Neuropsychological decline up to 20 years before incident mild cognitive impairment

Abstract

INTRODUCTION:

Some Alzheimer disease biomarker studies found amyloid changes 20 years or more in advance of expected symptoms while cognitive changes lagged for over a decade, but this apparent lag might reflect the sensitivities of the biomarker and cognitive assays used. How far in advance of incident amnestic mild cognitive impairment (MCI) does cognition begin to decline?

METHODS:

Longitudinal neuropsychological study of an apolipoprotein E e4 enriched cohort of cognitively normal individuals at entry. Linear mixed models for MCI converters (n=65) and nonconverters (n=719) fitted for each neuropsychological measure; annual changes compared between groups before and after linear model intersections (inflection points).

RESULTS:

34 of 35 cognitive measures and 9 of 18 behavioral measures declined faster post-inflection in the MCI converters; the earliest cognitive inflection point was nearly 20 years in advance of MCI diagnosis.

DISCUSSION:

The preclinical duration of cognitive and behavioral changes approaches the earliest reported biomarker changes.

1.0. Background

The transition from normal to pathological cognitive aging is riddled with many questions and controversies. For clinicians, is a modestly suboptimal performance on a mental status test age appropriate or pathological? A related research concern, especially pressing with the rise of preclinical therapeutic trials for Alzheimer’s disease (AD), is determining how long before symptoms develop does pathology begin? Some biomarker studies in autosomal dominant AD (ADAD) kindreds^1,2 and in late onset AD (LOAD) cohorts³ found changes related to amyloid 20 years or more in advance of expected symptoms while cognitive changes lagged for over a decade. But might this apparent lag in cognition reflect the relative sensitivity of the various biomarker and cognitive assays used in such studies?

The first evidence of preclinical AD derived from clinical-pathological correlations of cognitively normal elders whose brains harbored cortical amyloid^4,5. Similar traces of AD pathology have been identified in younger apolipoprotein E (APOE) e4 carriers⁶. More recently, biomarker studies in living patients have shown cerebral amyloid burden becomes more prevalent and pronounced with increasing age even in the absence of symptomatic memory loss^7,8. Mild cognitive impairment (MCI) represents the earliest symptomatic stage typically of a neurodegenerative dementia and most often AD^9,10. MCI was originally anchored to memory but has since been expanded to include essentially any cognitive profile that is thought to represent pathology beyond that of normal aging while still admitting that non-neuropathological causes including normal aging and depression for example are possible in a nontrivial number of cases⁹.

Whether due to pure AD, another degenerative disease, or a mix of degenerative and vascular pathologies, as a clinical syndrome, MCI is important to recognize as it portends further decline in the majority of patients for whom it is hoped earlier intervention might slow or prevent disease progression to the stage of dementia and ultimately death. Herein we present evidence that cognition begins to decline roughly 20 years before the diagnosis of incident MCI, much earlier than previous studies have shown.

2.0. Methods

2.1. Subjects.

From January 1, 1994 through June 30, 2016, cognitively normal residents of Maricopa County age 21 years and older were recruited through local media ads, underwent APOE genotyping and longitudinal neuropsychological assessment every two years. APOE genotype was determined using Taqman Single Nucleotide Polymorphism assays. All APOE e4 homozygotes were matched by age, sex, and education to one ε4 heterozygote (HTZ; all with the ε3/4 genotype) and two ε4 non-carriers. Additional heterozygous persons and non-carriers who were otherwise eligible for enrollment were also recruited so that roughly half of the entire cohort represents matched quartos while remaining members were unmatched but otherwise fulfilled entry criteria. Screening tests included a medical history (including review of medications in all cases and medical records when provided), neurological examination, the Folstein Mini-Mental Status Exam (MMSE), Hamilton Depression (Ham-D) Rating Scale, Functional Activities Questionnaire (FAQ), Instrumental Activities of Daily Living (IADL), and Structured Psychiatric Interview for DSM-III-R. We excluded anyone with potentially confounding medical, neurologic, or psychiatric problems (essentially any condition that might adversely affect cognitive abilities such as end stage organ disease, stroke, or active major depression). Anyone with a history, self-reported or documented, of a psychotic disorder or psychiatric hospitalization was excluded. As expected most participants expressed a personal interest in AD and were introspective about whether their episodes of forgetting meant anything. Such concern was not exclusionary, but subjective memory complaints severe enough to motivate clinical evaluation were exclusionary even if objective testing revealed no impairment. No one met published criteria for MCI⁹, AD¹¹, other forms of dementia or major depressive disorder¹². Entry criteria included scores of at least 27 on the MMSE (with at least 1 of 3 on the recall subtest), 10 or less on the Ham-D, and perfect scores on the FAQ and IADL.

