This is an open-access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work, first published in the Journal of Medical Internet Research, is properly cited. The complete bibliographic information, a link to the original publication on https://www.jmir.org/, as well as this copyright and license information must be included.
Human voice has increasingly been recognized as an effective indicator for the detection of cognitive disorders. However, the association of acoustic features with specific cognitive functions and mild cognitive impairment (MCI) has yet to be evaluated in a large community-based population.
This study aimed to investigate the association between acoustic features and neuropsychological (NP) tests across multiple cognitive domains and evaluate the added predictive power of acoustic composite scores for the classification of MCI.
This study included participants without dementia from the Framingham Heart Study, a large community-based cohort with longitudinal surveillance for incident dementia. For each participant, 65 low-level acoustic descriptors were derived from voice recordings of NP test administration. The associations between individual acoustic descriptors and 18 NP tests were assessed with linear mixed-effect models adjusted for age, sex, and education. Acoustic composite scores were then built by combining acoustic features significantly associated with NP tests. The added prediction power of acoustic composite scores for prevalent and incident MCI was also evaluated.
The study included 7874 voice recordings from 4950 participants (age: mean 62, SD 14 years; 4336/7874, 55.07% women), of whom 453 were diagnosed with MCI. In all, 8 NP tests were associated with more than 15 acoustic features after adjusting for multiple testing. Additionally, 4 of the acoustic composite scores were significantly associated with prevalent MCI and 7 were associated with incident MCI. The acoustic composite scores can increase the area under the curve of the baseline model for MCI prediction from 0.712 to 0.755.
Multiple acoustic features are significantly associated with NP test performance and MCI, which can potentially be used as digital biomarkers for early cognitive impairment monitoring.
Alzheimer disease (AD) is a chronic neurodegenerative disease characterized behaviorally by memory loss, language impairment, motor problems, loss of executive function, and emotional distress, which can progress to severe levels. There are currently no definitive disease-modifying treatment methods [
At present, diagnosis relies largely on some combination of clinical examination [
Producing speech is a cognitively complex task [
Applying the findings of earlier research to a general population, however, is difficult due to the small sample sizes and use of cognitive assessment protocols that are not sufficiently comprehensive. Further, voice analyses that include linguistic features are difficult to generalize to other languages. There remains a paucity of research determining the relationship between acoustic features and NP tests that span across multiple cognitive domains. In addition, a comprehensive characterization of acoustic features that are associated with incident MCI is warranted. The objective of this study was to investigate the association of acoustic features and different NP test scores across cognitive domains and how they compare in identifying prevalent and incident MCI in the Framingham Heart Study (FHS) community-based cohort.
The original sample included 9253 observations from 5189 participants who completed at least one NP assessment that was voice recorded. A subset of participants had multiple recordings over the course of the study period. Each digital voice recording and the corresponding NP tests were treated as 1 observation. Exclusion criteria included those observations with missing education information (n=492), prevalent dementia (n=313), flagged as potential MCI but have not gone through dementia review (n=551), and those whose voice recording was less than 10 minutes in length (n=23).
The Institutional Review Board of the Boston University Medical Campus approved the procedures and protocols of the Framingham Heart Study (FHS is H-32132). All participants provided written informed consent.
The details of FHS NP test administration have been reported previously [
Cognitive domain and corresponding neuropsychological (NP) tests.
