A Novel Explainable AI Method to Assess Associations between Temporal Patterns in Patient Trajectories and Adverse Outcome Risks: Analyzing Fitness as a Risk Factor of ADRD.

We present a novel explainable artificial intelligence (XAI) method to assess the associations between the temporal patterns in the patient trajectories recorded in longitudinal clinical data and the adverse outcome risks, through explanations for a type of deep neural network model called Hybrid Value-Aware Transformer (HVAT) model. The HVAT models can learn jointly from longitudinal and non-longitudinal clinical data, and in particular can leverage the time-varying numerical values associated with the clinical codes or concepts within the longitudinal data for outcome prediction. The key component of the XAI method is the definitions of two derived variables, the temporal mean and the temporal slope, which are defined for the clinical concepts with associated time-varying numerical values. The two variables represent the overall level and the rate of change over time, respectively, in the trajectory formed by the values associated with the clinical concept. Two operations on the original values are designed for changing the values of the two derived variables separately. The effects of the two variables on the outcome risks learned by the HVAT model are calculated in terms of impact scores and impacts. Interpretations of the impact scores and impacts as being similar to those of odds ratios are also provided. We applied the XAI method to the study of cardiorespiratory fitness (CRF) as a risk factor of Alzheimer's disease and related dementias (ADRD). Using a retrospective case-control study design, we found that each one-unit increase in the overall CRF level is associated with a 5% reduction in ADRD risk, while each one-unit increase in the changing rate of CRF over time is associated with a 1% reduction. A closer investigation revealed that the association between the changing rate of CRF level and the ADRD risk is nonlinear, or more specifically, approximately piecewise linear along the axis of the changing rate on two pieces: the piece of negative changing rates and the piece of positive changing rates.

Enhancing Clinical Data Analysis by Explaining Interaction Effects between Covariates in Deep Neural Network Models.

Deep neural network (DNN) is a powerful technology that is being utilized by a growing number and range of research projects, including disease risk prediction models. One of the key strengths of DNN is its ability to model non-linear relationships, which include covariate interactions. We developed a novel method called interaction scores for measuring the covariate interactions captured by DNN models. As the method is model-agnostic, it can also be applied to other types of machine learning models. It is designed to be a generalization of the coefficient of the interaction term in a logistic regression; hence, its values are easily interpretable. The interaction score can be calculated at both an individual level and population level. The individual-level score provides an individualized explanation for covariate interactions. We applied this method to two simulated datasets and a real-world clinical dataset on Alzheimer's disease and related dementia (ADRD). We also applied two existing interaction measurement methods to those datasets for comparison. The results on the simulated datasets showed that the interaction score method can explain the underlying interaction effects, there are strong correlations between the population-level interaction scores and the ground truth values, and the individual-level interaction scores vary when the interaction was designed to be non-uniform. Another validation of our new method is that the interactions discovered from the ADRD data included both known and novel relationships.

Hybrid Value-Aware Transformer Architecture for Joint Learning from Longitudinal and Non-Longitudinal Clinical Data.

Transformer is the latest deep neural network (DNN) architecture for sequence data learning, which has revolutionized the field of natural language processing. This success has motivated researchers to explore its application in the healthcare domain. Despite the similarities between longitudinal clinical data and natural language data, clinical data presents unique complexities that make adapting Transformer to this domain challenging. To address this issue, we have designed a new Transformer-based DNN architecture, referred to as Hybrid Value-Aware Transformer (HVAT), which can jointly learn from longitudinal and non-longitudinal clinical data. HVAT is unique in the ability to learn from the numerical values associated with clinical codes/concepts such as labs, and in the use of a flexible longitudinal data representation called clinical tokens. We have also trained a prototype HVAT model on a case-control dataset, achieving high performance in predicting Alzheimer's disease and related dementias as the patient outcome. The results demonstrate the potential of HVAT for broader clinical data-learning tasks.

