Review of Artificial Intelligence for Clinical Use in Alzheimer's Disease and Related Dementias
Open accessAbstract
As the U.S. population ages, Alzheimer's disease and related dementias (ADRD) cases are increasing, resulting in long wait times for specialist care. We review state-of-the-art artificial intelligence (AI) applications in ADRD care, from streamlining clinical diagnosis to pioneering novel digital biomarkers. Near-term AI applications include neuroimaging interpretation, conversational agents for patient interviews, and digital cognitive assessments. Large language models show promise as collaborative partners, helping clinicians interpret complex data while supporting patients and caregivers. Emerging digital biomarkers-speech analysis, passive monitoring through wearable devices, electronic health record analysis, and multiomics-offer potential for continuous monitoring to detect cognitive decline years before traditional assessments. Despite the acceleration of AI innovation, most of these systems are inaccessible in clinical practice. Implementation bottlenecks include limited external validation, technical challenges, model biases, infrastructure, and regulatory requirements. This review aims to help neurologists navigate this rapidly evolving AI landscape and prepare for implementation in ADRD care.
Integrity signals
Registry
Cites
Statistics
Claims the paper makes
127 of 176 cited sources checked · 5 supported · 2 flagged · accuracy not yet measured
- Premisecitation
Dementia diagnosis delays are greater among racial and ethnic minorities.
“with greater delays among racial and ethnic minorities.6”
Quote in the paperNot a substantial claim1 of 1 sources checked- cites [6] source
The Introduction (page unavailable) reports differences in the timing of Alzheimer’s diagnosis between ethnoracial groups. It does not say that racial and ethnic minorities experience greater diagnostic delays, so the direction of the claimed disparity is unsupported. (LLM, conf=0.78)
Experimental — accuracy not yet measured
- Premisecitation
Retrieval augmented generation can ground a large language model in current clinical diagnostic criteria for all neurodegenerative diseases.
“Retrieval augmented generation can be used to ground the LLM in the most up-to-date clinical diagnostic criteria for all neurodegenerative diseases.23”
Quote in the paperNot a substantial claim1 of 1 sources checked- cites [23] source
The Methods sections describe AMIE’s chain-of-reasoning and few-shot instructions, but do not address retrieval-augmented generation as a way to ground an LLM in current diagnostic criteria. Searches for retrieval augmentation, up-to-date criteria, and neurodegenerative diseases found no passage supporting the claim, particularly its scope of all neurodegenerative diseases. (LLM, conf=0.76)
Experimental — accuracy not yet measured
- Premisecitation
A diagnostic pipeline that mimics a physician's clinical reasoning, including chain-of-thought prompting with large language models, has shown performance improvements.
“A diagnostic pipeline that mimics the clinical process/reasoning of a physician, including chain of thought reasoning with LLM prompts, has shown performance boosts.22”
Quote in the paperNot a substantial claim1 of 1 sources checked- cites [22] source
The Discussion addresses reporting gaps and inconsistent performance data for FDA-approved AI devices, not a physician-like diagnostic pipeline or improvements from LLM prompting. Searches for chain-of-thought reasoning and diagnostic-pipeline performance found no passage addressing the attributed claim; no page number was provided for the indexed passages. (LLM, conf=0.77)
Experimental — accuracy not yet measured
- Premisecitation
Post-diagnosis management challenges are increasing because new therapeutics require intensive monitoring.
“After diagnosis, challenges with management are growing as new therapeutics with intensive requirements for monitoring are emerging, adding both urgency and complexity.13,14”
Quote in the paperNot a substantial claim2 of 2 sources checked- cites [13] source
The Randomization and Intervention section describes scheduled MRIs, additional imaging every 4 to 6 weeks when abnormalities were detected, and criteria for interrupting and resuming donanemab. This supports intensive monitoring for one therapy, but the trial does not establish that post-diagnosis management challenges are increasing as new therapies emerge. Page numbers were not available in our copy. (LLM, conf=0.85)
Experimental — accuracy not yet measured - cites [14] source
The Introduction discusses the difficulty of diagnosing and monitoring early disease progression, while the Conclusion discusses challenges in neuroimaging-based classification; neither addresses post-diagnosis management. Searches for new therapeutics, intensive monitoring, and treatment-related management found no passage linking emerging therapies to growing management challenges. Page numbers were not provided in our copy. (LLM, conf=0.76)
Experimental — accuracy not yet measured
- Premisecitation
AI-powered conversational agents, especially those using large language models, are showing the capabilities needed to gather an effective patient history.
“AI-powered conversational agents are demonstrating the capabilities needed to gather an effective patient history, particularly using LLMs.”
Quote in the paperNot a substantial claim1 of 2 sources checked- cites [26] source
- cites [27] source
The Design and Methods section describes interviews with physicians, while the Results discuss obtaining histories from family members of patients with dementia; no page numbers are available in our copy. Searches for AI-powered conversational agents, LLMs, chatbots, and automated patient-history gathering found no passage addressing the claim. (LLM, conf=0.78)
Experimental — accuracy not yet measured
- Premisecitation
New care models demand additional resources that the current workforce is ill-equipped to support.
“new care models such as those proposed by the Centers for Medicare and Medicaid Services15 demand additional resources that our current workforce is ill-equipped to support.”
Quote in the paperNot a substantial claim1 of 1 sources checked- cites [15] source
The paper’s Stage 1: Individual Quality Control section describes an Alzheimer’s disease genetic association study, not care delivery. Searches for care models, the Centers for Medicare and Medicaid Services, resources, and workforce capacity found no passage addressing the claim. (LLM, conf=0.78)
Experimental — accuracy not yet measured
- Premisecitation
Traditional machine learning methods are considerably useful for analyzing unified data modalities such as biomarkers, genetic information, neuroimaging, and connected speech.
“traditional ML paradigms have considerable utility in analyzing unified data modalities, including biomarkers,16 genetic information (including emerging polygenic risk scores),17 neuroimaging,18 and connected speech.19”
Quote in the paperNot a substantial claim4 of 4 sources checked- cites [16] source
- cites [17] source
- cites [18] source
The Introduction discusses deep learning for neurological diagnosis, prognosis, and treatment, but does not address traditional machine learning applied to the listed data modalities. The Conclusions mention medical imaging, but not the claimed utility across biomarkers, genetic information, neuroimaging, and connected speech. Searches for connected speech and the other named modalities found no supporting passage in our copy. (LLM, conf=0.76)
Experimental — accuracy not yet measured - cites [19] source
Citation 19 is attached to the claim about analyzing connected speech. The Introduction discusses prior AI image analysis, and the Methods describe giving clinical summaries to ChatGPT and Google Bard; neither addresses machine learning analysis of connected speech. Searches for connected speech and speech analysis found no supporting passage in our copy. (LLM, conf=0.78)
Experimental — accuracy not yet measured
- Premisecitation
Average wait times for dementia specialists are projected to exceed 40 months by 2027.
“The average wait times for dementia specialists are projected to exceed 40 months by 2027”
Quote in the paperNot a substantial claim1 of 1 sources checked- cites [12] source
The Eligibility Criteria and Discussion describe a lecanemab trial, not access to dementia specialists. Searches for specialist wait times, 40 months, and 2027 found no projection in our copy. (LLM, conf=0.76)
Experimental — accuracy not yet measured
- Premisecitation
In primary care, more than half of dementia diagnoses are delayed until moderate or advanced stages.
“more than 50% of dementia diagnoses are delayed until moderate or advanced stages in primary care,4,5”
Quote in the paperSubstantial claim2 of 2 sources checked- cites [4] source
The Results (Prevalence of missed diagnosis of dementia; page unavailable) show that primary-care diagnostic sensitivity was lowest for mild dementia, consistent with delayed recognition. But the reported sensitivities vary across studies, and the Discussion says no study definitively established the prevalence of missed or delayed diagnosis. The paper does not establish that more than 50% of diagnoses occur only at moderate or advanced stages. (LLM, conf=0.86)
Experimental — accuracy not yet measured - cites [5] source
The Abstract reports that 55.2% of participants with dementia had no reported history of a clinical cognitive evaluation by a physician. The Discussion says more severe dementia was associated with a greater likelihood of evaluation. Neither passage establishes that more than half of diagnoses in primary care were delayed until moderate or advanced stages; the study measured whether an evaluation had occurred, not the stage at diagnosis. (LLM, conf=0.88)
Experimental — accuracy not yet measured
- Premisecitation
Empirical evidence supporting the usefulness of large language models in characterizing neurodegenerative disease continues to emerge.
“Empirical evidence supporting the utility of LLMs in neurodegenerative disease characterization continues to emerge.”
Quote in the paperNot a substantial claim2 of 2 sources checked- cites [20] source
The Discussion reports that an LLM outperformed physicians on a measure of processing medical data and clinical reasoning, supporting a general claim about clinical utility. The Methods describe 20 clinical cases, but our copy does not identify evidence about characterizing neurodegenerative disease; searches for neurodegenerative disease and related diagnoses found none. (LLM, conf=0.85)
Experimental — accuracy not yet measured - cites [21] source
The Results (§3.4; page unavailable) report improved LLM performance on biomedical true/false and multiple-choice questions using KG-RAG, providing empirical support for utility in disease-related tasks. The paper does not establish that these results apply specifically to neurodegenerative disease characterization; searches for “neurodegenerative” and “Alzheimer’s disease” found no such passage. (LLM, conf=0.84)
Experimental — accuracy not yet measured
- Premisecitation
Rural areas face delays for dementia specialists that are three times longer than urban areas.
“with rural areas facing three times longer delays compared to urban regions.12”
Quote in the paperNot a substantial claim1 of 1 sources checked- cites [12] source
The Eligibility Criteria and Discussion describe a lecanemab trial, not access to dementia specialists. Searches for rural–urban differences and specialist waiting times found no passage in our copy reporting threefold longer delays in rural areas. (LLM, conf=0.75)
Experimental — accuracy not yet measured
- Premisecitation
GPT-4 produced neuropathologic differential diagnoses that correlated highly with experts when using human-curated clinical summaries.
“provided neuropathologic differential diagnoses with high correlation to experts using human-curated clinical summaries.21”
Quote in the paperNot a substantial claim1 of 1 sources checked- cites [21] source
Section 2.6 describes GPT-4 answering validation prompts and compares its responses with ground truth, but does not describe neuropathologic differential diagnoses. Searches for neuropathologic diagnoses and human-curated clinical summaries found no passage reporting a high correlation with experts. The cited paper’s copy therefore does not establish the attributed finding. (LLM, conf=0.74)
Experimental — accuracy not yet measured
- Premisecitation
Logistic regression models are widely used in research for binary classification, particularly to distinguish healthy controls from individuals with Alzheimer's disease using MRI.
“Notably, logistic regression models are finding widespread research application in binary classification scenarios, particularly in distinguishing between healthy controls and individuals with Alzheimer’s disease using MRI.18”
Quote in the paperNot a substantial claim1 of 1 sources checked- cites [18] source
The Introduction and Conclusions (pages not provided) discuss large language models and neurology examinations, not logistic regression for MRI-based Alzheimer’s classification. Searches for logistic regression, healthy controls, Alzheimer’s disease, and MRI found no passage addressing the attributed claim. (LLM, conf=0.78)
Experimental — accuracy not yet measured
- Premisecitation
Large language models hold significant promise for directly assisting patients and caregivers throughout the dementia journey.
“Beyond supporting clinicians, LLMs also hold significant promise for directly assisting patients and caregivers throughout the dementia journey.”
Quote in the paperSubstantial claim1 of 1 sources checked- cites [48] source
The Discussion describes artificial intelligence–driven decision support for primary-care workflows, not LLMs directly assisting patients or caregivers. Searches for LLMs and direct patient or caregiver support found no passage addressing that claim; the Background’s discussion of care-partner engagement does not connect it to LLMs. No page numbers were available in our copy. (LLM, conf=0.76)
Experimental — accuracy not yet measured
- Premisecitation
The emergence of deep learning architectures, particularly convolutional neural networks, has revolutionized neuroimaging analysis, including MRI and PET scan interpretation.
“The emergence of deep learning architectures, particularly convolutional neural networks (CNNs), has revolutionized neuroimaging analysis, including MRI and PET scan interpretation.76”
Quote in the paperNot a substantial claim1 of 1 sources checked- cites [76] source
The Introduction discusses convolutional neural networks only in relation to speech-based classification of Parkinson’s medication states; the Discussion likewise addresses deep learning for speech analysis. A search for MRI, PET, and neuroimaging found no passage addressing whether CNNs have transformed scan interpretation. The paper therefore neither supports nor contradicts the cited claim. (LLM, conf=0.76)
Experimental — accuracy not yet measured
- Premisecitation
PENSIEVE-AI uses an image-based deep learning model to detect mild cognitive impairment and dementia from touch-screen tablet drawing tasks.
“PENSIEVE-AI uses an image-based deep learning model to detect mild cognitive impairment (MCI) and dementia based on touch-screen tablet drawing tasks,60”
Quote in the paperNot a substantial claim1 of 1 sources checked- cites [60] source
The Discussion (page unavailable) describes classifying Parkinsonian motor severity from short videos of gait and finger tapping. The Methods (page unavailable) mention an iPad used to record those videos, not touch-screen drawing tasks. Searches for PENSIEVE-AI, mild cognitive impairment, dementia, and tablet drawing found no passage addressing the attributed claim. (LLM, conf=0.76)
Experimental — accuracy not yet measured
- Premisecitation
Growing evidence indicates that VR-based cognitive assessments exhibit ecological validity by simulating real-world tasks and aligning with established neuropsychological and functional measures.
“There is growing evidence that VR-based cognitive assessments exhibit ecological validity by simulating real-world tasks and aligning with established neuropsychological and functional measures.65,66”
Quote in the paperNot a substantial claim1 of 2 sources checked- cites [65] source
The Introduction, Material and Methods, and Discussion describe MRI-based dementia diagnosis and comparisons with radiologists, not VR-based cognitive assessments. Searches for virtual reality, ecological validity, simulated real-world tasks, and alignment with neuropsychological or functional measures found no passage addressing the claim; page numbers were unavailable. (LLM, conf=0.76)
Experimental — accuracy not yet measured - cites [66] source
- Premisecitation
As cognitive abilities decline, patients may find AI companions more accessible than traditional educational resources.
“As cognitive abilities decline, patients may find AI companions more accessible than traditional educational resources,49”
Quote in the paperNot a substantial claim1 of 1 sources checked- cites [49] source
The Introduction discusses the accessibility and practical limitations of remote digital cognitive assessments, not AI companions or educational resources (page not provided). Searches for AI companions and for their accessibility relative to traditional educational resources found no passage addressing the claim. (LLM, conf=0.75)
Experimental — accuracy not yet measured
- Premisecitation
A growing number of studies are using AI to evaluate eye tracking data during digital cognitive and oculomotor tasks.
“Beyond task performance, a growing number of studies are also using AI to evaluate eye tracking data during digital cognitive and oculomotor tasks.64”
Quote in the paperNot a substantial claim1 of 1 sources checked- cites [64] source
The Introduction’s acquisition-protocol section describes resting-state fMRI with eyes-open fixation, while its temporal-preprocessing section describes automated classification of fMRI components. Searches for eye tracking, oculomotor tasks, and AI analysis of gaze data found no passage addressing the attributed claim. Our copy provides no basis to assess whether a growing number of studies use AI for that purpose. (LLM, conf=0.76)
Experimental — accuracy not yet measured
- Premisecitation
Most conversational agents have been tested only in text-based, low-stakes interactions.
“Most have been tested only with text-based, low-stakes interactions,33–38”
Quote in the paperNot a substantial claim4 of 6 sources checked- cites [33] source
The Simulation input parameters section describes testing ChatGPT with text prompts drawn from survey questionnaires. It does not establish that most conversational agents have been tested only this way, or characterize those interactions as low-stakes. No page number is available in our copy. (LLM, conf=0.78)
Experimental — accuracy not yet measured - cites [34] source
The Methods (Data) describe testing language models on text-based doctor–patient dialogue summarization, which supports the text-based portion of the claim. The Discussion calls clinical summarization a critical task; it does not characterize the testing as low-stakes or establish what most conversational agents have been tested on. (LLM, conf=0.82)
Experimental — accuracy not yet measured - cites [35] source
The Introduction and Limitations describe evaluating clinical-note summarization models on simulated, text-based doctor–patient conversations rather than live encounters. But the paper does not study conversational agents or establish that most have been tested this way. Its Introduction also warns that omitted facts or hallucinations could affect patient outcomes, so it does not characterize the clinical task as low-stakes. (LLM, conf=0.83)
Experimental — accuracy not yet measured - cites [36] source
- cites [37] source
- cites [38] source
The Discussion (page not provided) says previous LLM studies were limited to multiple-choice questions or simple open-ended fact retrieval, supporting the characterization of some testing as text-based and limited in scope. It does not establish that most conversational agents have been tested only in low-stakes interactions. (LLM, conf=0.78)
Experimental — accuracy not yet measured
- Premisecitation
Large language models can act as collaborative partners that help clinicians interpret familiar clinical information by incorporating current evidence and highlighting subtle patterns that might be missed in busy settings.
“LLMs can serve as collaborative partners that help clinicians interpret familiar clinical information (e.g., neuroimaging, laboratory results, and clinical notes) by incorporating the latest evidence-based information and highlighting subtle patterns that might be overlooked in busy clinical settings.45–47”
Quote in the paperNot a substantial claim3 of 3 sources checked- cites [45] source
The Conclusion (page unavailable) discusses web-based cognitive assessments for screening and remote testing, not large language models assisting clinicians with clinical information. Searches for LLMs, clinical notes, evidence-based interpretation, and subtle patterns found no passage addressing the attributed claim. (LLM, conf=0.75)
Experimental — accuracy not yet measured - cites [46] source
The Discussion describes digital cognitive-assessment tools, EMR guidance, and RN support for clinicians under time pressure. Searches for large language models and for AI interpretation of neuroimaging, laboratory results, and clinical notes found no passage addressing the claimed LLM capabilities; the paper’s clinical-support workflow does not establish them. (LLM, conf=0.78)
Experimental — accuracy not yet measured - cites [47] source
The Introduction and Methods discuss a cognitive composite for detecting and monitoring decline in Alzheimer’s disease and related disorders. Searches for large language models, clinical decision support, and the claimed use of current evidence or subtle clinical patterns found no passage addressing those capabilities. Page numbers are unavailable in our copy. (LLM, conf=0.78)
Experimental — accuracy not yet measured
- Premisecitation
Further studies show promise for AI applications in optimizing cognitive test administration and scoring across diverse cultures and health literacy levels.
