Correlates of Conversion from Mild Cognitive Impairment to Dementia with Lewy Bodies: Data from the National Alzheimer’s Coordinating Center

Abstract

Background:

Many patients with dementia with Lewy bodies (DLB) miss out on the best standards of care and psychosocial support due to diagnostic delays or inaccuracies following symptom onset.

Objective:

This study seeks to identify baseline characteristics in individuals with mild cognitive impairment (MCI) that correlate with eventual conversion to DLB or Alzheimer’s disease (AD).

Methods:

Baseline neuropsychological and neuropsychiatric data were analyzed in National Alzheimer’s Coordinating Center participants who completed the Uniform Data Set between 2006 and 2020 and subsequently converted from MCI to DLB or AD (n = 1632).

Results:

Only 6% of participants with MCI converted to DLB. Among those who converted to DLB, multidomain amnestic MCI (aMCI) was the most common subtype at study entry. As part of logistic regression analyses, odds ratios (ORs) were estimated for conversion to DLB versus AD based on study-entry characteristics, adjusting for age, sex, education, and years to diagnosis. The strongest predictors of conversion to DLB (p ≤ 0.0001) were nonamnestic MCI versus aMCI (OR 8.2, CI [5.0, 14]), multidomain MCI versus single-domain MCI (OR 2.7, CI [1.7. 4.2]), male sex (OR 4.2, CI [2.5, 7.1]), and presence of nighttime behaviors (OR 4.4 CI [2.8, 6.9]).

Conclusion:

A diagnosis of prodromal DLB should be considered in individuals with MCI who present with prominent executive/visuospatial deficits, neuropsychiatric symptoms, and less memory impairment. Early diagnosis of DLB may guide treatment planning, including the avoidance of antipsychotic medications in patients who develop psychotic symptoms, caregiver support, and initiation of early treatment(s) once medications become available.

INTRODUCTION

The variable clinical presentation of dementia with Lewy bodies (DLB) presents challenges in accurate clinical diagnosis such that nearly half of patients with DLB experience an average delay in diagnosis of 18 months or more after first reporting their symptoms to a physician [1]. As one report found, patients are almost twice as likely to receive a diagnosis of probable DLB in a specialty clinic (1 case in 13) as in a primary care clinic (1 case in 25), exemplifying the difficulties in obtaining a timely and cost-effective DLB diagnosis [2]. Indeed, in its early stages, DLB is often misdiagnosed as Alzheimer’s disease (AD) and/or other psychiatric disorders (such as depression or psychosis), and these misdiagnoses have troubling treatment implications [3]. For example, if patients do not receive a DLB diagnosis and develop psychosis or agitation, their primary care physicians may prescribe antipsychotics; yet more than 80% of patients with DLB who are treated with antipsychotics experience harmful side effects, ranging from increased confusion and sedation to irreversible parkinsonism and life-threatening severe neuroleptic sensitivity [4].

The current research consensus criteria for prodromal DLB includes mild cognitive impairment (MCI), delirium onset, and psychiatric presentation [5]. These criteria may be useful for research purposes, but there is insufficient evidence to formally apply these criteria for MCI-DLB in the clinical setting. Likewise, the proposed biomarkers in the assessment of MCI-DLB of polysomnography, DaTscans, and metaiodobenzylguanidine scans are not feasible to obtain in many clinical settings.

Given that inaccurate or delayed DLB diagnoses can result in adverse consequences for patients, caregivers, and the medical system, it is vital to look elsewhere for other indicators that may differentiate DLB early in the course of illness. Past studies found that spontaneous parkinsonism, fluctuations in cognition, visual hallucinations, and REM behavior sleep disorder (RBD) can develop within the predementia stage more often in those who transition to DLB compared to those who transition to AD [6–12]. Furthermore, patients with DLB have different cognitive domains affected compared to AD, with DLB showing more prominent visuospatial deficits and fewer memory deficits than AD [3, 13]. Moreover, some investigators have suggested that the presence of nonamnestic mild cognitive impairment (naMCI) rather than amnestic MCI (aMCI) may be helpful in identifying prodromal DLB [14] or that individuals with symptoms in a single cognitive domain may be less likely to develop DLB than individuals with symptoms in multiple cognitive domains [15]. However, few large studies have been conducted to directly compare the rates that naMCI, aMCI, and their various single- and multidomain subgroups convert to DLB. The purpose of this study is thus to compare the baseline clinical and neuropsychiatric characteristics of MCI, in participants in the National Alzheimer’s Coordinating Center (NACC) who were diagnosed with MCI and then subsequently converted to DLB or AD.

