Predictors of Life Expectancy in Autopsy-Confirmed Alzheimer’s Disease

Abstract

Background:

Life expectancy (LE) following Alzheimer’s disease (AD) is highly variable. The literature to date is limited by smaller sample sizes and clinical diagnoses.

Objective:

No study to date has evaluated predictors of AD LE in a retrospective large autopsy-confirmed sample, which was the primary objective of this study.

Methods:

Participants (≥50 years old) clinically and neuropathologically diagnosed with AD were evaluated using National Alzheimer’s Coordinating Center (N = 1,401) data. Analyses focused on 21 demographic, medical, neuropsychiatric, neurological, functional, and global cognitive predictors of LE at AD dementia diagnosis. These 21 predictors were evaluated in univariate analyses. Variables found to be significant were then entered into a forward multiple regression. LE was defined as months between AD diagnosis and death.

Results:

Fourteen predictors were significant in univariate analyses and entered into the regression. Seven predictors explained 27% of LE variance in 764 total participants. Mini-Mental State Examination (MMSE) score was the strongest predictor of LE, followed by sex, age, race/ethnicity, neuropsychiatric symptoms, abnormal neurological exam results, and functional impairment ratings. Post-hoc analyses revealed correlations of LE were strongest with MMSE ≤12.

Conclusion:

Global cognitive functioning was the strongest predictor of LE following diagnosis, and AD patients with severe impairment had the shortest LE. AD patients who are older, male, white, and have more motor symptoms, functional impairment, and neuropsychiatric symptoms were also more likely have shorter LE. While this model cannot provide individual prognoses, additional studies may focus on these variables to enhance predictions of LE in patients with AD.

INTRODUCTION

The prevalence and number of deaths attributed to Alzheimer’s disease (AD) is projected to rise significantly over the next 30 years [1]. The estimated life expectancy (LE) following an AD diagnosis is an often-asked question by patients and families to aid with planning, future care needs, and end-of-life decisions. However, estimates of LE in AD usually range between 3 and 12 years after symptom onset [2], making it difficult to provide reliable LE estimates to patients and families. Several risk factors for shorter survival following symptom onset have been identified, including older age at diagnosis [3, 4], male sex [2, 4], non-Hispanic white race/ethnicity [5], dementia severity [2, 4], neuropsychiatric symptoms [2, 6], apolipoprotein E ε 4 status [7], and comorbid medical conditions [8]. However, despite this research, there is no consensus regarding the relative contributions of each of these LE predictors.

Over the last 25 years, at least twenty predictors of LE in AD have been identified [3, 4, 6, 7, 9–13]. Most studies have evaluated only a few predictors in each model, partially due to database limitations (e.g., sample size, variables collected, etc.). As a result, the relative contributions of examined variables may be misrepresented due to unexplained variance. For example, investigations that examine the association between LE and just a few predictors (e.g., age, sex, motor signs, and Mini-Mental State Exam [MMSE]) may overestimate the contribution of these factors when failing to include other variables (e.g., cardiovascular conditions, seizures, neuropsychiatric symptoms, etc.). Larger investigations that have the sample size and design to examine more predictors have substantial variability in design and variables included [8]. For example, studies have differed in setting (community based versus memory disorder research centers), age inclusion criteria, sample size, clinical diagnoses (AD versus all-dementia), and duration of follow-up (for a list of reviewed literature, see Table 1). Some conflicting findings have also been reported regarding the role of neuropsychiatric symptoms, global cognitive deterioration, vascular risk factors, and functional dependence in relation to LE [2]. Another potentially significant limitation of most studies has been the reliance upon clinical diagnoses of AD, which are about 80% accurate in specialized dementia centers when compared to autopsy classification [14]. Only one study to date has utilized an autopsy-confirmed sample to evaluate LE in those with AD [7], but the main focus of that investigation was to evaluate mortality differences between those with AD and Lewy body disease, rather than examining specific predictors for LE in AD. Utilizing autopsy-confirmed data reduces inclusion of misdiagnosed subjects, which is especially important considering that some non-AD neurodegenerative conditions can have shorter LE compared to AD [7].

Table 1.