“Further studies show promise of AI applications in optimizing cognitive test administration and scoring across diverse cultures and health literacy levels.59,60”
Quote in the paperNot a substantial claim1 of 2 sources checked- cites [59] source
- cites [60] source
The Discussion describes an AI pipeline for classifying Parkinsonian motor severity from videos; it mentions cognitive scores only as possible inputs to improve diagnosis. The Methods say a neurologist conducted the MoCA assessment. Our copy does not address AI administration or scoring of cognitive tests across cultures or health-literacy levels (searched for cognitive testing, MoCA, culture, and health literacy). (LLM, conf=0.75)
Experimental — accuracy not yet measured
- Premisecitation
The dual capability of supporting clinicians and patients and families positions large language models as comprehensive tools that could address multiple challenges across dementia care.
“This dual capability—supporting both clinicians as well as patients and their families—positions LLMs as comprehensive tools that could address multiple challenges across the continuum of dementia care.48”
Quote in the paperNot a substantial claim1 of 1 sources checked- cites [48] source
The Discussion describes TabCAT-BHA as a digital cognitive assessment for primary care and mentions future AI-driven decision support. It does not address large language models or their proposed dual support for clinicians and patients and families across dementia care (searched for LLMs and AI support for patients and families). (LLM, conf=0.76)
Experimental — accuracy not yet measured
- Premisecitation
The Novoic app collects speech recordings from story recall tasks and analyzes them with a fully automated AI-based transcription and analysis pipeline to detect Alzheimer's disease.
“the Novoic app collects speech recordings from story recall tasks and analyzes them with a fully automated AI-based transcription and analysis pipeline for detection of Alzheimer’s disease,61”
Quote in the paperNot a substantial claim1 of 1 sources checked- cites [61] source
The Background and Method sections describe a PET imaging pipeline for processing scans, not a speech-analysis app. Searches for Novoic, story recall, speech transcription, and AI-based analysis found no passage addressing the attributed claim. The indexed copy provides no page number for these sections. (LLM, conf=0.78)
Experimental — accuracy not yet measured
- Premisecitation
Linus Health developed a brief digital cognitive battery that used AI models to identify subtypes of mild cognitive impairment.
“Linus Health also developed a brief digital cognitive battery that used AI models to identify subtypes of MCI.63”
Quote in the paperNot a substantial claim1 of 1 sources checked- cites [63] source
The Background section identifies this paper as an overview of ADNI MRI, and the Quantitative Analyses section discusses MRI-derived measures. Searches for Linus Health, a brief digital cognitive battery, AI models, and MCI subtypes found no passage addressing the attributed claim. (LLM, conf=0.76)
Experimental — accuracy not yet measured
- Premisecitation
Large language models can summarize information collected from the electronic health record and interview into a history of present illness or outline, helping clinicians quickly identify concerning patterns.
“Once valuable information is collected from the EHR and interview, LLMs can summarize it into a history of present illness or outline, allowing a clinician to quickly identify concerning patterns.40”
Quote in the paperNot a substantial claim1 of 1 sources checked- cites [40] source
The “Clinical record summarization” subsection (Method > Large Language Models for dementia; page not provided) says LLMs can summarize patient–doctor conversations and clinical notes, potentially saving clinicians time. It does not say they turn combined EHR and interview information into a history of present illness or outline, or that the result helps clinicians quickly identify concerning patterns. (LLM, conf=0.84)
Experimental — accuracy not yet measured
- Premisecitation
Conversational agents can perform worse with non-English speaking people.
“can perform worse with non-English speaking people,39”
Quote in the paperNot a substantial claim1 of 1 sources checked- cites [39] source
The Discussion (page unavailable) says the study did not evaluate multiturn dialogue and that its assessment of bias and equity was limited. Searches for non-English speakers, multilingual performance, and language barriers found no passage comparing conversational-agent performance by users’ language. (LLM, conf=0.74)
Experimental — accuracy not yet measured
- Premisecitation
Convolutional neural networks trained on structural MRI data have shown particular prowess in detecting subtle patterns of cortical thinning associated with various neurodegenerative conditions.
“CNNs trained on structural MRI data have shown particular prowess in detecting subtle patterns of cortical thinning associated with various neurodegenerative conditions.18”
Quote in the paperNot a substantial claim1 of 1 sources checked- cites [18] source
The Introduction discusses deep learning in neurology generally, while the Conclusions discuss LLM performance and possible future medical-imaging capabilities. Searches for CNNs, structural MRI, cortical thinning, and neurodegeneration found no passage supporting the specific claim in our copy. (LLM, conf=0.76)
Experimental — accuracy not yet measured
- Premisecitation
Various passive data types-physical activity, movement, driving, sleep, speech, eye tracking, facial expressions, technology use, writing, texting, and typing dynamics-have been shown to distinguish people with neurodegenerative disease from those without.
“For example, changes in physical activity and movement patterns, driving behavior, sleep patterns, everyday speech, eye tracking, facial expressions, technology use, writing, texting, and keyboard typing dynamics have all been shown to differentiate those with and without neurodegenerative disease.111–119”
Quote in the paperSubstantial claim6 of 9 sources checked- cites [111] source
The Methods (§2.4) describe EHR diagnoses, including sleep disorders, rather than passive measurements of sleep patterns or the other listed behaviors. The Results (§3) identify clinical predictors of ADRD, but do not report whether the listed passive data distinguish people with neurodegenerative disease from those without. Searches for the listed behaviors and passive digital measurements found no passage addressing that claim. (LLM, conf=0.76)
Experimental — accuracy not yet measured - cites [112] source
The Methods (Data collection and instruments used; page unavailable) describe clinician-administered functional, balance, and cognitive assessments, while the Discussion (Medical perspective; page unavailable) identifies functional scales as differentiating features. Neither passage addresses whether passively collected activity, driving, sleep, speech, eye-tracking, facial-expression, technology-use, or typing data distinguish people with and without neurodegenerative disease; searches for those terms and passive digital biomarkers found no such passage. (LLM, conf=0.72)
Experimental — accuracy not yet measured - cites [113] source
Split into 2 atoms — 0 confirmed, 0 partially supported, 0 disputed, 0 misattributed, 2 unverifiable. Expand the claim to see per-atom verdicts.
Experimental — accuracy not yet measured - cites [114] source
- cites [115] source
The Methods (Data; page unavailable) say that records from both Parkinson’s patients and healthy controls were used to train the model, but only Parkinson’s cases were subsequently analyzed for subtypes. The Results (Subtype characteristics; page unavailable) discuss clinical motor and sleep measures that characterize those subtypes, not passive data that distinguish people with neurodegenerative disease from those without. Searches for the listed passive data types and case–control discrimination found no passage establishing the attributed claim. (LLM, conf=0.76)
Experimental — accuracy not yet measured - cites [116] source
The Introduction and the section on constructing temporal trajectories describe clinical EHR data used to study progression from mild cognitive impairment to Alzheimer’s disease. Our copy does not report that the listed passive data types distinguish people with neurodegenerative disease from those without (searched for the listed data types and for passive-data differentiation). (LLM, conf=0.78)
Experimental — accuracy not yet measured - cites [117] source
- cites [118] source
The Results (Note Clustering; page not provided) mention movement and sleep symptoms in clinical notes, but compare subtypes among people with Alzheimer disease and related dementias—not people with and without neurodegenerative disease. Searches for driving, speech, eye tracking, technology use, and typing dynamics found no passage showing that these passive data types distinguish the two groups. (LLM, conf=0.76)
Experimental — accuracy not yet measured - cites [119] source
- Premisecitation
Recent machine learning work on real-world EHRs indicates that inexpensive, routinely collected data can identify patients at higher risk of future Alzheimer's disease.
“Recent work applying ML to real-world EHRs shows that inexpensive, routinely collected data can flag patients at elevated risk for future Alzheimer’s disease.”
Quote in the paperSubstantial claim1 of 2 sources checked- cites [123] source
The Materials and Methods (Data; page unavailable) describe multi-omics molecular data, and the Discussion (page unavailable) describes machine learning applied to those data. Searches for EHRs and inexpensive, routinely collected data found no passage showing that such data can flag patients at risk of future Alzheimer’s disease. (LLM, conf=0.75)
Experimental — accuracy not yet measured - cites [124] source
- Premisecitation
Increased availability of speech data via mobile, telehealth, and ambient voice technologies is speeding progress in this area.
“Growing access to speech data through mobile platforms, telehealth, and ambient voice technology is accelerating progress in this domain.105,106”
Quote in the paperNot a substantial claim2 of 2 sources checked- cites [105] source
The Introduction discusses growing computer and smartphone use, while the Methods and Discussion describe monitoring computer use and other activities in the home. Searches for speech data, telehealth, and ambient voice technology found no passage linking increased access to such data with faster progress. (LLM, conf=0.76)
Experimental — accuracy not yet measured - cites [106] source
Section 9 discusses machine-learning analysis of emotion in speech, while Section 13 describes mobile tools that expand access to cognitive screening. Neither passage says that speech data are becoming more available through mobile platforms, telehealth, or ambient voice technology, or that such availability is accelerating progress. Searches for speech, voice, and ambient technology found no passage establishing that connection. (LLM, conf=0.76)
Experimental — accuracy not yet measured
- Premisecitation
AI can help solve this, potentially identifying data-driven patterns of behavioral change linked to AD and other neurodegenerative diseases from passive digital data.
“AI represents a powerful part of the solution, with potential to identify data-driven patterns of behavioral change associated with AD and other neurodegenerative diseases through passive digital data.121”
Quote in the paperSubstantial claim1 of 1 sources checked- cites [121] source
The Highlights (pp. 2–3) say AI can detect undiscovered patterns in dementia genetics and omics data. Examples of best practice (p. 8) suggests combining those data with wearable or other streaming data, but does not say AI can identify patterns of behavioral change from passive digital data. That specific application is not supported by the cited paper. (LLM, conf=0.84)
Experimental — accuracy not yet measured
- Premisecitation
Speech is increasingly recognized as a promising digital biomarker for early detection and monitoring of Alzheimer's disease and related dementias.
“As such, it is increasingly recognized as a promising digital biomarker for the early detection and monitoring of ADRD.81,82”
Quote in the paperSubstantial claim2 of 2 sources checked- cites [81] source
The Discussion (§5.4; page unavailable) describes speech as a sensitive marker of very early neurodegeneration and proposes automated speech analysis for future prescreening. The study concerns frontotemporal degeneration, however, not ADRD broadly; it uses cross-sectional data and says longitudinal analysis is still in progress, so it does not establish monitoring. (LLM, conf=0.84)
Experimental — accuracy not yet measured - cites [82] source
The Abstract presents speech analysis as a promising, non-invasive way to detect Alzheimer’s disease early, and §1.1 describes language changes as relevant biomarkers. The review does not establish speech-based monitoring of disease progression; it focuses on detection and classification of Alzheimer’s disease rather than ADRD broadly. (LLM, conf=0.84)
Experimental — accuracy not yet measured
- Rests on an assumption: Speech provides rich cognitive and behavioral information by combining linguistic content and vocal motor features.
- Premisecitation
Combining blood-based -omics data with neuropathology and imaging in a multi-modal AI model has identified AD subtypes and co-pathology.
“different types of blood-based -omics data have been combined with neuropathology and imaging data in a multi-modal AI model to identify subtypes of AD and co-pathology.136,137”
Quote in the paperNot a substantial claim1 of 2 sources checked- cites [136] source
The ADNI dataset section describes an analysis of T1-weighted MRI scans from AD patients and healthy controls. Searches for blood-based omics, neuropathology, co-pathology, and AD subtypes found no passage describing the claimed multimodal model or its findings. (LLM, conf=0.76)
Experimental — accuracy not yet measured - cites [137]
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- Premisecitation
AI models applied to genomic data have produced polygenic risk scores that beat traditional linear-model risk scores.
“AI models have already been applied to genomic data to derive polygenic risk scores for different diseases that outperform traditional risk scores generated from linear models,135”
Quote in the paperNot a substantial claim1 of 1 sources checked- cites [135] source
The References list works on genomic machine learning, but no passage found in our copy says that AI-derived polygenic risk scores outperform scores from linear models. Searches for polygenic risk scores, genomic disease prediction, and linear-model comparisons did not locate that claim; the Results comparison with linear models concerns civil war prediction, not genomics. (LLM, conf=0.76)
Experimental — accuracy not yet measured
- Interpretationcitation
AD-BERT highlighted terms such as "memory," "MCI," and "difficulty recalling dates," suggesting it picks up linguistic signs of emerging cognitive decline.
“This model highlighted phrases such as “memory,” “MCI,” and “difficulty recalling dates,” indicating that it captures linguistic cues of emerging cognitive decline.”
Quote in the paperNot a substantial claim1 of 1 sources checked- cites [130] source
The target-discovery section (section 2; page unavailable) discusses natural-language processing in general, and the clinical-trial section (section 7; page unavailable) discusses classification using imaging, cognitive trajectories, and biomarkers. Searches for AD-BERT and “difficulty recalling dates” found no passage reporting the highlighted phrases or showing that this model captures linguistic cues of emerging cognitive decline. (LLM, conf=0.76)
Experimental — accuracy not yet measured
- Premisecitation
Unstructured data can also be used for subphenotyping, as shown by a recent study identifying three ADRD subtypes.
“Unstructured data can also be used for subphenotyping, as shown in the identification of three ADRD subtypes in a recent study.131”
Quote in the paperNot a substantial claim1 of 1 sources checked- cites [131] source
Our copy discusses AI in drug discovery, including candidate molecules and clinical trials (Abstract and section F; page numbers unavailable). Searches for ADRD subtypes, subphenotyping, and unstructured patient data found no passage addressing the claimed identification of three subtypes. (LLM, conf=0.75)
Experimental — accuracy not yet measured
- Premisecitation
Advanced NLP models like AD-BERT are starting to improve EHR-based AD prediction by mining free-text clinical notes.
“Advanced NLP models, such as AD-BERT, are beginning to boost EHR-based prediction of Alzheimer’s disease by mining the free-text of clinical notes.130”
Quote in the paperNot a substantial claim1 of 1 sources checked- cites [130] source
Section 2 discusses NLP for extracting information from published literature for target discovery, while section 9.3 discusses potential uses of electronic medical records in dementia trials. Neither addresses AD-BERT or improvement in Alzheimer’s prediction from free-text clinical notes. Searches for AD-BERT, BERT, and clinical-note prediction found no supporting passage in our copy; page numbers were unavailable. (LLM, conf=0.73)
Experimental — accuracy not yet measured
- Premisecitation
AD-BERT achieved an AUROC of 0.883 in an external cohort for predicting whether an MCI patient would ever convert to AD.
“AD-BERT had impressive accuracy for predicting whether an MCI patient would ever convert to AD (AUROC: 0.883 in an external cohort).”
Quote in the paperNot a substantial claim1 of 1 sources checked- cites [130] source
Section 7, “Clinical Trial Design and Machine Learning” (page unavailable), discusses machine-learning approaches to patient classification and trial enrichment, but does not report an AD-BERT result. Searches for AD-BERT, the 0.883 AUROC, and external-cohort prediction of MCI-to-AD conversion found no passage addressing the claim. (LLM, conf=0.75)
Experimental — accuracy not yet measured
- Premisecitation
Another study used functional scales (iADLs/ADLs) and common EHR labs to distinguish MCI (AUC 0.75) and dementia (AUC 0.96) from controls.
“Complementing these findings, another study incorporated functional scales (iADLs/ADLs) and common labs from the EHR to separate MCI (AUC: 0.75) and dementia (AUC: 0.96) from controls.125”
Quote in the paperNot a substantial claim1 of 1 sources checked- cites [125] source
The section ‘vERnet-B recognizes BRCA1 pathogenicity in an independent dataset’ reports AUCs for classifying BRCA1 variants, not people with MCI or dementia. Searches for iADLs/ADLs, EHR labs, MCI, and dementia found no passage addressing the claimed study or its AUCs of 0.75 and 0.96; no page numbers were provided in our copy. (LLM, conf=0.78)
Experimental — accuracy not yet measured
- Premisecitation
Traditional machine learning and deep learning approaches have successfully differentiated people with AD, MCI, FTD (including PPA), or PD from healthy controls.
“Both traditional ML approaches (e.g., support vector machines, logistic regression, XGBoost) and deep learning methods (e.g., convolutional and recurrent neural networks, transformers) have shown success in distinguishing individuals with AD, MCI, FTD, including PPA, or PD from healthy controls.93–96”
Quote in the paperNot a substantial claim3 of 4 sources checked- cites [93] source
The Video analysis section describes plans to develop and validate classification and deep-learning algorithms, not successful discrimination from healthy controls. The Speech analysis section discusses feature extraction, but reports no results for AD, MCI, FTD/PPA, or PD. Searches for the named model types and for disease-versus-control classification found no passage establishing the attributed claim. (LLM, conf=0.76)
Experimental — accuracy not yet measured - cites [94] source
The Statistical analysis section describes regression and group comparisons of smartphone cognitive-test scores, not machine-learning or deep-learning classification of diagnoses. The Cognitive test scores section reports performance differences by FTD disease severity, but does not address whether the named methods distinguish AD, MCI, FTD/PPA, or PD from healthy controls. Searches for machine-learning methods and diagnostic classification found no passage establishing the attributed claim. (LLM, conf=0.74)
Experimental — accuracy not yet measured - cites [95] source
- cites [96] source
In Opportunity 2, the paper discusses digital phenotyping for Parkinson’s disease, and its references include a study titled “High-accuracy detection of early Parkinson’s Disease.” Its discussion of machine learning does not establish that the listed traditional and deep-learning methods distinguished AD, MCI, FTD/PPA, or PD from healthy controls. No page numbers are available in our copy. (LLM, conf=0.75)
Experimental — accuracy not yet measured
- Premisecitation
Unsupervised AI applied to EHRs has characterized dementia heterogeneity by identifying distinct subtypes and progression patterns in AD and PD.