MATERIALS AND METHODS

Longitudinal data were collected through the National Institute of Aging (NIA)–funded Alzheimer’s Disease Centers (ADCs) of the NACC from 2006 to 2020. All participants completed the Uniform Data Set (UDS), an assessment that includes the short Neuropsychiatric Inventory (NPI-Q, for evaluating the presence or absence of specific neuropsychiatric domains), Trail Making Test (TMT) Part A and Part B, Logical Memory test, Digit Span tests, Wechsler Adult Intelligence Scale (WAIS), Boston Naming Test, Verbal Fluency tests, and Unified Parkinson’s Disease Rating Scale (UPDRS), which were together designed to characterize individuals with mild AD and MCI in comparison to nondemented aging. However, due to the changes in the protocol between NACC versions, the UPDRS and neuropsychological battery were not consistently collected.

Executive dysfunction was defined as the difference between TMT tests (Part B minus Part A), with higher values indicating more dysfunction. MCI, DLB, and AD diagnoses, including MCI domains, were established according to Petersen et al. [16–18], McKeith et al. [19–23], and NIA/Reagan criteria [24], respectively, by a group of two or more that typically included clinicians, a neuropsychologist, and the examining physician [25]. In these analyses, we identified 1) all participants who received a diagnosis of MCI at NACC enrollment or during follow-up and 2) who had at least one additional follow-up visit after the diagnosis of MCI. Given that the diagnosis of MCI represents the beginning of participants’ inclusion in these analyses, we refer to this as MCI at study entry, and we identify the assessments obtained at the MCI diagnosis as “baseline” assessments. We subsequently classified participants into four diagnostic categories based on their clinical diagnosis during the last available follow-up visit: AD, DLB, other dementias, and continued MCI. Only participants who were subsequently diagnosed as either DLB or AD, after the age of 50 were selected for these analyses.

Bivariate associations between dementia diagnosis and demographic or study entry characteristics were assessed using two-sample t-tests for continuous variables or Fisher’s test for categorical variables. Multivariate associations were assessed using logistic regression of DLB versus AD conversion. The first model had age at dementia diagnosis, years to diagnosis, sex, and years of education as covariates. Age at diagnosis was modeled as a two-degree orthogonal polynomial to allow for nonlinear effects. Next, the associations between DLB conversion and neuropsychological and neuropsychiatric variables at study entry were assessed in a series of logistic regressions with separate models for each study-entry variable and with age at dementia diagnosis, years to diagnosis, sex, and education as covariates. Given that multidomain MCI could be a surrogate for advanced MCI disease, we carried out a second logistic regression model of DLB conversion on MCI domain using only participants with incident MCI at study entry as an exploratory analysis. To test whether associations between DLB conversion and neuropsychiatric symptoms were independent of MCI status, we carried out a set of logistic regressions of DLB conversion on neuropsychiatric symptoms as described above but with MCI type and MCI domain as additional covariates. Hypothesis testing was performed using likelihood ratio tests with the Benjamini-Hochberg correction to maintain a false discovery rate of 5%. Model results are summarized with estimated marginal means and standard errors (SEs) for % DLB conversion and odds ratios (ORs) for conversion to DLB versus AD by category for discrete variables and by change equivalent to the interquartile range for continuous variables; all ORs are accompanied by Wald-based 95% confidence intervals (CIs). For neuropsychological and neuropsychiatric variables, ORs greater than 1.0 indicate that conversion to DLB was associated with worse outcomes on these variables, and ORs less than 1.0 indicate that conversion to AD was associated with worse outcomes.