Data were reviewed at each visit and MCI suspected by consensus of two behavioral neurologists (RJC and BKW) and a neuropsychologist (DECL) in those individuals who endorsed symptoms of memory loss (corroborated by a close informant), and exhibited objective decline from previous performance that consistently affected multiple neuropsychological tests within the same cognitive domain^13,14. MCI suspects were then invited to complete a clinical neurological evaluation that included standard laboratory and radiological assessments following which a formal diagnosis was made and provided to the patient. The time between completion of neuropsychological testing and ultimate MCI diagnosis varied but was typically between one and two months.

All individuals gave their written, informed consent to participate in the study and have the results of the APOE test withheld from them which was approved by the Mayo Clinic Institutional Review Board.

2.2. Neuropsychological Assessment.

A comprehensive neuropsychological battery was administered every one to two years, and is summarized in the Table 1. Our initial 1994 battery was expanded in 1998. The National Alzheimer’s Coordinating Committee issued the Uniform Data Set (UDS) in 2004 and so UDS tests were added then. The neuropsychological tests (cognitive and behavioral) used within our battery are detailed in reference 15, and encompass four broadly defined cognitive domains (memory, executive skills, language, and visuospatial skills), as well as behavioral measures. Regarding subjective cognition, participants and their informants completed the paired Multidimensional Assessment of Neurodegenerative Symptoms Questionnaire including both self (MANS-Self) and Observer (MANS-Observer) versions¹⁶. The MANS are paired self and informant based questionnaires comprised of 87 questions that assess changes over the preceding year in daily habits, personality, and motor functioning. It employs a quantitative scale for rating the frequency of a symptom from 0 (never) to 4 (routinely), with intermediate values of 1 (once), 2 (occasionally), and 3 (more than monthly); scores can range from 0-344 with higher scores indicating more frequent and severe symptoms.

Table 1.

