CDR, LM) at Year 1.

Table 3.

Specific components of the criteria met at Year 1 for Stable CN and Stable MCI groups.

	Stable CN (N=392)	Stable MCI (N=659)
	N / %	N / %
1 Component (Total)	4 / 1.1%	15 / 2.3%
MMSE only	0 / 0.0%	13 / 2.0%
CDR only	3 / 0.8%	1 / 0.2%
LM only	1 / 0.3%	1 / 0.2%
2 Components (Total)	53 / 13.5%	215 / 32.6%
MMSE + LM	26 / 6.6%	32 / 4.9%
CDR + LM	0 / 0.0%	35 / 5.3%
MMSE + CDR	27 / 6.9%	148 / 22.5%
3 Components
MMSE + CDR + LM	335 / 85.5%	429 / 65.1%

Note. CN=Cognitively Normal; MCI=Mild Cognitive Impairment; MMSE = Mini-Mental State Exam; CDR=Clinical Dementia Rating; LM=Logical Memory II, Delayed Recall of one paragraph. CN had 11 participants and MCI had 2 participants who were missing data for one or more elements of the criteria, but were given an ADNI diagnosis. The MCI MMSE + LM only group included 10 participants with CDR = 1.0 and 22 with CDR = 0.

Given the very low baseline-to-Year 1 reversion rate (3.0%), we examined specific components of ADNI’s MCI criteria (MMSE, CDR, LM) to determine if there was a subgroup of stable MCI participants who might have been better characterized as CN at Year 1 (i.e., possible “reverters,” not classified as such). At baseline, 99.0% of MCI participants met all three required components of the MCI criteria that were examined. At Year 1, however, about one-third (34.9%) of stable MCI participants did not meet all three components of the criteria (1 component n=15, 2.3%; 2 components n=215, 32.6%). Of the participants who did not meet all three criteria (N=230), the component that was most frequently not met at Year 1 was LM (64.3%; 148 out of 230 MCI participants). Had those participants who did not perform below the LM cutoffs at Year 1 (MMSE+CDR only group) been instead diagnosed as CN at Year 1, the ADNI criteria would have shown a MCI-to-CN reversion rate of 21.8% (see Figure 1). This is a conservative estimate using the E-MCI criteria and not including the 15 participants (14 of whom did not meet the LM cutoffs) who only met one of the components of the MCI criteria at Year 1.

Figure 1. Comparison of meta-analysis, ADNI-based diagnosis, and theoretical ADNI reversion at Year 1.

Dark Green bar represents the average reversion rate across studies from a recent meta-analysis [6]. Light Green bar represents the baseline to Year 1 reversion rate based on the diagnoses in ADNI. Blue bar represents the theoretical ADNI reversion rate if the MMSE+CDR only group was classified as CN at Year 1 instead of MCI. Red box represents the 14% and 31% rate of reversion for clinical (14%) and community-based (31%) samples, respectively found by a recent meta-analysis [6].

3.3.2. Biomarker analyses

Since the LM component of the criteria was the most frequently unmet component at Year 1 among “stable” MCI participants, we examined AD-related CSF and genetic susceptibility markers and compared the MMSE+CDR only group (did not meet LM criteria at Year 1) to stable CN (CN at both baseline and Year 1) as well as stable MCI participants who met all three criteria at Year 1 (n=429).

Chi-squared analyses indicated significant differences between stable CN, MMSE+CDR only, and stable MCI participants on proportions of participants who were positive for Aβ [Χ²(2)=59.89, p<.001], p-tau/Aβ ratio [Χ²(2)=74.24, p<.001], t-tau/Aβ ratio [Χ²(2)=77.66, p<.001], and at least one APOE ε4 allele [Χ²(2)=57.88, p<.001]. Follow-up analyses revealed that only the stable MCI group differed from stable CN and MMSE+CDR only groups on AD biomarkers (all ps<.001), whereas the stable CN and MMSE+CDR only groups did not differ on any AD biomarkers (p-values ranged from .202 to .992; see Figure 2).