Cognitive domain | NP test |
Verbal memory |
Logical Memory—Immediate Recall Logical Memory—Delayed Recall Logical Memory—Recognition Paired Associate Learning—Immediate Recall Paired Associate Learning—Delayed Recall Paired Associate Learning—Recognition |
Visual memory |
Visual Reproduction—Immediate Recall Visual Reproduction—Delayed Recall Visual Reproduction—Recognition |
Attention and concentration |
Digit Span—Forward Trail Making Test A |
Executive function |
Digit Span—Backward Trail Making Test B |
Abstract reasoning |
Similarities |
Language |
Boston Naming Test—30-item version |
Visuoperceptual organization |
Hooper Visual Organization Test |
Verbal fluency |
Controlled Oral Word Association Test Category Naming Test—Animal |
Since 2005, the FHS has been digitally recording all spoken responses during NP test administration, which encompasses the verbal interactions between the tester and the participant. This study included digital voice recordings obtained from September 2005 to March 2020. OpenSMILE software (version 2.1.3) [
The cognitive status of FHS participants included assessments by NP tests. For those identified with possible cognitive impairment, NP tests were administered on average about every 1 to 2 years. When potential cognitive impairment decline was present, a clinical review was conducted by a panel with at least one neurologist and one neuropsychologist. MCI diagnosis was determined by the review panel, which required that the participant exhibit evidence of a decline in cognitive performance in 1 or more cognitive domains, have no records indicating functional decline, and do not meet the criteria for dementia [
To compare the difference between demographics and standard NP test scores in MCI and normal control groups, Wilcoxon rank sum test was used for continuous variables [
A set of acoustic composite scores was generated by regressing each NP test against the group of acoustic features that were significantly associated with each NP test. The acoustic composite score is a weighted combination of acoustic features. The weight of each acoustic feature in the composite score was derived by training a linear mixed-effects effect model. For participant
where
The association of normalized acoustic composite scores with prevalent MCI was assessed by logistic regression models. Based on the regression coefficients, the odds ratios (ORs) and 95% CIs were estimated.
To determine the relationship between acoustic composite scores and incident MCI, participants whose age at the voice recording was <60 years (n=2718) and those with prevalent MCI (n=222) were excluded. The first observation of each participant was included in this analysis. The association between acoustic composite scores with incident MCI was quantified by Cox proportional hazards models (censored at the last date of contact or death) [
We further evaluated the added predictability of the acoustic composite score for incident MCI. Receiver operating characteristic (ROC) analysis was performed to estimate the area under the curve (AUC) using a random forest model. A baseline model was constructed using age, sex, and education as predictors. A second model was constructed using these predictors and additional acoustic composite scores that were found to be significantly related to specific NP tests. The mean AUC of 10-fold cross-validation was computed for each model for comparison. We also performed a secondary analysis by including NP tests and clinical risk factors in the prediction of incident MCI. The statistical analyses were performed using Python software (version 3.9.7; Python Software Foundation).
Our study included 7874 observations from 4950 participants of FHS (age: mean 62, SD 14 years; 4336/7874, 55.07% women; 4279/7874, 54.34% self-reported college-level education or higher). Most participants (2657/4950, 53.68%) had 1 voice recording. Some participants (1775/4950, 35.86%) had 2 recordings, and the remaining participants (518/4950, 10.46%) had 3 or more recordings. Among these observations, 453 of these observations were diagnosed with MCI. The details of sample characteristics are shown in
We examined the association of acoustic features with NP tests. As shown in
Acoustic composite scores were also generated using the significant acoustic features for each NP test. As shown in