Up to 20% of participants were missing some of the neuropsychological assessments. Multiple imputation of these missing measures was performed using the multiple imputation by chained equations (MICE) algorithm. All study characteristics included in the models (including DLB conversion) were also used in the imputation, along with other study entry characteristics that were potentially associated with the missing measures (e.g., ethnicity, Clinical Dementia Rating score, and Geriatric Depression Scale score). Predictive mean matching and logistic regression were used to impute continuous and binary measures respectively. Twenty imputed data sets were generated and checked for algorithm convergence, and models were reapplied on each imputed dataset. Estimates and likelihood ratio tests were obtained by pooling the results based on Rubin’s rules. Analysis was performed using R–3.6.2 [26] and packages emmeans [27], rms [28], tidyverse [29], and mice [30].

RESULTS

An initial sample of 1,697 participants received a diagnosis of MCI at NACC enrollment or at some point during follow-up. We excluded from our analyses one participants who remained MCI at their last visit, 12 developed non-DLB or non-AD dementias, and 52 had other dementia diagnoses.

The remaining 1,632 participants had a mean number of years from study entry to dementia diagnosis of 2.7 (range 0.5–13.9) and were split nearly evenly between males and females. Ninety-five of these participants developed DLB (6%) and 1,537 developed AD (94%), including 149 participants with naMCI who converted to DLB (n = 33) or AD (n = 116) and 1,483 with aMCI who converted to DLB (n = 62) or AD (n = 1,421). Of the 95 participants who converted to DLB as defined by their final diagnosis, 23 (24%) were initially diagnosed with AD, 1 (1%) was given a non-AD/non-DLB dementia diagnosis, and the remaining 71 (75%) had an initial diagnosis of DLB. Conversely, of the 1,546 participants who converted to AD as defined by their final diagnosis, 1,523 (99%) were initially diagnosed with AD, 8 (< 1%) were given a non-AD/non-DLB dementia diagnosis, and the remaining 6 (< 1%) had an initial diagnosis of DLB. Thus, for the majority of NACC participants, the initial and final diagnoses were consistent.

Examining the subtypes more closely within participants who converted to DLB and AD, we found that naMCI was more common in participants who converted to DLB than in participants who converted to AD (35% versus 8%, p < 0.0001; see Fig. 1). We also found that participants who converted to DLB and AD had a similar prevalence of multidomain aMCI at study entry (48% versus 43%), whereas participants who converted to DLB had a lower prevalence of single-domain aMCI at study entry than participants who converted to AD (17% versus 50%). Thus, among participants who developed DLB, the most frequent MCI subtype at study entry was multidomain aMCI, whereas among participants who developed AD, single-domain and multidomain aMCI were similar in prevalence at study entry. There were also strong associations between conversion to DLB and executive function and night behaviors. Additional sample characteristics are presented in Table 1.

Fig. 1.

Proportion of conversion to DLB or AD by study entry MCI type.

Table 1.

Summary statistics for MCI at study entry (n = 1,632) by eventual DLB or AD diagnosis