Figure 2. Proportion of stable CN, MMSE+CDR only, and stable MCI groups with positive AD biomarkers.

***p<.001; For the CSF markers (Aβ, p-tau/Aβ, t-tau/Aβ), the stable CN group had n=276, CDR + MMSE only group had n=110, and stable MCI group had n=324.

3.3.3. Cognition and depressive symptom analyses

There were significant group differences in LM [F(2, 954)=732.64, p<.001], MMSE [Kruskal-Wallis Χ²(2)=201.63, p<.001], and GDS [Kruskal-Wallis Χ²(2)=85.45, p<.001] scores. LM performance at Year 1 did not differ between the stable CN (mean=13.66, SD=4.01) and MMSE+CDR only (mean=13.71, SD=2.15) groups (p>.99), but both groups performed better than the stable MCI group (mean=4.93, SD=3.46, ps<.001). MMSE scores did not differ between the stable CN (mean=28.96, SD=1.30) and MMSE+CDR only (mean=28.80, SD=1.43) groups (p=.879). However, the stable MCI group had a lower MMSE score (mean=27.37, SD=1.78) than both the stable CN (p<.001) and MMSE+CDR only (p<.001) groups. GDS scores, however, differed between the stable CN (mean=0.95, SD=1.42) and both the MMSE+CDR only (mean=1.66, SD=1.83) and stable MCI (mean=1.92, SD=1.96) groups (both ps<.001), but the MMSE+CDR only and stable MCI groups did not differ from one another (p=.263).

3.3.4. Progression to dementia

Differential rates of progression to dementia were examined between groups. Compared to the stable CN group, Cox regressions showed that both the MMSE+CDR only group (HR: 5.98, 95% CI: 2.07, 17.25, p=.001) and, to a much greater extent, the stable MCI group (HR: 36.15, 95% CI: 14.78, 88.37, p<.001) had greater risk of progression to dementia. Compared to the MMSE+CDR only group, the stable MCI group had greater risk of progression to dementia (HR: 6.04, 95% CI: 3.26, 11.19, p<.001). Kaplan-Meier curves and numbers of events/persons at risk are shown in Figure 3.

Figure 3. Kaplan-Meier curves for stable CN, MMSE+CDR only, and stable MCI group progression to dementia.

3.4. Post hoc analysis of change in memory

Change in objective memory performance from baseline-to-Year 1 by group was examined. For LM, there was a significant Group × Time interaction [F(2, 975)=188.97, p<.001, η_p²=.279] such that the stable MCI group declined and stable CN and MMSE+CDR only groups significantly improved from baseline-to-Year 1 (all ps<.01); however, the MMSE+CDR only group demonstrated accelerated improvement (p<.001) and no longer differed from the stable CN group at Year 1 (see Supplementary Figure 1). For AVLT delayed recall, there was a small, but significant, Group × Time interaction [F(2, 974)=8.49, p<.001, η_p²=.017], but the stable CN and MMSE+CDR only groups did not differ in AVLT performance and did not have a significant improvement from baseline-to-Year 1 (ps>.05); instead, the stable MCI group significantly declined (see Supplementary Figure 2).

4. Discussion

The current study found that there was significant diagnostic stability of participants initially identified as CN and MCI across assessments, as would be generally expected over the course of a relatively short period of time. However, the reversion rate of 3% appears to be substantially lower than data reported in meta-analyses showing typical reversion rates of 18% [5] and 24% [6]. Furthermore, 3% is notably lower than the most conservative estimates of reversion in clinical samples of 8% [5] and 14% [6], despite the fact that these meta-analyses included several studies that used similar criteria to those used in ADNI (i.e., Mayo/Petersen [8] or International Working Group [9] criteria).

