Dr. Rhoda Au E-Mail
Professor, Anatomy & Neurobiology, Boston University School of Medicine, Boston, MA, USA;
Professor, Epidemiology, Boston University School of Public Health, Boston, MA, USA
Research Keywords: cognition, neuropsychological testing, dementia, Alzheimer’s disease, brain aging, brain health, digital phenotyping, digital biomarkers
Precision medicine brings the promise of individualized treatments but doesn’t go far enough. In the U.S., 86% of healthcare related concerns are centered on chronic diseases, the majority of which take years, if not decades, to progress to diagnosis. If we refocus healthcare objectives from detecting and treating disease to optimizing health and identifying indices of change within the realm of normal, available interventions may alter the trajectory of symptoms so dramatically that the diseases never emerge. In this special issue, we seek to highlight how internet connected devices, smartphone applications and advanced analytic methods allow tracking and analyzing health-related behaviors in ways that have not been possible before. These emergent technologies will challenge current gold standards and fuel innovations that will enable solutions that move the primary focus of healthcare from precision medicine to a broader emphasis on precision health.
Keywords: Alzheimer’s disease, brain health, early detection, algorithms, smartphone applications, interconnected devices, advanced computational analytics, disease stratification, precision, digital biomarkers, digital health technologies
Smartphone technology is increasingly used by the public to assess sleep. Specific features of some sleep-tracking applications are comparable to actigraphy in objectively monitoring sleep. The clinical utility of smartphone apps should be investigated further to increase access to convenient means of monitoring sleep.
Smartphone and subjective sleep measures were administered to 29 community-dwelling healthy adults [aged 20–67, Mean (M) = 26.8; 18 women, 11 men], and actigraphy to 19 of them. Total sleep time (TST) and sleep efficiency were measured with actigraphy and the Sleep Time App (Azumio Inc.). Sleep diaries captured subjective TST and sleep efficiency, and the Epworth Sleepiness Scale and Pittsburgh Sleep Quality Index provided self-report data. An exit questionnaire was administered to examine App feasibility and likelihood of future use.
The App significantly overestimated TST when compared to actigraphy. There was no significant difference in sleep efficiency between methodologies. There was also no significant difference between TST recorded through the App and through sleep diaries. Participants’ self-reported ease of use of the smartphone App positively correlated with likelihood of future use.
Based on the current findings, future research is needed to investigate the utility and feasibility of multiple smartphone applications in monitoring sleep in healthy and clinical populations.
Although clinicians primarily diagnose dementia based on a combination of metrics such as medical history and formal neuropsychological tests, recent work using linguistic analysis of narrative speech to identify dementia has shown promising results. We aim to build upon research by Thomas JA & Burkardt HA et al. (J Alzheimers Dis. 2020;76:905–2) and Alhanai et al. (arXiv:1710.07551v1. 2020) on the Framingham Heart Study (FHS) Cognitive Aging Cohort by 1) demonstrating the predictive capability of linguistic analysis in differentiating cognitively normal from cognitively impaired participants and 2) comparing the performance of the original linguistic features with the performance of expanded features.
Data were derived from a subset of the FHS Cognitive Aging Cohort. We analyzed a sub-selection of 98 participants, which provided 127 unique audio files and clinical observations (n = 127, female = 47%, cognitively impaired = 43%). We built on previous work which extracted original linguistic features from transcribed audio files by extracting expanded features. We used both feature sets to train logistic regression classifiers to distinguish cognitively normal from cognitively impaired participants and compared the predictive power of the original and expanded linguistic feature sets, and participants’ Mini-Mental State Examination (MMSE) scores.
Based on the area under the receiver-operator characteristic curve (AUC) of the models, both the original (AUC = 0.882) and expanded (AUC = 0.883) feature sets outperformed MMSE (AUC = 0.870) in classifying cognitively impaired and cognitively normal participants. Although the original and expanded feature sets had similar AUC, the expanded feature set showed better positive and negative predictive value [expanded: positive predictive value (PPV) = 0.738, negative predictive value (NPV) = 0.889; original: PPV = 0.701, NPV = 0.869].