	DLB (n = 95)	AD (n = 1,537)	p *
Age at diagnosis (y)	77.1 ± 7.4 (55–97)	79.9 ± 8.8 (53–110)	0.0027
Years to diagnosis	2.4 ± 2.0 (0.9–10.8)	2.8 ± 2.1 (0.4–13.9)	0.089
MCI status at study entry			< 0.0001
naMCI, single domain	14 (15%)	71 (5%)
naMCI, multidomain	19 (20%)	45 (3%)
aMCI, single domain	16 (17%)	766 (50%)
aMCI, multidomain	46 (48%)	655 (43%)
Sex (male)	77 (81%)	733 (48%)	< 0.0001
Education (y)	16.3 ± 3.5 (3–28)	15.5 ± 3.2 (0–30)	0.0078
APOE ε4 allelea	31/68 (46%)	713/1321 (54%)	0.22
UPDRS total scoreb	8.9 ± 10.3 (0–43)	2.7 ± 4.6 (0–45)	< 0.0001
Trail Making Testc (sec)
Part A	54 ± 30 (18–150)	45 ± 22 (14–150)	0.0002
Part B	181 ± 86 (31–300)	146 ± 77 (33–300)	< 0.0001
Executive dysfunction tertilec		< 0.0001
Lower	16 (17%)	477 (33%)
Upper	49 (53%)	464 (32%)
Logical Memory (number of story units recalled)d
Immediate recall	8.2 ± 4.5 (0–18)	8.0 ± 4.1 (0–25)	0.72
Delayed recall	6.4 ± 4.5 (0–18)	5.1 ± 4.5 (0–21)	0.016
Digit Span (number of correct trials)e
Forward	8.0 ± 2.1 (4–12)	7.9 ± 2.0 (1–12)	0.66
Backward	5.4 ± 2.1 (2–12)	5.9 ± 2.0 (2–12)	0.040
WAIS (score)f	32.9 ± 12.3 (4–68)	36.4 ± 11.4 (0–74)	0.013
Boston Naming (total score)g	24.9 ± 4.7 (9–30)	24.5 ± 4.8 (0–30)	0.50
Verbal Fluency (total number named)h
Animals	15.0 5.5 (3–36)	15.1 4.8 (2–35)	0.85
Vegetables	9.0 4.2 (2–26)	10.3 3.7 (0–26)	0.0012
NPI item (presence)i
Delusions	9/92 (10%)	53/1470 (4%)	0.0087
Hallucinations	10/92 (11%)	9/1470 (1%)	< 0.0001
Agitation	22/91 (24%)	267/1470 (18%)	0.16
Depression	28/92 (30%)	308/1470 (21%)	0.036
Anxiety	32/92 (35%)	302/1470 (21%)	0.0023
Euphoria	3/92 (3%)	27/1469 (2%)	0.42
Apathy	30/92 (33%)	248/1470 (17%)	0.0004
Disinhibition	12/92 (13%)	129/1470 (9%)	0.19
Irritability	32/91 (35%)	390/1469 (27%)	0.088
Aberrant motor	12/92 (13%)	93/1470 (6%)	0.028
Aberrant nighttime behavior	48/92 (52%)	273/1467 (19%)	< 0.0001
Changes in appetite	23/92 (25%)	177/1470 (12%)	0.0010

Means ± SD (range) or frequency (%). SD: standard deviation.

^*t-test (age at diagnosis) or Fisher’s test.

^aMissing data for participants with DLB (n = 27) and AD (n = 216)

^bmissing data for participants with DLB (n = 20) and AD (n = 217)

^cmissing data for participants with DLB (n = 3) and AD (n = 98)

^dmissing data for participants with DLB (n = 21) and AD (n = 259)

^emissing data for participants with DLB (n = 21) and AD (n = 243)

^fmissing data for participants with DLB (n = 25) and AD (n = 299)

^gmissing data for participants with DLB (n = 21) and AD (n = 249)

^hmissing data for participants with DLB (n = 1) and AD (n = 69)

ⁱmissing data for participants with DLB (n = 4) and AD (n = 70).

In a multivariable logistic regression analysis of DLB versus AD on age at diagnosis (modeled as a two-degree polynomial), sex, education, and years to diagnosis, male participants had a roughly 4 times higher risk of developing DLB than female participants (OR 4.2, 95% CI [2.5, 7.1], p < 0.0001). Participants who completed 12 years of education had a slightly lower risk of developing DLB than participants who completed 16 years of education (OR 0.81, 95% CI [0.62, 1.1], p = 0.19). The risk of developing DLB versus AD began to decline for NACC participants after age 80 (age 75 versus age 85: OR 1.6, 95% CI [1.1, 2.3], p = 0.023; see Fig. 2). By contrast, less association was observed between years to diagnosis and conversion to DLB (4 years versus 1 year: OR 1.2, 95% CI [0.81, 1.7], p = 0.47). Age at diagnosis was modeled as a two-degree orthogonal polynomial.

Fig. 2.

Log odds (95% CI) of DLB versus AD by age at diagnosis from a logistic regression analysis of DLB versus AD diagnosis on age at diagnosis, sex, education, and years to diagnosis.

