Risk Aversion and Alzheimer’s Disease in Old Age

Abstract

Objective:

To test the hypothesis that late-life risk aversion is partly a prodromal sign of dementia.

Design:

Longitudinal clinical-pathologic cohort study.

Setting:

Participants’ residences in Chicago area.

Participants:

A total of 874 older persons without dementia at enrollment.

Measurements:

At baseline, risk aversion was assessed with questions involving choices between certain smaller rewards and uncertain larger rewards. At annual intervals thereafter, participants underwent evaluations that included cognitive testing and diagnosis of mild cognitive impairment (MCI) and dementia. At death, a neuropathologic examination was done to quantify common pathologies linked to dementia.

Results:

Risk aversion at study onset ranged from 0.05 to 0.91 (mean = 0.32, SD = 0.31). During a mean of 4.6 years of follow-up, 123 (of 874) developed dementia. Higher risk aversion was associated with higher dementia incidence (hazard ratio [HR] = 2.08, 95% confidence interval [CI]: 1.18, 3.65) and more rapid decline in episodic (estimate= −0.062, SE = 0.019, t[3677]=−3.33, p<.001) and semantic (estimate = −0.039, SE = 0.015, t[3655]=−2.61, p = .009) memory but not in other cognitive systems. Of 702 people without cognitive impairment at baseline, 223 developed incident MCI. Higher risk aversion was associated with higher incidence of MCI (HR = 2.10, 95% CI: 1.34, 3.29) and more rapid episodic memory decline. In 181 neuropathologically examined individuals, higher risk aversion was associated with higher levels of plaques, tangles, and cerebral amyloid angiopathy.

Conclusion:

The results support the hypothesis that high risk aversion in old age is partly an early sign of the pathology of Alzheimer’s disease.

INTRODUCTION

Numerous economics, behavioral economics, and neuroeconomics studies have shown that individual preferences related to risk and time play a critical role in human decision making and predict economic outcomes, financial and healthcare decisions, and even health and social behaviors (1–5). Risk aversion, the tendency to prefer a certain (i.e., safe) outcome over an uncertain (i.e., risky) but potentially better outcome, is an important driver of behavior and influences a host of behaviors relevant for maintaining independence, health and wellbeing across the lifespan. Individuals who are risk averse tend to invest more conservatively (e.g., invest in low yield but safe options such as Treasury bonds instead of higher yield but riskier options such as stocks), prefer positions of employment with high stability but limited opportunity for advancement (1,5), are less likely to self-employ (5), and exhibit poorer financial and healthcare decision making ability (our DM ref—already in your ref list). Individuals who are risk averse also are less likely to engage in harmful health behaviors such as cigarette smoking, excessive alcohol use, and seatbelt non-use (3,4), and tend to marry sooner than those who are less risk averse (6). These associations are thought to be due in part to the tendency for risk averse persons to prefer “safe” choices over uncertain or unsafe choices, in some cases limiting exposure to key learning opportunities, and risk aversion shares features with personality characteristics such as harm avoidance. Interestingly, however, increasing evidence suggests that risk aversion changes with age, such that older persons are less likely to take risks compared to younger persons (5, 6).

The bases of these age-related differences in risk aversion are uncertain. We recently reported that a prior history of more rapid cognitive decline predicted a higher level of risk aversion even among older persons without dementia (11) and that risk aversion is related to functional connectivity in some brain regions known to be impacted early in dementia (12). These observations suggest that increased risk aversion in old age may be related to late-life cognitive disorders such as dementia and its precursor, mild cognitive impairment (MCI), either as a risk factor or disease consequence.

The present study examines the association between risk aversion and dementia. Analyses are based on older persons without dementia at the time of risk aversion assessment. They subsequently had annual evaluations that included cognitive testing and classification of dementia and MCI, allowing us to test the hypotheses that higher risk aversion predicts higher incidence of dementia and MCI and faster rate of cognitive decline. Participants who died underwent a brain autopsy and uniform neuropathologic examination, allowing us to test the hypothesis that higher risk aversion is partly a consequence of dementia related to pathologic processes.