Linguistic analysis has been shown to be a potentially powerful tool for clinical use in classifying cognitive impairment. This study expands the work of several others, but further studies into the plausibility of speech analysis in clinical use are vital to ensure the validity of speech analysis for clinical classification of cognitive impairment.
Impaired sleep quality and sleep oxygenation are common sleep pathologies. This study assessed the impact of these abnormalities on white matter (WM) integrity in an epidemiological cohort.
The target population was the Framingham Heart Study Generation-2/Omni-1 Cohorts. Magnetic resonance imaging (diffusion tensor imaging) was used to assess WM integrity. Wearable digital devices were used to assess sleep quality: the (M1-SleepImageTM system) and the Nonin WristOx for nocturnal oxygenation. The M1 device collects trunk actigraphy and the electrocardiogram (ECG); sleep stability indices were computed using cardiopulmonary coupling using the ECG. Two nights of recording were averaged.
Stable sleep was positively associated with WM health. Actigraphic periods of wake during the sleep period were associated with increased mean diffusivity. One marker of sleep fragmentation which covaries with respiratory chemoreflex activation was associated with reduced fractional anisotropy and increased mean diffusivity. Both oxygen desaturation index and oxygen saturation time under 90% were associated with pathological directions of diffusion tensor imaging signals. Gender differences were noted across most variables, with female sex showing the larger and significant impact.
Sleep quality assessed by a novel digital analysis and sleep hypoxia was associated with WM injury, especially in women.
Reduced pre-operative cognitive functioning in older adults is a risk factor for postoperative complications, but it is unknown if preoperative digitally-acquired clock drawing test (CDT) cognitive screening variables, which allow for more nuanced examination of patient performance, may predict lengthier hospital stay and greater cost of hospital care. This issue is particularly relevant for older adults undergoing transcatheter aortic valve replacement (TAVR), as this surgical procedure is chosen for intermediate-risk older adults needing aortic replacement. This proof of concept research explored if specific latency and graphomotor variables indicative of planning from digitally-acquired command and copy clock drawing would predict post-TAVR duration and cost of hospitalization, over and above age, education, American Society of Anesthesiologists (ASA) physical status classification score, and frailty.
Form January 2018 to December 2019, 162 out of 190 individuals electing TAVR completed digital clock drawing as part of a hospital wide cognitive screening program. Separate hierarchical regressions were computed for the command and copy conditions of the CDT and assessed how a-priori selected clock drawing metrics (total time to completion, ideal digit placement difference, and hour hand distance from center; included within the same block) incrementally predicted outcome, as measured by R2 change significance values.
Above and beyond age, education, ASA physical status classification score, and frailty, only digitally-acquired CDT copy performance explained significant variance for length of hospital stay (9.5%) and cost of care (8.9%).
Digital variables from clock copy condition provided predictive value over common demographic and comorbidity variables. We hypothesize this is due to the sensitivity of the copy condition to executive dysfunction, as has been shown in previous studies for subtypes of cognitive impairment. Individuals undergoing TAVR procedures are often frail and executively compromised due to their cerebrovascular disease. We encourage additional research on the value of digitally-acquired clock drawing within different surgery types. Type of cognitive impairment and the value of digitally-acquired CDT command and copy parameters in other surgeries remain unknown.
Prior research employing a standard backward digit span test has been successful in operationally defining neurocognitive constructs associated with the Fuster’s model of executive attention. The current research sought to test if similar behavior could be obtained using a cross-modal mental manipulation test.
Memory clinic patients were studied. Using Jak-Bondi criteria, 24 patients were classified with mild cognitive impairment (MCI), and 33 memory clinic patients did not meet criteria for MCI (i.e. non-MCI). All patients were assessed with the digital version of the WRAML-2 Symbolic Working Memory Test-Part 1, a cross-modal mental manipulation task where patients hear digits, but respond by touching digits from lowest to highest on an answer key. Only 4 and 5-span trials were analyzed. Using an iPad, all test stimuli were played; and, all responses were obtained with a touch key. Only correct trials were analyzed. Average time to complete trials and latency for each digit was recorded.