Overview of variables previously found to relate to life expectancy and their descriptions

Predictor	Description
Age [2,3,6, 8,31]	Age of diagnosis*
Disease duration prior to diagnosis [6]	Age of diagnosis minus clinician estimated age of onset (y)*
Sex [6, 8, 11]	Male/Female*
Race/Ethnicity [5]	Non-Hispanic Caucasian or Minority*
Education [32]	Years of education (maximum = 20)*
Marital Status [33]	Married/Not Married*
Body Mass Index [9]	[Weight (lbs) × 703] / [height (in)2]
Smoking [34]	Years of smoking tobacco*
Hypertension [2]	Current or past history of hypertension*
Diabetes [2, 8]	Current or past history of diabetes*
FDA-approved medication for AD [11]	Current use of tacrine, donepezil, rivastigmine, galantamine, or memantine
Alcohol misuse [35]	Clinically significant impairment occurring over a 12-month period impacting work, driving, legal, or social functioning*
Incontinence [36]	Urinary or fecal incontinence*
Abnormal Neurological Exam [8]	Presence of focal motor deficits, gait disturbance, or eye movement abnormalities suggestive of CNS disorder based on physical or neurological exam findings
Stroke [37]	Any history of stroke*
Seizures [38]	Any history of seizures*
Mini-Mental State Exam (MMSE) [6, 8, 11]	Score on MMSE (0 to 30)
Functional activities questionnaire (FAQ) [11]	10-item rating of functional impairment (0 to 30) over the past 4 weeks completed by informant. Items are related to financial management, shopping, hobbies, cooking, tracking current events, functional memory, and transportation.
Geriatric Depression Scale (GDS) [39]	15-item GDS total score (0 to 15)
Neuropsychiatric Inventory Questionnaire (NPI-Q) [10]	NPI-Q total score (0 to 36) completed by informant, rated over the last month. The measure contains 12 items rated on a 0 (not present) to 4 (severe) Likert scale: delusions, hallucinations, agitation/aggression, depression/dysphoria, anxiety, elation/euphoria, apathy, disinhibition, irritability, psychomotor disturbance, nighttime behaviors, and appetite.
No. of Apolipoprotein ε4 (APOE ε4) alleles [7]	0, 1, or 2 APOE ε4 alleles

^*Collected by the clinician(s), based on participant/informant report, medical records, and/or observation.

The primary aim of this study was to examine a comprehensive set of LE risk factors and identify the relative contribution of each for predicting reduced life expectancy in AD. We used the National Alzheimer’s Coordinating Center (NACC) dataset to examine demographic, psychiatric, functional, general cognitive, and health/medical factors as potential predictors of LE in a large autopsy-confirmed AD sample. The use of autopsy-data ensured accurate diagnosis and had the advantage of knowing the date of death for all participants. To our knowledge, this was the first study to examine a comprehensive set of LE risk factors in a large autopsy-confirmed AD sample.

METHODS

Participants

Data used in this study were collected from participants within NACC’s Uniform Data Set (UDS) and Neuropathology Data Set (NPDS) who visited Alzheimer Disease Research Centers (ADRCs) between September 2005 to August 2015. NACC aggregates demographic, clinical, health, cognitive, and autopsy data from National Institute of Aging funded ADRCs across the United States, and research using NACC data was approved by the University of Washington Human Subjects Division. All participants consent to have their clinical data used in NACC. Participants who were at least 50 years old, clinically diagnosed with primary AD [15, 16], and had AD neuropathologic change [17] defined as Braak neurofibrillary tangle stage ≥5 and frequent neuritic plaques were included. Both clinical and neuropathological criteria were used to improve the generalizability of our findings into clinical settings. Additionally, in order to maximize our sample size and keep our results as generalizable as possible, we did not exclude those with comorbid neuropathologic diagnoses, as comorbid and mixed pathologies are increasingly recognized in aging populations [18]. Clinical diagnoses of AD were made by either a single clinician or a multidisciplinary consensus panel utilizing revised NINCDS-ADRDA work group criteria [15, 16].

Measures

Predictor variables were selected based on review of the relevant literature relating to LE and/or disease progression after dementia diagnosis. After a review of the literature, 21 predictor variables were identified for examination (see Table 1 for an outline of their description, collection procedures, and citations of relevant literature). LE was defined as the number of months between date of AD diagnosis (i.e., date of visit in which participant was diagnosed with dementia due to AD) and date of death. Because our primary variable of interest was LE following diagnosis, we attempted to account for the variability in time between when a person began to show clinical symptoms and the time of presentation into clinic since this could have implications on estimating LE. To this end, number of years from estimated symptom onset to the date of AD diagnosis was used as a measure of disease duration prior to presentation.