Hybrid Value-Aware Transformer Architecture for Joint Learning from Longitudinal and Non-Longitudinal Clinical Data.

Transformer is the latest deep neural network (DNN) architecture for sequence data learning that has revolutionized the field of natural language processing. This success has motivated researchers to explore its application in the healthcare domain. Despite the similarities between longitudinal clinical data and natural language data, clinical data presents unique complexities that make adapting Transformer to this domain challenging. To address this issue, we have designed a new Transformer-based DNN architecture, referred to as Hybrid Value-Aware Transformer (HVAT), which can jointly learn from longitudinal and non-longitudinal clinical data. HVAT is unique in the ability to learn from the numerical values associated with clinical codes/concepts such as labs, and also the use of a flexible longitudinal data representation called clinical tokens. We trained a prototype HVAT model on a case-control dataset, achieving high performance in predicting Alzheimer’s disease and related dementias as the patient outcome. The result demonstrates the potential of HVAT for broader clinical data learning tasks.

Neurocognitive predictors of financial capacity across the dementia spectrum: Normal aging, mild cognitive impairment, and Alzheimer's disease.

Financial capacity is a complex instrumental activity of daily living critical to independent functioning of older adults and sensitive to impairment in patients with amnestic mild cognitive impairment (MCI) and Alzheimer's disease (AD). However, little is known about the neurocognitive basis of financial impairment in dementia. We developed cognitive models of financial capacity in cognitively healthy older adults (n = 85) and patients with MCI (n = 113) and mild AD (n = 43). All participants were administered the Financial Capacity Instrument (FCI) and a neuropsychological test battery. Univariate correlation and multiple regression procedures were used to develop cognitive models of overall FCI performance across groups. The control model (R2 = .38) comprised (in order of entry) written arithmetic skills, delayed story recall, and simple visuomotor sequencing. The MCI model (R2 = .69) comprised written arithmetic skills, visuomotor sequencing and set alternation, and race. The AD model (R2 = .65) comprised written arithmetic skills, simple visuomotor sequencing, and immediate story recall. Written arithmetic skills (WRAT-3 Arithmetic) was the primary predictor across models, accounting for 27% (control model), 46% (AD model), and 55% (MCI model) of variance. Executive function and verbal memory were secondary model predictors. The results offer insight into the cognitive basis of financial capacity across the dementia spectrum of cognitive aging, MCI, and AD.

Commentary on "a roadmap for the prevention of dementia II. Leon Thal Symposium 2008." Innovations in care that advance Alzheimer's disease drug development.

Advancing the development of drugs for the prevention and treatment of Alzheimer's disease (AD) is dependent on the ability of investigators to identify, recruit, and retain appropriate subjects in clinical trials. Innovations in care that link primary-care providers with AD researchers can help overcome barriers to early, specific diagnosis and access to research studies. Collaborative care provides a new paradigm for the mutual benefit of patients, providers, and AD research. Recommendations to achieve this goal include funding clinical centers of excellence in AD, linked with community physicians to utilize clinical care and initial evaluations as early entry points for patients into AD research, and funding mini-fellowships for community physicians. Reimbursement for dementia care should be expanded to include periodic cognitive assessments for at-risk individuals, medically directed dementia education, and diagnostic imaging and biomarkers. These innovations can simultaneously improve the translation of research advances, and benefit AD research.

Correlation Between the Movement Disorder Society's Unified Parkinson's Disease Rating Scale and Nonmotor Scales in Patients with Parkinson's Disease.