Groups did not differ when average time to complete 4-span trials was calculated. MCI patients displayed slower latency, or required more time to re-order the 1st and 3rd digits. Regression analyses, primarily involving initial and latter response latencies, were associated with better, but different underlying neuropsychological abilities. Almost no 5-span analyses were significant.
This cross-modal test paradigm found no difference for total average time. MCI patients generated slower 1st and 3rd response latency, suggesting differences in time allocation to achieve correct serial order recall. Moreover, different neuropsychological abilities were associated with different time-based test components. These data extend prior findings using a standard backward digit span test. Differences in time epochs are consistent with constructs underlying the model of executive attention and help explain mental manipulation deficits in MCI. These latency measures could constitute neurocognitive biomarkers that track emergent disease.
Despite its high frequency of occurrence, mild traumatic brain injury (mTBI), or concussion, is difficult to recognize and diagnose, particularly in pediatric populations. Conventional methods to diagnose mTBI primarily rely on clinical questionnaires and sometimes include neuroimaging or pencil and paper neuropsychological testing. However, these methods are time consuming, require administration/interpretation from health professionals, and lack adequate test sensitivity and specificity. This study explores the use of BrainCheck Sport, a computerized neurocognitive test that is available on iPad, iPhone, or computer desktop, for mTBI assessment. The BrainCheck Sport Battery consists of 6 gamified traditional neurocognitive tests that assess areas of cognition vulnerable to mTBI such as attention, processing speed, executing functioning, and coordination.
We administered BrainCheck Sport to 10 participants diagnosed with mTBI at the emergency department of Children’s hospital or local high school within 96 h of injury, and 115 normal controls at a local high school. Statistical analysis included Mann-Whitney U test, chi-square tests, and Hochberg tests to examine differences between the mTBI group and control group on each assessment in the battery. Significant metrics from these assessments were used to build a logistic regression model that distinguishes mTBI from control participants.
BrainCheck Sport was able to detect significant differences in Coordination, Stroop, Immediate/Delayed Recognition between normal controls and mTBI patients. Receiver operating characteristic (ROC) analysis of our logistic regression model found a sensitivity of 84% and specificity of 81%, with an area under the curve of 0.884.
BrainCheck Sport has potential in distinguishing mTBI from control participants, by providing a shorter, gamified test battery to assess cognitive function after brain injury, while also providing a method for tracking recovery with the opportunity to do so remotely from a patient’s home.
Human voice contains rich information. Few longitudinal studies have been conducted to investigate the potential of voice to monitor cognitive health. The objective of this study is to identify voice biomarkers that are predictive of future dementia.
Participants were recruited from the Framingham Heart Study. The vocal responses to neuropsychological tests were recorded, which were then diarized to identify participant voice segments. Acoustic features were extracted with the OpenSMILE toolkit (v2.1). The association of each acoustic feature with incident dementia was assessed by Cox proportional hazards models.
Our study included 6, 528 voice recordings from 4, 849 participants (mean age 63 ± 15 years old, 54.6% women). The majority of participants (71.2%) had one voice recording, 23.9% had two voice recordings, and the remaining participants (4.9%) had three or more voice recordings. Although all asymptomatic at the time of examination, participants who developed dementia tended to have shorter segments than those who were dementia free (P < 0.001). Additionally, 14 acoustic features were significantly associated with dementia after adjusting for multiple testing (P < 0.05 / 48 = 1 × 10–3). The most significant acoustic feature was jitterDDP_sma_de (P = 7.9 × 10–7), which represents the differential frame-to-frame Jitter. A voice based linear classifier was also built that was capable of predicting incident dementia with area under curve of 0.812.
Multiple acoustic and linguistic features are identified that are associated with incident dementia among asymptomatic participants, which could be used to build better prediction models for passive cognitive health monitoring.