Cognitive and Behavioral Measures

Test	Score	Major Domain	Year Added
Folstein Minimental State Exam	0-30	General	1994
Mattis Dementia Rating Scale	0-144	General	1998
WAIS-R Information	Age Scaled Score	General	1994
WAIS-R Vocabulary	Age Scaled Score	Language	1994
WAIS-R Similarities	Age Scaled Score	Language	1994
Boston Naming Test	0-60	Language	1994
Token Test	0-44	Language	1994
Judgement of Line Orientation	0-30	Visuospatial	1994
Facial Recognition Test	0-50	Visuospatial	1994
WAIS-R Block Design	Age Scaled Score	Visuospatial	1994
Complex Figure Test-copy	0-36	Visuospatial	1994
Digit span forward length (from WAIS-R)	Forward Digit Span (absolute number)	Executive	2004 (UDS)
Digit Span backward length (from WAIS-R)	Backward Digit Span (absolute number)	Executive	2004 (UDS)
WAIS-R Mental Arithmetic	Age Scaled Score	Executive	1994
Paced Auditory Serial Attention Task-3 second	0-50	Executive	1998
Paced Auditory Serial Attention Task-2 second	0-50	Executive	1998
WAIS-R Digit Symbol Substitution	Age Scaled Score	Executive	1994
Trail Making Test Part A	Seconds to Complete	Executive	2004 (UDS)
Trail Making Test Part B	Seconds to Complete	Executive	2004 (UDS)
Controlled Oral Word Association Test	Total Number of Words	Executive	1994
Animals Fluency	Total Number of Animals Named	Executive	2004 (UDS)
Vegetable Fluency	Total Number of Vegetables Named	Executive	2004 (UDS)
Wisconsin Card Sorting Test-Total Errors	Total Errors	Executive	1998
Wisconsin Card Sorting Test-Categories Completed	Categories Completed	Executive	1998
Wisconsin Card Sorting Test-Perseverative Errors	Number of Perseverative Errors	Executive	1998
Complex Figure Test-recall	0-36	Memory	1994
Auditory Verbal Learning Test-Total Learning	Number Words Over 5 Trials	Memory	1994
Auditory Verbal Learning Test-% Recall	% Words Recalled at 30 Minutes	Memory	1994
Auditory Verbal Learning Test-Long Term Memory	Number of Words Recalled at 30 Minutes	Memory	1994
Visual Retention Test	Total Correct	Memory	1994
Selective Reminding Test-Immediate Free Recall	Total Correct Without Delay	Memory	1998
Selective Reminding Test-Delayed Free Recall	Total Correct After Delay	Memory	1998
Wechsler Memory Scale-R Logical Memory-Immediate Recall	Absolute Number Story Items	Memory	2004 (UDS)
Wechsler Memory Scale-R Logical Memory-Delayed Recall	Absolute Number Story Items	Memory	2004 (UDS)
Subjective Cognition and Social Standing
Multidimensional Assessment of Neurodegenerative Symptoms-Informant	0-348	Subjective Cognitive Change	1996
Multidimensional Assessment of Neurodegenerative Symptoms-Self	0-348	Subjective Cognitive Change	1996
Behavioral Measures
Geriatric Depression ScaleTotal	0-30	Depression	2004 (UDS)
Hamilton Depression Scale	0-49	Depression	1994
Beck Depression Inventory	0-67	Depression	1994
Personality Assessment Inventory
PAI_CSc_Somatization	T score	Somatization	1998
-PAI_SOM_Conversion	T score	Somatization	1998
-PAI_SOM_Hypocondriasis	T score	Somatization	1998
-PAI_SOM_Somatization	T score	Somatization	1998
PAI_CSc_Anxiety	T score	Anxiety	1998
-PAI_ANX_A	T score	Anxiety	1998
-PAI_ANX_C	T score	Anxiety	1998
-PAI_ANX_P	T score	Anxiety	1998
PAI_CSc_Anxiety Related Disorders	T score	Anxiety Related Disorders	1998
-PAI_ARD_Obessive	T score	Anxiety Related Disorders	1998
-PAI ARD Phobia	T score	Anxiety Related Disorders	1998
-PAI_ARD_Trauma	T score	Anxiety Related Disorders	1998
PAI_CSc_Depression	T score	Depression	1998
-PAI_DEP_Affective	T score	Depression	1998
-PAI_DEP_Cognitive	T score	Depression	1998
-PAI_DEP_Physiological	T score	Depression	1998
PAI_CSc_Mania	T score	Mania	1998
-PAI_MAN_Affective	T score	Mania	1998
-PAI_MAN_Grandiosity	T score	Mania	1998
-PAI_MAN_Irritability	T score	Mania	1998
PAI_CSc_Paranoia	T score	Paranoia	1998
-PAI_PAR_Hypervigilance	T score	Paranoia	1998
-PAI_PAR_Persecution	T score	Paranoia	1998
-PAI_PAR_Resentment	T score	Paranoia	1998
PAI_CSc_Schizophrenia	T score	Schizophrenia	1998
-PAI_SCZ_Psychotic Experiences	T score	Schizophrenia	1998
-PAI_SCZ_Social Detachment	T score	Schizophrenia	1998
-PAI_SCZ_Thought Disorder	T score	Schizophrenia	1998
PAI_TSc_Aggression	T score	Aggression	1998
-PAI_AGG_Attitude	T score	Aggression	1998
-PAI_AGG_Physical	T score	Aggression	1998
-PAI_AGG_Vverbal	T score	Aggression	1998
PAI_CSc_Antisocial	T score	Antisocial	1998
-PAI_ANT_Antisocial Behaviors	T score	Antisocial	1998
-PAI_ANT_Egocentricity	T score	Antisocial	1998
-PAI_ANT_Stimulus Seeking	T score	Antisocial	1998
PAI_CSc_Borderline	T score	Borderline	1998
-PAI_BOR_Affective Instability	T score	Borderline	1998
-PAI_BOR_Identity Problems	T score	Borderline	1998
-PAI_BOR_Negative Relationships	T score	Borderline	1998
-PAI_BOR_Self Harm	T score	Borderline	1998
PAI_CSc_Drugs	T score	Drugs	1998
PAI_CSc_Alcohol	T score	Alcohol	1998
PAI_TSc_Nonsupport	T score	Nonsupport	1998
PAI_TSc_Treatment Rejection	T score	Treatment Rejection	1998
PAI_TSc_Traumatic Stress	T score	Stress	1998
PAI_TSc_Suicidal ideation	T score	Suicide	1998
PAI_ISc_Dominance	T score	Dominance	1998
PAI_ISc_Warmth	T score	Warmth	1998

WAIS-R=Wechsler Adult Intelligence Scale-Revised; CSc=Clinical Scale; TSc=Treatment Scale; ISc=Interpersonal Scale

2.3. Statistical Analysis

Analyses were performed by BTL, ACD, YC, and YS. In order to determine how far in advance of incident MCI cognitive decline begins, we assessed the neuropsychological trajectories between individuals diagnosed with incident amnestic MCI during the study period (converters) and those remaining cognitively normal (nonconverters). For each of the neuropsychological measures, we constructed a linear mixed model accounting for the cross-sectional and longitudinal nature of the data to compare the longitudinal trajectories between converters and nonconverters. Quadratic terms were initially considered, however their addition did not outperform first-order linear modeling. Within the fitted linear model, the age at which the two neuropsychological measure trajectories intersect was considered an ‘inflection point’ whereby converter and nonconverter paths diverge. A confidence interval for the inflection point was generated using 500 bootstrapped samples. The inflection point was then used as a marker to form a piecewise linear mixed effects regression model used to assess pre- and post-inflection effects (before and after the converter and nonconverter trajectories diverge) and allow for comparison of annual change (slopes) between groups. Regression contrast analysis within the piecewise linear mixed model framework was used to compare the annual change in neuropsychological measure scores between converters and nonconverters, both before the inflection point and after the inflection point, separately. Additionally, we performed cross sectional comparisons at five and ten years prior to the age of MCI conversion. The average conversion age of the 65 converters was 73, so for the matched set of comparable nonconverters we used those who were within two years of ages 68 and 63 respectively.