We then performed association analysis of acoustic composite scores with prevalent MCI.
We further examined the association of acoustic composite scores with incident MCI by restricting the analysis to 2010 participants who were aged ≥60 years. Among them, 145 participants have incident MCI. As shown in
The added predictive power of
Baseline characteristics.
Variable | Total observation (N=7874) | MCIa (n=453) | NCb (n=7421) | |||
Age (years), mean (SD) | 62 (14) | 81 (8) | 61 (14) | <.001 | ||
|
.84 | |||||
|
Women | 4336 (55.07) | 252 (55.63) | 4084 (55.03) |
|
|
|
Men | 3538 (44.93) | 201 (44.37) | 3337 (44.97) |
|
|
|
<.001 | |||||
|
No high school | 202 (2.57) | 53 (11.70) | 149 (2.01) |
|
|
|
High school | 1443 (18.33) | 118 (26.05) | 1295 (17.45) |
|
|
|
Some college | 1950 (24.77) | 134 (29.58) | 1816 (24.47) |
|
|
|
College and higher | 4279 (54.34) | 148 (32.67) | 4161 (56.07) |
|
|
|
||||||
|
LMie | 12.35 (3.62) | 8.53 (3.76) | 12.58 (3.48) | <.001 | |
|
LMdf | 11.36 (3.83) | 6.93 (4.11) | 11.62 (3.65) | <.001 | |
|
LMrg | 9.52 (1.28) | 8.59 (1.72) | 9.57 (1.23) | <.001 | |
|
VRih | 8.61 (2.91) | 4.48 (2.23) | 8.85 (2.76) | <.001 | |
|
VRdi | 7.91 (3.17) | 3.11 (2.30) | 8.19 (2.99) | <.001 | |
|
VRrj | 3.11 (1.01) | 1.89 (1.06) | 3.18 (0.96) | <.001 | |
|
PASik | 14.45 (3.58) | 10.02 (2.79) | 14.71 (3.45) | <.001 | |
|
PASdl | 8.56 (1.47) | 6.56 (1.60) | 8.68 (1.38) | <.001 | |
|
PASrm | 9.82 (0.64) | 8.83 (1.74) | 9.88 (0.45) | <.001 | |
|
DSfn | 6.71 (1.31) | 6.06 (1.20) | 6.75 (1.30) | <.001 | |
|
DSbo | 4.92 (1.30) | 4.12 (1.01) | 4.97 (1.30) | <.001 | |
|
SIMp | 16.83 (3.61) | 12.63 (4.30) | 17.08 (3.40) | <.001 | |
|
BNT30q | 27.22 (2.81) | 23.66 (4.14) | 27.43 (2.56) | <.001 | |
|
TrailsAr | 0.42 (0.15) | 0.66 (0.21) | 0.40 (0.14) | <.001 | |
|
TrailsBs | 0.85 (0.34) | 1.54 (0.50) | 0.82 (0.29) | <.001 | |
|
HVOTt | 3.26 (0.15) | 3.06 (0.22) | 3.27 (0.13) | <.001 | |
|
FASu | 39.85 (12.52) | 28.76 (11.68) | 40.50 (12.26) | <.001 | |
|
CNT_Animalv | 19.48 (5.68) | 12.22 (4.37) | 19.91 (5.46) | <.001 |
aMCI: mild cognitive impairment.
bNC: normal control.
cSignificant associations were claimed if
dNP: neuropsychological.
eLMi: Logical Memory—Immediate Recall.
fLMd: Logical Memory—Delayed Recall.
gLMr: Logical Memory—Recognition.
hVRi: Visual Reproduction—Immediate Recall.
iVRd: Visual Reproduction—Delayed Recall.
jVRr: Visual Reproduction—Recognition.
kPASi: Paired Associate Learning—Immediate Recall.
lPASd: Paired Associate Learning—Delayed Recall.
mPASr: Paired Associate Learning—Recognition.
nDSf: Digit Span—Forward.
oDSb: Digit Span—Backward.
pSIM: Similarities.
qBNT30: Boston Naming Test—30-item version.
rTrailsA: Trail Making Test A.
sTrailsB: Trail Making Test B.
tHVOT: Hooper Visual Organization Test.
uFAS: Controlled Oral Word Association Test.
vCNT_Animal: Category Naming Test—Animal.
The most significant acoustic feature for each neuropsychological (NP) test.
NP test | Significant acoustic features, n | The most significant acoustic feature | Effect size | SE | |
LMib | 7 | audSpec_Rfilt_sma [25] | 0.0490 | 0.0095 | 2.7 × 10–7 |
LMdc | 3 | audSpec_Rfilt_sma [25] | 0.0402 | 0.0094 | 1.9 × 10–5 |
LMrd | 3 | audSpec_Rfilt_sma [23] | 0.0397 | 0.0108 | 2.3 × 10–4 |
VRie | 49 | mfcc_sma [11] | 0.1409 | 0.0082 | 8.4 × 10–66 |
VRdf | 43 | mfcc_sma [11] | 0.1137 | 0.0082 | 3.7 × 10–44 |
VRrg | 10 | pcm_fftMag_spectralRollOff75.0_sma | –0.0358 | 0.0095 | 1.7 × 10–4 |
PASih | 0 | N/Ai | N/A | N/A | N/A |
PASdj | 0 | N/A | N/A | N/A | N/A |
PASrk | 7 | audSpec_Rfilt_sma [1] | –0.0709 | 0.0112 | 2.3 × 10–10 |
DSfl | 44 | audSpec_Rfilt_sma [6] | 0.0898 | 0.0107 | 4.8 × 10–17 |
DSbm | 30 | audSpec_Rfilt_sma [5] | 0.0624 | 0.0110 | 1.2 × 10–8 |
SIMn | 24 | pcm_fftMag_spectralRollOff75.0_sma | –0.0530 | 0.0084 | 2.4 × 10–10 |
BNT30o | 23 | mfcc_sma [2] | 0.0433 | 0.0069 | 3.2 × 10–10 |
TrailsAp | 15 | pcm_fftMag_spectralSkewness_sma | –0.0363 | 0.0075 | 1.4 × 10–6 |
TrailsBq | 1 | pcm_fftMag_spectralSkewness_sma | –0.0269 | 0.0074 | 3.1 × 10–4 |
HVOTr | 5 | F0final_sma | –0.0472 | 0.0093 | 3.6 × 10–7 |
FASs | 26 | mfcc_sma [2] | 0.0534 | 0.0073 | 3.6 × 10–13 |
CNT_Animalt | 34 | mfcc_sma [2] | 0.0715 | 0.0082 | 2.6 × 10–18 |
aSignificant associations were claimed if
bLMi: Logical Memory—Immediate Recall.