METHODS

Participants

All study participants are from the Rush Memory and Aging Project, an ongoing longitudinal clinical-pathologic cohort study that began in 1997 and involves older persons from the Chicago metropolitan region who agreed to annual clinical evaluations and a brain autopsy at death (13). Risk aversion was added to the evaluations in 2010 as part of a substudy on decision making. Participants signed informed consent forms for the parent study and the decision making substudy, which were approved by the institutional review board of Rush University Medical Center.

At the time of these analyses (11/08/2017), of 1,146 parent study participants potentially eligible for the decision making substudy (Figure 1), 1,059 had been assessed, 60 had not yet completed the assessment, and 27 had refused (92.4% rate of participation in substudy). Three persons had some missing decision making data, and we excluded 62 persons with dementia. Of the remaining 994 persons eligible at the decision making substudy baseline, 27 died before they could be followed and 45 had not yet reached their first follow-up date. This left 922 people eligible for follow-up and follow-up data were available on 874 (94.8%). During the observation period, 4,708 annual clinical evaluations were scheduled in survivors of this group, 4,551 of the evaluations (96.7%) were sufficiently complete to allow classification of the presence or absence of dementia and 4,644 (98.6%) yielded a valid composite measure of global cognition. They had a mean age of 81.3 (SD = 7.6) at baseline and a mean of 15.3 years of education (SD = 3.0); 77.1% were women and 91.2% were white and not Latino.

Figure 1.

Flow diagram showing the composition of the groups used in analyses.

During follow-up, 214 persons died and 197 had a brain autopsy (92.1% rate of participation in brain autopsy), with full neuropathologic data available in 181 individuals (Figure 1). They died at a mean age of 90.7 (SD = 6.3) and had a mean of 4.1 years of follow-up (SD = 1.6) and 14.9 years of education (SD = 2.9); 74.6% were women and 96.7% were white and not Latino.

Clinical Evaluation

Participants had annual evaluations with a medical history, neurological examination, and cognitive testing. Based on the evaluation (and without reference to previously collected data), participants were classified with respect to dementia and MCI. The diagnosis of dementia required a history of cognitive decline and impairment in at least two cognitive domains (14). Persons who did not meet dementia criteria but had cognitive impairment were diagnosed with MCI. Further information on these MCI diagnostic criteria and their relation to clinical and pathologic outcomes is published elsewhere (13).

Assessment of Cognitive Function

A battery of 19 cognitive tests was administered annually. One aim of the testing was to support clinical classification. To this end, an algorithm was developed using educationally adjusted scores on 11 tests to rate impairment in five domains. After examining scores on all 19 tests, a neuropsychologist reviewed the algorithmic ratings of cognitive domains, agreed or disagreed with each rating, and in the event of disagreement, provided a new rating (and reasons for disagreement) (15).

The second aim was to assess change in cognitive abilities. For this aim, we excluded 2 tests with skewed distributions (Mini-Mental State Examination, Complex Ideational Material) leaving 17 tests. In analyses, we used a composite measure of global cognition based on all 17 tests and, supported by factor analyses in this (16) and other (17, 18) cohorts, composite measures of episodic memory (based on 7 tests: immediate and delayed recall of Logical Memory Story A and East Boston Story; Word List Memory, Word List Recall and Word List Recognition), semantic memory (based on 3 tests: Boston Naming Test, Verbal Fluency, Word Reading), working memory (based on 3 tests: Digit Span Forward, Digit Span Backward, Digit Ordering), perceptual speed (based on 2 tests: Symbol Digit Modalities Test, Number Comparison), and visuospatial ability (based on 2 tests: Standard Progressive Matrices, Judgment of Line Orientation). Raw test scores were converted to z scores, using the baseline mean and SD, and z scores of component tests were averaged to yield the composite cognitive measures.

Assessment of Risk Aversion

Risk aversion was assessed with 10 questions that involved choosing between a smaller certain reward and a larger uncertain reward (19, 20). These questions were piloted in our cohort and deemed to have adequate psychometric properties and be suitable for use in a large epidemiologic study (6, 11, 21). Questions were in the following format: “Would you prefer $15 for sure, OR a coin toss in which you will get $ (an amount larger than $15) if you flip heads or nothing if you flip tails?” The potential gambling gain varied from $20 to $300 across questions. An index of risk aversion was derived from the 10 responses as described in the Statistical Analysis section.