This very low reversion rate appeared to be at least partly related to incomplete application of the components of ADNI’s cognitive diagnostic criteria (MMSE, CDR, LM) at Year 1, as only 65.1% of those considered stable MCI continued to meet the cutoffs for all three components of the criteria at Year 1. The largest portion of participants who did not meet all three components of the criteria at follow-up was a group who met the MMSE and CDR components, but not LM (22.5% of the stable MCI sample). These MMSE+CDR only participants performed above the MCI cutoffs on the LM component of the criteria. Because the MMSE cutoff was the same for both ADNI CN and MCI classifications, it appears that ADNI diagnosticians weighted the CDR more heavily than LM, despite the CDR’s reliance on subjective reporting from the participant and study partner. ADNI’s diagnostic reliance on the CDR is also embedded within these three-operationalized components of the criteria; only the CDR has completely mutually exclusive categories for identifying CN versus MCI participants. As mentioned above, the MMSE cutoffs do not discriminate between CN and MCI participants (the criteria completely overlap). For LM, there is considerable overlap between CN and E-MCI cut-offs for ADNI GO and 2. Thus, the diagnosis of individuals with objective test performance that falls in this “border-zone” (i.e., CN or E-MCI based on MMSE and LM scores) relies solely on the more subjective CDR ratings.

These findings are consistent with our previous work showing that about one-third of ADNI-defined MCI participants have cognitive functioning [12], AD biomarkers [12–14,16], and everyday functioning [15] that is more consistent with CN than MCI participants (i.e., “false-positive” MCI diagnosis). The current study adds to our previous work by showing that, when more weight is given to more subjective (CDR) than objective (LM) information in the diagnostic decision-making process, there is a greater propensity for diagnostic misclassification, including artificially high rates of MCI stability.

Inconsistencies between the subjective reporting by the participant or study partner on the CDR and objective memory performance on LM (i.e., over-reporting cognitive or functional difficulties on the CDR in the context of intact objective performance) may be at least partially related to depressive symptoms [19–21]. Despite normal performance on objective memory (LM) and global cognition (MMSE) tests at Year 1, participants in the MMSE+CDR only group reported more depressive symptoms than the stable CN group. However, the severity of their depressive symptoms did not differ from symptoms of the stable MCI group. Although the severity of the depressive symptoms reported by both the MMSE+CDR only and stable MCI groups were well below clinically significant levels, the difference in reported symptoms at Year 1 between the MMSE+CDR only and the stable CN groups may contribute to the inconsistency between LM performance and CDR score for the MMSE+CDR only group.

When the MMSE+CDR only group was compared to stable MCI participants who met all three components of the MCI criteria and stable CN participants, results showed that the MMSE+CDR only group had a lower proportion of positive AD biomarkers (CSF and genetic susceptibility) at baseline than the stable MCI group, and they did not differ from the stable CN participants on any of these biofluid and genetic susceptibility markers. A similar pattern of findings was found for objective memory performance and global cognition at Year 1 in that the stable CN and MMSE+CDR only groups both had better performance than the stable MCI group but did not differ from one another. These findings suggest that the MMSE+CDR only group would have been better characterized as CN (and with low biomarker positivities) and not MCI at Year 1. Had the MMSE+CDR only participants been diagnosed as CN at Year 1 follow-up (i.e., “reverter”), the ADNI reversion rate would have been at least 21.8%, which is consistent with two recent meta-analyses showing rates of reversion at 18% [5] and 24% [6].

When the rates of progression to dementia over 5 years were examined, the stable MCI participants progressed to dementia at a significantly faster rate than the stable CN or the MMSE+CDR only groups. While the MMSE+CDR only group progressed to dementia faster than the stable CN group, this risk was much lower than that of the stable MCI group, consistent with the differences in biomarker positivity. Prior work suggests that MCI participants that revert to CN status ultimately progress to dementia at a faster rate than those initially diagnosed as CN [4]. The hazard ratio for risk of dementia in the MMSE+CDR only group (HR=5.98) is consistent with previous studies examining the risk of progression to dementia in “reverters” compared to baseline normal participants (e.g., HRs of 6.6 in the Mayo study [4] and 6.4 in the Sydney Memory and Ageing Study [22]).