The associations between conversion to DLB and study-entry MCI type, executive function, and night behaviors were maintained in logistic regression analyses after adjusting for age at diagnosis, sex, education, and years to diagnosis (p < 0.0001 for each association; see Table 2). More particularly, 21% of participants who had naMCI, 9% of participants in the highest tertile of TMT executive function, and 13% of participants with night behaviors converted to DLB, whereas only 3% of participants with aMCI (OR 8.2, CI [5.0, 14]), 2% of participants in the lowest tertile of TMT executive function (OR 4.2, CI [2.1, 8.5]), and 3% of participants without night behaviors (OR 4.4, CI [2.8, 6.9]) converted to DLB. Conversion to DLB was also associated with multidomain MCI (versus single-domain MCI; p ≤ 0.0001), slower times on the Trail Making Test Parts A and B (p ≤ 0.0001), poorer performance on the WAIS (p ≤ 0.0001), and the presence of apathy, delusions, anxiety, and appetite disturbances (p < 0.05) at study entry, although these effects were smaller in magnitude than the differences observed with MCI status, executive function, and night behaviors (see Table 2). Conversion to AD was associated with worse performance in the delayed recall section of the Logical Memory test (OR 0.54, 95% CI [0.37, 0.77], p = 0.0037). The associations between conversion to DLB and the neuropsychiatric symptoms described above were maintained in models with additional adjustments for MCI type and MCI domain (data not shown). Pooled estimates from logistic regression on the 20 imputed datasets produced similar findings (see Supplementary Table 1).

Table 2.

Estimated percentages of conversion to DLB and odds ratios (ORs) of conversion to DLB versus AD

Study-entry characteristic*	Mean percent ± SE	OR (95% CI)†	p ‡	p adjusted§
Age at diagnosis (y)
85	3.7±0.6	1.0	0.010	< 0.0001
75	5.8±0.9	1.6 (1.1, 2.3)
Sex
Female	2.5±0.6	1.0	< 0.0001	< 0.0001
Male	9.6±1.3	4.2 (2.5, 7.1)
Education (y)
16	5.1±0.7	1.0	0.13	0.19
12	4.1±0.8	0.81 (0.62, 1.1)
Years to diagnosis
4	4.6 ± 0.8	1.0	0.40	0.47
1	5.4 ± 0.9	1.2 (0.81, 1.7)
MCI type¶
Amnestic status
Amnestic	3.2±0.6	1.0	< 0.0001	< 0.0001
Nonamnestic	21.4±4.1	8.2 (4.9, 13.7)
Domain
Single	2.5±0.6	1.0	< 0.0001	< 0.0001
Multiple	6.4±1.1	2.7 (1.7, 4.3)
Trail Making Test
Part A time (sec)
30	3.8±0.7	1.0	< 0.0001	< 0.0001
60	6.4±0.9	1.7 (1.4, 2.2)
Part B time (sec)
90	3.3±0.6	1.0	< 0.0001	< 0.0001
180	5.8±0.9	1.8 (1.4, 2.3)
Executive function tertile
Lower	2.4±0.7	1.0	< 0.0001	< 0.0001
Upper	9.4±1.6	4.2 (2.1, 8.5)
Logical Memory (story units recalled)
Immediate recall
11	5.2±0.9	1.0	0.19	0.26
5	4.1±0.8	0.79 (0.55, 1.1)
Delayed recall
8	5.6±1.0	1.0	0.0011	0.0037
1	3.1±0.7	0.54 (0.37, 0.77)
Digit Span (trials correct)
Forward
9	4.5±0.8	1.0	0.98	0.98
7	4.5±0.8	1.0 (0.78, 1.3)
Backward
7	3.8±0.7	1.0	0.029	0.053
4	5.7±1.1	1.5 (1.0, 2.3)
WAIS (score)
45	3.3±0.7	1.0	0.0022	0.0067
25	6.7±1.3	2.1 (1.3, 3.3)
Boston Naming (total score)
28	4.3±0.8	1.0	0.63	0.65
22	4.6±0.8	1.1 (0.78, 1.5)
Verbal Fluency (total number named)
Animals
18	4.8±0.8	1.0	0.29	0.37
12	5.5±0.9	1.2 (0.88, 1.5)
Vegetables
12	4.7 ± 0.8	1.0	0.11	0.17
8	5.7 ± 0.9	1.2 (0.95, 1.6)
Neuropsychiatric symptoms#
Delusions
Absence	4.7 ± 0.7	1.0	0.015	0.029
Presence	12.4 ± 4.2	2.8 (1.3, 6.2)
Agitation
Absence	4.9±0.8	1.0	0.58	0.63
Presence	5.6±1.4	1.2 (0.69, 1.9)
Depression
Absence	4.5±0.7	1.0	0.038	0.064
Presence	7.3±1.6	1.7 (1.0, 2.7)
Anxiety
Absence	4.4±0.7	1.0	0.012	0.025
Presence	7.7±1.6	1.8 (1.2, 2.9)
Apathy
Absence	4.4±0.7	1.0	0.009	0.022
Presence	8.2±1.8	1.9 (1.2, 3.1)
Disinhibition
Absence	4.9±0.8	1.0	0.44	0.50
Presence	6.3±2	1.3 (0.68, 2.5)
Irritability
Absence	4.8±0.8	1.0	0.38	0.47
Presence	5.8±1.2	1.2 (0.78, 1.9)
Aberrant motor
Absence	4.8±0.7	1.0	0.063	0.10
Presence	9.0±2.8	2.0 (1.0, 3.8)
Aberrant nighttime behavior
Absence	3.2±0.6	1.0	< 0.0001	< 0.0001
Presence	12.6±2.2	4.4 (2.8, 6.9)
Changes in appetite
Absence	4.4 ± 0.7	1.0	0.0025	0.0067
Presence	9.7 ± 2.3	2.3 (1.4, 3.9)