cLMd: Logical Memory—Delayed Recall.
dLMr: Logical Memory—Recognition.
eVRi: Visual Reproduction—Immediate Recall.
fVRd: Visual Reproduction—Delayed Recall.
gVRr: Visual Reproduction—Recognition.
hPASi: Paired Associate Learning—Immediate Recall.
iN/A: not applicable.
jPASd: Paired Associate Learning—Delayed Recall.
kPASr: Paired Associate Learning—Recognition.
lDSf: Digit Span—Forward.
mDSb: Digit Span—Backward.
nSIM: Similarities.
oBNT30: Boston Naming Test—30-item version.
pTrailsA: Trail Making Test A.
qTrailsB: Trail Making Test B.
rHVOT: Hooper Visual Organization Test.
sFAS: Controlled Oral Word Association Test.
tCNT_Animal: Category Naming Test—Animal.
Association between acoustic composite scores and corresponding neuropsychological tests.
Acoustic composite score | Effect size | SE | |
acoustic_LMib | 0.0579 | 0.0094 | 6.6 × 10–10 |
acoustic_LMdc | 0.0310 | 0.0095 | 1.1 × 10–3 |
acoustic_LMrd | 0.0358 | 0.0105 | 6.8 × 10–4 |
acoustic_VRie | 0.1510 | 0.0086 | 3.3 × 10–69 |
acoustic_VRdf | 0.1079 | 0.0086 | 6.5 × 10–36 |
acoustic_VRrg | –0.0291 | 0.0098 | 3.0 × 10–3 |
acoustic_PASrh | 0.0841 | 0.0114 | 1.3 × 10–13 |
acoustic_DSfi | 0.1298 | 0.0097 | 1.8 × 10–40 |
acoustic_DSbj | 0.0553 | 0.0102 | 6.2 × 10–8 |
acoustic_SIMk | 0.0719 | 0.0089 | 5.1 × 10–16 |
acoustic_BNT30l | 0.0458 | 0.0071 | 1.4 × 10–10 |
acoustic_TrailsAm | 0.0408 | 0.0088 | 3.0 × 10–6 |
acoustic_TrailsBn | –0.0269 | 0.0075 | 3.1 × 10–4 |
acoustic_HVOTo | 0.0284 | 0.0090 | 1.7 × 10–3 |
acoustic_FASp | 0.0827 | 0.0079 | 1.4 × 10–25 |
acoustic_CNT_Animalq | 0.0529 | 0.0098 | 6.5 × 10–8 |
aSignificant associations were claimed if
bLMi: Logical Memory—Immediate Recall.
cLMd: Logical Memory—Delayed Recall.
dLMr: Logical Memory—Recognition.
eVRi: Visual Reproduction—Immediate Recall.
fVRd: Visual Reproduction—Delayed Recall.
gVRr: Visual Reproduction—Recognition.
hPASr: Paired Associate Learning—Recognition.
iDSf: Digit Span—Forward.
jDSb: Digit Span—Backward.
kSIM: Similarities.
lBNT30: Boston Naming Test—30-item version.
mTrailsA: Trail Making Test A.
nTrailsB: Trail Making Test B.
oHVOT: Hooper Visual Organization Test.
pFAS: Controlled Oral Word Association Test.
qCNT_Animal: Category Naming Test—Animal.
Association between acoustic composite scores and prevalent mild cognitive impairment.