The reasons for what appears to be a greater focus on the CDR than the LM performance at Year 1 follow-up are unclear, although bias (e.g., expectancy, confirmation, anchoring) might be a possibility if the diagnosticians had any awareness of a participant’s prior year’s diagnosis when deciding on their Year 1 diagnosis. This notion is consistent with past work showing that bias may negatively impact diagnostic decisions if researchers are not blinded to past diagnoses [3]. Furthermore, based on the protocol manuals available on the ADNI website (http://adni.loni.usc.edu/methods/documents/), the protocol appears to be vague regarding how to diagnose participants at follow-up visits as well as how to handle inconsistent results among the components of the classification criteria. An additional factor that likely made classification difficult was the inconsistency in the LM criteria between ADNI 1 and ADNI GO/ADNI 2. In ADNI GO and ADNI 2, LM was used to distinguish E-MCI from L-MCI. Thus, participants were classified as MCI using slightly different criteria between ADNI waves. The current investigation used the E-MCI criteria for all participants since this was more lenient (i.e., LM cutoff for E-MCI is more easily met than the cutoff for L-MCI). However, this approach provides the most conservative estimate of individuals that did not meet all three components of the MCI criteria, and it is likely that even fewer participants who were considered by ADNI to be stable MCI from baseline-to-Year 1 would have met the MCI LM criteria at Year 1 had the L-MCI criteria been applied.

The findings that a large proportion of ADNI MCI participants were not accurately identified as “reverters” and instead retained a potentially “false positive” diagnosis of MCI could have significant clinical and research implications. Maintaining a diagnosis of MCI when objective testing suggests reversion may have clinical and research consequences. From a psychological perspective, being consistently diagnosed with MCI despite having normal memory performance has unknown, yet potentially detrimental effects. Prior work has shown that knowledge of one’s own APOE ε4 allele status, for example, may have a negative effect on both subjective and objective memory performance relative to APOE ε4 carriers without knowledge of their genotype [23]. Additionally, there are significant implications of including these potential “reverters” in analyses that examine the longitudinal trajectories of AD biomarkers. We found the MMSE+CDR only group to have very similar AD biomarker profiles as stable CN participants. Therefore, participants who are mislabeled as “stable MCI” but do not actually meet criteria for MCI may be negatively impacting our understanding of AD biomarker trajectories or washing out effects of biomarkers as predictors of progression to dementia.

One potential difficulty in relying on objective memory performance in serial assessment is the accounting for possible practice effects. Previous work has identified significant practice effects of memory tests within ADNI [24], so we speculated that perhaps ADNI diagnosticians considered the potential for practice effects when diagnosing MCI at Year 1 despite normal LM performance. However, examination of the baseline-to-Year 1 change in LM and AVLT memory measures showed that stable CN and MMSE+CDR only groups did not differ at Year 1 on either measure and did not differ on AVLT performance at baseline. Together, these data suggest that the MMSE+CDR only change in performance from baseline-to-Year 1 is more likely regression to the mean than an elevated practice effect given the much smaller practice effect (likely not due to ceiling effects) of the stable CN group on LM and the similar performances at both baseline and Year 1 on the AVLT. Notably, these findings add to the literature that highlight the importance of using multiple objective neuropsychological measures in a diagnosis of MCI [10,25]. There has been considerable work showing that, within a neurologically normal population, the proportion of individuals with at least one impaired score is high [26–28], again emphasizing the importance on balancing reliability of impairment through the use of more than one test in a domain.

It is important to note that our study focused on the one year of follow-up for examining diagnostic stability. Future work examining the stability, reversion, and progression rates over longer follow-up periods in more detail will expand upon the present findings. Additionally, future work should examine the precision with which other large aging studies adhere to their diagnostic criteria and how diagnostic decisions are made when the criteria are mixed. This type of detailed inspection of diagnostic criteria may be particularly relevant to clinical trials because false positive errors and artificial diagnostic stability could mask potential results [29].