Study characteristics are presented using means and standard deviations (SD), frequencies, and relative risks (RR). Mean values and frequency distributions were compared between converters and nonconverters using t-tests and chi-squared tests, respectively. Kaplan-Meier (KM) method was used to estimate time to MCI diagnosis within groups. Logrank tests and Cox proportional hazards regression were used to compare time to MCI diagnosis between groups. Statistical modeling was conducted using the statistical software SAS version 9.4 (SAS Institute Inc., Cary, NC).

3.0. Results

Over a mean duration of 9.5 years (range 1.1-20.5 years) 65 individuals were diagnosed with incident amnestic MCI at a mean age of 73 (S.D. 6.8) years. (An additional patient who was an APOE e4 homozygote developed a nonamnestic dysexecutive form of MCI that was later found to reflect an unsuspected progranulin mutation with autopsy confirmed frontotemporal lobar degeneration in addition to AD, and was excluded from this analysis). 719 remained clinically normal over a mean observational period of 7.2 (S.D. 6.5) years. Demographic data are summarized in table 2. Figures 1a and 1b shows the Kaplan-Meier curves by APOE e4 carrier status and genotype respectively. APOE e4 carriers had roughly four times the risk of MCI diagnosis than non-carriers (HR 3.96; 95%CI: 2.25, 6.96). Men had over twice the risk of MCI diagnosis than women (HR 2.25; 95%CI: 1.42, 3.58). Sixteen experienced clinically significant decline in additional cognitive domains (as well as memory) including executive (n=10), visuospatial (n=7), and/or language (n=5). Of these 16, 12 were e4 carriers. 151 members of the entire cohort had an amyloid PET scan. Of those with MCI, it was positive in 6 of 12, and of the nonconverters it was positive in 24 of 139.

Table 2.

Entry Demographics

	Converters (N=65)	Nonconverters (N=719)	p
Age at entry (years) – mean (SD)	63.6 (7.4)	56.1 (11.6)	<0.001
Education (years) – mean (SD)	15.8 (2.8)	15.8 (2.5)	0.94
Time to diagnosis/duration followup (years) – mean (SD)	9.5 (4.5)	7.2 (6.5)	0.006
Women – n (%)	35 (53.8%)	533 (74.1%)	<0.001
APOE e4 carriers – n (%)	49 (75.4%)	287 (40.4%)	<0.001
Folstein Minimental State Exam – mean (SD)	29.3 (0.9)	29.7 (0.7)	<0.001
Hamilton Depression Scale – mean (SD)	2.5 (3.0)	2.5 (3.3)	0.87

Figure 1.

Kaplan-Meier curves of time to mild cognitive impairment contrasting a) APOE e4 carriers and noncarriers; and b) by APOE genotype.

48 have had subsequent followup from their MCI diagnosis ranging from one to 12.5 years and all have shown further decline. Four developed symptoms suggesting dementia with Lewy bodies: dream enactment behavior in three and mild parkinsonism in one. All four were APOE e4 carriers (three were e4 homozygotes). Two others had a stroke between eight and 16 years after MCI diagnosis, and both died. Ten in all have since died an average of 8.5 years after MCI diagnosis, and seven came to autopsy with AD confirmed in all but one case in who cerebral rarefaction was the major finding nearly six years after MCI was diagnosed. Two had additional findings of synucleinopathy, neither of who exhibited clinical parkinsonism (one was an APOE e4 carrier). Among the MCI nonconverters six developed Parkinson’s disease, three of who were e4 carriers.