Neuropathologic Examination

Most persons with dementia and MCI have a mixture of neuropathologic conditions. The aim of the neuropathologic examination, therefore, was to obtain quantitative measures of common neurologic conditions implicated in late-life loss of cognition. We followed a standard protocol for brain removal and sectioning and preservation of tissue (22).

There were 5 postmortem markers of cerebrovascular disease. We cut the cerebral hemispheres coronally into 1-cm slabs which were examined for gross infarcts. We assessed microinfarcts in 9 regions (6 cortical, 2 subcortical, 1 midbrain) using hematoxylin and eosin stain. Gross and microscopic infarcts were treated as present or absent in analyses. Amyloid-beta immunostaining in 4 regions was used to assess cerebral amyloid-beta angiopathy. We rated amyloid-beta deposition in the meningeal and parenchymal vessels of each region on a 5-point scale and averaged regional ratings to yield a continuous composite score (23). Visual inspection of the vessels in the circle of Willis was used to rate arteriosclerosis, and arteriolar sclerosis was assessed histologically from hematoxylin and eosin stained sections of the anterior basal ganglia (24). In analyses, moderate/severe ratings of atherosclerosis and arteriolar sclerosis were treated as present and less severe ratings were treated as absent.

There were 4 postmortem markers of neurodegenerative disease. A modified Bielschowsky silver stain was used to identify neurofibrillary tangles, neuritic plaques, and diffuse plaques in 5 brain regions, with regional scores of each pathology standardized and averaged to provide a continuous composite measure of plaques and tangles, the pathologic hallmarks of Alzheimer’s disease (AD) (25). TDP-43 cytoplasmic inclusions were assessed in 6 brain regions using monoclonal antibodies to phosphorylated TDP-43 (p5409/410; 1:100) and regional ratings were averaged to yield a continuous summary measure of TDP-43 pathology (26). Hippocampal sclerosis was defined as severe neuronal loss and gliosis in the hippocampus or subiculum (26). Lewy bodies were assessed in 6 brain regions using a monoclonal antibody to alpha-synuclein (27). Hippocampal sclerosis and Lewy bodies were treated as present or absent in analyses.

Statistical Analysis

The risk aversion coefficient was estimated using participants’ responses to the 10 risk aversion questions, as previously described (6, 11, 21, 28). For participant i with a risk aversion coefficient γ_i, the expected utility of the gamble at the jth question, $U_{ij}^G$, is defined by the function:

$$U_{ij}^G=\frac{0.5\times Gai{n}_j^{1-{\gamma }_i}}{1-{\gamma }_i}$$ Equation 1:

where Gain_j is the winning dollar amount at the jth gamble. The safe option payoff for ith participant at jth question has the expected utility:

$$U_{ij}^S=\frac{Saf{e}_j^{1-{\gamma }_i}}{1-{\gamma }_i}$$ Equation 2:

where Safe_j is the safe gain for the jth question. If observed outcomes in the trials are Y_ij and the decision of choosing the gamble is Y_ij = 1, the probability P(Y_ij = 1) depends on the difference between expected utility of the gamble and safe option. Therefore, the odds of choosing the gamble over safe option were calculated as:

$$\frac{P\left(Y_{ij}=1\right)}{1-P\left(Y_{ij}=1\right)}=e^{U_{ij}^G-U_{ij}^S}$$ Equation 3:

Here e is the exponential function. A positive $U_{ij}^G-U_{ij}^S$ suggests that a participant favored the gamble (i.e., odds greater than 1). γ_i was estimated using a nonlinear mixed model, specified as the following

$$\mathit{\text{Logit}}\left(P\left(Y_{ij}=1\right)\right)=\frac{0.5\times Gai{n}_j^{1-f^{-1}\left({\beta }_0+{\mu }_i\right)}-Saf{e}_j^{1-f^{-1}\left({\beta }_0+{\mu }_i\right)}}{1-f^{-1}\left({\beta }_0+{\mu }_i\right)}$$ Equation 4:

Here we parameterized γi using f⁻¹(β₀ + μ_i) where β₀ was a constant; μ_i was the random effect associated with i th participant and was assumed to have a normal distribution with mean 0 and variance σ²; and the link function f = logit. γ_i was computed using the maximum likelihood estimate of β₀ and empirical Bayes estimates of the random effects μ_i.