“Unsupervised AI approaches have also been applied to EHRs to characterize dementia heterogeneity, identifying distinct disease subtypes and progression patterns for patients with Alzheimer’s disease and Parkinson’s disease.126–129”
Quote in the paperNot a substantial claim3 of 4 sources checked- cites [126] source
The Results and Discussion describe classifying APP, PSEN1, and PSEN2 variants using genetic, bioinformatic, and cell-based evidence. Searches for unsupervised AI, EHRs, disease subtypes, and progression patterns found no passage addressing the attributed claim about Alzheimer’s and Parkinson’s disease. (LLM, conf=0.76)
Experimental — accuracy not yet measured - cites [127] source
The Methods (Screening of the main dementia genes) and Discussion describe genetic screening and characterization of frontotemporal dementia, not unsupervised AI applied to electronic health records. Searches for EHR-based clustering, Alzheimer’s or Parkinson’s subtypes, and progression patterns found no passage addressing the attributed claim; page numbers were not available in our copy. (LLM, conf=0.76)
Experimental — accuracy not yet measured - cites [128] source
The AI for Real-World Pharmacogenomics section discusses using EHR data for phenotype extraction in psychiatric pharmacogenomics, but does not address dementia subtypes or progression. Searches for dementia, Alzheimer’s and Parkinson’s disease, and unsupervised EHR clustering found no passage supporting the attributed finding. Page numbers were not available in our copy. (LLM, conf=0.76)
Experimental — accuracy not yet measured - cites [129] source
- Premisecitation
A barrier to clinical use of passive digital biomarkers has been insufficient computing resources to store and automatically process and analyze the large, continuously collected data.
“One barrier to the clinical implementation of passive digital biomarkers, however, has been the lack of computing resources for storing and automatically processing and analyzing the large amount of data collected from passive digital tools that continuously monitor behavior over long periods of time.120”
Quote in the paperNot a substantial claim1 of 1 sources checked- cites [120] source
The Study Design and Strengths and Limitations sections describe a retrospective Parkinson disease cohort and limitations in recording clinical symptoms. A search of our copy for passive digital biomarkers and computing, storage, and automated-processing barriers found no passage addressing the claimed barrier to clinical implementation. The copy is not marked as truncated. (LLM, conf=0.76)
Experimental — accuracy not yet measured
- Premisecitation
Even in dementia syndromes where language symptoms are not dominant, such as behavioral variant frontotemporal dementia, automated speech analysis can detect subtle linguistic and acoustic patterns reflecting underlying cognitive, behavioral, or socio-emotional changes.
“Even in dementia syndromes where language symptoms are not dominant, such as the behavioral variant of frontotemporal dementia, automated speech analysis can detect subtle linguistic and acoustic patterns that reflect underlying cognitive, behavioral, or socio-emotional changes.91,92”
Quote in the paperNot a substantial claim2 of 2 sources checked- cites [91] source
The Introduction says that linguistic and acoustic speech features have been used to classify dementia, and the Results report acoustic-feature classification of Alzheimer’s disease. The paper does not establish that these methods detect patterns in behavioral-variant frontotemporal dementia or that the patterns reflect behavioral or socio-emotional changes. (LLM, conf=0.86)
Experimental — accuracy not yet measured - cites [92] source
Split into 2 atoms — 0 confirmed, 2 partially supported, 0 disputed, 0 misattributed, 0 unverifiable. Expand the claim to see per-atom verdicts.
Experimental — accuracy not yet measured
- Premisecitation
New methods have demonstrated strong feasibility for unobtrusively collecting objective, naturalistic behavioral data from older adults using worn or embedded sensors.
“Emerging methods have shown excellent feasibility to unobtrusively capture objective and naturalistic behavioral data in older adults through the use of sensors that are either worn or embedded within objects that people use regularly, such as smartwatches, smartphones, and other placeable sensors (e.g., a GPS sensor placed in one’s car).109”
Quote in the paperSubstantial claim1 of 1 sources checked- cites [109] source
The Results section (page not available) describes wearable monitors, embedded home sensors, and GPS loggers collecting behavioral data in everyday settings. The Discussion (page not available) describes passive monitoring but also notes barriers to older adults’ acceptance. The review does not establish excellent feasibility across these methods for older adults generally. (LLM, conf=0.84)
Experimental — accuracy not yet measured
- Premisecitation
Many AI techniques need sophisticated computational infrastructure that may not be available in typical clinical settings.
“Many AI techniques require sophisticated computational infrastructure that may not be readily available in typical clinical settings.16”
Quote in the paperNot a substantial claim1 of 1 sources checked- cites [16] source
Section 3.2 reports computation times for individual dementia-diagnosis algorithms, but does not say they require sophisticated infrastructure or that such infrastructure is unavailable in typical clinics. The Introduction and section 5.1 discuss barriers to clinical implementation involving generalizability, training data, and algorithm availability instead. Searches for hardware, computing resources, and infrastructure found no passage addressing the attributed claim; page numbers were not available in our copy. (LLM, conf=0.72)
Experimental — accuracy not yet measured
- Interpretationnone given
AI has not yet been integrated into routine clinical care for Alzheimer's disease and related dementias.
“While AI has yet to be integrated into routine ADRD clinical care, recent developments suggest this may soon change.”
Quote in the paperSubstantial claim - Interpretationnone given
Traditional AI methods, such as machine learning models, have made significant progress in classifying disease states.
“Traditional AI methods (i.e., ML models) have made significant progress with the classification of disease states.”
Quote in the paperNot a substantial claim - Interpretationreasoning
These sophisticated models demonstrate remarkable precision in identifying disease-specific features, such as brain atrophy patterns characteristic of Alzheimer's disease.
“These sophisticated models demonstrate remarkable precision in identifying disease-specific features, such as patterns of brain atrophy characteristic of Alzheimer’s disease.”
Quote in the paperNot a substantial claim - Premisecitation
Recent studies have shown that short videos captured on consumer-grade technology can detect Parkinsonism.
“Recent studies have shown that short videos captured on consumer-grade technology can detect Parkinsonism.69,70”
Quote in the paperNot a substantial claim0 of 2 sources checked - Premisecitation
Support vector machines have shown particular efficacy in identifying imaging biomarkers associated with frontotemporal dementia.
“For example, support vector machines have demonstrated particular efficacy in identifying imaging biomarkers associated with frontotemporal dementia.75”
Quote in the paperNot a substantial claim0 of 1 sources checked- cites [75] source
- Interpretationcitation
Remote, self-administered cognitive tasks delivered on participants' own devices are proving practical and psychometrically sound in Alzheimer's disease research.
“Remote, self-administered cognitive tasks delivered on participants’ own devices are also proving both practical and psychometrically sound in AD research.”
Quote in the paperSubstantial claim - Interpretationreasoning
Caregivers can receive personalized guidance on managing behavioral symptoms or navigating healthcare systems.
“while caregivers can receive personalized guidance on managing behavioral symptoms or navigating healthcare systems.”
Quote in the paperNot a substantial claim - Interpretationreasoning
Digital cognitive assessments can enable AI integration because they capture potentially meaningful performance nuances, such as trial-level data and response times, that traditional paper-based tests cannot record or quickly interpret.
“Digital cognitive assessments can enable AI integration as they can capture potentially meaningful nuances in performance (e.g., trial-level data, response times) that are not possible to record or quickly interpret with traditional paper-based cognitive tests.”
Quote in the paperSubstantial claim - Interpretationreasoning
Ensuring that generated summaries are consistently comprehensive and accurate will be challenging, especially because hallucinations remain a concern with large language models.
“It will be challenging to ensure that generated summaries are consistently comprehensive and accurate, particularly as hallucinations (fabricated information) are still a concern with LLMs.”
Quote in the paperNot a substantial claim - Interpretationreasoning
These standardization efforts have greatly aided in advancing reliable AI imaging research.
“These standardization efforts have greatly aided in advancing reliable AI imaging research.”
Quote in the paperNot a substantial claim- Rests on an assumption: Consistent imaging protocols are necessary for AI model accuracy because systematic variation can substantially affect performance.
- Interpretationreasoning
Data quality variations may introduce confounding covariates and technical biases that can inflate reported AUROCs.
“High variations in data quality may also introduce confounding covariates (e.g., age, detection bias, indication bias) and technical biases (e.g., extent of missingness, imputation approaches, ontology mappings) that can also contribute to overly optimistic AUROCs reported.”
Quote in the paperNot a substantial claim - Interpretationreasoning
Speech is among the most clinically viable digital biomarkers because it is scalable, ecologically valid, easy to collect longitudinally, and can potentially be captured passively in real-world settings.
“Although most studies to date are cross-sectional and research-focused—and important ethical challenges must be addressed prior to clinical implementation107—speech remains one of the most clinically viable digital biomarkers given its scalability, ecological validity, ease of longitudinal collection, and potential for passive capture in real-world settings.”
Quote in the paperSubstantial claim- Rests on an assumption: Most speech studies to date are cross-sectional and research-focused.
- Premisecitation
AI approaches are especially suited to high-dimensional -omics data for clinical use because they can uncover nonlinear relationships among biological features that standard linear or logistic analyses struggle to detect.
“AI-based approaches are uniquely suited to leveraging such high-dimensional -omics data for potential clinical use, given their ability to reveal nonlinear relationships across different biological features that are difficult to observe by standard linear or logistic association analyses.134”
Quote in the paperSubstantial claim0 of 1 sources checked- cites [134] source
- Interpretationreasoning
Automated speech analysis can capture deviations in speech and language specific to dementia syndromes.
“Automated speech analysis can capture deviations in speech and language specific to dementia syndromes.”
Quote in the paperSubstantial claim - Interpretationnone given
The field has limited external validation, with most studies done in single institutions or highly selected research cohorts.
“The field suffers from limited external validation, with most studies conducted in single institutions or highly selected research cohorts.”
Quote in the paperNot a substantial claim - Premisecitation
Use of AI in dementia detection and diagnosis among marginalized populations may be hindered by mistrust and privacy concerns, even when tools are designed with equity and fairness in mind.
“Utilization of AI technologies in dementia detection and diagnosis among marginalized populations may also be hampered by mistrust and concerns about privacy as well as other factors, even when these tools are designed with an emphasis on equity and fairness.160”
Quote in the paperNot a substantial claim0 of 1 sources checked- cites [160]
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- Interpretationreasoning
AI could worsen existing brain health inequities if used incorrectly.
“AI has the potential to exacerbate existing brain health inequities if leveraged incorrectly.”
Quote in the paperNot a substantial claim - Premisecitation
This can cause a systematic lack of data from medical records used to train AI, reducing performance for these populations.
“This can lead to a systematic lack of data from medical records used to train AI, degrading performance for these populations.161”
Quote in the paperSubstantial claim0 of 1 sources checked- cites [161]
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- Interpretationreasoning
Implementation of these emerging digital biomarkers may take longer because they remain largely research-stage and need extensive validation before clinical deployment.
“However, implementation may have a longer time horizon as these innovative approaches remain largely research-stage and require extensive validation before clinical deployment.”
Quote in the paperSubstantial claim - Premisecitation
Prior work has shown that inadequate representation of certain ethnoracial groups can lead to unfairness in ML models predicting progression along the AD continuum for ADNI participants.
“Prior work has demonstrated lack of adequate representation of certain ethnoracial groups can lead to unfairness (fairness being the absence of favor or prejudice toward a group based on group characteristics) among ML models predicting progression along the AD continuum for ADNI participants.162”
Quote in the paperNot a substantial claim0 of 1 sources checked- cites [162]
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- Interpretationnone given
Implicit biases in healthcare and research settings can become encoded in algorithms.
“Implicit biases present in healthcare and research settings, from subjective decision-making to stigmatizing language and stereotypes, can be encoded into algorithms.”
Quote in the paperSubstantial claim
- May combine two claims. 54
- Classified as probably not a substantial claim. 20
May combine two claims.
PremisecitationSome studies have used regression models to forecast continuous cognitive scores like MMSE or to examine links with imaging and biomarker measures.
“While classification tasks are common in automated speech analysis in neurodegenerative research, several studies have employed regression models to predict continuous cognitive outcomes (e.g., MMSE)97,98 or to explore relationships with imaging and biomarker measures.99–101”
Quote in the paperNot a substantial claim5 of 5 sources checked- cites [97] source
The Results (page unavailable) report a study linking MRI markers with mobile assessments of semantic-memory performance, supporting the imaging-relationship portion of the claim. The cited paper does not establish that this relationship was examined using regression models or automated speech analysis. The MMSE-prediction portion of the citing sentence is attached to different citations (97–98). (LLM, conf=0.82)
Experimental — accuracy not yet measured - cites [98] source
The Speech and language section reports speech-based classification and a correlation between pauses and episodic memory, but does not describe regression predicting MMSE or relating speech measures to imaging or biomarkers. The Future work section proposes imaging and genomic measures for validation; it does not report studies that used regression models to examine those relationships. Our copy provides no page numbers for these passages. (LLM, conf=0.72)
Experimental — accuracy not yet measured - cites [99] source
The Results describe a PD-versus-control classifier and correlations between typing features and UPDRS motor-symptom scores, not regression predicting cognitive scores or examining imaging or biomarker measures. The Discussion explicitly presents regression to estimate motor-symptom severity as a future possibility, rather than something this study did (page numbers unavailable in our copy). (LLM, conf=0.85)
Checked by a reasoning model — may be wrong.
Experimental — accuracy not yet measured - cites [100] source
Methods > Statistical analysis (page unavailable) says the study used linear mixed regression models to examine longitudinal cognitive change, and Methods > Longitudinal cognitive data lists the cognitive scores analyzed. The source uses accelerometer activity data, not automated speech analysis; it does not identify MMSE among its outcomes or report relationships with imaging measures. The citing sentence therefore fits the regression-on-cognitive-scores portion but not those specific examples or context. (LLM, conf=0.85)
Experimental — accuracy not yet measured - cites [101] source
The Methods (page unavailable) describe using GPS driving features to predict preclinical Alzheimer disease defined by a CSF biomarker, so the paper does examine a relationship with a biomarker measure. But it uses a random-forest classifier with a binary outcome, not a regression model, and it does not analyze speech or predict continuous cognitive scores such as MMSE. (LLM, conf=0.83)
Experimental — accuracy not yet measured
May combine two claims.
PremisecitationA recent Mayo Clinic study highlighted an AI model that can detect nine different neurodegenerative phenotypes from FDG-PET images, with initial validation from ADNI yielding an AUC of 0.93.
“A recent example is work from the Mayo Clinic highlighting an AI model that can detect nine different neurodegenerative phenotypes from FDG-PET images, with initial validation from ADNI with AUC of 0.93.80”
Quote in the paperNot a substantial claim1 of 1 sources checked- cites [80] source
The Methods (Participants; page not provided) identify the study cohort as coming from the UCSF Memory and Aging Center, not the Mayo Clinic. The Results (Automated Speech and Language Analysis; page not provided) describe a speech-feature model distinguishing frontotemporal dementia groups, not an FDG-PET model detecting nine phenotypes. The Discussion says external validation remains future work; it does not report ADNI validation with an AUC of 0.93. (LLM, conf=0.88)
Checked by a reasoning model — may be wrong.
Experimental — accuracy not yet measured
May combine two claims.
PremisecitationAI capabilities have advanced at unprecedented rates over the past 2 to 3 years, alongside a sharp increase in FDA-cleared or approved AI-based medical devices.
“AI capabilities have advanced at unprecedented rates over the past 2 to 3 years, coinciding with a sharp increase in FDA-cleared or approved AI-based medical devices.24,25”
Quote in the paperNot a substantial claim2 of 2 sources checked- cites [24] source
The Introduction and Discussion passages (page numbers unavailable) address telemedicine adoption during COVID-19, not advances in AI capabilities or FDA-cleared or approved AI-based devices. Searches for the device-approval phrase and related AI/FDA terms found no passage addressing the claim. (LLM, conf=0.76)
Experimental — accuracy not yet measured - cites [25] source
The Methods and Discussion passages (pages not provided) concern mail, e-mail, and telephone follow-up in hand surgery research. Searches for recent AI advances and FDA-cleared or approved AI medical devices found no passage addressing either part of the claim. (LLM, conf=0.78)
Experimental — accuracy not yet measured
May combine two claims.
PremisecitationMost primary care providers report lacking time and confidence for the diagnostic process and prefer referring patients to specialists.
“most PCPs cite a lack of time and confidence for the diagnostic process,7,8 preferring to refer patients to specialists.”
Quote in the paperSubstantial claim1 of 2 sources checked- cites [7] source
- cites [8] source
The Results report low confidence in several diagnostic tasks and that 54% of surveyed PCPs referred at least 75% of patients with suspected neurocognitive disorders to specialists for a full workup. The Discussion mentions time as a commonly reported systems-level issue, but this study does not establish that most respondents cited lack of time as a reason for referral. (LLM, conf=0.84)
Experimental — accuracy not yet measured
Probably not a substantial claim: may not pass S3 (load-bearing).
PremisecitationThe number of people with Alzheimer's disease and related dementias in the U.S. is rapidly increasing.
“As the U.S. population ages, the number of people with Alzheimer’s disease and related dementias (ADRD) is rapidly increasing,1,2”
Quote in the paperNot a substantial claim2 of 2 sources checked- cites [1] source
The Abstract projects that the number of people in the United States with AD dementia will rise from 4.7 million in 2010 to 13.8 million in 2050. The Results link this growth to an aging population. The paper estimates AD dementia, however, not the broader category of Alzheimer’s disease and related dementias (ADRD). (LLM, conf=0.86)
Experimental — accuracy not yet measured - cites [2] source
The Results (page unavailable) report that global prevalence of Alzheimer’s disease and other dementias rose 117% from 1990 to 2016. They also report a 39% increase in U.S. DALYs from all neurological disorders combined, but do not give a U.S.-specific increase in the number of people with dementia. The Discussion (page unavailable) links rising numbers of non-communicable neurological disorders to population ageing and growth; it does not establish the claim specifically for U.S. dementia cases. (LLM, conf=0.83)
Experimental — accuracy not yet measured
Probably not a substantial claim: may not pass S3 (load-bearing).