Statistical analyses

To create the most parsimonious model, all potential predictors were first examined in univariate analyses. Exploratory univariate analyses (i.e., T-tests or one-way ANOVA for categorical groups and scale variables and Pearson correlations for pairs of scale variables) were conducted to identify which variables were associated with LE. Variables with significant associations in these univariate analyses were included in the final model. Significant associations in univariate analyses were defined as p < 0.08 for hypotheses of equal means and R ≥ 0.10 and p < 0.08 for hypotheses of association. A p < 0.08 was used to be more inclusive for variables that approached conventional levels of significance due to the exploratory nature of this step. The correlation minimum was set at R ≥ 0.10 to reduce the chances of a spurious relationship due to large sample sizes, as significance in Pearson correlations is greatly influenced by sample sizes [19]. All significant variables were then entered into a forward sequential multiple regression, with p < 0.05 defined as the point of entry and p ≥ 0.05 as the point of removal. Assumptions for Gaussian normal distribution, linearity, multicollinearity, heteroskedasticity, and multivariate outliers were reviewed, and missing data were removed case-wise for the purposes of these analyses. All statistical analyses were conducted using IBM SPSS versions 25 and 26. Missing data was excluded list-wise. For a flow chart of the variables evaluated and statistical procedures used, see Fig. 1.

Fig. 1.

Methodology flowchart.

RESULTS

Initial sample and exploratory univariate analyses

There were 1,418 participants that met initial inclusion criteria. Of those, 11 were excluded due to being outliers on symptom duration (years from onset to diagnosis > ± 3.29 SD from mean) and another five due to being outliers for number of years of cigarette smoking (years > ± 3.29 SD from mean). One additional participant was an outlier for both symptom duration and smoking. Of the 21 univariate analyses, 14 predictors were significantly associated with life-expectancy and entered into the forward regression model. Demographic predictors included age at AD diagnosis, race/ethnicity, sex, and marital status. Health predictors included history of diabetes, seizures, stroke, and if subjects were taking FDA-approved AD medication. Other predictors included incontinence, abnormal neurological exam, symptom duration, and scores on the Functional Activity Questionnaire (FAQ), MMSE, and Neuropsychiatric Inventory Questionnaire (NPI-Q). Results of univariate analyses and available sample sizes for each analysis can be found in Table 2.

Table 2.

Univariate analyses

Categorical variables	LE M-difference in months (SE)	p	n
Male sex	−05.60 (1.66)	0.001*	1,401
Non-Hispanic Caucasian	−05.94 (3.21)	0.060	1,351
Married/Not-married	−03.57 (1.86)	0.055	1,401
Hypertension	−00.72 (1.67)	0.666	1,387
Diabetes	−06.54 (2.75)	0.019*	1,389
FDA-Approved AD Medication	+05.69 (2.08)	0.007*	1,382
Seizures	−34.02 (3.55)	< 0.001**	1,354
Stroke	−13.78 (3.27)	< 0.001**	1,382
Incontinence	−21.65 (1.70)	< 0.001**	1,387
Abnormal Neurological Exam	−14.39 (1.69)	< 0.001**	1,365
Alcohol Misuse	−02.03 (3.60)	0.590	1,384
Apolipoprotein E ε4 Alleles	00.33 (2.08)	0.988	1,299
Scale variables	Correlation coefficient	p	N
Age at Diagnosis	−0.164	< 0.001**	1,401
Education (y)	+0.004	0.870	1,387
Symptom Duration	−0.279	< 0.001**	1,401
Body Mass Index	+0.062	0.038*	1,140
Smoking (y)	−0.042	0.129	1,289
Mini-Mental State Exam Score	+0.443	< 0.001**	1,227
Functional Activities Questionnaire	−0.427	< 0.001**	982
Geriatric Depression Scale	−0.051	0.102	1,029
Neuropsychiatric Inventory Questionnaire	−0.181	< 0.001**	1,322

LE, life expectancy,

^*< 0.05,

^**< 0.001

Final sample characteristics

The final model consisted of 764 participants with data on all 14 predictors. Life expectancy ranged from 1 to 131 months after AD diagnosis (M = 53.41, SD = 30.01). Most participants were diagnosed with AD dementia at their first visit (95%; N = 724). The sample was predominately non-Hispanic Caucasian (92%) and well-educated (M education = 15.3 years). Forty-eight participants were Black/African American, 8 were Asian, 5 were Hispanic, and 4 were other/undefined. Participants were evenly split between males (49%) and females (51%). The mean MMSE score was 16.1 with a standard deviation of 8.4 and ranged from 0 to 30 at time of clinical diagnosis. At autopsy, frequent neuropathological comorbidities included cerebral amyloid angiopathy (33% mild, 30% moderate, and 16% severe), infarcts/lacunes (17%) and Lewy bodies (43%), though only 13% had neocortical Lewy bodies suggesting mixed AD and Lewy body disease according to McKeith criteria [20]. Other less frequent comorbidities included Pick’s disease (N = 2), Corticobasal degeneration (N = 1), progressive supranuclear palsy (N = 3), and single or multiple hemorrhages (N = 23). See Table 3 for a description of the final sample characteristics.