Next, the association of risk aversion with subsequent risk of dementia was assessed in Cox proportional hazards models. These and subsequent models were adjusted for the potentially confounding effects of age, sex, and education. Because most participants were white and not Latino, we did not adjust for race/ethnicity. The associations of risk aversion with level of global cognition and rate of global cognitive change were assessed in a mixed-effects model. To determine whether the association of risk aversion differed across cognitive domains, we repeated the mixed-effects models separately for each of the 5 cognitive domain measures. We repeated the proportional hazards model and mixed-effects models excluding the 172 individuals who had MCI at baseline. In the 181 individuals with postmortem neuropathologic data, we regressed the last valid risk aversion score before death on each neuropathologic marker in separate analyses adjusted for age at death, sex, and education. We repeated the analyses using the mean risk aversion score as the outcome.

RESULTS

Risk aversion scores at baseline ranged from 0.05 to 0.91 (mean = 0.32, SD = 0.31, skewness = 0.77) with higher values denoting higher risk aversion. Cronbach’s coefficient alpha was .90 indicating an adequate level of internal consistency. Older age, less education, and female gender were associated with higher level of risk aversion (Table 1). Global cognition had a modest negative correlation with risk aversion, with about 3% shared variance between the measures.

Table 1.

Correlations at baseline among key study variables^*

	Age	Education	Risk aversion	Global cognition	Episodic memory	Semantic memory	Working memory	Perceptual speed	Visuospatial ability
Age		−.13	.11	−.34	−.25	−.25	−.13	−.40	−.18
Education			−.11	.35	.21	.38	.25	.22	.39
Risk aversion				−.17	−.07	−.13	−.15	−.17	−.18
Global cognition					.83	.78	.66	.75	.58
Episodic memory						.54	.36	.42	.29
Semantic memory							.44	.56	.45
Working memory								.41	.37
Perceptual memory									.40

^*Pearson correlations; all p ≤ .001

Risk Aversion and Incident Dementia (n = 874)

During a mean of 4.6 years of follow-up (SD = 2.1), 128 persons developed dementia. For ages less than 75, the incidence rate was 3.8 per 1000 person-years (2 cases in 529.6 person-years); for ages 75 and older, the rate was 39.8 (126 cases in 3164.9 person-years). The 128 with incident dementia were older (t[209.1] = 8.9, p<.001) and less educated (t[872] = 2.3, p = .022) than the 746 persons who did not develop dementia but they had a similar percent of women (χ²[1] = 0.6, p = .45) (Table 2, second and third columns).

Table 2.

Characteristics of subgroups at baseline

Characteristic	No dementia at baseline	Incident dementia	No incident dementia	No MCI at baseline	Incident MCI	No incident MCI
Number	874	128	746	702	223	479
Age	81.3 (7.6)	85.7 (5.8)	80.6 (7.6)	80.6 (7.7)	84.3 (6.3)	78.8 (7.6)
Education	15.3 (3.0)	14.8 (3.1)	15.4 (3.0)	15.3 (3.0)	15.1 (3.0)	15.5 (3.0)
Sex, % women	77.1	79.7	76.7	78.1	77.1	78.5
Risk aversion	0.32 (0.31)	0.40 (0.31)	0.31 (0.30)	0.31 (0.31)	0.37 (0.32)	0.28 (0.30)
Global cognition	0.234 (0.519)	−0.295 (0.454)	0.323 (0.474)	0.369 (0.439)	0.101 (0.415)	0.493 (0.393)
Episodic memory	0.321 (0.658)	−0.359 (0.632)	0.435 (0.590)	0.507 (0.508)	0.274 (0.517)	0.614 (0.466)
Semantic memory	0.237 (0.611)	−0.264 (0.682)	0.320 (0.556)	0.342 (0.539)	0.095 (0.556)	0.453 (0.492)
Working memory	0.129 (0.735)	−0.217 (0.679)	0.187 (0.729)	0.219 (0.721)	0.032 (0.696)	0.306 (0.717)
Perceptual speed	0.762 (0.761)	−0.356 (0.715)	0.248 (0.734)	0.277 (0.732)	−0.121 (0.744)	0.458 (0.651)
Visuospatial ability	0.230 (0.767)	−0.098 (0.820)	0.284 (0.745)	0.333 (0.705)	0.028 (0.731)	0.472 (0.647)