^*Study-entry characteristics estimated from logistic regression of DLB conversion on study-entry characteristics and then adjusted for age at dementia diagnosis, years to diagnosis, sex, and education.

^†ORs for conversion to DLB versus AD by category for discrete variables and by selected values for continuous variables. Continuous neuropsychological values were selected to correspond with change equivalent to the interquartile range. ORs > 1 indicate a higher risk of DLB versus AD, whereas ORs < 1 indicate a higher risk of AD versus DLB.

^‡Hypothesis testing was performed using likelihood ratio tests.

^§Adjustment to maintain a false discovery rate of 5% using the Benjamini-Hochberg correction.

^¶Nonamnestic versus amnestic MCI and single-domain versus multidomain MCI were modeled as covariates in a single logistic regression.

^#An insufficient number of participants had presence of hallucinations or elation to analyze in a logistic regression model.

In an exploratory analysis using only incident MCI, five participants (5%) who received their first diagnosis of multidomain MCI after enrollment in the NACC (n = 102) converted to DLB compared to four participants (2%) who received their first diagnosis of single-domain MCI after enrollment in the NACC (n = 176; OR 2.2, CI [0.46, 11], Fisher’s exact test p = 0.30).

DISCUSSION

MCI is a heterogeneous disorder [31, 32], and its classification by subtypes may be useful in future research studies and for improved clinical management of prodromal neurodegenerative disorders [33]. For example, in a study at the Mayo Alzheimer’s Disease Research Center of 211 participants who converted to DLB or AD, Ferman et al. suggest a potential link between naMCI and a future diagnosis of probable DLB [14]. The NACC represents an excellent setting to explore this possibility, as we have here produced the largest published study to analyze the conversion to DLB and AD in participants with either na/aMCI or single-/multidomain MCI subtypes at baseline. To that end, our findings demonstrate that a higher percentage of individuals who developed DLB presented with naMCI at study entry compared to those who developed AD (35% versus 8%).

Nevertheless, the majority of the participants in our study converted to AD, regardless of MCI subtype. This is consistent with past studies [34], but our observed rate of conversion to DLB from naMCI was much lower than the rate reported by Ferman et al. (22% versus 67%) [14]. This contrast may reflect population differences between the NACC, which enrolls participants from Alzheimer Disease Centers (ADCs) across the United States, and the Mayo Clinic, which has internationally recognized expertise in the investigation of sleep disorders and dementia; indeed, 80% of the Mayo Clinic participants who developed DLB had probable RBDs at baseline. That is, because ADCs focus on memory disorders, they may be less likely to include individuals who primarily present with psychiatric, sleep, and/or executive dysfunction, and these potentially underrepresented individuals may be more likely to convert to DLB. This possibility of selection bias is endemic to all investigations that use data generated from specialty research centers, and it should thus be considered prior to generalizing our current study findings.