Table 3.

Final Sample Characteristics (N = 764)

	n (%)
Sex
Females	387 (51)
Males	377 (49)
Race/Ethnicity
Caucasian	704 (92)
African American	48 (6)
Hispanic	5 (< 1)
Asian	8 (1)
Other	4 (< 1)
Education
< 12	34 (5)
12	159 (21)
13 to 15	124 (16)
16	217 (28)
17+	224 (29)
Marital Status
Married	521 (68)
Not Married	243 (32)
Age of AD Diagnosis
< 65	133 (17)
66 to 74	186 (24)
75 to 84	332 (44)
85+	113 (15)
Symptom Onset to Diagnosis (y)
< 2	32 (4)
2 to 4	153 (20)
4 to 6	191 (25)
6 to 8	154 (20)
8+	234 (31)
Age of Death
< 65	64 (8)
66 to 74	165 (22)
75 to 84	273 (36)
85+	265 (35)
Life Expectancy (months)
< 12	58 (8)
12 to 36	194 (25)
37 to 60	215 (28)
61 to 84	185 (24)
85 to 108	81 (11)
109+	31 (4)
Co-morbid Neuropathology*
Neocortical Lewy bodies	97 (13)
Infarcts/Lacunes	135 (17)
Hemorrhage(s)	23 (3)
CAA (mod. to sev.)	355 (47)
FTLD Tauopathy	13 (2)
Pick’s disease	2 (< 1)
Corticobasal degeneration	1 (< 1)
Progressive supranuclear palsy	3 (< 1)
FTLD tau-negative pathology	18 (2)

Sx, symptom; FTLD, frontotemporal lobar degeneration; CAA, cerebral amyloid angiopathy.

^*Neuropathological characteristics have a significant degree (~30%) of missing data for assessment of FTLD pathology due to changes in neuropathological procedures of the dataset.

Multivariate analysis

LE at the time of diagnosis was initially predicted by 14 factors that involved measures of sociodemographics, medical history, neurological exam, and cognitive status. Of the 14 variables entered into the forward entry multiple regression, seven were found to be significant predictors: MMSE, sex, age of diagnosis, race/ethnicity, NPI-Q, FAQ, and abnormal neurological exam. These seven predictors explained 27% of the variance in LE, F (1,756) = 40.7, p < 0.001, R² = 0.274. Each point increase on the MMSE was associated with an additional 1.2 months of LE. LE for males was 8.6 months shorter than females. Ten years of age was associated with 4 months LE. Non-Hispanic Caucasians had 12.9 months less survival time. Every 10 points on the NPI-Q was associated with 5 months of LE, as was an abnormal neurological exam. Lastly, 10 points on the FAQ was associated with 3.5 months of LE. Steps of entry, and coefficient results can be found in Tables 4 and 5.

Table 4.

Forward entry regression model summary

Model	R	R2	Adjusted R2	SE	Change statistics
					R2 Change	F Change	p
1	0.451a	0.204	0.203	26.80	0.204	194.85	< 0.001
2	0.473b	0.224	0.222	26.48	0.020	19.82	< 0.001
3	0.494c	0.244	0.241	26.15	0.020	19.85	< 0.001
4	0.505d	0.255	0.251	25.98	0.011	11.19	0.001
5	0.513e	0.263	0.258	25.85	0.008	8.58	0.003
6	0.519f	0.270	0.264	25.75	0.007	6.95	0.009
7	0.523g	0.274	0.267	25.70	0.004	4.15	0.042

^aPredictors: Constant, MMSE.

^bPredictors: Constant, MMSE, Sex.

^cPredictors: Constant, MMSE, Sex, Age of Diagnosis.

^dPredictors: Constant, MMSE, Sex, Age of Diagnosis, Race/Ethnicity.

^ePredictors: Constant, MMSE, Sex, Age of Diagnosis, Race/Ethnicity, NPI-Q.

^fPredictors: Constant, MMSE, Sex, Age of Diagnosis, Race/Ethnicity, NPI-Q, Abnormal Neurological Exam.

^gPredictors: Constant, MMSE, Sex, Age of Diagnosis, Race/Ethnicity, NPI-Q, Abnormal Neurological Exam, FAQ.