PremisecitationGPT-4 surpassed average human performance on a neurology board-style examination.
“GPT-4 exceeded mean human performance on a neurology board-style examination20”
Quote in the paperNot a substantial claim1 of 1 sources checked- cites [20] source
The Results report that GPT-4 scored significantly higher than attending physicians and residents on the R-IDEA clinical-reasoning measure. But the Methods describe simulated clinical cases involving internal medicine physicians, not a neurology board-style examination; the claimed examination setting is unsupported. Page numbers are unavailable in our copy. (LLM, conf=0.88)
Experimental — accuracy not yet measured
Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.
PremisecitationU.S. dementia cases are projected to double by 2060, reaching 1 million annually.
“U.S. cases projected to double by 2060 to 1 million annually.3”
Quote in the paperNot a substantial claim1 of 1 sources checked- cites [3] source
May combine two claims.
PremisecitationLarge language models can enable conversational agents to interview patients efficiently and empathetically, adjust explanations to a patient's education level, and speak in most patients' preferred languages.
“LLMs can enable conversational agents to interview patients efficiently and empathetically,26 adjust explanations to a patient’s education level, and speak in most patients’ preferred languages.27”
Quote in the paperNot a substantial claim1 of 2 sources checked- cites [26] source
- cites [27] source
The Discussion (page unavailable) reports that families needing translators faced delays and that few non-Latino physicians spoke Spanish or had onsite translators. It does not address whether LLM agents can speak patients’ preferred languages or tailor explanations to education level. Searches for those capabilities found no supporting passage. (LLM, conf=0.77)
Experimental — accuracy not yet measured
May combine two claims.
PremisecitationDemand for dementia specialists far exceeds supply, and the gap is widening.
“the demand for dementia specialists greatly exceeds the available supply, with the gap continuing to widen.9–11”
Quote in the paperSubstantial claim1 of 3 sources checked- cites [9] source
- cites [10] source
- cites [11] source
The Base Case section identifies dementia specialists as the main capacity constraint and projects the number of patients waiting to rise from 0.8 million in 2025 to about 5.4 million in 2032. But it also says queues and wait times subsequently decline, so the paper does not support an unqualified claim that the gap continues to widen. Page numbers are unavailable in our copy. (LLM, conf=0.88)
Experimental — accuracy not yet measured
Probably not a substantial claim: may not pass S3 (load-bearing).
PremisecitationAfter age 55, lifetime dementia risk is 48% for women and 35% for men.
“with lifetime dementia risk reaching 48% for women and 35% for men after age 55”
Quote in the paperNot a substantial claim1 of 1 sources checked- cites [3] source
May combine two claims.
PremisecitationVoice-based conversational agents could interact with patients by phone, overcoming writing and typing barriers of traditional questionnaires and increasing access for patients with lower socioeconomic status, limited English proficiency, or advanced age.
“More recent advancements with voice-based conversational agents28 could interact with patients over the phone to overcome barriers with the writing and typing typically required of traditional questionnaires, and also increase access for patients who are of lower socioeconomic status, limited English proficiency, or elderly.29–32”
Quote in the paperNot a substantial claim3 of 5 sources checked- cites [28] source
Split into 2 atoms — 0 confirmed, 2 partially supported, 0 disputed, 0 misattributed, 0 unverifiable. Expand the claim to see per-atom verdicts.
Experimental — accuracy not yet measured - cites [29]
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- cites [30] source
Split into 2 atoms — 0 confirmed, 2 partially supported, 0 disputed, 0 misattributed, 0 unverifiable. Expand the claim to see per-atom verdicts.
Experimental — accuracy not yet measured - cites [31] source
- cites [32] source
Split into 2 atoms — 0 confirmed, 2 partially supported, 0 disputed, 0 misattributed, 0 unverifiable. Expand the claim to see per-atom verdicts.
Experimental — accuracy not yet measured
May combine two claims.
PremisecitationCurrent cognitive assessment challenges include cost, clinician time, required training even for common tests, and language and cultural biases that complicate interpretation in medically under-served populations.
“Current challenges include cost, time required for clinicians to administer tests, specific training even for the most common assessments (e.g., MMSE, MoCA), and language and cultural biases that complicate interpretation in medically under-served populations.6,32,51”
Quote in the paperSubstantial claim3 of 3 sources checked- cites [6] source
The section on ethnoracial factors in disease detection says linguistic and cultural factors can impair interpretation of cognitive tests, discusses cost-effectiveness, and notes that neuropsychologists need training to work with minority groups. It does not identify clinician administration time as a challenge or say that administering common tests such as the MMSE or MoCA requires specific training. Page numbers are unavailable in our copy. (LLM, conf=0.87)
Experimental — accuracy not yet measured - cites [32] source
The Introduction and Communication Inefficiency sections discuss provider workload, delays, and the need to explain health information to older adults. They do not address the cost, administration time, training requirements, or language and cultural biases of cognitive assessments. Searches for MMSE, MoCA, and cognitive-assessment barriers found no relevant passage in our copy. (LLM, conf=0.75)
Experimental — accuracy not yet measured - cites [51] source
The Introduction (page unavailable) describes common brief tests, including MoCA and MMSE, as labor-intensive and requiring trained professionals; it also says literacy and language demands may limit their usefulness in underserved communities. The Methods > Measures section gives MoCA’s administration time as about 15–20 minutes, and the Discussion > Implications of findings describes potential staff-cost savings from the new test. The paper does not clearly establish cost as a barrier for existing tests or say that language and cultural biases complicate interpretation in medically underserved populations specifically. (LLM, conf=0.84)
Experimental — accuracy not yet measured
May combine two claims.
PremisecitationInitiatives such as the Alzheimer's Disease Imaging Initiative have established best practices for acquisition and preprocessing, and large-scale studies such as the UK Biobank have advanced standardization for diverse MRI techniques.
“Initiatives like the Alzheimer’s disease imaging initiative (ADNI) have established best practices for acquisition and preprocessing, while large-scale studies such as the UK Biobank have advanced standardization for diverse MRI techniques.71–74”
Quote in the paperNot a substantial claim4 of 4 sources checked- cites [71] source
The Results section’s “Brain Imaging and Picture Description Tasks” passage discusses cerebral blood flow during a picture-description task, not MRI acquisition or preprocessing standards. Searches for ADNI, UK Biobank, and MRI standardization found no passage in our copy addressing either attributed claim. (LLM, conf=0.76)
Experimental — accuracy not yet measured - cites [72] source
The Discussion’s ‘Other modalities’ section (page unavailable) mentions MRI only as a data type used in some reviewed studies; its ‘Pre-processing’ section discusses speech and text data. Searches for ADNI, UK Biobank, and MRI acquisition and preprocessing found no passage addressing either initiative’s role in MRI standardization. (LLM, conf=0.76)
Experimental — accuracy not yet measured - cites [73] source
The Introduction (page unavailable) describes an analysis of linguistic data from the Framingham Heart Study, and the Discussion mentions brain imaging only as a biomarker used to predict disease progression. Searches for ADNI, UK Biobank, and MRI acquisition and preprocessing standards found no passage addressing the attributed claim. (LLM, conf=0.76)
Experimental — accuracy not yet measured - cites [74] source
The Review protocol and Conclusion sections describe a systematic review of connected speech elicited by picture description, not MRI acquisition or preprocessing. Searches of our copy for ADNI and UK Biobank MRI standardization found no passage addressing either claim. (LLM, conf=0.76)
Experimental — accuracy not yet measured
May combine two claims.
PremisecitationThe Altoida app delivers augmented reality tasks and analyzes functional motor behaviors collected via accelerometer, gyroscope, and touch screen interactions to detect mild cognitive impairment with AI models.
“the Altoida app delivers augmented reality tasks and analyzes functional motor behaviors collected via accelerometer, gyroscope, and touch screen interactions to detect MCI with AI models.62”
Quote in the paperNot a substantial claim1 of 1 sources checked- cites [62] source
The ADNI PET Core at 20 discusses PET imaging and its relationship to cognitive decline (ADNI-1 and ADNI-GO/2/3 sections; page unavailable). Searches for Altoida, augmented-reality tasks, device sensors, and AI-based MCI detection found no passage addressing the app or its methods. (LLM, conf=0.76)
Experimental — accuracy not yet measured
May combine two claims.
PremisecitationPromising emerging solutions include AI confidence scores and secondary AI models that critically evaluate summaries.
“Emerging promising solutions include the adoption of AI confidence scores and secondary AI models that critically evaluate the summaries.41–43”
Quote in the paperNot a substantial claim3 of 3 sources checked- cites [41] source
The Discussion (page unavailable) recommends more interpretable AI models and describes ways the virtual companion limits exposure to inaccuracies. It does not address AI confidence scores or secondary AI models that critically evaluate summaries; searches for those terms and paraphrases found no supporting passage. (LLM, conf=0.75)
Experimental — accuracy not yet measured - cites [42] source
The sections on designing AI AgeTech and the Conclusion discuss care partnerships and a humanistic design model, not confidence scores or secondary AI models evaluating summaries. Searches for those terms and for AI verification of summaries found no relevant passage in our copy. Page numbers were not available. (LLM, conf=0.76)
Experimental — accuracy not yet measured - cites [43] source
The Results section on beliefs about practice improvement discusses decision-support tools for dementia assessment and care, not AI-generated summaries. Searches for AI confidence scores and secondary models evaluating summaries found no passage addressing either proposed solution in our copy. (LLM, conf=0.76)
Experimental — accuracy not yet measured
May combine two claims.
PremisecitationAI transcription and summarization are already used during visits, as healthcare institutions widely adopt ambient AI scribes.
“AI transcription and summarization are already being used during the visit as healthcare institutions are widely adopting ambient AI scribes.44”
Quote in the paperNot a substantial claim1 of 1 sources checked- cites [44] source
The Introduction discusses digital tools for cognitive assessment, and the Discussion describes AI research trends in that field. Neither passage addresses ambient AI scribes, transcription or summarization during visits, or their adoption by healthcare institutions (searched those terms and clinical documentation). (LLM, conf=0.76)
Experimental — accuracy not yet measured
Probably not a substantial claim: may not pass S3 (load-bearing).
PremisecitationThe landscape of digital cognitive assessments continues to evolve, with emerging tools increasingly incorporating AI to enhance administration, scoring, and feedback of results.
“The landscape of digital cognitive assessments continues to evolve, with emerging tools increasingly incorporating AI to enhance the administration, scoring, and feedback of results.67”
Quote in the paperNot a substantial claim1 of 1 sources checked- cites [67] source
The Results section, ‘AI-augmented clinician assessments’ (page unavailable), reports that the model improved clinicians’ diagnostic assessments using AI-assisted confidence scores. The Discussion (page unavailable) describes probability scores generated from inputs that can include cognitive test results. These passages support AI-assisted interpretation, but do not establish that digital cognitive assessment tools increasingly use AI to administer or score tests or provide feedback on results. (LLM, conf=0.83)
Experimental — accuracy not yet measured
May combine two claims.
PremisecitationA recent scoping review found encouraging adherence rates among research participants and moderate correlations with gold-standard neuropsychological batteries, with r approximately 0.53 to 0.70.
“A recent scoping review58 showed encouraging adherence rates among research participants and moderate correlations with gold-standard neuropsychological batteries (r = approximately 0.53–0.70).”
Quote in the paperNot a substantial claim1 of 1 sources checked- cites [58] source
The Results section on remotely administered assessments reports promising adherence in short studies, including 80% completing all daily tasks in one study, while cautioning that attrition is a problem in longer fully remote studies. The paper also discusses validation against conventional cognitive measures, but the passages in our copy do not establish the claimed correlation range of approximately r = 0.53–0.70. (LLM, conf=0.84)
Experimental — accuracy not yet measured
May combine two claims.
PremisecitationIncorporating large language models into virtual reality has enabled engaging virtual companions for persons living with dementia, making virtual psychometrists particularly promising.
“Incorporation of LLMs in virtual reality has enabled engaging virtual companions for persons living with dementia,49 which makes virtual psychometrists particularly promising.”
Quote in the paperNot a substantial claim1 of 1 sources checked- cites [49] source
The Results section describes an augmented-reality cognitive task and speech-based assessments, but does not address LLM-powered virtual companions for people living with dementia. Searches for large language models, virtual companions, and virtual psychometrists found no passage supporting the claim. The cited paper’s copy is not truncated; the claim is simply not addressed there. (LLM, conf=0.72)
Experimental — accuracy not yet measured
May combine two claims.
PremisecitationSeveral validated digital cognitive assessment tools, with and without AI integration, are now available for clinic or remote home use, though their levels of validation vary.
“More broadly, several validated digital cognitive assessment tools with and without AI integration are now available for clinic or remotely at home, though with varying levels of validation.52,53”
Quote in the paperNot a substantial claim1 of 2 sources checked- cites [52] source
The Methods (Study assessments; page not specified) describe a remotely administered, smartphone-based cognitive test, and the Discussion reports its AI-enhanced screening performance. This supports the paper as an example of a studied remote digital tool, but it does not establish that several digital tools with and without AI are available for clinic or home use, or compare their levels of validation. (LLM, conf=0.84)
Experimental — accuracy not yet measured - cites [53]
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
May combine two claims.
PremisecitationSubstantial work has developed AI models to aid neurological diagnoses from multi-sequence brain MRI or PET, and more recent efforts involve multimodal AI models incorporating clinical, laboratory, and imaging data to differentiate dementia etiologies.
“Substantial work has been done to develop AI models that can aid in neurological diagnoses based on multi-sequence brain MRI or PET, and more recent efforts involve the development of multimodal AI models that incorporate clinical, laboratory, and imaging data for differentiating etiologies of dementia.77–79”
Quote in the paperNot a substantial claim2 of 3 sources checked- cites [77] source
Split into 2 atoms — 0 confirmed, 0 partially supported, 0 disputed, 0 misattributed, 2 unverifiable. Expand the claim to see per-atom verdicts.
Experimental — accuracy not yet measured - cites [78]
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- cites [79] source
Split into 2 atoms — 0 confirmed, 0 partially supported, 0 disputed, 0 misattributed, 2 unverifiable. Expand the claim to see per-atom verdicts.
Experimental — accuracy not yet measured
Probably not a substantial claim: may not pass S3 (load-bearing).
PremisecitationVirtual Reality is a particularly promising emerging tool that can provide an immersive and engaging experience with almost limitless possibilities for testing cognition.
“A particularly promising emerging tool is Virtual Reality, which can be an immersive and engaging experience with almost limitless possibilities for testing cognition.65”
Quote in the paperNot a substantial claim1 of 1 sources checked- cites [65] source
The Introduction and Discussion address machine-learning classification of MRI scans for dementia diagnosis, not virtual reality as a tool for testing cognition. Searches for virtual reality, immersive experiences, and cognitive testing found no passage addressing the attributed claim; the paper's copy is not truncated. (LLM, conf=0.76)
Experimental — accuracy not yet measured
May combine two claims.
PremisecitationThe conversational abilities of large language models create opportunities for companionship, helping patients understand their diagnosis and treatment options, and guiding caregivers through complex care decisions.
“The conversational abilities of these models create opportunities for providing companionship, helping patients understand their diagnosis and treatment options, and guiding caregivers through complex care decisions.48”
Quote in the paperSubstantial claim1 of 1 sources checked- cites [48] source
The Discussion mentions future AI-driven decision support using electronic health record data, but not conversational large language models. The Methods describe diagnosis education delivered by a nurse or social worker, not by a model. Searches for companionship, conversational abilities, patient explanations, and caregiver decision guidance found no passage addressing the claim. (LLM, conf=0.76)
Experimental — accuracy not yet measured
May combine two claims.
PremisecitationThe TabCAT-Brain Health Assessment is an iPad-based in-clinic assessment with immediate scoring that outperforms traditional paper-based assessments, has been adopted organically across several UCSF-affiliated primary care clinics, and is free for primary care use.
“the TabCAT-Brain Health Assessment developed at UCSF is an iPad-based in-clinic assessment that includes immediate scoring, outperforms traditional paper-based assessments, has had organic adoption across several UCSF-affiliated PCP clinics, and is free for primary care use.54–57”
Quote in the paperNot a substantial claim3 of 4 sources checked- cites [54] source
Split into 2 atoms — 0 confirmed, 0 partially supported, 0 disputed, 0 misattributed, 2 unverifiable. Expand the claim to see per-atom verdicts.
Experimental — accuracy not yet measured - cites [55] source
- cites [56] source
Split into 2 atoms — 0 confirmed, 0 partially supported, 0 disputed, 0 misattributed, 2 unverifiable. Expand the claim to see per-atom verdicts.
Experimental — accuracy not yet measured - cites [57] source
Split into 2 atoms — 0 confirmed, 0 partially supported, 0 disputed, 0 misattributed, 2 unverifiable. Expand the claim to see per-atom verdicts.
Experimental — accuracy not yet measured
May combine two claims.
PremisecitationIn Alzheimer's disease, early language changes include word-finding difficulties, reduced lexical diversity, and vague or empty speech.
“In AD, early language changes include word-finding difficulties, reduced lexical diversity, and vague or empty speech.83–85”
Quote in the paperNot a substantial claim1 of 3 sources checked- cites [83]
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- cites [84] source
- cites [85] source
The Background describes loss of discourse complexity and connected speech as symptoms of cognitive decline associated with AD. The Methods measure lexical variety and poverty of content, but the cited passages do not establish reduced lexical diversity or word-finding difficulties as early changes in AD; nor do they specifically describe vague or empty speech as an early change. No page numbers are available in our copy. (LLM, conf=0.82)
Experimental — accuracy not yet measured
May combine two claims.