There is a small subset of ADNI participants with autopsy data, and it would be ideal to determine that the MMSE+CDR only group did not go on to develop AD pathology, although autopsy-confirmation of individuals who were CN or MCI at baseline is not without issues. Most older adults with normal cognition or mild cognitive changes will live many years past the date of an initial observation, and autopsies will capture new pathology that may have developed after that baseline observation. However, reliance on CSF biomarkers is also not without limitations. While CSF Aβ_1-42 and tau-Aβ ratios (t-tau/Aβ, p-tau/Aβ) have shown strong concordance with Aβ PET and level of cognitive impairment [18,30], CSF AD biomarkers are limited in that there is no way to determine regional patterns that may be evident on PET. Further, CSF Aβ_1-42 levels have been shown to be decreased in other neurodegenerative processes (i.e., Lewy body dementia) [31,32]; similarly, CSF tau levels have been shown to be elevated in non-AD pathologies (e.g., Creutzfeldt-Jakob disease, stroke) [33]. Despite these limitations, the ability to incorporate CSF AD markers that correspond to their baseline assessment is a relative strength of the ADNI study.

The current study, consistent with prior ADNI work, showed a very low MCI-to-CN reversion rate over one year. It also demonstrated that when all features of the criteria were not met, the more subjective CDR was weighted more heavily than the objective memory measure (LM) in the diagnostic decision-making process, resulting in an artificially low reversion rate. These findings are further supported through biofluid and genetic markers, progression rates, and memory changes that all suggest that baseline MCI participants with normal memory scores at Year 1, despite a CDR score of 0.5, would have likely been better classified as CN at Year 1. Our future directions for this work include examination of diagnostic stability, reversion, and progression rates in ADNI using MCI criteria that are based on actuarial neuropsychological test performances [10,11]. The use of objective neuropsychological scores in a way that balances sensitivity and reliability (i.e., two impaired cognitive scores within a cognitive domain, rather than one impaired LM score) may provide a method to more precisely identify stable MCI participants and differentiate them from true “reverters” who are cognitively normal at follow-up.

Highlights.

• ADNI has a very low one-year MCI-to-normal reversion rate of 3%

• 22.5% of subjects classified as stable MCI did not have impaired memory at Year 1

• Their biomarkers were also more similar to those of cognitively normal subjects

• Year 1 MCI diagnosis appears to be driven more by the CDR than memory performance

• Had the memory cutoffs been consistently applied, the reversion rate would be 21.8%

Research in Context.

1.Systematic Review: The authors reviewed studies (e.g., using PubMed) related to MCI diagnostic stability and reversion. MCI-to-normal reversion rates vary widely, depending on the study and MCI criteria. The reversion rate previously reported in ADNI was lower than all other studies included in a recent meta-analysis.

2.Interpretation: Results suggest that the low reversion rate in ADNI may be driven by weighting the more subjective CDR more heavily than the objective memory test, resulting in the continuation of the MCI diagnosis at one-year follow-up, despite normal memory and biomarkers. Had those subjects with normal memory performance been classified as cognitively normal at follow-up, the ADNI reversion rate would be more consistent with the literature.

3.Future directions: Future work will examine MCI-to-normal reversion using MCI criteria based on objective neuropsychological test performances. We will determine whether a comprehensive neuropsychological approach provides an improved method for differentiating stable MCI participants from “reverters” who are cognitively normal at follow-up.

Abbreviations

AD

Alzheimer’s disease

ADNI

Alzheimer’s Disease Neuroimaging Initiative

ANOVA

Analysis of Variance

APOE

Apolipoprotein E

Aβ

Amyloid-β1–42

CDR

Clinical Dementia Rating

CN

Cognitively normal

CSF

Cerebrospinal Fluid

E-MCI

Early MCI

GDS

Geriatric Depression Scale

L-MCI

Late MCI

LM

Logical Memory

MCI

Mild Cognitive Impairment

MMSE

Mini Mental State Exam

NINCDS/ADRDA

National Institute of Neurological and Communication Disorders and Stroke/Alzheimer’s Disease and Related Disorders Association

p-tau

hyperphosphorylated tau

t-tau

total tau