We tested for the hypothesized relation of risk aversion to incident dementia in a proportional hazards model adjusted for age, education, and sex. Higher level of risk aversion was associated with higher risk of developing dementia (hazard ratio [HR] = 2.08, 95% confidence interval [CI]: 1.18, 3.65). Figure 2 demonstrates this effect across the interquartile range: a typical participant with a high level of risk aversion (dashed red line; score = 0.059, 75^th percentile) had an approximately 48% higher hazard of developing dementia than a typical participant with a low level of risk aversion (solid line; score = 0.05, 25^th percentile).

Figure 2.

Cumulative risk of incident dementia at the upper (dashed red line, 75^th percentile) vs lower solid black line, 25^th percentile) ends of the interquartile range of risk aversion, from a proportional hazards model adjusted for age, education, and sex.

Risk Aversion and Cognitive Decline (n=874)

At baseline, those who subsequently developed dementia had higher risk aversion and lower global cognition than those who did not develop dementia (second and third columns of Table 2). Therefore, we constructed a mixed-effects model with terms for risk aversion, age, sex, and education. This approach allowed us to assess whether risk aversion had an association with rate of cognitive decline, the principal manifestation of dementia, after accounting for its association with baseline cognitive function. In this analysis, higher level of risk aversion was associated with lower baseline level of global cognition (mean estimate of additional loss in cognition per 1-unit increase in risk aversion = −0.179, SE = 0.051, t[869]=−3.54, p<.001) as expected. With this association accounted for, higher risk aversion was associated with more rapid global cognitive decline (mean estimate of change in cognitive decline rate per 1-unit increase in risk aversion = −0.041, SE = .0015, t[3765]=−2.79, p=.005.

To determine whether the association of risk aversion with cognitive decline varied across domains of function, we repeated the model with measures of specific cognitive functions (Table 3). Higher risk aversion was associated with more rapid decline in episodic and semantic memory (as indicated by the significant interactions of time with risk aversion in Table 3) but not with change in other cognitive domains.

Table 3.

Relation of risk aversion to change in different cognitive functions in those without dementia at study onset (n = 874) and those without cognitive impairment at study onset (n = 704)

Analytic Group	Model Term	Episodic Memory Estimate (SE) t[df] P	Semantic Memory Estimate (SE) t[df] P	Working Memory Estimate (SE) t[df] P	Perceptual Speed Estimate (SE) t[df] P	Visuospatial Ability Estimate (SE) t[df] P
No dementia at baseline	Time	−0.056 (0.009) −6.31[3677] <.001	−0.063 (0.007) −8.70[3655] <.001	−0.044 (0.007) −5.99[3766] <.001	−0.106 (0.007) −14.75[3574] <.001	−0.033 (0.007) −4.63[3605] <.001
	Risk Aversion	−0.075 (0.068) −1.11[861] .268	−0.163 (0.060) −2.72[857] .007	−0.260 (0.074) −3.52[869] <.001	−0.266 (0.074) −3.59[850] <.001	−0.209 (0.073) −2.87[852] .004
	Time x Risk Aversion	−0.062 (0.019) −3.33[3677] <.001	−0.039 (0.015) −2.61[3655] .009	−0.024 (0.015) −1.52[3766] .129	−0.006 (0.015) −0.37[3574] .708	−0.009 (0.015) −0.60[3605] .546
No cognitive impairment at baseline	Time	−0.040 (0.009) −4.36[3040] <.001	−0.051 (0.007) −7.69[3023] <.001	−0.034 (0.008) −4.44[3083] <.001	−0.097 (0.007) −13.07[2969] <.001	−0.030 (0.007) −4.24[2992] <.001
	Risk Aversion	−0.040 (0.058) −0.68[691] .495	−0.156 (0.060) −2.61[688] .009	−0.222 (0.081) −2.74[697] .006	−0.249 (0.077) −3.21[683] .001	−0.233 (0.074) −3.14[685] .002
	Time x Risk Aversion	−0.061 (0.019) –3.17[3040] .002	−0.025 (0.014) −1.81[3023] .071	−0.031 (0.016) −1.94[3083] .052	0.001 (0.016) 0.04[2969] .967	0.005 (0.015) 0.34[2992] .737