Acoustic composite score | Odds ratio (95% CI) | |
acoustic_LMib | 1.09 (0.94-1.26) | 2.6 × 10–1 |
acoustic_LMdc | 1.14 (0.99-1.31) | 7.4 × 10–2 |
acoustic_LMrd | 1.23 (1.08-1.40) |
|
acoustic_VRie | 1.05 (0.92-1.19) | 4.7 × 10–1 |
acoustic_VRdf | 1.07 (0.94-1.21) | 3.2 × 10–1 |
acoustic_VRrg | 0.94 (0.80-1.10) | 4.6 × 10–1 |
acoustic_PASrh | 0.9 (0.81-0.99) | 3.6 × 10–2 |
acoustic_DSfi | 1.17 (1.04-1.32) | 1.1 × 10–2 |
acoustic_DSbj | 0.94 (0.83-1.07) | 3.5 × 10–1 |
acoustic_SIMk | 0.94 (0.84-1.06) | 3.1 × 10–1 |
acoustic_BNT30l | 0.92 (0.82-1.04) | 2.0 × 10–1 |
acoustic_TrailsAm | 1.12 (0.98-1.28) | 9.6 × 10–2 |
acoustic_TrailsBn | 0.69 (0.59-0.81) |
|
acoustic_HVOTo | 0.91 (0.81-1.03) | 1.4 × 10–1 |
acoustic_FASp | 0.72 (0.64-0.81) |
|
acoustic_CNT_Animalq | 0.70 (0.61-0.80) |
|
aSignificant associations were claimed if
bLMi: Logical Memory—Immediate Recall.
cLMd: Logical Memory—Delayed Recall.
dLMr: Logical Memory—Recognition.
eVRi: Visual Reproduction—Immediate Recall.
fVRd: Visual Reproduction—Delayed Recall.
gVRr: Visual Reproduction—Recognition.
hPASr: Paired Associate Learning—Recognition.
iDSf: Digit Span—Forward.
jDSb: Digit Span—Backward.
kSIM: Similarities.
lBNT30: Boston Naming Test—30-item version.
mTrailsA: Trail Making Test A.
nTrailsB: Trail Making Test B.
oHVOT: Hooper Visual Organization Test.
pFAS: Controlled Oral Word Association Test.
qCNT_Animal: Category Naming Test—Animal.
Association between acoustic composite scores and incident mild cognitive impairment.
Acoustic composite score | Hazard ratio (95% CI) | |
acoustic_LMib | 0.60 (0.47-0.77) |
|
acoustic_LMdc | 0.76 (0.59-0.97) | 2.9 × 10–2 |
acoustic_LMrd | 0.74 (0.61-0.91) | 3.9 × 10–3 |
acoustic_VRie | 1.28 (1.10-1.48) |
|
acoustic_VRdf | 1.25 (1.08-1.44) |
|
acoustic_VRrg | 0.44 (0.33-0.59) |
|
acoustic_PASrh | 1.11 (0.95-1.30) | 2.0 × 10–1 |
acoustic_DSfi | 1.11 (0.96-1.29) | 1.6 × 10–1 |
acoustic_DSbj | 1.09 (0.93-1.27) | 2.9 × 10–1 |
acoustic_SIMk | 1.37 (1.16-1.61) |
|
acoustic_BNT30l | 1.23 (1.06-1.43) | 6.4 × 10–3 |
acoustic_TrailsAm | 0.75 (0.61-0.93) | 7.9 × 10–3 |
acoustic_TrailsBn | 2.03 (1.58-2.60) |
|
acoustic_HVOTo | 0.78 (0.67-0.91) |
|
acoustic_FASp | 0.87 (0.76-1.01) | 6.1 × 10–2 |
acoustic_CNT_Animalq | 0.85 (0.70-1.02) | 8.6 × 10–2 |
aSignificant associations were claimed if
bLMi: Logical Memory—Immediate Recall.
cLMd: Logical Memory—Delayed Recall.
dLMr: Logical Memory—Recognition.
eVRi: Visual Reproduction—Immediate Recall.
fVRd: Visual Reproduction—Delayed Recall.
gVRr: Visual Reproduction—Recognition.
hPASr: Paired Associate Learning—Recognition.
iDSf: Digit Span—Forward.
jDSb: Digit Span—Backward.
kSIM: Similarities.
lBNT30: Boston Naming Test—30-item version.
mTrailsA: Trail Making Test A.
nTrailsB: Trail Making Test B.
oHVOT: Hooper Visual Organization Test.
pFAS: Controlled Oral Word Association Test.
qCNT_Animal: Category Naming Test—Animal.