Table 5.

Regression coefficients

							95% Confidence interval for B
Model		B	SE	Beta	t	p	Lower	Upper
1	Constant	27.37	2.10		13.02		23.24	31.50
	MMSE	1.62	0.12	0.45	13.96	<0.001	1.39	1.85
2	Constant	32.24	2.35		13.73	<0.001	27.63	36.84
	MMSE	1.58	0.12	0.44	13.74	<0.001	1.35	1.80
	Sex	−8.56	1.92	−0.14	−4.45	<0.001	−12.33	−4.78
3	Constant	67.51	8.25		8.18		51.32	83.71
	MMSE	1.57	0.11	0.44	13.81	<0.001	1.35	1.79
	Sex	−10.07	1.93	−0.17	−5.22	<0.001	−13.86	−6.29
	Age of Diagnosis	−0.46	0.10	−0.14	−4.46	<0.001	−0.66	−0.26
4	Constant	65.17	8.23		7.92	<0.001	49.02	81.31
	MMSE	1.59	0.11	0.44	14.06		1.37	1.81
	Sex	−9.42	1.93	−0.16	−4.89	<0.001	−13.20	−5.64
	Age of Diagnosis	−0.45	0.10	−0.14	−4.38	<0.001	−0.65	−0.25
	Race/Ethnicity	11.76	3.52	0.11	3.35	0.001	4.86	18.66
5	Constant	71.15	8.44		8.44		54.59	87.71
	MMSE	1.51	0.12	0.42	13.12	<0.001	1.29	1.74
	Sex	−9.29	1.92	−0.16	−4.85	<0.001	−13.06	−5.53
	Age of Diagnosis	−0.46	0.10	−0.15	−4.58	<0.001	−0.66	−0.27
	Race/Ethnicity	13.12	3.53	0.12	3.72	<0.001	6.19	20.05
	NPI-Q	−0.62	0.21	−0.10	−2.93	0.003	−1.03	−0.20
6	Constant	73.44	8.45		8.69		56.86	90.02
	MMSE	1.43	0.12	0.40	12.03	<0.001	1.20	1.66
	Sex	−8.78	1.92	−0.15	−4.58	<0.001	−12.55	−5.02
	Age of Diagnosis	−0.46	0.10	−0.14	−4.50	<0.001	−0.65	−0.26
	Race/Ethnicity	12.86	3.52	0.12	3.66	<0.001	5.95	19.76
	NPI-Q	−0.60	0.21	−0.09	−2.84	0.005	−1.01	−0.18
	Abnormal Neurological Exam	−5.30	2.01	−0.09	−2.64	0.009	−9.24	−1.35
7	Constant	80.65	9.14		8.82		62.71	98.60
	MMSE	1.22	0.16	0.34	7.67	<0.001	0.91	1.53
	Sex	−8.66	1.92	−0.14	−4.52	<0.001	−12.42	−4.89
	Age of Diagnosis	−0.41	0.10	−0.13	−4.03	<0.001	−0.62	−0.21
	Race/Ethnicity	12.87	3.51	0.12	3.67	<0.001	5.98	19.75
	NPI-Q	−0.52	0.21	−0.08	−2.43	0.016	−0.94	−0.10
	Abnormal Neurological Exam	−5.09	2.01	−0.08	−2.54	0.011	−9.03	−1.15
	FAQ	−0.34	0.17	−0.09	−2.04	0.042	−0.67	−0.01

B, unstandardized beta; Beta, standardized beta.

Additional MMSE post-hoc analyses

Post-hoc analyses were conducted on MMSE as a predictor of LE due to MMSE explaining the majority of variance in the model (20% of the 27% total variance). MMSE was stratified into severe (MMSE ≤ 12, n = 237), moderate (MMSE = 13 to 20, n = 243), and mild levels of cognitive impairment (MMSE > 20, n = 284). One-way ANOVA revealed that LE differed between these three groups, F = 86.534 (2, 761), p < 0.001. Tukey’s post-hoc test revealed significant differences (ps < 0.001) in LE between mild (M = 35.8, SD = 26.6), moderate (M = 54.53, SD = 27.23), and severe (M = 67.14, SD = 27.49) cognitive impairment. Pearson correlations were used to examine the linear association between MMSE score and LE for each stratified MMSE group. When analyzing the data in this fashion, only the lowest range of MMSE scores (i.e., MMSE ≤ 12) was significantly associated with LE, R = 0.265, p < 0.001. MMSE scores ranging from 13 to 20 (R = 0.088, p = 0.086) and > 20 (R = 0.066, p = 0.132) did not significantly correlate with LE. In addition, because the MMSE is also heavily weighted by orientation items (i.e., 10 of 30 possible points), post-hoc analyses analyzing the association between LE and orientation versus non-orientation items were conducted. Pearson correlations evaluating the linear association between LE and MMSE items were significant for both orientation (R = 0.413, p < 0.001) and non-orientation (R = 0.428, p < 0.001) items.