Supplementary Table 1 shows the neuropsychological and behavioral data at entry for the originally well matched APOE e4 carrier and noncarrier groups as well as the inflection point baseline data for the MCI converter and nonconverter groups at which point differences in age and memory performance have become evident. Of 56 tested neuropsychological measures, the inflection point and subsequent piecewise linear mixed models could be estimated for 53 outcomes. Table 3 summarizes the results for these 35 cognitive and 18 behavioral measures (Supplementary Table 2 is an expanded version with the complete data set). 34 of 35 cognitive neuropsychological scores including both subjective cognition measures, all three general intellect, all four language, all four visuospatial, all nine memory, and 12 of 13 executive measures showed a significantly steeper slope of decline post-inflection in the MCI converters relative to the nonconverters with the sole exception being a reverse digit span test although this may reflect the significantly greater rate of decline pre-inflection in the MCI converters. Figure 2 summarizes the inflection point ages for each cognitive measure. The earliest cognitive inflection point was nearly 20 years in advance of MCI diagnosis. Multiple declarative memory tests declined early, and generally had the narrowest confidence intervals. Inflection point ages of delayed recall scores were as follows: Auditory Verbal Learning Test (AVLT) Long Term Memory (LTM) 53.4 years, AVLT % recall 54 years, Selective Reminding Test (SRT) delayed free recall 54.8 years, Complex Figure Test (CFT) recall 55 years, Wechsler Memory Scale-Revised (WMS-R) logical memory delayed recall 58.6 years. Immediate recall (and related measures) inflection points were as follows: AVLT Total Learning 55.2 years, SRT immediate free recall 55.4 years, Visual Retention Test 55.9 years, WMS-R logical memory immediate recall 58.6 years. The inflection age of self-reported subjective cognition nearly overlapped logical memory delayed recall while informant-based change lagged self-observation by another two years. The earliest inflection point, however, was for the CFT-copy, age 53.3 years while the next earliest visuospatial test, the Judgement of Line Orientation (JLO) was two years later.

Table 3.

Trajectories (Slopes) of Cognitive and Behavioral Measures Before and After Inflection Points of MCI Converters and Nonconverters