The receiver operating characteristic curves of 2 models to predict incident mild cognitive impairment. AUC: area under the curve.
Relating acoustic features with NP test performance is potentially a novel way for screening at the preclinical stages of AD and other dementias. This paper clarifies the relationship between comprehensive acoustic features and NP test performance on large cohort data. Representations relative spectra–style filtered auditory spectrum (spectral), MFCC (cepstral), and magnitude of spectral features (spectral) are 3 categories of acoustic features that were significantly associated with NP test performance. Representations relative spectra–style filtered auditory spectrum is a filtered representation of an audio signal that is robust to additive and convolutional noise [
Results could expand current evidence regarding the predictive ability of digital voice on MCI that are critical to monitor early cognitive decline. The added predictive ability of acoustic features was evaluated by constructing random forest models with baseline features and additional acoustic composite scores. The model with baseline features and 7 acoustic composite scores corresponding to LMi, VRi, VRd, VRr, SIM, TrailsB, and HVOT tests could achieve an AUC of 0.755 for incident MCI prediction. Monitoring acoustic features outside of the clinical settings offers a more convenient way to aid in the assessment of cognitive health than traditional methods. Increasing evidence suggests that the human voice can be a predictor of cognitive decline before a clinical diagnosis of AD is made [
Notably, the association between acoustic features and a standard epidemiologic NP test procedure was examined based on participants from a community-based cohort with a diverse range of ages and health conditions. The large volume of voice data provides a more robust representation of participants. Each voice recording lasts, on average, around an hour and contains a wealth of information. The longitudinal collection of data provides a great opportunity to assess the cognitive health of participants throughout the entire course of the disease and prospectively reveals a temporal relationship between acoustic features and MCI. It is worth to noting that 4 of the acoustic composite scores (
This study also has some limitations. First, the use of NP tests to diagnose MCI may have led to some circularity and an overestimation of the diagnosis performance [
We examined the association of acoustic features with specific cognitive functions—prevalent and incident MCI—in a large community-based population. Overall, this study’s establishment of a relationship between MCI risk and human voice features provides foundational evidence for an alternative cognitive assessment approach that is cost-effective and easy to administer for detecting cognition-related disorders. Multiple acoustic features were significantly associated with NP test performance and MCI and could be potentially used as a digital biomarker for early cognitive impairment monitoring.
Supplemental tables and figure.
Alzheimer disease
area under the curve
Controlled Oral Word Association Test
Framingham Heart Study
Hooper Visual Organization Test
low-level descriptor
Logical Memory—Immediate Recall
mild cognitive impairment
Mel-frequency cepstral coefficient VRi: Visual Reproduction—Immediate Recall
neuropsychological
odds ratio
receiver operating characteristic
Similarities
Trail Making Test B
Visual Reproduction—Delayed Recall
Visual Reproduction—Recognition
We acknowledge the Framingham Heart Study (FHS) participants for their dedication. This study would not be possible without them. We also thank the researchers in the FHS for their efforts over the years in the examination of subjects.
This work was supported by the National Heart, Lung, and Blood Institute (contract N01-HC-25195) and by grants from the National Institute on Aging (AG-008122, AG-16495, AG-062109, AG-049810, AG-068753, AG054156, and U01AG068221) and the National Institute of Neurological Disorders and Stroke (NS017950). It was also supported by Defense Advanced Research Projects Agency (contract FA8750-16-C-0299) and Pfizer, Inc. This work was also supported by the grants from the Alzheimer’s Association (AARG-NTF-20-643020) and the American Heart Association (20SFRN35360180). The views expressed in this manuscript are those of the authors and do not necessarily represent the views of the National Institutes of Health or the US Department of Health and Human Services. The funding agencies had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
The derived acoustic features could be requested through a formal research application to the Framingham Heart Study [
RA is a scientific advisor to Signant Health and a consultant to Biogen and the Davos Alzheimer's Collaborative.