Due to these findings, and the wide range of cognitive impairment in our sample, we elected to conduct a sensitivity analysis to evaluate whether the final predictors were significant at various levels of cognitive impairment (i.e., dementia severity). We did this by comparing the final model predictors in separate regressions in these three MMSE groups: severe (MMSE ≤ 12, n = 237), moderate (MMSE = 13 to 20, n = 243), and mild (MMSE > 20, n = 284). As can be seen in Table 6, predictors remained significant at severe and moderate levels of cognitive impairment, but only sex remained significant and had a similar effect (approximately 8 months shorter survival time) at milder levels of cognitive impairment.

Table 6.

Models and predictors of LE by dementia severity

							95% Confidence interval for B
Model	Predictors	B	SE	Beta	t	p	Lower	Upper
MMSE ≤ 12	(Constant)	118.15	19.15		6.17	<0.001	80.43	155.87
n = 237	Age of AD Diagnosis	−0.47	0.17	−0.17	−2.78	0.006	−00.79	−0.14
F = 7.260	Race/Ethnicity	13.37	5.75	0.14	2.33	0.021	02.05	24.70
p < 0.001	Gender	−8.76	3.32	−0.16	−2.64	0.009	−15.30	−2.22
R2 = 0.159	FAQ	−1.19	0.58	−0.13	−2.03	0.044	−02.34	−0.03
	Abnormal Neurological Exam	−8.81	3.32	−0.16	−2.65	0.008	−15.35	−2.27
	NPI-Q	−0.70	0.31	−0.14	−2.28	0.023	−01.31	−0.10
MMSE=13–20	(Constant)	107.27	14.10		7.61	<0.001	79.49	135.05
n = 243	Age of AD Diagnosis	−0.39	0.18	−0.14	−2.16	0.032	−00.75	−00.03
F = 6.252	Race/Ethnicity	12.22	6.02	0.13	2.03	0.044	00.36	24.09
p < 0.001	Gender	−8.79	3.44	−0.16	−2.56	0.011	−15.57	−02.02
R2 = 0.115	FAQ	−0.65	0.28	−0.15	−2.28	0.023	−01.21	−00.09
	Abnormal Neurological Exam	−9.90	3.47	−0.17	−2.85	0.005	−16.74	−03.06
	NPI-Q	−0.37	0.44	−0.05	−0.83	0.406	−01.24	00.50
MMSE ≥ 21	(Constant)	98.27	15.08		6.52	<0.001	68.58	127.96
n = 284	Age of AD Diagnosis	−0.28	0.19	−0.09	−1.45	0.149	−0.66	00.10
F = 2.749	Race/Ethnicity	11.68	6.63	0.10	1.76	0.079	−01.36	24.72
p = 0.013	Gender	−8.37	3.33	−0.15	−2.52	0.012	−14.92	−01.82
R2 = 0.056	FAQ	−0.32	0.24	−0.09	−1.37	0.171	−00.79	00.14
	Abnormal Neurological Exam	2.75	3.70	0.04	0.74	0.459	−04.55	10.04
	NPI-Q	−0.52	0.41	−0.08	−1.28	0.201	−1.32	0.28

DISCUSSION

While the literature on LE in AD is extensive and study designs vary widely, this was the first large-scale comprehensive investigation of previously identified risk factors in an autopsy-confirmed sample. Overall level of cognitive impairment, as measured by the MMSE, accounted for the greatest variance in a model that originally included 14 factors. The six other predictors that accounted for the remaining variance of the model, were sex, age of diagnosis, race/ethnicity, neuropsychiatric symptoms (as measured by the NPI-Q), abnormal neurological exam, and functional abilities (as measured by the FAQ). Despite these findings, only 27% of LE variance was explained, suggesting that more explanatory variance is needed to use this model for individual predictions. Importantly, analyses also suggested that predictors vary by severity of dementia, as the majority of the significant predictors, besides male sex, were not predictive of LE at milder levels of dementia (i.e., MMSE > 20). As such, results from the remaining predictors, including age of diagnosis, race/ethnicity, neuropsychiatric symptoms, abnormal neurological exam, and functional abilities do not appear to generalize to individuals with milder dementia. Overall, attempting to predict LE in those with AD is extremely complex, differs by dementia severity, and may be dependent on a number of factors that are not available in the NACC dataset (e.g., sudden onset causes of death, genetics, lifestyle factors, etc.). Nonetheless, our findings are in keeping with prior studies using clinical samples, and additionally identify variables that may be of special interest in future studies.