Research suggests that Alzheimer’s disease (AD) is heterogeneous with numerous subtypes. Through a proprietary interactive ML system, several underlying biological mechanisms associated with AD pathology were uncovered. This paper is an introduction to emerging analytic efforts that can more precisely elucidate the heterogeneity of AD.
A public AD data set (GSE84422) consisting of transcriptomic data of postmortem brain samples from healthy controls (n = 121) and AD (n = 380) subjects was analyzed. Data were processed by an artificial intelligence platform designed to discover potential drug repurposing candidates, followed by an interactive augmented intelligence program.
Using perspective analytics, six perspective classes were identified: Class I is defined by TUBB1, ASB4, and PDE5A; Class II by NRG2 and ZNF3; Class III by IGF1, ASB4, and GTSE1; Class IV is defined by cDNA FLJ39269, ITGA1, and CPM; Class V is defined by PDE5A, PSEN1, and NDUFS8; and Class VI is defined by DCAF17, cDNA FLJ75819, and SLC33A1. It is hypothesized that these classes represent biological mechanisms that may act alone or in any combination to manifest an Alzheimer’s pathology.
Using a limited transcriptomic public database, six different classes that drive AD were uncovered, supporting the premise that AD is a heterogeneously complex disorder. The perspective classes highlighted genetic pathways associated with vasculogenesis, cellular signaling and differentiation, metabolic function, mitochondrial function, nitric oxide, and metal ion metabolism. The interplay among these genetic factors reveals a more profound underlying complexity of AD that may be responsible for the confluence of several biological factors. These results are not exhaustive; instead, they demonstrate that even within a relatively small study sample, next-generation machine intelligence can uncover multiple genetically driven subtypes. The models and the underlying hypotheses generated using novel analytic methods may translate into potential treatment pathways.
This commentary is the product of a concerted effort to understand the needs, barriers, and gaps in the field of speech and language biomarkers for Alzheimer’s disease (AD). It distills interviews, surveys, and extensive correspondence with global leaders in the areas of dementia research, clinical trials, linguistics, and data analytics into an idealized clinical-study design for the harmonized collection of voice recordings. The ultimate goal of the effort is to democratize the ongoing speech and language analytics efforts by making such rich datasets available to the wider research ecosystem.
Quantitative analysis of brain single photon emission computed tomography (SPECT) perfusion imaging is dependent on normative datasets that are challenging to produce. This study investigated the combination of SPECT neuroimaging from a large clinical population rather than small numbers of controls. The authors hypothesized this “population template” would demonstrate noninferiority to a control dataset, providing a viable alternative for quantifying perfusion abnormalities in SPECT neuroimaging.
A total of 2, 068 clinical SPECT scans were averaged to form the “population template”. Validation was three-fold. First, the template was imported into SPECT brain analysis software, MIMneuro®, and compared against its control dataset of 90 individuals through its region and cluster analysis tools. Second, a cohort of 100 cognitively impaired subjects was evaluated against both the population template and MIMneuro®’s normative dataset to compute region-based metrics. Concordance and intraclass correlation coefficients, mean square deviations, total deviation indices, and limits of agreement were derived from these data to measure agreement and test for noninferiority. Finally, the same patients were clinically read in CereMetrix® to confirm that expected perfusion patterns appeared after comparison to the template.
MIMneuro®’s default threshold for normality is ± 1.65 z-score and this served as our noninferiority margin. Direct comparison of the template to controls produced no regions that exceeded this threshold and all clusters identified were far from statistically significant. Agreement measures revealed consistency between the softwares and that CereMetrix® results were noninferior to MIMneuro®, albeit with proportional bias. Visual analysis also confirmed that expected perfusion patterns appeared when individual scans were compared to the population template within CereMetrix®.
The authors demonstrated a population template was noninferior to a smaller control dataset despite inclusion of abnormal scans. This suggests that our patient-based population template can serve as an alternative for identifying and quantifying perfusion abnormalities in brain SPECT.