		Trajectory Slope Overall			Trajectory Slope Before AOI			Trajectory Slope After AOI
	Age of Inflection	MCI Nonconverters	MCI Converters	p	MCI Nonconverters	MCI Converters	p	MCI Nonconverters	MCI Converters	p
General Intellect
Folstein MMSE	57.2	−0.005	−0.106	<0.001	0.0042	−0.017	0.59	−0.008	−0.113	<0.001
Dementia Rating Scale	59.4	−0.02	−0.55	<0.001	0.0564	−0.08	0.34	−0.059	−0.594	<0.001
WAIS Information	54.7	0.062	−0.038	<0.001	0.0562	0.1097	0.60	0.0631	−0.043	<0.001
Language
WAIS Vocabulary	58.8	0.0501	−0.031	<0.001	0.0798	−0.038	0.05	0.0377	−0.031	<0.001
WAIS Similar	53.5	0.0962	0.0278	<0.001	0.1041	0.1673	0.74	0.0948	0.0248	<0.001
Bostn Naming Test	58.1	−0.018	−0.411	<0.001	0.0862	−0.199	0.02	−0.055	−0.431	<0.001
Token Test	63.0	−0.008	−0.071	0.002	0.0034	−0.022	0.63	−0.018	−0.084	0.008
Visuospatial
Judgement of Line Orientation	55.3	−0.029	−0.246	<0.001	0.0208	0.3924	0.15	−0.042	−0.264	<0.001
Facial Recognition Test	56.7	−0.071	−0.267	<0.001	−0.015	0.0192	0.89	−0.089	−0.28	<0.001
WAIS Block Design	57.2	0.0442	−0.104	<0.001	0.1043	0.1087	0.96	0.026	−0.12	<0.001
Complex Figure Test-Copy	53.3	−0.02	−0.251	<0.001	0.0015	0.0803	0.80	−0.024	−0.258	<0.001
Executive
WAIS Digit Span Backward	64.8	−0.007	−0.05	0.045	0.0051	−0.197	<0.001	−0.018	−0.025	0.77
WAIS Arithmetic	57.5	−0.01	−0.14	<0.001	0.0227	−0.012	0.65	−0.021	−0.151	<0.001
PASAT 3 Second	59.5	0.0716	−0.9	<0.001	0.3661	0.422	0.88	−0.085	−1.058	<0.001
PASAT 2 Second	58.3	0.2999	−0.4	<0.001	0.6485	0.6922	0.93	0.1438	−0.494	<0.001
WAIS Digit Symbol Items	55.5	−0.474	−1.245	<0.001	−0.175	−11.58	0.001	−0.528	−1.196	<0.001
TMT-A (seconds)	55.1	0.1415	0.7659	0.001	−0.083	0.8082	0.91	0.193	0.7655	0.003
TMT-B (seconds)	58.1	0.8977	4.6749	<0.001	0.3074	1.6676	0.81	1.0913	4.7389	<0.001
COWAT	63.7	0.1998	−0.315	<0.001	0.3982	0.0375	0.02	0.0051	−0.456	<0.001
Animal Fluency (UDS)	61.5	−0.115	−0.647	<0.001	0.0113	−0.663	0.07	−0.182	−0.644	<0.001
Vegetable Fluency (UDS)	55.2	−0.098	−0.368	<0.001	0.0351	0.8062	0.75	−0.126	−0.372	<0.001
WCST Categories	61.8	−0.069	−0.179	<0.001	−0.026	−0.106	0.11	−0.103	−0.194	<0.001
WCST Errors	60.8	0.714	1.9292	<0.001	0.3292	1.6364	0.03	0.9706	1.9737	<0.001
WCST Persev Errors	59.8	0.3414	1.0579	<0.001	0.106	0.7815	0.09	0.4775	1.0892	<0.001
Memory
Complex Figure Test-Recall	55.0	0.159	−0.491	<0.001	0.2803	0.8931	0.10	0.1323	−0.546	<0.001
AVLT Total Learning	55.2	−0.087	−1.046	<0.001	0.1763	0.3401	0.73	−0.145	−1.105	<0.001
AVLT Recall	54.0	−0.043	−2.668	<0.001	0.1821	−2.867	0.08	−0.088	−2.663	<0.001
AVLT Long Term Memory	53.4	−0.003	−0.388	<0.001	0.0688	−0.302	0.19	−0.015	−0.39	<0.001
Visual Retention Test	55.9	−0.032	−0.193	<0.001	0.0113	0.102	0.34	−0.044	−0.209	<0.001
SRT Immediate Free Recall	55.4	0.2512	−1.452	<0.001	0.4728	1.6022	0.13	0.193	−1.551	<0.001
SRT Delayed Free Recall	54.8	0.0509	−0.213	<0.001	0.0843	0.5869	0.06	0.0416	−0.234	<0.001
Logical Memory Immediate Recall (UDS)	58.6	−0.003	−0.467	<0.001	0.0442	−0.683	0.14	−0.019	−0.461	<0.001
Logical Memory Delayed Recall (UDS)	58.4	0.013	−0.554	<0.001	0.057	−0.558	0.26	−0.001	−0.553	<0.001
Subjective Cognition
MANS Informant	60.7	0.0605	2.6147	<0.001	−0.435	1.6059	0.02	0.3557	2.756	<0.001
MANS Self	58.6	0.0466	2.2717	<0.001	−0.468	0.1588	0.59	0.2474	2.4463	<0.001
Depression
Geriatric Depression Scale (UDS)	58.1	−0.018	0.2008	<0.001	−0.117	0.2547	0.49	0.0106	0.1994	0.002
Hamilton Depression Scale	56.3	−0.026	0.0615	<0.001	−0.011	−0.061	0.74	−0.03	0.0685	<0.001
Beck Depression Inventory	65.6	0.0271	0.1623	0.001	−0.016	0.1469	0.04	0.0912	0.1703	0.17
PAI-CSC-DEP	60.3	0.0827	0.4877	<0.001	−0.043	0.3706	0.06	0.1621	0.5065	<0.001
Personality Assessment Inventory
PAI-CSC-DEP	60.3	0.0827	0.4877	<0.001	−0.043	0.3706	0.06	0.1621	0.5065	<0.001
PAI-CSC-ANT	56.2	−0.105	−0.036	0.15	−0.236	0.0487	0.31	−0.064	−0.04	0.64
PAI-CSC-ANX	59.2	−0.038	0.3315	<0.001	−0.106	0.0407	0.47	−0.004	0.3655	<0.001
PAI-CSC-ARD	59.9	−0.015	0.1766	0.003	−0.139	0.2391	0.08	0.0572	0.1677	0.13
PAI-CSC-BOR	58.7	−0.115	0.1028	<0.001	−0.299	−0.174	0.55	−0.028	0.1311	0.006
PAI-CSC-DRG	59.0	−0.085	0.0331	0.11	−0.165	0.2274	0.15	−0.041	0.0121	0.51
PAI-CSC-PAR	63.6	−0.061	0.102	0.005	−0.133	−0.068	0.61	0.0219	0.1667	0.05
PAI-CSC-SCZ	61.6	−0.061	0.2334	<0.001	−0.141	−0.071	0.65	0.0027	0.3041	<0.001
PAI-CSC-SOM	62.1	0.15	0.4068	<0.001	0.0955	0.3173	0.14	0.1967	0.4302	0.001
PAI-ISC-DOM	57.3	−0.125	−0.239	0.06	−0.066	−0.005	0.84	−0.147	−0.255	0.10
PAI-ISC-WRM	68.6	−0.008	−0.171	0.01	0.0425	−0.05	0.34	−0.161	−0.318	0.17
PAI-TSC-AGG	54.6	−0.078	0.0591	0.01	−0.203	−0.01	0.68	−0.047	0.0611	0.06
PAI-TSC-NON	68.2	0.0584	0.1958	0.07	−0.012	0.0089	0.85	0.2624	0.4015	0.26
PAI-TSC-STR	62.4	−0.089	0.126	0.007	−0.187	−0.009	0.36	0.0055	0.1639	0.10
PAI-TSC-SUI	68.5	0.0203	0.1837	0.008	−0.008	0.1602	0.06	0.1066	0.2114	0.32

Figure 2.