MMSE performance was the strongest predictor of LE following AD diagnosis, accounting for approximately 20% of the 27% explained variance. Importantly, however, post-hoc analyses revealed that only those with the lowest scores (i.e., MMSE < 13) had a significant correlation with LE, suggesting that the predictive ability of MMSE is attenuated in those with less severe dementia at the time of diagnosis. These findings are consistent with a 2004 study by Larson et al., which evaluated 521 community-based individuals with AD in the Seattle area. The authors found that LE was associated with a MMSE score of less than 17 and that those with a 5-point decline one year after AD diagnosis had shorter LE (HR: 1.60), suggesting those with more advanced global cognitive impairment and rapid global cognitive decline were at risk of shorter LE [8]. Furthermore, additional research has shown that the MMSE is not a particularly sensitive or specific measure to predict disease progression in those with MCI. A recent Cochrane review found that baseline MMSE had a sensitivity ranging from 27 to 89% and a specificity ranging from 32% to 90% in predicting progression from MCI to AD dementia [21], suggesting that the MMSE is not an adequate predictor of AD progression in those with early cognitive impairments. Interestingly, although it may be expected that those with severe dementia would perform worse on orientation items, we found similar, moderately strong associations between LE and both orientation and non-orientation items in this sample. Importantly, we also found that severity of dementia (as measured by the MMSE) had a major impact on predictive factors, as most predictors only remained significant at lower levels of cognitive functioning, suggesting that results do not generalize to individuals with milder dementia. Overall, it will be important for future investigations to examine the role of cognitive impairment in predicting LE in those with milder dementia or those with MCI using more sensitive neuropsychological measures. While most studies have not examined comprehensive neuropsychological measures as LE predictors, some studies have found that memory [10], verbal fluency [22], and executive functioning [23] are associated with LE, suggesting these particular domains, along with global cognitive screening measures, may be important predictors for LE determinations.

Demographic factors were also associated with LE, as male sex, later age of diagnosis, and non-Hispanic Caucasian race/ethnicity were associated with reduced survival in moderate to severe dementia. Only male sex was predictive of LE in those with milder dementia. Our findings are in line with a series of investigations finding that male sex [6, 8] and advanced age [2, 3, 6, 8, 31] were moderate to strong predictors of LE, but the research regarding the influence of race and ethnicity on LE is limited. In a recent study, Mayeda and colleagues reviewed over 350,000 medical records and identified over 60,000 incident dementia cases for a 14-year period. Findings suggested that non-Hispanic Caucasians had the shortest survival times at a median of 3.1 years after a diagnosis of dementia, followed by American Indians (3.4 years), African Americans (3.7 years), Hispanics (4.1 years), and Asian Americans (4.4 years) [5]. Importantly, their findings were consistent with mortality patterns in those without dementia, suggesting that similar factors may be driving mortality differences in those with and without dementia for each ethnicity [5]. Our results echoed Mayeda’s findings in that we found that non-Hispanic whites had reduced survival times compared to minorities, even after accounting for other variables in a multivariate model. However, our sample was 95% non-Hispanic Caucasian, with the majority of our minority participants being African-American, forcing dichotomization of groups into minority versus non-minority status. Thus, it is difficult to know the extent of the association between race/ethnicity and LE based on these results alone, and our sample characteristics limit generalizability to diverse minority populations. Additional replication and evaluation of race/ethnicity when evaluating LE is needed, along with investigations of intersecting diversity factors such as sex by race interactions, quality of education, socioeconomic status, and rural versus non-rural communities (where access to specialized dementia care may differ).

Our results suggest that abnormal neurological exam findings, defined as focal motor deficits, gait disturbance, and/or eye movement abnormalities, were associated with ~5 months of reduced LE in individuals with moderate to severe dementia. Abnormal motor signs are not uncommon in AD, occurring in approximately 1/3 of cases [24] and even in preclinical manifestations [25]. Extra-pyramidal signs were among the first predictors of LE that were examined in those with dementia and AD [6, 8]. The presence of motor symptoms may suggest the presence of comorbid pathology, or a mixed pathology presentation, which has been found to reduce survival times during the disease course [7, 26]. Alternatively, the presence of motor signs may represent higher levels of disease burden and or involvement of frontosubcortical networks, as the presence of motor features in AD has been associated with worse functional end-points [27]. This may even be more likely considering the level of cognitive impairment observed in the present sample, which suggests a greater severity of overall disease burden.