PremisecitationAcoustic features in Alzheimer's disease may include slower articulation rate, increased pauses, and reduced prosody, though these are typically more subtle than in disorders with prominent motor speech involvement such as Parkinson's disease or nonfluent variant primary progressive aphasia, which is additionally marked by agrammatism.
“Acoustic features in AD may include slower articulation rate, increased pauses, and reduced prosody,86 though these are typically more subtle than in disorders with prominent motor speech involvement, such as Parkinson’s disease87 or the nonfluent variant of primary progressive aphasia (PPA), which is additionally marked by agrammatism.88”
Quote in the paperNot a substantial claim3 of 3 sources checked- cites [86] source
Split into 2 atoms — 0 confirmed, 2 partially supported, 0 disputed, 0 misattributed, 0 unverifiable. Expand the claim to see per-atom verdicts.
Experimental — accuracy not yet measured - cites [87] source
Split into 2 atoms — 0 confirmed, 1 partially supported, 0 disputed, 0 misattributed, 1 unverifiable. Expand the claim to see per-atom verdicts.
Experimental — accuracy not yet measured - cites [88] source
Probably not a substantial claim: may not pass S3 (load-bearing).
PremisecitationIn the semantic variant of primary progressive aphasia, spontaneous speech is fluent but marked by semantic errors, reduced specificity, and loss of content due to degradation of conceptual knowledge.
“while in the semantic variant PPA, spontaneous speech is fluent but marked by semantic errors, reduced specificity, and loss of content due to degradation of conceptual knowledge.90”
Quote in the paperNot a substantial claim1 of 1 sources checked- cites [90] source
The Introduction discusses speech features in Alzheimer’s disease and mild cognitive impairment, not the semantic variant of primary progressive aphasia. Searches for that variant and its purported speech characteristics found no passage in our copy addressing the claim. (LLM, conf=0.76)
Experimental — accuracy not yet measured
May combine two claims.
PremisecitationGradient boosting trees and random-forest models have predicted ADRD up to five years ahead (AUC > 0.85) or AD seven years ahead (AUC 0.72) using only diagnosis codes, prescriptions, and common lab results.
“Recent studies have used gradient boosting trees and random-forest models to predict ADRD up to 5 years in advance (AUC > 0.85) or to predict AD 7 years in advance (AUC: 0.72), the latter using only diagnosis codes, prescriptions, and common laboratory results.123,124”
Quote in the paperSubstantial claim1 of 2 sources checked- cites [123] source
The Introduction and Materials and Methods describe a multi-omics approach to AD progression and subtyping, not predictions from diagnosis codes, prescriptions, and common laboratory results. Searches for gradient boosting, random forests, the five- and seven-year horizons, and the stated AUC values found no passage addressing the attributed results. (LLM, conf=0.75)
Experimental — accuracy not yet measured - cites [124] source
Probably not a substantial claim: may not pass S3 (load-bearing).
PremisecitationIn the logopenic variant of primary progressive aphasia, speech is characterized by frequent pauses, word-finding failures, and impaired repetition.
“In the logopenic variant PPA, speech is characterized by frequent pauses, word-finding failures, and impaired repetition,89”
Quote in the paperNot a substantial claim1 of 1 sources checked- cites [89] source
The Introduction (page unavailable) describes frequent pauses in people with mild cognitive impairment, not in people with logopenic-variant primary progressive aphasia. Searches for logopenic PPA and impaired repetition found no passage addressing that variant’s speech characteristics; the cited paper therefore does not verify the attributed claim. (LLM, conf=0.76)
Experimental — accuracy not yet measured
May combine two claims.
PremisecitationNewer methods create high-dimensional embeddings from audio or text that capture complex patterns, giving a more holistic view than individual features but reducing transparency.
“Additionally, recent approaches have generated high-dimensional representations—known as embeddings—that capture nuanced relationships within audio (e.g., wav2vec2) or text (e.g., BERT) to extract complex patterns,102–104 enabling a more holistic approach to speech analysis than using individual features, yet at the cost of transparency.”
Quote in the paperNot a substantial claim3 of 3 sources checked- cites [102] source
The Methods (page unavailable) describe actigraphy, cognitive evaluations, and neuropathology, while the statistical analyses concern rest-activity rhythms and cognitive decline. Searches for embeddings, wav2vec2, BERT, audio, text, and speech analysis found no passage addressing the attributed claim. (LLM, conf=0.78)
Experimental — accuracy not yet measured - cites [103] source
The Methods section describes five individually quantified speech markers, and the Discussion calls for investigating additional markers. Neither passage addresses high-dimensional audio or text embeddings, whether they provide a more holistic analysis, or their effect on transparency (searched for embeddings, wav2vec2, BERT, and interpretability). (LLM, conf=0.75)
Experimental — accuracy not yet measured - cites [104] source
The Discussion’s “Comparison to Prior Work” section concerns passive smartphone data and digital phenotyping, not speech embeddings. Searches for wav2vec2, BERT, audio or text embeddings, and transparency found no passage in our copy addressing the attributed claim. The available passages have no page numbers. (LLM, conf=0.74)
Experimental — accuracy not yet measured
May combine two claims.
PremisecitationSuch technologies can gather multi-domain clinical data on cognitive, behavioral, social, and physical function without burdening patients, enabling ADRD digital biomarker development.
“These technologies can capture multi-domain clinical data reflecting cognitive, behavioral, social, and physical function without any added patient burden, allowing for the development of digital biomarkers of ADRD.110”
Quote in the paperSubstantial claim1 of 1 sources checked- cites [110] source
The Main and Methods (pages not supplied) describe using existing EHR data, including clinical conditions and measurements, to predict AD risk; the Discussion describes use before more costly assessments. The paper does not establish that these data capture all four stated domains of function without any added patient burden, or that the approach develops digital biomarkers of ADRD. (LLM, conf=0.82)
Experimental — accuracy not yet measured
Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.
InterpretationcitationAI-informed treatment approaches, though early in development, may become essential for optimizing dementia care but will require rigorous validation and clinical testing for safety and reliability.
“AI-informed treatment approaches, while still in early development, may become essential tools for optimizing dementia care but will need to be paired with rigorous validation and clinical testing to ensure safety and reliability.145”
Quote in the paperNot a substantial claim1 of 1 sources checked- cites [145] source
The Introduction (page unavailable) says ML algorithms can inform clinical decisions but warns that biased Alzheimer disease progression predictions could lead to improper treatment recommendations or care. The Conclusions (page unavailable) call for mitigating bias. The paper evaluates prediction-model fairness; it does not say AI-informed treatments may become essential to dementia care or call specifically for clinical testing to establish their safety and reliability. (LLM, conf=0.84)
Experimental — accuracy not yet measured
May combine two claims.
PremisecitationNew AI models trained on evolutionary constraints, structural features, or functional data are increasingly used to clarify VUS significance and could help reclassify VUSs in known ADRD risk genes.
“Emerging AI models trained on evolutionary constraints, structural features, or functional data are increasingly used to resolve the significance of VUSs,138,139 and could help reclassify VUSs in known ADRD risk genes (e.g., APP, PSEN1/2, MAPT).140,141”
Quote in the paperNot a substantial claim3 of 4 sources checked- cites [138] source
The Precision medicine and confidence limits section discusses AI-assisted risk prediction, but not genetic variant interpretation. Searches for VUS reclassification, APP, PSEN1/2, MAPT, and evolutionary or structural features found no passage in our copy addressing whether AI could reclassify variants in ADRD risk genes. (LLM, conf=0.76)
Experimental — accuracy not yet measured - cites [139] source
- cites [140] source
Appendix B discusses general benefits and risks of pretrained AI models, not genetic variant interpretation. Searches for VUS reclassification, ADRD risk genes, evolutionary constraints, structural features, and functional data found no passage addressing the attributed claim in our copy. (LLM, conf=0.78)
Experimental — accuracy not yet measured - cites [141] source
The cited paper discusses explainable AI and model risk (Section 4, p. 9), not the use of AI to assess variants of uncertain significance. Searches for VUSs, ADRD risk genes, and evolutionary, structural, or functional data found no passage addressing their reclassification. (LLM, conf=0.75)
Experimental — accuracy not yet measured
May combine two claims.
PremisecitationExternal validation is needed; one review found that only 2 of 431 digital biomarker studies included validation at outside institutions.
“There is a need for external validation, with one review finding that only 2 of 431 digital biomarker studies included validation at outside institutions.122”
Quote in the paperSubstantial claim1 of 1 sources checked- cites [122] source
The Materials and methods (Evaluation metrics; page unavailable) describe tenfold cross-validation of a genome-based risk model, not a review of digital biomarker studies. Searches for external validation, outside institutions, and the 2-of-431 figure found no passage in our copy addressing the attributed finding or stating the claimed need for external validation. (LLM, conf=0.76)
Experimental — accuracy not yet measured
May combine two claims.
Interpretationnone givenAI can extract useful information from large text volumes, parse scanned documents, and synthesize electronic health record data to obtain relevant history or generate targeted assessment questions.
“Even before the patient is seen, AI can extract value from large volumes of text, parsing scanned documents and synthesizing electronic health record (EHR) data to extract relevant history or generate targeted questions for assessment.”
Quote in the paperNot a substantial claimMay combine two claims.
Interpretationnone givenMany AI-based tools are now available for clinical use, but the supporting evidence varies widely and many implementation challenges prevent widespread adoption.
“Many AI-based tools can now be used in the clinic; however, the level of evidence supporting their clinical use varies widely, and many implementation challenges prevent widespread use.”
Quote in the paperSubstantial claimProbably not a substantial claim: may not pass S3 (load-bearing), S4 (based).
InterpretationreasoningAI could likely help address the limitations of brief office visits for diagnosing neurodegenerative diseases.
“AI could likely help with these limitations.”
Quote in the paperNot a substantial claim- Rests on an assumption: Brief office visits create significant obstacles for diagnosing neurodegenerative diseases.
- Rests on an assumption: Physicians with limited time may have difficulty reviewing lengthy charts or gathering comprehensive histories from distressed patients or caregivers.
Probably not a substantial claim: may not pass S3 (load-bearing).
Interpretationnone givenAs new AI tools become validated and enter clinical practice, their analytical outputs will become part of the complex data landscape that clinicians must interpret.
“Importantly, as new AI tools become validated and enter clinical practice, their analytical outputs will become part of the complex data landscape that clinicians must interpret.”
Quote in the paperNot a substantial claimMay combine two claims.
PremisecitationNew multimodal AI models have made significant progress in describing video content, such as surgeries, making video data a promising method for AI decision support with further research.
“New multimodal AI models have made significant progress in describing video content, such as surgeries, making video data a promising method for AI decision support with further research.68”
Quote in the paperNot a substantial claim0 of 1 sources checked- cites [68] source
May combine two claims.
InterpretationreasoningAI scribes still make mistakes, but clinical implementation has advanced because clinicians can check and correct summaries before signing notes.
“These AI scribes still make mistakes, but clinical implementation has moved forward as clinicians can feasibly check over the summaries and correct them before signing their notes.”
Quote in the paperNot a substantial claimProbably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.
Interpretationnone givenDigital cognitive assessments can evaluate clinical manifestations of neurodegenerative disease or other cognitive disorders, but clinical follow-up is always needed to make a diagnosis.
“Digital cognitive assessments can be used to evaluate for clinical manifestations of neurodegenerative disease or other cognitive disorders, but clinical follow-up is always needed to make a diagnosis.”
Quote in the paperNot a substantial claimProbably not a substantial claim: may not pass S3 (load-bearing).
InterpretationreasoningLarge language model capabilities could translate and integrate AI-derived insights, explaining in natural language what various AI tools detected and how those findings fit with traditional clinical data to inform diagnosis and management.
“LLM capabilities have the potential to translate and integrate these AI-derived insights, explaining in natural language what various AI tools have detected and how these findings fit together with traditional clinical data to inform diagnosis and management decisions (Fig. 1).”
Quote in the paperNot a substantial claimProbably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.
InterpretationreasoningThe complexity of neuroimaging interpretation in neurodegenerative diseases has made it an active area of AI research, with investigators exploring automated approaches to detect subtle changes that may be missed in routine clinical reading.
“The complexity of neuroimaging interpretation in neurodegenerative diseases has made it an active area of AI research, with investigators exploring automated approaches to detect subtle changes that may be missed in routine clinical reading.”
Quote in the paperNot a substantial claim- Rests on an assumption: MRI and PET neuroimaging are central to diagnosing neurodegenerative disease etiologies, excluding treatable causes, assessing vascular contributions, and observing atrophy or biomarker patterns.
May combine two claims.
Interpretationnone givenAlthough an in-person physical exam is ideal, physical exam findings for these syndromes can often be uncovered through video visits, increasing accessibility for patients who cannot travel to see a specialist.
“While an in-person physical exam is ideal, physical exam findings for these syndromes can often be uncovered through video visits, which can increase accessibility for patients who cannot feasibly travel to see a specialist.”
Quote in the paperNot a substantial claim- Rests on an assumption: Physical examination remains important for diagnosis, including motor findings in several neurodegenerative conditions.
- Rests on an assumption: The absence of motor findings in Alzheimer's disease is also diagnostically important.
Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.
InterpretationreasoningConversational agents will likely need a monitoring system to detect emergency situations during interviews, which raises implementation challenges.
“and likely will require a monitoring system that detects emergency situations during these interviews, raising implementation challenges.”
Quote in the paperNot a substantial claimMay combine two claims.
PremisecitationPassive monitoring with digital health technology complements traditional neuropsychological and neurologic evaluations and helps overcome their limitations.
“Passive monitoring via digital health technology is a complementary approach to overcome the limitations of traditional neuropsychological and neurologic evaluations.108”
Quote in the paperSubstantial claim0 of 1 sources checked- cites [108] source
May combine two claims.
InterpretationreasoningA major caveat is that most EHR studies use ICD-9/10 codes for progression labels instead of biomarker-confirmed diagnoses, and diagnostic misclassification in routine practice would cap reported AUROCs.
“A key caveat is that most EHR studies rely on ICD-9/10 codes for progression labels rather than biomarker-confirmed diagnoses, and any diagnostic misclassification in routine practice would place an upper bound on reported AUROCs.”
Quote in the paperSubstantial claimMay combine two claims.
InterpretationreasoningPassive data collection is low-burden and scalable, and with further validation it has strong potential as a clinical diagnostic and monitoring tool.
“Passive approaches to data collection are also low-burden and scalable, with great potential to be utilized as clinical diagnostic and monitoring tools with continued validation.”
Quote in the paperSubstantial claim- Rests on an assumption: Passive monitoring methods can detect subtle neurodegenerative changes in real time, even before symptoms are noticeable.
May combine two claims.
InterpretationcitationBecause neurodegeneration is chronic, index dates are often heterogeneous across syndromes, which can lead AI models to learn physician behaviors that predict later diagnosis in downstream visits.
“Furthermore, given that neurodegeneration is a chronic disease process, there is often heterogeneity in identifying an index date, particularly for different neurodegenerative syndromes. This can lead to AI models that excel in learning physician behaviors (e.g., ordering for a cognitive test) that predict subsequent diagnosis in downstream visits (e.g., ICD code for dementia given).132”
Quote in the paperNot a substantial claim0 of 1 sources checked- cites [132] source
May combine two claims.
PremisecitationKey ADRD community concerns include obtaining informed consent from patients with diminished decision-making capacity, protecting privacy beyond traditional de-identification, and maintaining transparency about AI decision-making.
“Key concerns for the ADRD community include ensuring informed consent from patients with diminished decision-making capacity, protecting patient privacy beyond traditional de-identification methods, and maintaining transparency about AI decision-making processes.152–154”
Quote in the paperNot a substantial claim0 of 3 sources checked- cites [152]
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- cites [153]
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- cites [154]
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
Probably not a substantial claim: may not pass S3 (load-bearing).
Interpretationnone givenLegal liability for patient harm involving AI systems remains unclear.
“Legal liability remains unclear regarding responsibility when AI systems contribute to patient harm.”
Quote in the paperNot a substantial claimMay combine two claims.
PremisecitationOptimistic performance in research studies may sometimes stem from methodological issues such as data leakage and selection bias, which helps explain why many models fail in prospective real-world testing.
“Furthermore, the optimistic performance reported in research studies may at times reflect methodological issues, including data leakage and selection bias, explaining why many models fail when tested prospectively in real-world settings.148–150”
Quote in the paperSubstantial claim0 of 3 sources checked- cites [148]
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- cites [149]
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- cites [150]
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
May combine two claims.
Premisenone givenIn some early-onset or atypical dementia cases, genetic variants of unknown significance may be undiagnosed pathogenic causes, but their uncertain classification limits clinical actionability.
“In certain early-onset or atypical dementia cases, the presence of genetic variants of unknown significance (VUSs) may represent undiagnosed pathogenic causes of disease, but their uncertain classification limits clinical actionability.”
Quote in the paperNot a substantial claimMay combine two claims.
PremisecitationRacial and ethnic minority patients and those from socioeconomically disadvantaged backgrounds receive more fragmented medical care and may have limited access to online patient portals.
“Additionally, racial and ethnic minority patients and those from socioeconomically disadvantaged backgrounds receive more fragmented medical care,160 and may have limited access to online patient portals.”
Quote in the paperSubstantial claim0 of 1 sources checked- cites [160]
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
May combine two claims.
PremisecitationBias can enter through nonrepresentative health data, lack of diverse perspectives in algorithm development, and inadequate oversight of AI-based decisions.
“Bias can be introduced through nonrepresentative health data, the absence of diverse perspectives in algorithm development, and inadequate oversight of AI-based decisions.158”
Quote in the paperSubstantial claim0 of 1 sources checked- cites [158]
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
May combine two claims.
Interpretationnone givenAlthough AI models can aid dimensionality reduction by identifying important features, interpretability remains a challenge.
“While AI models can help with this dimensionality reduction, for example, by identifying the most important features to focus on, interpretability has been a challenge.”
Quote in the paperNot a substantial claimProbably not a substantial claim: may not pass S4 (based).