Background: The Movement Disorder Society's Unified Parkinson's Disease Rating Scale (MDSUPDRS), Scales for Outcomes in Parkinson's Disease-Autonomic (SCOPA-AUT), Mayo Sleep Questionnaire, Epworth Sleepiness Scale, and Neuropsychiatric Inventory Questionnaire (NPI-Q) are validated instruments for assessing signs and symptoms of Parkinson's disease (PD). Objective: We sought to determine whether responses on the MDS-UPDRS correlate with responses to other scales used in patients with PD. Design: Study subjects were enrolled in the Arizona Study of Aging and Neurodegenerative Disorders (AZSAND). Participants were selected if they had completed all scales within a one-month window. Spearman's rank correlation coefficients were calculated. Results: A total of 96 eligible subjects were identified. High correlation (r-values) was found between the SCOPA-AUT and MDS-UPDRS excessive saliva (0.73; p<0.001), constipation (0.62; p<0.001), and swallowing (0.59; p<0.001) questions. The r-values for the NPI-Q and MDS-UPDRS depression and anxiety questions were 0.53 (p<0.001), and 0.67 (p<0.001). Conclusion: MDS-UPDRS correlates well with some but not all questions from the SCOPA-AUT and NPI-Q. This work emphasizes the importance of employing multiple methods for assessing nonmotor symptoms in patients with PD.

Brain Lewy-Type Synucleinopathy Density Is Associated with a Lower Prevalence of Atherosclerotic Cardiovascular Disease Risk Factors in Patients with Parkinson's Disease1.

BACKGROUND: Some epidemiology studies suggest that atherosclerotic cardiovascular disease (ASCVD) risk factors increase the risk of developing Parkinson's disease (PD). However, conflicting data suggest lower rates of ASCVD in PD. OBJECTIVE: The objective of this study is to determine, with data from a longitudinal clinicopathological study, whether ASCVD risk factors are associated with a PD diagnosis and/or increased brain or peripheral load of Lewy-type synucleinopathy (LTS). METHODS: All subjects were followed to autopsy and neuropathological examination in the Arizona Study of Aging and Neurodegenerative Disorders (AZSAND). Multivariable regression models, including age, gender, and smoking history, were used to investigate the association of a PD diagnosis or brain or submandibular gland LTS load with ASCVD risk factors. RESULTS: 150 subjects were included (PD n = 60, controls n = 90). Univariable comparisons and regression models showed a general trend to inverse associations. The multivariable odds ratio (OR) of brain LTS load for carotid artery disease was 0.93 (95% CI: 0.86 to 0.98; p = 0.02), for anticoagulant use 0.95 (95% CI: 0.90 to 0.99; p = 0.04) and for abnormal heart weight 0.96 (95% CI: 0.92 to 0.99; p = 0.01). Composite clinical and overall (clinical + pathology composite risk scores) composite risk scores were also significantly lower in the PD subjects (p = 0.0164 and 0.0187, respectively). Submandibular gland LTS load was not significantly related to ASCVD conditions. CONCLUSIONS: This study shows associations of higher brain LTS with lower prevalence of both clinical and pathological indices of ASCVD in PD subjects versus age-similar controls. We suggest that this is due to α-synuclein pathology-induced sympathetic denervation in PD.

Awareness of deficits in financial abilities in patients with mild cognitive impairment: going beyond self-informant discrepancy.

OBJECTIVE: Self and informant reports of functional abilities are weighted heavily in diagnostic decision making regarding mild cognitive impairment (MCI). However, it is unclear whether patients with MCI are fully aware and provide reliable estimates of their functional status. In this study, the authors used three different approaches to examine accuracy of self-report of financial abilities among patients with MCI. DESIGN: Cross-sectional, case-comparison group study. SETTING: University medical center. PARTICIPANTS: Seventy-four patients with MCI and their informants, and 73 cognitively healthy older adults and their informants. MEASUREMENTS: The authors compared MCI patients' report of their financial abilities with their performance on an objective measure of financial capacity. The authors also compared informant reports of patients' abilities with patients' objective test performance, and informant reports with patients' self-report. RESULTS: The authors found that the discrepancy between self-report and objective performance was higher among MCI patients compared with the cognitively healthy older adults on the financial domains of Checkbook Management, Bank Statement Management, and Bill Payment, and on overall financial capacity. The authors also found that MCI patients with poorer global cognition overestimated their financial abilities whereas those with higher depressive symptoms underestimated their financial abilities. Overall, MCI patients were better at estimating their financial abilities than their informants. CONCLUSIONS: Patients with MCI are not fully aware of deficits in their financial abilities. Both cognitive impairment and depression impact MCI patients' self-reported functioning. In addition, MCI informants misestimate patients' financial abilities. This raises concerns about the widespread use of informant report as the gold standard against which to evaluate patient self-report of functioning.