The presence and severity of neuropsychiatric symptoms, as measured by the NPI-Q, was associated with reduced LE in AD in those with moderate to severe dementia, though this accounted for the smallest effect on LE. Hallucinations and delusions have been found to be a significant predictor of LE in previous studies [10, 28], but other neuropsychiatric symptoms are less studied. For example, it is unknown if other symptoms such as apathy, appetite changes, and nighttime behaviors may be predictors of LE. The mechanisms underlying the disease process that leads to hallucinations in AD are largely unknown, but a recent review identified several possible neural mechanisms including atrophy, cortical thinning, network disruption, occipital periventricular hyperintesities, and right frontal hypoperfusion [29]. Alternatively, the presence of these factors may simply reflect greater disease severity and/or comorbid pathology, similar to abnormal neurological signs and MMSE performance. Nonetheless, our results suggest that additional investigation of neuropsychiatric symptoms among AD patients is needed to evaluate this relationship.

Functional status as measured by the FAQ was found to be a significant predictor of LE in those with moderate to severe dementia, with 10 points on the FAQ corresponding to 3.5 months of reduced LE after diagnosis. Previous investigations also found that decline in instrumental activities of daily living is associated with reduced LE. In a 2014 retrospective study by Wattmo and colleagues, 791 deceased individuals were diagnosed with mild to moderate AD. They found that more impaired activities of daily living and self-maintenance (i.e., bathing, toileting, etc.) were associated with reduced survival. While this study utilized a different measure of functional abilities [30], all areas assessed (e.g., cooking, finances, transportation, medication management, etc.) overlap considerably with the FAQ. Our study would suggest that similar to neuropsychiatric symptoms, the impact of functional changes, while an important predictor, likely accounts for a smaller effect on LE than global cognitive impairment, though our model cannot speak to the interaction between functional and cognitive change, which may be an important area of future study.

While this study is the largest comprehensive autopsy-confirmed investigation of LE in AD to date, we acknowledge several limitations. First, even with a multifactorial model that included a vast array of predictors, only 27% of the variance in LE was explained, suggesting that additional variables may be needed to improve predictability. We also did not have cause of death available in the dataset, which may additionally have improved predictability. Second, the generalizability of our findings to individuals with diverse backgrounds is limited, given the vast majority of NACC participants are well-educated, non-Hispanic Caucasian individuals. Since minority status was associated with longer survival times, it is important to consider the impact of race and ethnicity, educational attainment, and related factors in future investigations. Third, although we attempted to account for symptom duration and/or timing of dementia diagnosis as a risk factor, the majority of participants were diagnosed with dementia at their baseline NACC visit, and the onset of AD symptoms prior to clinical evaluation is unknown. Given the frequency of severe cognitive impairment as measured by the MMSE, some individuals may have been suffering from symptoms for a longer period than reported, and thus may have skewed our findings to result in shorter life expectancy after diagnosis. We may have observed in this phenomenon in our post-hoc analyses, when we found that most predictors were not significant at milder levels of dementia. Along these lines, because of the severity of cognitive impairment in this sample, our predictors do not appear to generalize to those with milder dementia, and future studies are needed to further explore predictors of LE in this particular population. Finally, we acknowledge the inherent limitations of self/informant report for many of the NACC variables examined. Despite the fact that many subjects in the NACC dataset have informants for collateral information (and the FAQ and NPI-Q are informant-based), relying only on self-report of patients and informants rather than confirmed medical documentation of vascular and health factors reduce confidence in negative findings among these variables (e.g., diabetes, hypertension, etc.).

Conclusions

To our knowledge, this is the largest-ever autopsy-confirmed investigation of LE after an AD diagnosis. Out of 21 potential predictors, we found 7 predictors explained approximately 27% of the variance in LE in a sample of 764 individuals with autopsy-confirmed AD. Global cognitive status at the time of clinical diagnosis had the largest effect on LE, with the strongest association occurring in those with MMSE ≤ 12. We also found that global cognitive status had a major impact on other predictors, as only sex was a significant predictor of LE across severe, moderate, and mild stages of dementia. In those with moderate to severe dementia, we also found age, race, functional decline, neuropsychiatric symptoms, and abnormal neurological exam results were significant predictors of LE. Prediction of LE is a complex endeavor, and the current results suggest several factors that may be important for future investigations.