Interpretationnone givenEmerging digital biomarkers-speech analysis, passive monitoring, and advanced EHR mining-have transformative potential for continuous, unobtrusive monitoring that could detect cognitive decline years before traditional assessments.
“Emerging digital biomarkers—including speech analysis, passive monitoring, and advanced EHR mining—hold transformative potential for continuous, unobtrusive monitoring that could detect cognitive decline years before traditional assessments.”
Quote in the paperNot a substantial claim- Rests on an assumption: Passive monitoring methods can detect subtle neurodegenerative changes in real time, even before symptoms are noticeable.
May combine two claims.
Interpretationnone givenPatient safety risks from AI include misdiagnosis, inappropriate treatment recommendations, and adverse psychological effects from AI interactions.
“Patient safety risks include potential misdiagnosis, inappropriate treatment recommendations, and adverse psychological effects from AI interactions.”
Quote in the paperNot a substantial claimMay combine two claims.
InterpretationreasoningIt is likely that AI models using multi-modal -omics data will eventually be developed for clinical use and will outperform single or small-panel biomarkers for risk stratification, diagnosis, staging, prognostication, and tailored treatment selection in neurodegenerative diseases.
“Despite these challenges, it is likely that AI models that incorporate multi-modal -omics data will eventually be developed for clinical use that prove superior to single or small panel biomarker measurements for risk stratification, diagnosis, staging, prognostication, and tailored treatment selection in neurodegenerative diseases.”
Quote in the paperNot a substantial claimProbably not a substantial claim: may not pass S3 (load-bearing).
Interpretationnone givenInfrastructure requirements, such as standardized data collection protocols and computational resources, may create additional barriers, especially in resource-limited healthcare settings.
“Infrastructure requirements, including standardized data collection protocols and computational resources, may pose additional barriers, particularly for resource-limited healthcare settings.”
Quote in the paperNot a substantial claimMay combine two claims.
Interpretationnone givenCurrent limitations for AI and -omics in the clinic include high costs of comprehensive biological testing, lack of standardization across labs, and technical difficulties from high-dimensional data that can cause overfitting and poor generalization.
“Current limitations include the high cost of comprehensive biological testing, the lack of standardization across laboratories, and technical challenges that come with analyzing a huge amount of data where the number of measurements far exceeds the number of patients studied, leading to potential overfitting of the model and poor generalization across different groups of patients.”
Quote in the paperNot a substantial claimProbably not a substantial claim: may not pass S4 (based). May combine two claims.
Interpretationnone givenNear-term applications likely involve AI-enhanced processing of familiar data types that fit existing workflows: AI-assisted neuroimaging interpretation, conversational agents interviewing patients and caregivers, and digital cognitive assessments offering standardized, scalable alternatives to traditional testing.
“It is likely that the relatively near-term applications involve AI-enhanced processing of familiar data types that integrate with existing workflows: AI-assisted neuroimaging interpretation, conversational agents interviewing patients and caregivers, and digital cognitive assessments offering standardized, scalable alternatives to traditional testing.”
Quote in the paperNot a substantial claimClassified as probably not a substantial claim. May combine two claims.
Interpretationnone givenTechnical challenges include making models interpretable for clinical decisions, addressing biases that could worsen health disparities, and creating robust safeguards against AI hallucinations in high-stakes medical decisions.
“Technical challenges include ensuring model interpretability for clinical decision-making, addressing biases that could exacerbate health disparities, and developing robust safeguards against AI hallucinations in high-stakes medical decisions.”
Quote in the paperSubstantial claimMay combine two claims.
PremisecitationModel outputs can be hard to interpret and therefore hard to apply in clinical care.
“The output of models can be difficult to interpret and, therefore, difficult to apply to clinical care.146,147”
Quote in the paperNot a substantial claim0 of 2 sources checked- cites [146]
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- cites [147]
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
May combine two claims.
Interpretationnone givenMany models are trained on small, homogeneous datasets from single academic centers, which likely limits their generalizability to diverse clinical populations.
“Many models are trained on small, homogeneous datasets from single academic centers, likely limiting their generalizability to diverse clinical populations.”
Quote in the paperSubstantial claimMay combine two claims.
PremisecitationNumerous initiatives have created frameworks and guidelines to ensure AI applications are fair, appropriate, valid, effective, and safe, which will require transparency about technical performance and organizational capacity to manage deployment risks.
“In response, numerous initiatives have developed frameworks and guidelines to ensure AI applications are fair, appropriate, valid, effective, and safe, which will at least in part require transparency with technical performance as well as organizational capacity to manage risks associated with deploying this technology.151”
Quote in the paperNot a substantial claim0 of 1 sources checked- cites [151]
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
May combine two claims.
PremisecitationUnlike drugs with unclear but static mechanisms, AI performance can change over time, making explainability and interpretability important so humans can spot model deterioration or drift at the point of care.
“In contrast to approving drugs with unclear but static mechanisms of action, AI performance can change over time, raising the importance of explainability and interpretability to ensure humans can identify signs of model deterioration or drift at the point of care.155–157”
Quote in the paperSubstantial claim0 of 3 sources checked- cites [155]
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- cites [156]
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- cites [157]
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
The paper relies on this without support.
Many primary care providers and general neurologists have difficulty keeping up with rapid advances in the field.
“Many PCPs and general neurologists struggle to keep pace with the rapid advances in the field, which includes maintaining awareness of research studies that could potentially benefit their patients and the ADRD field.”
The paper relies on this without support.
Brief office visits create significant obstacles for diagnosing neurodegenerative diseases.
“The constraints of brief office visits present significant challenges for diagnosing neurodegenerative diseases.”
The paper relies on this without support.
Physicians with limited time may have difficulty reviewing lengthy charts or gathering comprehensive histories from distressed patients or caregivers.
“Physicians, faced with limited time, may struggle to effectively review a lengthy patient chart or gather comprehensive patient histories from distressed patients or their caregivers.”
The paper relies on this without support.
Physical examination remains important for diagnosis, including motor findings in several neurodegenerative conditions.
“Physical examination continues to be important in obtaining a diagnosis, including the motor findings associated with Lewy body disease, corticobasal syndrome, progressive supranuclear palsy, amyotrophic lateral sclerosis, and many other conditions.”
The paper relies on this without support.
The absence of motor findings in Alzheimer's disease is also diagnostically important.
“In addition, the absence of motor findings in AD is also important.”
The paper relies on this without support.
MRI and PET neuroimaging are central to diagnosing neurodegenerative disease etiologies, excluding treatable causes, assessing vascular contributions, and observing atrophy or biomarker patterns.
“Neuroimaging with MRI and PET is a cornerstone of the diagnosis of etiologies of neurodegenerative disease, including the exclusion of readily treatable causes of dementia, assessing vascular etiologies of dementia, and observing spatial patterns of atrophy (MRI) or biomarker presence and location (PET).”
The paper relies on this without support.
Consistent imaging protocols are necessary for AI model accuracy because systematic variation can substantially affect performance.
“However, consistent imaging protocols are essential for AI model accuracy, as systematic variation can significantly impact performance.”
The paper relies on this without support.
Speech provides rich cognitive and behavioral information by combining linguistic content and vocal motor features.
“Speech is a rich source of cognitive and behavioral information, integrating both linguistic content and vocal motor features.”
The paper relies on this without support.
Most speech studies to date are cross-sectional and research-focused.
“Although most studies to date are cross-sectional and research-focused—and important ethical challenges must be addressed prior to clinical implementation107—speech remains one of the most clinically viable digital biomarkers given its scalability, ecological validity, ease of longitudinal collection, and potential for passive capture in real-world settings.”
The paper relies on this without support.
Passive monitoring methods can detect subtle neurodegenerative changes in real time, even before symptoms are noticeable.
“A growing number of studies suggest these passive monitoring methods can detect subtle changes in neurodegenerative disease as they are happening in real time, even before symptoms become noticeable to individuals or their loved ones.”
Cites
Lecanemab in early Alzheimer's disease
van Dyck, C. H., Swanson, C. J., Aisen, P., Bateman, R. J., Chen, C., Gee, M., Kanekiyo, M., Li, D., Reyderman, L., Cohen, S., et al. (2023) · New England Journal of Medicine · doi:10.1056/nejmoa2212948 · Cited by 5,678 · see citing works
Feigin, V. L., Nichols, E., Alam, T., Bannick, M. S., Beghi, E., Blake, N., Culpepper, W. J., Dorsey, E. R., Elbaz, A., Ellenbogen, R. G., et al. (2019) · The Lancet Neurology · doi:10.1016/s1474-4422(18)30499-x · Cited by 5,644 · see citing works
A guide to deep learning in healthcare
Esteva, A., Robicquet, A., Ramsundar, B., Kuleshov, V., DePristo, M., Chou, K., Cui, C., Corrado, G., Thrun, S., & Dean, J. (2019) · Nature Medicine · doi:10.1038/s41591-018-0316-z · Cited by 5,312 · see citing works
Harold, D., Abraham, R., Hollingworth, P., Sims, R., Gerrish, A., Hamshere, M. L., Pahwa, J. S., Moskvina, V., Dowzell, K., Williams, A., et al. (2009) · Nature Genetics · doi:10.1038/ng.440 · Cited by 3,326 · see citing works
Donanemab in Early Symptomatic Alzheimer Disease: The TRAILBLAZER-ALZ 2 Randomized Clinical Trial
Sims, J. R., Zimmer, J. A., Evans, C. D., Lu, M., Ardayfio, P., Sparks, J., Wessels, A. M., Shcherbinin, S., Wang, H., Monkul Nery, E. S., et al. (2023) · JAMA · doi:10.1001/jama.2023.13239 · Cited by 2,932 · see citing works
Artificial intelligence in healthcare
Yu, K.-H., Beam, A. L., & Kohane, I. S. (2018) · Nature Biomedical Engineering · doi:10.1038/s41551-018-0305-z · Cited by 2,817 · see citing works
Alzheimer disease in the United States (2010–2050) estimated using the 2010 census
Hebert, L. E., Weuve, J., Scherr, P. A., & Evans, D. A. (2013) · Neurology · doi:10.1212/wnl.0b013e31828726f5 · Cited by 2,642 · see citing works
Toward expert-level medical question answering with large language models
Singhal, K., Tu, T., Gottweis, J., Sayres, R., Wulczyn, E., Amin, M., Hou, L., Clark, K., Pfohl, S. R., Cole-Lewis, H., et al. (2025) · Nature Medicine · doi:10.1038/s41591-024-03423-7 · Cited by 1,734 · see citing works
Resting-state fMRI in the human connectome project
Smith, S. M., Beckmann, C. F., Andersson, J., Auerbach, E. J., Bijsterbosch, J., Douaud, G., Duff, E., Feinberg, D. A., Griffanti, L., Harms, M. P., et al. (2013) · NeuroImage · doi:10.1016/j.neuroimage.2013.05.039 · Cited by 1,502 · see citing works
Potential Biases in Machine Learning Algorithms Using Electronic Health Record Data
Gianfrancesco, M. A., Tamang, S., Yazdany, J., & Schmajuk, G. (2018) · JAMA Internal Medicine · doi:10.1001/jamainternmed.2018.3763 · Cited by 1,395 · see citing works
Leakage and the reproducibility crisis in machine-learning-based science
Kapoor, S., & Narayanan, A. (2023) · Patterns · doi:10.1016/j.patter.2023.100804 · Cited by 1,387 · see citing works
Artificial Intelligence Risk Management Framework (AI RMF 1.0)
Tabassi, E. (2023) · doi:10.6028/nist.ai.100-1 · Cited by 1,136 · see citing works
Missed and Delayed Diagnosis of Dementia in Primary Care: Prevalence and Contributing Factors
Bradford, A., Kunik, M. E., Schulz, P., Williams, S. P., & Singh, H. (2009) · Alzheimer Disease & Associated Disorders · doi:10.1097/wad.0b013e3181a6bebc · Cited by 969 · see citing works
Adapted large language models can outperform medical experts in clinical text summarization
Van Veen, D., Van Uden, C., Blankemeier, L., Delbrouck, J.-B., Aali, A., Bluethgen, C., Pareek, A., Polacin, M., Reis, E. P., Seehofnerová, A., et al. (2024) · Nature Medicine · doi:10.1038/s41591-024-02855-5 · Cited by 891 · see citing works
Disease variant prediction with deep generative models of evolutionary data
Frazer, J., Notin, P., Dias, M., Gomez, A., Min, J. K., Brock, K., Gal, Y., & Marks, D. S. (2021) · Nature · doi:10.1038/s41586-021-04043-8 · Cited by 791 · see citing works
Rathore, S., Habes, M., Iftikhar, M. A., Shacklett, A., & Davatzikos, C. (2017) · NeuroImage · doi:10.1016/j.neuroimage.2017.03.057 · Cited by 622 · see citing works
Ding, Y., Sohn, J. H., Kawczynski, M. G., Trivedi, H., Harnish, R., Jenkins, N. W., Lituiev, D., Copeland, T. P., Aboian, M. S., Mari Aparici, C., et al. (2019) · Radiology · doi:10.1148/radiol.2018180958 · Cited by 577 · see citing works
Babulal, G. M., Quiroz, Y. T., Albensi, B. C., Arenaza‐Urquijo, E., Astell, A. J., Babiloni, C., Bahar‐Fuchs, A., Bell, J., Bowman, G. L., Brickman, A. M., et al. (2019) · Alzheimer's & Dementia · doi:10.1016/j.jalz.2018.09.009 · Cited by 528 · see citing works
Digital biomarkers for Alzheimer’s disease: the mobile/wearable devices opportunity
Kourtis, L. C., Regele, O. B., Wright, J. M., & Jones, G. B. (2019) · npj Digital Medicine · doi:10.1038/s41746-019-0084-2 · Cited by 476 · see citing works
Toward clinical digital phenotyping: a timely opportunity to consider purpose, quality, and safety
Huckvale, K., Venkatesh, S., & Christensen, H. (2019) · npj Digital Medicine · doi:10.1038/s41746-019-0166-1 · Cited by 433 · see citing works
Ambient artificial intelligence scribes to alleviate the burden of clinical documentation.