Autoimmunity in Alzheimer's disease: increased levels of circulating IgGs binding Abeta and RAGE peptides.

Plasma samples derived from 33 Alzheimer's disease (AD) and 42 control participants were subjected to several steps to purify specific anti-(amyloid)Abeta IgGs. Affinity-purified IgGs binding the peptide Abeta1-42, a neurotoxic sequence derived from the trans-membrane amyloid precursor protein, exhibited nearly four-fold higher titers in AD patients compared with their control non-AD cohort. Affinity-purified IgGs binding a fragment of the receptor for advanced glycation end products (RAGE) likewise were increased nearly three-fold in AD individuals. Abeta and RAGE IgG titers were negatively correlated with cognitive status, i.e. the more cognitively impaired individuals tended to exhibit higher IgG titers. Abeta IgG titers were negatively correlated with age in the control group, but not with the AD group. Levels of circulating AB- and RAGE-like proteins were not different between AD and control participants, nor was there a relationship between individual IgG titers and the respective Abeta- and RAGE-like proteins. Freshly prepared leukocyte preparations were subjected to flow cytometric analysis. AD individuals exhibited significantly increased populations of cells expressing binding sites for monoclonal antibodies directed against Abeta (5.5-fold), betaAPP (3.5-fold), and RAGE (2.6-fold) relative to the control group. These findings confirm the presence of circulating IgGs specifically directed at proteins implicated in immunological processes linked to AD. The close relationship between titers for Abeta and RAGE IgGs suggests the possibility that the antibodies are being produced in response to a common mechanism or protein complex (with the respective epitopes) linked to the disease.

Mild cognitive impairment can be distinguished from Alzheimer disease and normal aging for clinical trials.

BACKGROUND: Mild cognitive impairment (MCI) represents a transitional state between the cognitive changes of normal aging and very early dementia and is becoming increasingly recognized as a risk factor for Alzheimer disease (AD). The Memory Impairment Study (MIS) is a multicenter clinical trial in patients with MCI designed to evaluate whether vitamin E or donepezil is effective at delaying the time to a clinical diagnosis of AD. OBJECTIVE: To describe the baseline characteristics of patients with MCI recruited for the MIS and compare them with those of elderly controls and patients with AD in another clinical trial. DESIGN: Descriptive and comparative study of patients with MCI participating in a multicenter clinical trial. SETTING: Memory disorder centers in the United States and Canada. PATIENTS: A total of 769 patients with MCI, 107 cognitively normal elderly controls, 122 patients with very mild AD (Clinical Dementia Rating [CDR] 0.5), and 183 patients with mild AD (CDR 1.0) were evaluated. Patients in the MIS met operational criteria for amnestic MCI. Controls were recruited in parallel with the MCI group, underwent the same assessments, and had a CDR of 0. MAIN OUTCOME MEASURES: Clinical, neuropsychologic, functional, neuroimaging, and genetic measures. RESULTS: Mean +/- SD Alzheimer's Disease Assessment Scale-Cognitive Subscale scores were 5.6 +/- 3.3 for controls, 11.3 +/- 4.4 for patients with MCI, 18.0 +/- 6.2 for the AD CDR 0.5 group, and 25.2 +/- 8.8 for the AD CDR 1.0 group. Compared with controls, patients with MCI were most impaired on memory tasks, with less severe impairments in other cognitive domains. Patients with MCI were more likely than controls but less likely than patients with AD to carry the apolipoprotein E epsilon4 allele. Patients with MCI had hippocampal volumes that were intermediate between those of controls and patients with AD. CONCLUSIONS: Patients with MCI had a predominant memory impairment with relative sparing of other cognitive domains and were intermediate between clinically normal individuals and patients with AD on cognitive and functional ratings. These results demonstrate the successful implementation of operational criteria for this unique group of at-risk patients in a multicenter clinical trial.