Note: From 10 separate mixed-effects models adjusted for age, sex, and education. SE, standard error; df, degrees of freedom.

Risk Aversion and Incident Mild Cognitive Impairment (n = 702)

To limit the potential effect of pre-existing cognitive impairment on results, we excluded the 172 individuals with MCI at baseline leaving 702 persons without manifest cognitive impairment. During follow-up, 223 individuals developed MCI. This indicates an MCI incidence rate of 18.1 per 1000 person-years for ages less than 75 and 98.8 per 1000 person-years for ages 75 and older. The 223 with incident MCI were older than the 479 who did not (t[516.8] 10.1, p<.001) but did not differ in education (t[700] = 1.4, p = .17) or percent of women (χ²[1] = 0.2, p = .68) (Table 2, fifth and sixth columns). In a proportional hazards model, higher risk aversion was associated with higher risk of MCI (HR = 2.10, 95% CI: 1.34, 3.29). Figure 3 shows that the hazard of developing MCI was approximately 46% higher in those at the upper (dashed red line, 75^th percentile) compared to the lower (solid black line, 25^th percentile) end of the interquartile range of risk aversion.

Figure 3.

Cumulative risk of mild cognitive impairment at the upper (dashed red line, 75^th percentile) vs lower (solid black line, 25^th percentile) ends of the interquartile range of risk aversion, from a proportional hazards model adjusted for age, education, and sex.

Risk Aversion and Cognitive Decline (n=702)

At baseline, those who subsequently developed MCI already differed in risk aversion and global cognition from those who did not develop MCI (Table 2, fifth and sixth columns). We constructed a mixed-effects model to determine whether risk aversion was related to global cognitive decline after adjusting for its association with baseline level of global cognition. With adjustment for the association of risk aversion with global cognition at baseline (estimate = −0.149, SE = 0.048, t[697]=−3.08, p = .002), higher risk aversion predicted more rapid global cognitive decline (estimate of change in cognitive decline rate per 1-unit change in risk aversion = −0.043, SE = 0.015, t[3086]=−2.89, p = .004). In subsequent analysis of specific cognitive domains, higher risk aversion was related to more rapid decline in episodic memory (Table 3).

Risk Aversion and Neuropathologic Burden (n= 181)

To assess whether risk aversion was also related to neuropathologic conditions traditionally linked to MCI and dementia, we conducted additional analyses on individuals with a completed neuropathologic examination (Figure 1). They died a mean of 9.2 months after their last clinical evaluation (SD = 8.6) and the brain was removed a mean of 9.5 hours after death (SD = 7.4). The composite measure of plaques and tangles ranged from 0.00 to 2.28 (mean = 0.71, SD = 0.54). Ratings of TDP-43 pathology ranged from 0.0 to 3.5 (mean = 0.57, SD = 0.88); 21.2% had Lewy bodies and 4.4% had hippocampal sclerosis. Cerebral amyloid angiopathy ratings ranged from 0.0 to 4.0 (mean = 1.1, SD = 1.1); 37.6% had at least 1 chronic gross infarct; 37.0% had at least 1 chronic microinfarct; 23.3% had atherosclerosis and 13.8% had arteriolar sclerosis.

In separate analyses, we regressed the last valid risk aversion score (mean = 0.36, SD = 0.32) on each neuropathologic marker controlling for age at death, sex, and education. Higher levels of the composite measure of plaques and tangles and the rating of cerebral amyloid angiopathy were related to higher level of risk aversion (Table 4). Results were comparable when analyses were repeated with the mean risk aversion score as the outcome rather than the last valid score.