Tierney, A. A., Gayre, G., Hoberman, B., Mattern, B., Ballesca, M., Kipnis, P., Liu, V., & Lee, K. (2024) · NEJM Catalyst · doi:10.1056/cat.23.0404 · Cited by 400 · see citing works
Implementing a screening and diagnosis program for dementia in primary care
Boustani, M., Callahan, C. M., Unverzagt, F. W., Austrom, M. G., Perkins, A. J., Fultz, B. A., Hui, S. L., & Hendrie, H. C. (2005) · Journal of General Internal Medicine · doi:10.1007/s11606-005-0103-7 · Cited by 374 · see citing works
Towards conversational diagnostic artificial intelligence
Tu, T., Schaekermann, M., Palepu, A., Saab, K., Freyberg, J., Tanno, R., Wang, A., Li, B., Amin, M., Cheng, Y., et al. (2025) · Nature · doi:10.1038/s41586-025-08866-7 · Cited by 366 · see citing works
Bron, E. E., Smits, M., van der Flier, W. M., Vrenken, H., Barkhof, F., Scheltens, P., Papma, J. M., Steketee, R. M. E., Méndez Orellana, C., Meijboom, R., et al. (2015) · NeuroImage · doi:10.1016/j.neuroimage.2015.01.048 · Cited by 352 · see citing works
Mueller, K. D., Hermann, B., Mecollari, J., & Turkstra, L. S. (2018) · Journal of Clinical and Experimental Neuropsychology · doi:10.1080/13803395.2018.1446513 · Cited by 336 · see citing works
Prognosis and neuropathologic correlation of clinical subtypes of Parkinson disease
De Pablo-Fernández, E., Lees, A. J., Holton, J. L., & Warner, T. T. (2019) · JAMA Neurology · doi:10.1001/jamaneurol.2018.4377 · Cited by 332 · see citing works
Opportunities and challenges in the collection and analysis of digital phenotyping data
Onnela, J.-P. “. (2020) · Neuropsychopharmacology · doi:10.1038/s41386-020-0771-3 · Cited by 321 · see citing works
Accuracy of dementia diagnosis: a direct comparison between radiologists and a computerized method
Kloppel, S., Stonnington, C. M., Barnes, J., Chen, F., Chu, C., Good, C. D., Mader, I., Mitchell, L. A., Patel, A. C., Roberts, C. C., et al. (2008) · Brain · doi:10.1093/brain/awn239 · Cited by 271 · see citing works
Practice constraints, behavioral problems, and dementia care: primary care physicians' perspectives
Hinton, L., Franz, C. E., Reddy, G., Flores, Y., Kravitz, R. L., & Barker, J. C. (2007) · Journal of General Internal Medicine · doi:10.1007/s11606-007-0317-y · Cited by 268 · see citing works
García, S. D. L. F., Ritchie, C. W., & Luz, S. F. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2010.06047 · Cited by 265 · see citing works
Savage, T., Nayak, A., Gallo, R., Rangan, E., & Chen, J. H. (2024) · npj Digital Medicine · doi:10.1038/s41746-024-01010-1 · Cited by 265 · see citing works
Four principles of explainable artificial intelligence
Phillips, P. J., Hahn, C. A., Fontana, P. C., Yates, A. N., Greene, K., Broniatowski, D. A., & Przybocki, M. A. (2021) · doi:10.6028/nist.ir.8312 · Cited by 262 · see citing works
Piau, A., Wild, K., Mattek, N., & Kaye, J. (2019) · Journal of Medical Internet Research · doi:10.2196/12785 · Cited by 252 · see citing works
Projected workforce of psychiatrists in the United States: a population analysis
Satiani, A., Niedermier, J., Satiani, B., & Svendsen, D. P. (2018) · Psychiatric Services · doi:10.1176/appi.ps.201700344 · Cited by 235 · see citing works
Current advances in digital cognitive assessment for preclinical Alzheimer's disease
Öhman, F., Hassenstab, J., Berron, D., Schöll, M., & Papp, K. V. (2021) · Alzheimer's & Dementia: Diagnosis, Assessment & Disease Monitoring · doi:10.1002/dad2.12217 · Cited by 214 · see citing works
Deep representation learning of electronic health records to unlock patient stratification at scale
Landi, I., Glicksberg, B. S., Lee, H.-C., Cherng, S., Landi, G., Danieletto, M., Dudley, J. T., Furlanello, C., & Miotto, R. (2020) · npj Digital Medicine · doi:10.1038/s41746-020-0301-z · Cited by 213 · see citing works
Linguistic markers predict onset of Alzheimer's disease
Eyigoz, E., Mathur, S., Santamaria, M., Cecchi, G., & Naylor, M. (2020) · EClinicalMedicine · doi:10.1016/j.eclinm.2020.100583 · Cited by 206 · see citing works
A Scoping Review of Reporting Gaps in FDA-Approved AI Medical Devices
Muralidharan, V., Adewale, B. A., Huang, C. J., Nta, M. T., Ademiju, P. O., Pathmarajah, P., Hang, M. K., Adesanya, O., Abdullateef, R. O., Babatunde, A. O., et al. (2024) · npj Digital Medicine · doi:10.1038/s41746-024-01270-x · Cited by 201 · see citing works
A generalist medical language model for disease diagnosis assistance
Liu, X., Liu, H., Yang, G., Jiang, Z., Cui, S., Zhang, Z., Wang, H., Tao, L., Sun, Y., Song, Z., et al. (2025) · Nature Medicine · doi:10.1038/s41591-024-03416-6 · Cited by 183 · see citing works
AI-based differential diagnosis of dementia etiologies on multimodal data
Xue, C., Kowshik, S. S., Lteif, D., Puducheri, S., Jasodanand, V. H., Zhou, O. T., Walia, A. S., Guney, O. B., Zhang, J. D., Poésy, S., et al. (2024) · Nature Medicine · doi:10.1038/s41591-024-03118-z · Cited by 182 · see citing works
Clinical reasoning of a generative artificial intelligence model compared with physicians
Cabral, S., Restrepo, D., Kanjee, Z., Wilson, P., Crowe, B., Abdulnour, R.-E., & Rodman, A. (2024) · JAMA Internal Medicine · doi:10.1001/jamainternmed.2024.0295 · Cited by 168 · see citing works
Effect of data leakage in brain MRI classification using 2D convolutional neural networks
Yagis, E., Atnafu, S. W., García Seco de Herrera, A., Marzi, C., Scheda, R., Giannelli, M., Tessa, C., Citi, L., & Diciotti, S. (2021) · Scientific Reports · doi:10.1038/s41598-021-01681-w · Cited by 161 · see citing works
Lifetime risk and projected burden of dementia
Fang, M., Hu, J., Weiss, J., Knopman, D. S., Albert, M., Windham, B. G., Walker, K. A., Sharrett, A. R., Gottesman, R. F., Lutsey, P. L., et al. (2025) · Nature Medicine · doi:10.1038/s41591-024-03340-9 · Cited by 153 · see citing works
Rauschecker, A. M., Rudie, J. D., Xie, L., Wang, J., Duong, M. T., Botzolakis, E. J., Kovalovich, A. M., Egan, J., Cook, T. C., Bryan, R. N., et al. (2020) · Radiology · doi:10.1148/radiol.2020190283 · Cited by 148 · see citing works
Self-citationPredicting dementia from spontaneous speech using large language models
Agbavor, F., & Liang, H. (2022) · PLOS Digital Health · doi:10.1371/journal.pdig.0000168 · Cited by 147 · see citing works
Unobtrusive measurement of daily computer use to detect mild cognitive impairment
Kaye, J., Mattek, N., Dodge, H. H., Campbell, I., Hayes, T., Austin, D., Hatt, W., Wild, K., Jimison, H., & Pavel, M. (2014) · Alzheimer's & Dementia · doi:10.1016/j.jalz.2013.01.011 · Cited by 145 · see citing works
Data-driven subtyping of Parkinson's disease using longitudinal clinical records: a cohort study
Zhang, X., Chou, J., Liang, J., Xiao, C., Zhao, Y., Sarva, H., Henchcliffe, C., & Wang, F. (2019) · Scientific Reports · doi:10.1038/s41598-018-37545-z · Cited by 137 · see citing works
Biomedical knowledge graph-optimized prompt generation for large language models
Soman, K., Rose, P. W., Morris, J. H., Akbas, R. E., Smith, B., Peetoom, B., Villouta-Reyes, C., Cerono, G., Shi, Y., Rizk-Jackson, A., et al. (2024) · Bioinformatics · doi:10.1093/bioinformatics/btae560 · Cited by 135 · see citing works
Majersik, J. J., Ahmed, A., Chen, I.-H. A., Shill, H., Hanes, G. P., Pelak, V. S., Hopp, J. L., Omuro, A., Kluger, B., & Leslie-Mazwi, T. (2021) · Neurology · doi:10.1212/wnl.0000000000012111 · Cited by 133 · see citing works
Connected speech features from picture description in Alzheimer's disease: a systematic review
Slegers, A., Filiou, R.-P., Montembeault, M., & Brambati, S. M. (2018) · Journal of Alzheimer’s Disease · doi:10.3233/jad-170881 · Cited by 124 · see citing works
Machine learning and digital biomarkers can detect early stages of neurodegenerative diseases
Chudzik, A., Śledzianowski, A., & Przybyszewski, A. W. (2024) · Sensors · doi:10.3390/s24051572 · Cited by 122 · see citing works
Yang, Z., Xu, X., Yao, B., Rogers, E., Zhang, S., Intille, S., Shara, N., Gao, G. G., & Wang, D. (2024) · Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies · doi:10.1145/3659625 · Cited by 120 · see citing works
Bernstein, A., Rogers, K. M., Possin, K. L., Steele, N. Z. R., Ritchie, C. S., Kramer, J. H., Geschwind, M., Higgins, J. J., Wohlgemuth, J., Pesano, R., et al. (2019) · BMC Health Services Research · doi:10.1186/s12913-019-4603-2 · Cited by 120 · see citing works
Self-citationFactors associated with cognitive evaluations in the United States
Kotagal, V., Langa, K. M., Plassman, B. L., Fisher, G. G., Giordani, B. J., Wallace, R. B., Burke, J. R., Steffens, D. C., Kabeto, M., Albin, R. L., et al. (2015) · Neurology · doi:10.1212/wnl.0000000000001096 · Cited by 119 · see citing works
Performance of large language models on a neurology board-style examination
Schubert, M. C., Wick, W., & Venkataramani, V. (2023) · JAMA Network Open · doi:10.1001/jamanetworkopen.2023.46721 · Cited by 115 · see citing works
Bridging the chasm between AI and clinical implementation
Aristidou, A., Jena, R., & Topol, E. J. (2022) · The Lancet · doi:10.1016/s0140-6736(22)00235-5 · Cited by 115 · see citing works
Li, Q., Yang, X., Xu, J., Guo, Y., He, X., Hu, H., Lyu, T., Marra, D., Miller, A., Smith, G., et al. (2023) · Alzheimer's & Dementia · doi:10.1002/alz.12967 · Cited by 113 · see citing works
Bin Sawad, A., Narayan, B., Alnefaie, A., Maqbool, A., Mckie, I., Smith, J., Yuksel, B., Puthal, D., Prasad, M., & Kocaballi, A. B. (2022) · Sensors · doi:10.3390/s22072625 · Cited by 113 · see citing works
McGreevey, J. D., Hanson, C. W., & Koppel, R. (2020) · JAMA · doi:10.1001/jama.2020.2724 · Cited by 112 · see citing works
Iturria-Medina, Y., Adewale, Q., Khan, A. F., Ducharme, S., Rosa-Neto, P., O’Donnell, K., Petyuk, V. A., Gauthier, S., De Jager, P. L., Breitner, J., et al. (2022) · Science Advances · doi:10.1126/sciadv.abo6764 · Cited by 100 · see citing works
Iakovakis, D., Hadjidimitriou, S., Charisis, V., Bostantzopoulou, S., Katsarou, Z., & Hadjileontiadis, L. J. (2018) · Scientific Reports · doi:10.1038/s41598-018-25999-0 · Cited by 92 · see citing works
GPS driving: a digital biomarker for preclinical Alzheimer disease
Bayat, S., Babulal, G. M., Schindler, S. E., Fagan, A. M., Morris, J. C., Mihailidis, A., & Roe, C. M. (2021) · Alzheimer's Research & Therapy · doi:10.1186/s13195-021-00852-1 · Cited by 91 · see citing works
Harnessing EHR data for health research
Tang, A. S., Woldemariam, S. R., Miramontes, S., Norgeot, B., Oskotsky, T. T., & Sirota, M. (2024) · Nature Medicine · doi:10.1038/s41591-024-03074-8 · Cited by 87 · see citing works
Self-citationTang, A. S., Rankin, K. P., Cerono, G., Miramontes, S., Mills, H., Roger, J., Zeng, B., Nelson, C., Soman, K., Woldemariam, S., et al. (2024) · Nature Aging · doi:10.1038/s43587-024-00573-8 · Cited by 86 · see citing works
Self-citationHuq, S. M., Maskeliūnas, R., & Damaševičius, R. (2022) · Disability and Rehabilitation Assistive Technology · doi:10.1080/17483107.2022.2146768 · Cited by 85 · see citing works
Artificial intelligence for drug discovery and development in Alzheimer's disease
Qiu, Y., & Cheng, F. (2024) · Current Opinion in Structural Biology · doi:10.1016/j.sbi.2024.102776 · Cited by 83 · see citing works
Racial and Socioeconomic Disparities in Access to Primary Care Among People With Chronic Conditions
Shi, L., Chen, C., Nie, X., Zhu, J., & Hu, R. (2014) · The Journal of the American Board of Family Medicine · doi:10.3122/jabfm.2014.02.130246 · Cited by 83 · see citing works
Speech- and language-based classification of Alzheimer's disease: a systematic review
Vigo, I., Coelho, L., & Reis, S. (2022) · Bioengineering · doi:10.3390/bioengineering9010027 · Cited by 83 · see citing works
Speech based detection of Alzheimer's disease: a survey of AI techniques, datasets and challenges
Ding, K., Chetty, M., Noori Hoshyar, A., Bhattacharya, T., & Klein, B. (2024) · Artificial Intelligence Review · doi:10.1007/s10462-024-10961-6 · Cited by 82 · see citing works
Jack, C. R., Arani, A., Borowski, B. J., Cash, D. M., Crawford, K., Das, S. R., DeCarli, C., Fletcher, E., Fox, N. C., Gunter, J. L., et al. (2024) · Alzheimer's & Dementia · doi:10.1002/alz.14166 · Cited by 82 · see citing works
Ramos, E. M., Dokuru, D. R., Van Berlo, V., Wojta, K., Wang, Q., Huang, A. Y., Deverasetty, S., Qin, Y., van Blitterswijk, M., Jackson, J., et al. (2020) · Alzheimer's & Dementia · doi:10.1002/alz.12011 · Cited by 80 · see citing works
Hindelang, M., Sitaru, S., & Zink, A. (2024) · JMIR Medical Informatics · doi:10.2196/56628 · Cited by 79 · see citing works
Performance and biases of large language models in public opinion simulation.
Qu, Y., & Wang, J. (2024) · Humanities and Social Sciences Communications · doi:10.1057/s41599-024-03609-x · Cited by 79 · see citing works
Alzheimer's disease digital biomarkers multidimensional landscape and AI model scoping review
Qi, W., Zhu, X., Wang, B., Shi, Y., Dong, C., Shen, S., Li, J., Zhang, K., He, Y., Zhao, M., et al. (2025) · npj Digital Medicine · doi:10.1038/s41746-025-01640-z · Cited by 78 · cited 1× · see citing works
“While the integration of AI analytics in passive digital biomarker research is still only being used in research settings, 122 this methodology could eventually be used to improve detection and monitoring of neurodegenerative disease-related changes in clinical practice, as well as to monitor…”
Koga, S., Martin, N. B., & Dickson, D. W. (2024) · Brain Pathology · doi:10.1111/bpa.13207 · Cited by 78 · see citing works
Psychological foundations of explainability and interpretability in artificial intelligence
Broniatowski, D. A. (2021) · doi:10.6028/nist.ir.8367 · Cited by 76 · see citing works
Polk, S. E., Öhman, F., Hassenstab, J., König, A., Papp, K. V., Schöll, M., & Berron, D. (2025) · npj Digital Medicine · doi:10.1038/s41746-025-01583-5 · Cited by 73 · see citing works
Artificial intelligence and multi-omics in pharmacogenomics: a new era of precision medicine
Zack, M., Stupichev, D. N., Moore, A. J., Slobodchikov, I. D., Sokolov, D. G., Trifonov, I. F., & Gobbs, A. (2025) · Mayo Clinic Proceedings: Digital Health · doi:10.1016/j.mcpdig.2025.100246 · Cited by 70 · see citing works
Ethics of artificial intelligence in medicine
Savulescu, J., Giubilini, A., Vandersluis, R., & Mishra, A. (2024) · Singapore Medical Journal · doi:10.4103/singaporemedj.smj-2023-279 · Cited by 68 · see citing works
Mao, C., Xu, J., Rasmussen, L., Li, Y., Adekkanattu, P., Pacheco, J., Bonakdarpour, B., Vassar, R., Shen, L., Jiang, G., et al. (2023) · Journal of Biomedical Informatics · doi:10.1016/j.jbi.2023.104442 · Cited by 67 · see citing works
Differences in response rates between mail, e-mail, and telephone follow-up in hand surgery research
Nota, S. P. F. T., Strooker, J. A., & Ring, D. (2014) · HAND · doi:10.1007/s11552-014-9618-x · Cited by 67 · see citing works
Introduction to large language models (LLMs) for dementia care and research
Treder, M. S., Lee, S., & Tsvetanov, K. A. (2024) · Frontiers in Dementia · doi:10.3389/frdem.2024.1385303 · Cited by 63 · see citing works
Digital biomarker‐based individualized prognosis for people at risk of dementia
Buegler, M., Harms, R., Balasa, M., Meier, I. B., Exarchos, T. P., Rai, L. A., Boyle, R. T., Tort‐Merino, A., Kozori, M., Lazarou, E., et al. (2020) · Alzheimer s & Dementia Diagnosis Assessment & Disease Monitoring · doi:10.1002/dad2.12073 · Cited by 62 · see citing works
An Empirical Study of Clinical Note Generation from Doctor-Patient Encounters
Ben Abacha, A., Yim, W.-W., Fan, Y., & Lin, T. (2023) · Proceedings of the 17th Conference of the European Chapter of the Association for Computational Linguistics · doi:10.18653/v1/2023.eacl-main.168 · Cited by 54 · see citing works
Lexical and Acoustic Speech Features Relating to Alzheimer Disease Pathology
Cho, S., Cousins, K. A. Q., Shellikeri, S., Ash, S., Irwin, D. J., Liberman, M. Y., Grossman, M., & Nevler, N. (2022) · Neurology · doi:10.1212/wnl.0000000000200581 · Cited by 51 · see citing works
Veneziani, I., Marra, A., Formica, C., Grimaldi, A., Marino, S., Quartarone, A., & Maresca, G. (2024) · Journal of Personalized Medicine · doi:10.3390/jpm14010113 · Cited by 49 · see citing works
Algorithmic fairness of machine learning models for Alzheimer disease progression
Yuan, C., Linn, K. A., & Hubbard, R. A. (2023) · JAMA Network Open · doi:10.1001/jamanetworkopen.2023.42203 · Cited by 49 · see citing works
Ostrand, R., & Gunstad, J. (2021) · Journal of Geriatric Psychiatry and Neurology · doi:10.1177/0891988720933358 · Cited by 49 · see citing works