Cognitive Test Score Ages of Inflection. The ages at which overall linear trajectories intersect for each cognitive measure between those who developed MCI and those who remained cognitively normal. Error bars represent 95% confidence intervals based on 500 bootstrapped samples.

Figure 3 summarizes the inflection point ages for each behavioral measure. Confidence intervals were wide for all. Three of four depression measures declined more steeply post-inflection in the MCI converters, one of which was the Personality Assessment Inventory (PAI) Clinical Scale for Depression. Another four PAI Clinical Scale scores also statistically significantly worsened post-inflection in the MCI converters including Somatization, Anxiety, Borderline, and Schizopheniform symptoms. The earliest inflection point (PAI Treatment Scale for Aggression [PAI-TSc-AGG]; p=0.06) nearly rivaled the earliest cognitive change 18 years prior to MCI diagnosis, though the confidence interval was very wide.

Figure 3.

Behavioral Test Score Ages of Inflection. The ages at which overall linear trajectories intersect for each behavioral measure showed a statistically significant change in slope between those who developed MCI and those who remained cognitively normal. Error bars represent 95% confidence intervals based on 500 bootstrapped samples.

Supplementary Table 3 shows the cross sectional comparisons five and ten years prior to MCI conversion. At ten years prior to conversion there were too few who had completed the Uniform Data Set tests and so these are omitted (including, for example Logical Memory). Otherwise, both comparisons found that all memory measures were lower in the eventual MCI converters, although the difference was greater at five than at ten years prior to diagnosis. In addition, at five years prior to diagnosis several non-memory measure differences reached statistical significance.

4.0. Discussion

The cognitive and behavioral ages of inflection identified in this longitudinal cohort study represent the earliest reported. This was likely facilitated by our young lower age limit of enrollment, the long duration of our observational period, and our analysis of incident diagnoses based upon within-subject decline. Our findings are consistent with and extend the findings of previous studies that have shown weaker memory performances at the preclinical stage of AD that predict subsequent dementia^17–21 with differences identified as much as 10 years before dementia diagnosis¹⁹, as well as with a meta-analysis of 47 studies showing preclinical decline in multiple cognitive domains²⁰. Multiple studies have found memory performance to be a sensitive predictor of MCI conversion to dementia^22–25. Our findings are also consistent with our previous reports of accelerated memory decline in healthy APOE e4 carriers^26,27 and extend them to a true “pre-MCI” stage. Our finding of a fourfold increase in amnestic MCI risk attributed to APOE e4 is consistent with previous studies²⁸, and our finding that male sex more than doubled MCI risk is consistent with the Olmsted County data^29,30. Regarding e2, despite its known protective effect against AD, carriers unexpectedly had a higher conversion rate than e4 heterozygotes, but the number of e2 converters was too small to draw further conclusions.

Nine memory measures from all five memory tests administered declined prior to diagnosis. Word list tests including the AVLT and SRT declined earlier than the logical memory test which is based on recalling a story, a finding shared by previous studies in both preclinical^17,19 and MCI based^22–25,31 cohorts. All memory measures were significantly lower as well in cross sectional comparisons five and ten years prior to MCI conversion, illustrating the sensitivity of word list memory paradigms that could be useful in preclinical AD therapeutic trials. Over the pre-MCI course, tests that were sensitive to other cognitive domains also declined, even though memory was the main clinical complaint. To what degree a specific test changed may reflect its “within-domain” sensitivity (for example, how sensitive a visuospatial test is to visuospatial impairment) or in some cases might instead reflect “cross-sensitivity” with memory. For example, the MMSE inflection point was roughly two years sooner than the Dementia Rating Scale (DRS). If we consider orientation, learning, and memory to all reflect memory, then memory makes up a higher proportion of the MMSE than the DRS. Tests can be considered as generally reflective of a given domain but in reality may reflect multiple domains to different degrees. Sixteen patients in this study developed clinically evident non-amnestic features (in addition to memory decline). A recent analysis of four large cohorts of AD dementia patients (that were not limited to incident cases) found that non-amnestic features were more likely in APOE e4 non-carriers³². Non-AD pathologies may partially account for our findings, though we found that APOE e4 carriers were heavily over-represented in our atypical group. Ten developed clinical synucleinopathies and another two were subclinical found at autopsy. Of these twelve, eight were e4 carriers, consistent with a recent large study of Lewy body related disease³³.