Temporoparietal hypometabolism in frontotemporal lobar degeneration and associated imaging diagnostic errors.

OBJECTIVE: To evaluate the cause of diagnostic errors in the visual interpretation of positron emission tomographic scans with fludeoxyglucose F 18 (FDG-PET) in patients with frontotemporal lobar degeneration (FTLD) and patients with Alzheimer disease (AD). DESIGN: Twelve trained raters unaware of clinical and autopsy information independently reviewed FDG-PET scans and provided their diagnostic impression and confidence of either FTLD or AD. Six of these raters also recorded whether metabolism appeared normal or abnormal in 5 predefined brain regions in each hemisphere-frontal cortex, anterior cingulate cortex, anterior temporal cortex, temporoparietal cortex, and posterior cingulate cortex. Results were compared with neuropathological diagnoses. SETTING: Academic medical centers. PATIENTS: Forty-five patients with pathologically confirmed FTLD (n=14) or AD (n=31). RESULTS: Raters had a high degree of diagnostic accuracy in the interpretation of FDG-PET scans; however, raters consistently found some scans more difficult to interpret than others. Unanimity of diagnosis among the raters was more frequent in patients with AD (27 of 31 patients [87%]) than in patients with FTLD (7 of 14 patients [50%]) (P=.02). Disagreements in interpretation of scans in patients with FTLD largely occurred when there was temporoparietal hypometabolism, which was present in 7 of the 14 FTLD scans and 6 of the 7 scans lacking unanimity. Hypometabolism of anterior cingulate and anterior temporal regions had higher specificities and positive likelihood ratios for FTLD than temporoparietal hypometabolism had for AD. CONCLUSIONS: Temporoparietal hypometabolism in FTLD is common and may cause inaccurate interpretation of FDG-PET scans. An interpretation paradigm that focuses on the absence of hypometabolism in regions typically affected in AD before considering FTLD is likely to misclassify a significant portion of FTLD scans. Anterior cingulate and/or anterior temporal hypometabolism indicates a high likelihood of FTLD, even when temporoparietal hypometabolism is present. Ultimately, the accurate interpretation of FDG-PET scans in patients with dementia cannot rest on the presence or absence of a single region of hypometabolism but rather must take into account the relative hypometabolism of all brain regions.

Validation of consensus panel diagnosis in dementia.

BACKGROUND: The clinical diagnosis of dementing diseases largely depends on the subjective interpretation of patient symptoms. Consensus panels are frequently used in research to determine diagnoses when definitive pathologic findings are unavailable. Nevertheless, research on group decision making indicates that many factors can adversely affect panel performance. OBJECTIVE: To determine conditions that improve consensus panel diagnosis. DESIGN: Comparison of neuropathologic diagnoses with individual and consensus panel diagnoses based on clinical scenarios only, fludeoxyglucose F 18 positron emission tomography images only, and scenarios plus images. SETTING: Expert and trainee individual and consensus panel deliberations using a modified Delphi method in a pilot research study of the diagnostic utility of fludeoxyglucose F 18 positron emission tomography. PATIENTS: Forty-five patients with pathologically confirmed Alzheimer disease or frontotemporal dementia. MAIN OUTCOME MEASURES: Statistical measures of diagnostic accuracy, agreement, and confidence for individual raters and panelists before and after consensus deliberations. RESULTS: The consensus protocol using trainees and experts surpassed the accuracy of individual expert diagnoses when clinical information elicited diverse judgments. In these situations, consensus was 3.5 times more likely to produce positive rather than negative changes in the accuracy and diagnostic certainty of individual panelists. A rule that forced group consensus was at least as accurate as majority and unanimity rules. CONCLUSIONS: Using a modified Delphi protocol to arrive at a consensus diagnosis is a reasonable substitute for pathologic information. This protocol improves diagnostic accuracy and certainty when panelist judgments differ and is easily adapted to other research and clinical settings while avoiding the potential pitfalls of group decision making.