In our NACC sample, we also observed different performance levels in various cognitive domains. For example, participants who converted to DLB had lower baseline visuospatial skills and higher baseline logical memory scores than participants who converted to AD. Likewise, participants who developed DLB were more likely to have an initial diagnosis of multidomain aMCI than an initial diagnosis of single-domain aMCI or naMCI. That rate of conversion from multidomain aMCI to DLB is consistent with the findings of several smaller studies in which the diagnosis of DLB was highly associated with multidomain MCI [14, 15, 35]. Some researchers have suggested that single- and multidomain MCI may exist on a continuum such that multidomain MCI is simply later in the disease course than single-domain MCI [36, 37]; this could mean that patients with prodromal DLB tend to present later in the disease course, after symptoms have emerged in more domains. As an exploratory analysis, we examined a small subsample of participants with single- and multidomain incident MCI (n = 278). Although only 9 of these participants converted to DLB, those with single-domain incident MCI had a lower rate of conversion to DLB than those with multidomain incident MCI. The potential association between incident multidomain MCI and DLB has been previously reported [15, 35], and when considered together with our other findings, it seems that the number or type of MCI domains (with observed baseline deficits) may be useful for helping to determine the risk of DLB in individuals with MCI.

Neuropsychiatric symptoms represent another baseline characteristic that may be useful for evaluating prodromal DLB or the risk of conversion to DLB. For instance, in a study by Donaghy et al. [7], participants who were characterized as MCI-DLB (i.e., participants who met criteria for MCI and either met two of the core clinical features of DLB or one of the core clinical features with an imaging biomarker) were more likely to experience neuropsychiatric symptoms than participants who were characterized as MCI-AD (i.e., participants who met criteria for MCI and had none of the core clinical features or biomarkers of DLB). Our study extends this work in a large longitudinal sample that included all participants with MCI regardless of the presence or absence of core clinical features or biomarkers of DLB by showing that participants who converted from MCI to DLB had more aberrant nighttime behaviors (i.e., more nighttime awakening, unintended early morning rising, and daytime sleepiness), changes in appetite, and other neuropsychiatric symptoms than participants who converted to AD. It is especially worth highlighting the sample differences between our study and Donaghy et al. in the context of the recently published criteria for prodromal DLB published by McKeith et al. [5]—that is, a number of the participants who converted from MCI to DLB in our study would not have met the Donaghy or McKeith criteria for prodromal DLB, as only 10% of our participants experienced visual hallucinations at baseline, and many lacked other core features of DLB at baseline or were not assessed for RBD or DLB biomarkers. Instead, the most frequently observed baseline neuropsychiatric symptoms in NACC participants who eventually converted to DLB were non-specific dementias, such as apathy, anxiety, nighttime behaviors, and appetite changes.

Our neuropsychiatric findings are consistent with past findings using NACC participants with MCI. For instance, although Rosenberg et al. found no differences in neuropsychiatric symptoms when comparing naMCI and aMCI in 2011, they did find that MCI participants with executive dysfunction experienced more neuropsychiatric symptoms than MCI participants without executive dysfunction [38]. In two other studies that used the NACC dataset, participants who were cognitively normal at baseline but had neuropsychiatric symptoms were more likely to convert to incident MCI or dementia than those who did not have neuropsychiatric symptoms [39, 40].

Given that MCI subtypes and/or neuropsychiatric symptoms appear insufficient for predicting conversion to DLB or AD by themselves, it may therefore be useful for future studies to develop profiles that indicate to clinicians whether a patient is more at risk for DLB, and ideally, such profiles will include both MCI subtypes and relevant neuropsychiatric and nighttime symptoms. For example, patients with MCI who (a) present with fewer memory deficits and more prominent neuropsychiatric symptoms and then (b) show more prominent executive or visuospatial deficits during cognitive testing screening have a profile that should alert clinicians to avoid prescribing antipsychotics, which are associated with severe adverse reactions in prodromal DLB. In these ways, MCI subtypes and neuropsychiatric features may help inform treatment decisions and improve the early identification of prodromal DLB. Attention to these clinical characteristics is particularly important given that the family members of patients with MCI may become narrowly focused on memory impairments and may overlook other impairments that may be useful in identifying prodromal DLB and thereby better improving and coordinating care.