Wearables, smartphones, and artificial intelligence for digital phenotyping and health
Perez-Pozuelo, I., Spathis, D., Clifton, E. A. D., & Cecilia Mascolo, C. (2020) · Elsevier eBooks · doi:10.1016/b978-0-12-820077-3.00003-1 · Cited by 49 · see citing works
Automatic Speech Classifier for Mild Cognitive Impairment and Early Dementia
Bertini, F., Allevi, D., Lutero, G., Montesi, D., & Calzà, L. (2021) · ACM Transactions on Computing for Healthcare · doi:10.1145/3469089 · Cited by 49 · see citing works
Hussain, S. A., Bresnahan, M., & Zhuang, J. (2024) · Ethnicity and Health · doi:10.1080/13557858.2024.2422848 · Cited by 48 · see citing works
Artificial intelligence for dementia drug discovery and trials optimization
Doherty, T., Yao, Z., Khleifat, A. A. L., Tantiangco, H., Tamburin, S., Albertyn, C., Thakur, L., Llewellyn, D. J., Oxtoby, N. P., Lourida, I., et al. (2023) · Alzheimer's & Dementia · doi:10.1002/alz.13428 · Cited by 48 · see citing works
A remote speech-based AI system to screen for early Alzheimer's disease via smartphones
Fristed, E., Skirrow, C., Meszaros, M., Lenain, R., Meepegama, U., Cappa, S., Aarsland, D., & Weston, J. (2022) · Alzheimer's & Dementia: Diagnosis, Assessment & Disease Monitoring · doi:10.1002/dad2.12366 · Cited by 47 · see citing works
Wearables, smartphones, and artificial intelligence for digital phenotyping and health
Perez-Pozuelo, I., Spathis, D., Clifton, E. A., & Mascolo, C. (2020) · Elsevier · doi:10.17863/cam.60922 · Cited by 47
García, A. M., Welch, A. E., Mandelli, M. L., Henry, M. L., Lukic, S., Torres Prioris, M. J., Deleon, J., Ratnasiri, B. M., Lorca-Puls, D. L., Miller, B. L., et al. (2022) · Neurology · doi:10.1212/wnl.0000000000200750 · Cited by 47 · see citing works
Faria, A. L., Latorre, J., Silva Cameirão, M., Bermúdez i Badia, S., & Llorens, R. (2023) · Frontiers in Psychology · doi:10.3389/fpsyg.2023.1233346 · Cited by 46 · see citing works
Artificial intelligence classifies primary progressive aphasia from connected speech
Rezaii, N., Hochberg, D., Quimby, M., Wong, B., Brickhouse, M., Touroutoglou, A., Dickerson, B. C., & Wolff, P. (2024) · Brain · doi:10.1093/brain/awae196 · Cited by 44 · see citing works
Winer, J. R., Lok, R., Weed, L., He, Z., Poston, K. L., Mormino, E. C., & Zeitzer, J. M. (2024) · Alzheimer's Research & Therapy · doi:10.1186/s13195-024-01411-0 · Cited by 44 · see citing works
Tong, L., George, B., Crotty, B. H., Somai, M., Taylor, B. W., Osinski, K., & Luo, J. (2022) · IPEM-Translation · doi:10.1016/j.ipemt.2022.100010 · Cited by 43 · see citing works
AI's potential to accelerate drug discovery needs a reality check
(2023) · Nature · doi:10.1038/d41586-023-03172-6 · Cited by 42 · see citing works
Automated analysis of lexical features in frontotemporal degeneration
Cho, S., Nevler, N., Ash, S., Shellikeri, S., Irwin, D. J., Massimo, L., Rascovsky, K., Olm, C., Grossman, M., & Liberman, M. (2021) · Cortex · doi:10.1016/j.cortex.2021.01.012 · Cited by 41 · see citing works
BHA-CS: A novel cognitive composite for Alzheimer's disease and related disorders
Tsoy, E., Erlhoff, S. J., Goode, C. A., Dorsman, K. A., Kanjanapong, S., Lindbergh, C. A., La Joie, R., Strom, A., Rabinovici, G. D., Lanata, S. C., et al. (2020) · Alzheimer's & Dementia: Diagnosis, Assessment & Disease Monitoring · doi:10.1002/dad2.12042 · Cited by 38 · see citing works
Self-citationDiaz-Asper, C., Hauglid, M. K., Chandler, C., Cohen, A. S., Foltz, P. W., & Elvevåg, B. (2024) · American Psychologist · doi:10.1037/amp0001195 · Cited by 35 · see citing works
Jagust, W. J., Koeppe, R. A., Rabinovici, G. D., Villemagne, V. L., Harrison, T. M., Landau, S. M., & Initiative, T. A. D. N. (2024) · Alzheimer's & Dementia · doi:10.1002/alz.14165 · Cited by 34 · see citing works
Mancuso, V., Sarcinella, E. D., Bruni, F., Arlati, S., Di Santo, S. G., Cavallo, M., Cipresso, P., & Pedroli, E. (2024) · Frontiers in Human Neuroscience · doi:10.3389/fnhum.2024.1380575 · Cited by 31 · see citing works
Artificial intelligence for dementia genetics and omics
Bettencourt, C., Skene, N., Bandres‐Ciga, S., Anderson, E., Winchester, L. M., Foote, I. F., Schwartzentruber, J., Botia, J. A., Nalls, M., Singleton, A., et al. (2023) · Alzheimer's & Dementia · doi:10.1002/alz.13427 · Cited by 30 · see citing works
Fristed, E., Skirrow, C., Meszaros, M., Lenain, R., Meepegama, U., Papp, K. V., Ropacki, M., & Weston, J. (2022) · Brain Communications · doi:10.1093/braincomms/fcac231 · Cited by 29 · see citing works
Van Egroo, M., van Someren, E. J. W., Grinberg, L. T., Bennett, D. A., & Jacobs, H. I. L. (2024) · Annals of Neurology · doi:10.1002/ana.26880 · Cited by 28 · see citing works
Pagán, V. M., McClung, K. S., & Peden, C. J. (2022) · Telemedicine and e-Health · doi:10.1089/tmj.2021.0412 · Cited by 28 · see citing works
Bernstein, A., Rogers, K. M., Possin, K. L., Steele, N. Z. R., Ritchie, C. S., Miller, B. L., & Rankin, K. P. (2019) · Journal of General Internal Medicine · doi:10.1007/s11606-019-05013-7 · Cited by 28 · see citing works
Self-citationSpeech patterns during memory recall relates to early tau burden across adulthood
Young, C. B., Smith, V., Karjadi, C., Grogan, S., Ang, T. F. A., Insel, P. S., Henderson, V. W., Sumner, M., Poston, K. L., Au, R., et al. (2024) · Alzheimer's & Dementia · doi:10.1002/alz.13731 · Cited by 27 · see citing works
Research on digital tool in cognitive assessment: a bibliometric analysis
Chen, L., Zhen, W., & Peng, D. (2023) · Frontiers in Psychiatry · doi:10.3389/fpsyt.2023.1227261 · Cited by 26 · see citing works
Confounding factors need to be accounted for in assessing bias by machine learning algorithms
Mukherjee, P., Shen, T. C., Liu, J., Mathai, T., Shafaat, O., & Summers, R. M. (2022) · Nature Medicine · doi:10.1038/s41591-022-01847-7 · Cited by 26 · see citing works
Automated analysis of written language in the three variants of primary progressive aphasia
Josephy-Hernandez, S., Rezaii, N., Jones, A., Loyer, E., Hochberg, D., Quimby, M., Wong, B., & Dickerson, B. C. (2023) · Brain Communications · doi:10.1093/braincomms/fcad202 · Cited by 25 · see citing works
Interpretable video-based tracking and quantification of Parkinsonism clinical motor states
Deng, D., Ostrem, J. L., Nguyen, V., Cummins, D. D., Sun, J., Pathak, A., Little, S., & Abbasi-Asl, R. (2024) · npj Parkinson's Disease · doi:10.1038/s41531-024-00742-x · Cited by 25 · see citing works
Barnard, L., Botha, H., Corriveau-Lecavalier, N., Graff-Radford, J., Dicks, E., Gogineni, V., Zhang, G., Burkett, B. J., Johnson, D. R., Huls, S. J., et al. (2025) · Neurology · doi:10.1212/wnl.0000000000213831 · Cited by 25 · see citing works
Kodali, M., Kadiri, S. R., & Alku, P. (2024) · Computer Speech & Language · doi:10.1016/j.csl.2023.101548 · Cited by 24 · see citing works
An explainable machine learning model of cognitive decline derived from speech
Chandler, C., Diaz‐Asper, C., Turner, R. S., Reynolds, B., & Elvevåg, B. (2023) · Alzheimer's & Dementia: Diagnosis, Assessment & Disease Monitoring · doi:10.1002/dad2.12516 · Cited by 24 · see citing works
Nasreen, S., Hough, J., & Purver, M. (2021) · Interspeech 2021 · doi:10.21437/interspeech.2021-1526 · Cited by 22 · see citing works
Automated detection of progressive speech changes in early Alzheimer's disease
Robin, J., Xu, M., Balagopalan, A., Novikova, J., Kahn, L., Oday, A., Hejrati, M., Hashemifar, S., Negahdar, M., Simpson, W., et al. (2023) · Alzheimer's & Dementia: Diagnosis, Assessment & Disease Monitoring · doi:10.1002/dad2.12445 · Cited by 21 · see citing works
Taylor, J. C., Heuer, H. W., Clark, A. L., Wise, A. B., Manoochehri, M., Forsberg, L., Mester, C., Rao, M., Brushaber, D., Kramer, J., et al. (2023) · Alzheimer's & Dementia: Diagnosis, Assessment & Disease Monitoring · doi:10.1002/dad2.12423 · Cited by 20 · see citing works
Lukic, S., Fan, Z., García, A. M., Welch, A. E., Ratnasiri, B. M., Wilson, S. M., Henry, M. L., Vonk, J., Deleon, J., Miller, B. L., et al. (2024) · Cortex · doi:10.1016/j.cortex.2023.12.013 · Cited by 19 · see citing works
Self-citationWeb-based cognitive assessment in older adults: Where do we stand?
Belleville, S., LaPlume, A. A., & Purkart, R. (2023) · Current Opinion in Neurology · doi:10.1097/wco.0000000000001192 · Cited by 18 · see citing works
König, A., Zeghari, R., Guerchouche, R., Duc Tran, M., Bremond, F., Linz, N., Lindsay, H., Langel, K., Ramakers, I., Lemoine, P., et al. (2021) · BMJ Open · doi:10.1136/bmjopen-2020-047083 · Cited by 17 · see citing works
Sheehy, L., Bouchard, S., Kakkar, A., El Hakim, R., Lhoest, J., & Frank, A. (2024) · Journal of Clinical Medicine · doi:10.3390/jcm13185574 · Cited by 16 · see citing works
Paolillo, E. W., Casaletto, K. B., Clark, A. L., Taylor, J. C., Heuer, H. W., Wise, A. B., Dhanam, S., Sanderson-Cimino, M., Saloner, R., Kramer, J. H., et al. (2024) · JMIR Aging · doi:10.2196/52831 · Cited by 16 · see citing works
Self-citationRichter-Laskowska, M., Sobotnicka, E., & Bednorz, A. (2025) · Scientific Reports · doi:10.1038/s41598-025-90460-y · Cited by 16 · see citing works
Libon, D. J., Swenson, R., Price, C. C., Lamar, M., Cosentino, S., Bezdicek, O., Kling, M. A., Tobyne, S., Jannati, A., Banks, R., et al. (2024) · Frontiers in Psychology · doi:10.3389/fpsyg.2024.1415629 · Cited by 16 · see citing works
DeepRisk: a deep learning approach for genome-wide assessment of common disease risk
Peng, J., Bao, Z., Li, J., Han, R., Wang, Y., Han, L., Peng, J., Wang, T., Hao, J., Wei, Z., et al. (2024) · Fundamental Research · doi:10.1016/j.fmre.2024.02.015 · Cited by 15 · see citing works
Co-creating humanistic AI AgeTech to support dynamic care ecosystems: a preliminary guiding model
Hwang, A. S., Tannou, T., Nanthakumar, J., Cao, W., Chu, C. H., Zeytinoglu Atici, C., Scane, K., Yu, A., Tsang, W., Chan, J., et al. (2024) · The Gerontologist · doi:10.1093/geront/gnae093 · Cited by 13 · see citing works
PENSIEVE-AI a brief cognitive test to detect cognitive impairment across diverse literacy
Liew, T. M., Foo, J. Y. H., Yang, H., Tay, S. Y., Koay, W. I., Yip, K. F., Ting, S. K. S., Narasimhalu, K., Li, W., Tan, C., et al. (2025) · Nature Communications · doi:10.1038/s41467-025-58201-x · Cited by 11 · see citing works
Vonk, J. M. J., Morin, B. T., Pillai, J., Rosado Rolon, D., Bogley, R., Baquirin, D. P., Ezzes, Z., Tee, B. L., de Leon, J., Wauters, L., et al. (2025) · Neurology · doi:10.1212/wnl.0000000000213556 · Cited by 10 · see citing works
Self-citationEnhancing early detection of cognitive impairment in primary care with the TabCAT-BHA
Breithaupt, A. G., Sideman, A. B., Goode, C., Hill‐Sakurai, L., Premkumar, M., Scheffler, A. W., Chodos, A., Tsoy, E., Rankin, K. P., Kramer, J. H., et al. (2025) · Alzheimer's & Dementia · doi:10.1002/alz.70437 · Cited by 10 · see citing works
Self-citationLi, C., Zhang, L., Zhuo, Z., Su, F., Li, H., Xu, S., Liu, Y., Zhang, Z., Xie, Y., Yu, X., et al. (2023) · Theranostics · doi:10.7150/thno.79362 · Cited by 10 · see citing works
Zhou, M., Tang, A. S., Zhang, H., Xu, Z., Ke, A. M. C., Su, C., Huang, Y., Mantyh, W. G., Jaffee, M. S., Rankin, K. P., et al. (2025) · Journal of Biomedical Informatics · doi:10.1016/j.jbi.2025.104820 · Cited by 9 · see citing works
Self-citationEvaluating pathogenicity of variants of unknown significance in APP, PSEN1, and PSEN2
Marsh, J. A., Huang, G., Bowling, K., Renton, A. E., Ziegemeier, E., Ball, T., Pottier, C., Cruchaga, C., Day, G. S., Bateman, R. J., et al. (2025) · Neurotherapeutics · doi:10.1016/j.neurot.2025.e00527 · Cited by 9 · see citing works
Thies, T., Mallick, E., Tröger, J., Baykara, E., Mücke, D., & Barbe, M. T. (2025) · npj Parkinson's Disease · doi:10.1038/s41531-025-00959-4 · Cited by 8 · see citing works
West, M., Cheng, Y., He, Y., Leng, Y., Magdamo, C., Hyman, B. T., Dickson, J. R., Serrano-Pozo, A., Blacker, D., & Das, S. (2025) · JMIR Aging · doi:10.2196/65178 · Cited by 8 · see citing works
Li, Q., Koehler, S., Koenig, A., Dyrba, M., Mallik, E., Linz, N., Priller, J., Spruth, E., Altenstein, S., Wiltfang, J., et al. (2025) · Journal of Alzheimer’s Disease · doi:10.1177/13872877251359967 · Cited by 6 · see citing works
AI-enabled Parkinson's disease screening using smile videos
Adnan, T., Islam, M. S., Lee, S., Wasifur Rahman Chowdhury, E. M., Tithi, S. D., Noshin, K., Islam, M. R., Sarker, I., Rahman, M. S., Schneider, R. B., et al. (2025) · NEJM AI · doi:10.1056/aioa2400950 · Cited by 5 · see citing works
de la Fuente Garcia, S., Ritchie, C. W., & Luz, S. (2020) · Journal of Alzheimer's Disease · doi:10.3233/jad-200888 · Cited by 5 · see citing works
Xu, J., Yin, R., Huang, Y., Gao, H., Wu, Y., Guo, J., Smith, G. E., DeKosky, S. T., Wang, F., Guo, Y., et al. (2023) · doi:10.1101/2023.07.27.23293270 · Cited by 5 · see citing works
Sideman, A. B., Nguyen, H. Q., Langer-Gould, A., Lee, E. A., Borson, S., Shen, E., Tsoy, E., Macias, M., Goode, C., Rankin, K., et al. (2024) · BMC Primary Care · doi:10.1186/s12875-024-02544-9 · Cited by 4 · see citing works
Self-citationKetata, I., & Ellouz, E. (2025) · Journal of Alzheimer’s Disease · doi:10.1177/13872877251389145 · Cited by 3 · see citing works
Harrison, T. M., Landau, S. M., Baker, S. L., Boswell, M., & Jagust, W. J. (2024) · Alzheimer's & Dementia · doi:10.1002/alz.095834 · Cited by 1 · see citing works
Flagged first, then by citations
Research works printed but not linked: 1
- 2024, Health Data, Technology, and Interoperability: Certification Program Updates, Algorithm Transparency, and Information Sharing, 1192
Cited by
Sorted by citations
Shateri, A., & Rezaei, S. (2026) · Romanian Journal of Neurology · doi:10.37897/rjn.2026.3.7
Application and Research Progress of Digital Biomarkers Related to Alzheimer’s Disease
黄, 乐. (2026) · Open Journal of Natural Science · doi:10.12677/ojns.2026.145058
Guo, Y., Ying, H., Wang, W., Li, H., Jia, G., & Sulaiman, R. (2026) · Acta Neurologica Belgica · doi:10.1007/s13760-026-03167-w
Smeriglio, E., Imbesi, M., Xiao, J., Smeriglio, A., & Trombetta, D. (2026) · Frontiers in Aging Neuroscience · doi:10.3389/fnagi.2026.1924338
AI Use in Dementia Care Requires Engagement with Patient and Caregiver Perspectives
Friend, J., Black, V., & Chiong, W. (2026) · The American Journal of Bioethics · doi:10.1080/15265161.2026.2715910
- Authors
- Breithaupt, A. G., Tang, A., Paolillo, E. W., Bibars, M., Johnson, E. C. B., Saloner, R., Possin, K. L., Windon, C. C., Hill-Jarrett, T. G., Giorgio, J., Rauschecker, A. M., Kwon, H., Vonk, J. M. J., & Pinheiro-Chagas, P.
- Affiliations
- Emory University (US)
- University of California, San Francisco (US)
- Georgia Institute of Technology (US)
- Department of Neurology, Memory and Aging Center, Weill Institute for Neuroscience, University of California, San Francisco, California, United States
- University of California, Berkeley (US)
- Corresponding author
- Breithaupt, Andrew G.
- ORCID
- Venue
- Seminars in Neurology
- Publisher
- Georg Thieme Verlag KG
- Published
- 2025-11-28
- Volume
- 46
- Issue
- 01
- Pages
- 105-116
- ISSN
- 0271-8235 (ISSN-L)
- 1098-9021
- PMC
- PMC13078895
- OpenAlex
- W4416806068
- Semantic Scholar
- 83ec00cae11d3cc216a1415809113e16e8205f8d
- Cited by
- 6
- References
- 136
- Citations by year
- 2026: 6
- FWCI
- 1.06
- Type
- journal article
- Language
- en
- Open access
- Locations
- Seminars in Neurology (publishedVersion)
- PubMed Central (publishedVersion)
- PubMed (publishedVersion)
- Funders
- National Institute on Aging of the National Institutes of Health (K23AG093166)
- Topics
- Health Sciences › Medicine › Psychiatry and Mental health › Dementia and Cognitive Impairment Research
- Health Sciences › Medicine › Health Informatics › Artificial Intelligence in Healthcare and Education
- Health Sciences › Health Professions › Health Information Management › Artificial Intelligence in Healthcare
- Keywords
- artificial intelligence, Alzheimer's disease and related dementias, diagnosis, digital biomarkers, neuroimaging interpretation, chatbots, digital cognitive assessment, large language models, speech analysis, wearable devices, electronic health records, multi-omics, cognitive decline detection, AI implementation barriers, model bias, regulatory requirements
- MeSH terms
- Alzheimer Disease, Artificial Intelligence, Dementia, Humans, Neuroimaging
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Claims read under definition version 3 (extract-v3) on 2026-09-28.
Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-09-28.
Cited sources are checked by gpt-6-sol.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