Table 4.

Association of postmortem neuropathologic markers with risk aversion*

Neuropathologic marker	Last valid score Estimate SE t[df] P	Mean score Estimate SE t[df] P
Plaques and tangles	0.114 0.047 2.45[176] 0.015	0.095 .0036 2.26[176] 0.010
TDP-43	0.020 0.029 0.71[176] 0.478	0.029 0.022 1.32[176] 0.188
Lewy bodies	−0.067 0.061 −1.11[174] 0.270	0.010 0.048 0.21[174] 0.833
Hippocampal sclerosis	−0.102 0.120 −0.85[176] 0.397	0.027 0.094 0.28[176] 0.778
Cerebral amyloid angiopathy	0.058 0.022 2.65[175] 0.009	0.056 0.017 3.37[175] <0.001
Chronic gross infarcts	0.058 0.051 1.13[176] 0.260	0.009 0.040 0.22[176] 0.823
Chronic microinfarcts	0.024 0.073 0.47[176] 0.846	0.031 0.040 0.79[176] 0.432
Atherosclerosis	0.014 0.073 0.19[176] 0.846	0.010 0.057 0.18[176] 0.860
Arteriolar sclerosis	−0.081 0.060 −1.35[175] 0.180	0.020 0.047 0.42[175] 0.674

Note: From 18 separate linear regression models adjusted for age at death, sex, and education. SE, standard error; df, degrees of freedom.

DISCUSSION

We assessed the association of risk aversion with adverse cognitive outcomes in more than 800 older persons. During follow-up, more than 100 persons developed dementia and more than 200 developed MCI. Higher level of risk aversion was associated with higher risk of developing MCI and dementia. Further, among those who died and had a neuropathologic examination, higher levels of plaques, tangles, and cerebral amyloid angiopathy were associated with higher risk aversion. The results suggest that high risk aversion in old age is at least partly an early sign of the pathology of AD.

In previous research, risk aversion has been shown to be higher in those with MCI compared to healthy controls (29). The present findings build on these results in two ways. First, they show in a prospective study that higher risk aversion predicts not only incident MCI but also incident dementia. Second, there was a direct association of markers of AD pathology (i.e., composite measure of plaques and tangles, rating of cerebral amyloid angiopathy) with higher level of risk aversion. These observations suggest that higher level of risk aversion in old age is an early sign of AD.

We also found that after accounting for the known association of risk aversion with level of cognition, higher risk aversion was associated with more rapid decline in episodic memory, arguably the primary clinical feature of AD (14). Risk aversion was marginally related to decline in semantic memory and unrelated to decline in working memory, perceptual speed, or visuospatial ability. These data link higher risk aversion with both the clinical and pathologic manifestations of AD.

These findings show that the consequences of AD extend beyond cognitive, sensory, and motor impairment to important aspects of behavior that are essential for decision making, including risk preferences. In addition, these changes in behavior apparently begin to emerge even among persons who have neuropathologic evidence of the disease but are not yet manifesting its traditional cognitive signs, suggesting that this consequence of Alzheimer’s disease is not limited to those who meet clinical criteria for dementia and MCI. Thus, the public health burden of AD may be greater than currently appreciated.

Study strengths and limitations should be noted. Rate of participation in clinical follow-up and brain autopsy was high. Previously established indices of cognition were used in analyses, reducing measurement error. Risk aversion was associated with both binary (i.e., dementia, MCI) and continuous (i.e., rate of cognitive decline) cognitive outcomes. Another study strength is that multiple postmortem markers were assessed, but a limitation is that this increased the risk of a Type I error in the clinical-pathologic correlations. Because individual differences in risk aversion are seen across the life span (10), it is likely that AD only explains a portion of the variability in risk aversion in old age. Participants were selected and not representative of older persons in the United States, suggesting the need for further research, particularly in racial and ethnic minorities.

HIGHLIGHTS.

• The study tested whether risk aversion is a sign of dementia.

• The main findings were that high risk aversion was related to dementia risk, memory decline, and postmortem markers of Alzheimer’s disease.

• We conclude that high risk aversion in old age is partly due to Alzheimer’s disease.