Elevated brain scyllo-inositol concentrations in patients with Alzheimer's disease.

in vivo (1)H MRS reveals reduced N-acetylaspartate (NAA) and elevated myo-inositol (mI) in patients with mild Alzheimer's disease (AD) and patients with amnestic mild cognitive impairment (MCI). We are unaware of studies that have documented abnormal scyllo-inositol (sI) levels in patients with AD or patients with MCI, although a previous MRS study in older adults has indicated that sI is a peak of interest to measure in AD. Fifteen patients with mild AD, 26 patients with amnestic MCI, and 19 healthy older adults were recruited to this study. All underwent (1)H MRS of the posterior cingulate gyrus of the brain using a 3 T MRI scanner. Increases in the sI/creatine (Cr) ratio were observed in patients with mild AD (P < 0.05). The mI/Cr ratio was raised in patients with mild AD (P < 0.01) and MCI (P < 0.05). Reduced NAA/Cr was detected in patients with mild AD (P < 0.05). The sI/Cr ratio correlated negatively (r = -0.60, P < 0.05) with a measure of clock drawing in patients with mild AD, indicating that impaired cognitive ability in AD is associated with higher concentrations of sI/Cr. In vivo measurement of sI/Cr in the posterior cingulate gyrus of patients with mild AD revealed increases compared with cognitively healthy older adults. Further research on the mechanisms of sI increase in AD is needed. Future studies on the longitudinal course of sI/Cr in MCI and AD appear warranted.

Executive function is associated with brain proton magnetic resonance spectroscopy in amnestic mild cognitive impairment.

Persons with amnestic mild cognitive impairment (MCI) show deficits on executive function measures, although the neuroanatomic basis of executive function in MCI is unknown. We investigated cognitive correlates of 3-tesla proton magnetic resonance spectroscopy (MRS) of the posterior cingulate gyrus in 26 MCI patients. Posterior cingulate ratio of myo-inositol to creatine (mI/Cr) was negatively correlated (-.51) with spontaneous clock drawing. This relationship was not attenuated after accounting for age, overall cognitive function, or memory performance. This finding suggests a role for the posterior cingulate in executive function in MCI. Proton MRS may offer a means to track neurometabolic changes associated with cognitive impairment in MCI.

Amnestic mild cognitive impairment: diagnostic outcomes and clinical prediction over a two-year time period.

Amnestic mild cognitive impairment (MCI) has been defined as a precursor to Alzheimer's disease (AD), although it is sometimes difficult to identify which persons with MCI will eventually convert to AD. We sought to predict MCI conversion to AD over a two-year follow-up period using baseline demographic and neuropsychological test data from 49 MCI patients. Using a stepwise discriminant function analysis with Dementia Rating Scale (DRS) Initiation/Perseveration and Wechsler Memory Scale, third edition (WMS-III) Visual Reproduction Percent Retention scores, we correctly classified 85.7% of the sample as either AD converters or MCI nonconverters, with 76.9% sensitivity and 88.9% specificity. Adding race, the presence of vascular risk factors, or cholinesterase inhibitor use to the analysis did not greatly change the classification rates obtained with neuropsychological test data. Examining neuropsychological test cutoff scores revealed that DRS Initiation/Perseveration scores below 37 and Visual Reproduction Percent Retention scores below 26% correctly identified AD converters with 76.9% sensitivity and 91.7% specificity. These results demonstrate that commonly administered neuropsychological tests identify persons with MCI at baseline who are at risk for conversion to AD within 1-2 years. Such methods could aid in identifying MCI patients who might benefit from early treatment, in providing prognostic information to patients, and identifying potential clinical trial participants.