Behavioral measures also changed during the pre-MCI period despite the absence of any clinically identified behavioral problems, though the confidence intervals were wide in all cases. The source of these behavioral changes is not certain but the median raphe³⁴ and locus ceruleus³⁵ are sites of early pathology, and the neurochemistry of agitation in AD dementia reflects an alteration of multiple monoaminergic neurochemical projection systems³⁶. Additionally, in patients experiencing visual hallucinations in the setting of dementia with Lewy bodies³⁷ as well as Parkinson’s disease³⁸ cortical GABA-ergic interneuron signaling is impaired in the absence of visual cortex synucleinopathy. Regarding depression, for which we found three of four tests declining more rapidly in MCI converters relative to nonconverters post-inflection, we have previously shown in a normal aging cohort that there was no differential acceleration in depression scores among e4 carriers³⁹. During the transition from normal aging to MCI, however, there is a change in personality and behavioral measures including depression scores⁴⁰. This temporal pattern suggests that depression, rather than being a risk factor, is more likely an early manifestation of neurodegeneration. Further observation will be needed to confirm whether these subclinical changes during the pre-MCI stage are predictive of subsequent behavioral disorders.

AD has been thought to be the single most common cause of MCI^9,10, but neuropathological studies of patients dying with MCI are few. Petersen et al found mild AD-like changes in 10 of 15 patients dying with amnestic MCI at a mean age of 88.9 years⁴¹. In a series of 34 subjects with amnestic MCI (aMCI; mean age at death 88 years) and 15 with nonamnestic MCI (mean age at death (mean age at death 83 years), Dugger et al reported that 53% in each group met neuropathological criteria for AD though the aMCI group had denser temporal lobe neurofibrillary tangle pathology⁴². In an autopsy series of 1337 patients derived from 4 different Alzheimer Disease Centers, 343 had MCI that remained stable until death an average of 4 years later (mean age at death 89.5 years) and of these, 72.3% had Braak stage III or higher, the majority of which had additional pathologies⁴³. Similarly, in a large autopsy series among patients dying at a mean age of 90 years, 95% had at least one age-related pathology, nearly 60% had three or more, and the rate of mixed pathology was higher in those with a clinical diagnosis of AD⁴⁴. Whether the extent of mixed pathology at these very advanced ages is representative of pathology at the incident MCI age of our subjects is unclear.

There are several limitations of our study. First, MCI is pathologically heterogeneous and while AD likely accounts for the majority of cases in our APOE e4 enriched cohort, patients who have come to autopsy often have had a mixed neuropathological picture and occasionally a primarily non-AD cause^42,43,45. The latter is uncommon, however, among amnestic presentations, and given that the age at death of autopsied MCI patients has been in the late 80s it is far from clear whether the neuropathology of MCI at the younger ages in our study would be the same. Second, the general lack of amyloid and tau biomarkers precludes our ability to determine the relative timing of cognitive and biomarker changes in these subjects. However, unlike most previous studies, we found the timing of cognitive decline rivaled the earliest reported biomarker changes and should prompt reconsideration of the relative sensitivities of the methods used in other studies. This is consistent with findings from the Alzheimer’s Disease Neuroimaging Initiative Cohort in a study modeling neurodegenerative disease progression. Of seven included variables including cerebrospinal fluid amyloid and tau, hippocampal volumes, and several cognitive measures, it was the AVLT LTM score that changed before all others⁴⁶. Third, as shown in Supplementary Table 1, although age at entry for APOE e4 carriers and noncarriers was the same at entry, once subjects were reclassified by clinical outcome, older age at entry was associated with greater risk for incident MCI resulting (which is not surprising) in fewer time points of data preceding the inflection point that separated converters and nonconverters. This actually raises the possibility that the age of inflection may be even earlier on some tests, and underscores the importance of enrolling younger individuals in preclinical studies. Fourth, our findings reflect group rather than individual analyses, so we cannot account for individual differences that would likely impact a patient’s subsequent course. This would include the possibility that some of those in the nonconverter group will eventually become converters. Judging from the width of the confidence intervals memory decline, as expected, remains not only an early but also the most consistent finding. Finally, the changes on behavioral measures were of no current clinical significance, but appeared to vary greatly between individuals, so their implications for potential behavioral disturbances will require further observation.

In summary, we found cognitive and behavioral changes begin nearly 20 years before incident MCI, a duration that approaches the earliest reported biomarker changes.

Research In Context.

Systematic Review:

In addition to evaluating studies related to preclinical Alzheimer’s disease (AD) through traditional sources we reviewed the history of current thinking from the discoveries leading to the amyloid hypothesis to the phylogeny of amyloid precursor protein (APP), and asked how to resolve the evident discrepancies between this highly conserved gene/protein, amyloid targeted treatment failures, and the incontrovertible genetic evidence linking APP alteration to the cause of AD.

Interpretation:

In contrast to previously reported and widely assumed findings of an extreme time lag between the earliest reported amyloid biomarker changes and cognitive decline, we found the earliest cognitive changes occur 20 years before incident mild cognitive impairment, rivaling the earliest reported biomarker changes challenging current disease models.

Future Directions:

Alternate pathophysiological models exploring a loss of a critical APP-related function (rather than a gain of toxicity) better fit this chronology and should be further explored to identify alternate therapeutic targets.