The two most significant known demographic risk factors for conversion to DLB are male sex and advanced age [41], and our findings clearly support these previous observations. Similar to past studies [42, 43], we found that participants who converted to DLB were more likely to be male than female (81% versus 19%, p < 0.0001), but the factors underlying these sex differences are unclear. We also found that participants who converted to DLB tended to have a younger age of dementia diagnosis than participants who converted to AD, with the risk of developing DLB versus AD beginning to decline after age 80. That contrast in age at diagnosis parallels the findings of Mueller et al. [44] and Price et al. [45], though neither of those studies reached statistical significance when comparing age at diagnosis in DLB and AD. Additionally, past studies have suggested that the range for age of onset in DLB extends into the mid-80s [46, 47], with Savica et al. [42] finding that the incidence of DLB peaks between the ages of 70 and 79 yet remains high for participants age 80 to 99.

Although our study clearly reinforces past findings on the importance of MCI subtypes, baseline neuropsychiatric features, sex, and age on the conversion to DLB or AD, there were also some key limitations to our work. As described above, because NACC participants were evaluated in research centers that primarily recruited participants with memory complaints, our findings are not necessarily generalizable to patients who present to movement or sleep disorder clinics; this may partly explain why the conversion rate from naMCI to DLB was lower in our study than in some past studies. In addition to potential selection bias, our study also faces the obstacle of bias related to participants who are lost to follow-up. That is, if participants who convert to DLB are more likely to be lost to follow-up, that right censoring could bias our results in a way that we cannot directly investigate using NACC data.

Another limitation of our study is that RBD-related data were not collected prior to NACC UDS version 3.0; this means that we were unable to incorporate this important diagnostic feature into our analyses. Although we did not specifically evaluate RBD, the NPI questions on aberrant nighttime behavior elicited related information on nighttime awakening, rising too early in the morning, and excessive daytime sleepiness. Finally, APOE genotyping is missing for most of our participants, and only a quarter of the NACC participants in this sample have undergone neuropathological assessments; this means that we were unable to consider genetic susceptibility factors or to confirm clinical diagnoses with neuropathological diagnoses. Although other studies have investigated genotype and/or neuropathological diagnoses in MCI and DLB, showing, for example, that the presence of an ε4 allele increases the odds of an amnestic presentation within NACC participants [48], to our knowledge such studies have either not included neuropsychiatric symptoms or have lacked sufficient sample sizes to correlate Lewy-related pathology with particular MCI subtypes.

Despite our study’s limitations, an important clinical consideration is that information on RBD and APOE genotypes are unlikely to be available in many primary care clinical settings, whereas the variables included in our analyses can be obtained via careful history and neurological examinations, both of which are feasible in most clinical settings. Similarly, the recently published criteria for the diagnosis of prodromal DLB may eventually provide clinicians with a roadmap to identifying DLB earlier in its disease course [5], but some of the core criteria and biomarkers outlined in those criteria require specialization beyond the standard toolset of many providers in primary care settings (e.g., DaTscans, polysomnography, positron emission tomography, or single-photon emission computerized tomography). Therefore, our findings suggest that there may be merit in homing in on specific neuropsychiatric and neuropsychological domains when evaluating the possibility of prodromal DLB. Further research is necessary to evaluate the diagnostic utility of specific combinations of signs and symptoms to enable early distinctions between DLB and AD. This could include, for example, prospective investigations in large MCI samples of checklists like the Lewy Body Composite Risk Score [49]. Indeed, large prospective longitudinal studies that include a combination of neuroimaging and cerebrospinal biomarkers, clinical and neuropsychological assessments, and neuropathological examination would greatly contribute to our understanding of the epidemiological, diagnostic, and prognostic value of neuropsychiatric features in MCI and thereby help us to better distinguish between DLB and AD. Such work will be critical when DLB-specific clinical trials and new treatment strategies become available.