Introduction: Mild cognitive impairment (MCI) affects an estimated 15 to 20 percent of adults over 65 and precedes Alzheimer’s disease in up to half of cases within five years, yet remains chronically underdiagnosed in primary care settings where brief, accessible screening tools are most needed. Speech produced during structured picture description tasks offers a non-invasive, scalable signal for early detection, but the field lacks open, reproducible pipelines validated in prospective clinical cohorts. We developed SPEECH-MARC, an open-source multimodal framework that extracts acoustic, linguistic, and scene-level contextual features from picture description speech to classify MoCA-defined cognitive status. Methods: In a prospective cohort of 127 community-dwelling older adults (cognitively unimpaired, CU: n=61; MCI: n=66) recruited through the UC San Diego Brain Health Virtual Reality Study, we extracted 192 candidate features spanning eGeMAPS acoustic parameters, lexical diversity indices, syntactic complexity measures, and an automated scoring framework for Information Content Units (ICUs) and Areas of Interest (AOIs) adapted for a culturally updated picture description stimulus. Classification used stratified 5-fold cross-validation with feature selection (k=20 features per fold) performed strictly within each training fold, maintaining an effective 1:5 feature-to-training-sample ratio appropriate for small-cohort machine learning. Results: Combined multimodal models achieved AUC=0.662 (95% CI: 0.563 to 0.755), outperforming acoustic-only (AUC=0.619), linguistic-only (AUC=0.603), and ICU/AOI-only (AUC=0.629) approaches. The strongest predictors were spectral flux variability, lexical diversity, and peripheral scene coverage. Mean word length showed the strongest continuous association with MoCA total score (r=0.31, p=0.0003). Multiple lexical and scene-description features showed FDR-corrected associations with category fluency, digit symbol, and design fluency scores measured independently in the same visit. Conclusion: Multimodal speech features extracted from a single picture description recording provide clinically interpretable discrimination of MoCA-defined cognitive status. Findings are cross-sectional and require longitudinal validation, but the feature patterns identified are consistent with known profiles of early cognitive decline. The framework is freely available as a Python package with a web-based interface requiring no programming expertise, intended to support research replication and pilot deployment in settings where access to formal neuropsychological testing is limited.
Background:Tertiary cancer centers were built for depth and nuance rather than speed, yet patients increasingly judge access by how quickly and with how little friction they can reach a clinician. This mismatch is widening as cancer demand rises and the workforce fails to scale, and added physical capacity alone cannot close it. Durable gains in access, quality, and cost instead require safe, evidence-based capabilities beginning in the patient’s home. Summary:This review outlines a personal health care (PHC) model in oncology that relocates the early work of evaluation from the clinic into the home and examines where the model holds and where it does not. Key Messages:The model is organized as five capability layers: sensing, deciding and acting, monitoring with governance and safety, escalation, and external integration. Patient-owned sensing at home, such as validated at-home urine testing and wearable smart devices, feeds clinician-supervised decision support, while the first in-person visit is preserved for treatment initiation. The clinician remains in the loop across all five layers, as a clinically trained reviewer can draw more decision-relevant information from a brief virtual encounter than an untrained intermediary following a fixed algorithm. For at-home diagnostics and artificial intelligence (AI) tools, the appropriate comparator is not the leading specialists but the large share of patients who currently reach no specialist at all. As the cost of cancer care is driven by the workforce rather than by buildings or equipment, relocating an encounter to the home does not by itself reduce cost, and only a lower-cost workforce meaningfully changes that trajectory.
US health care is not giving patients what they need. Over 100 million Americans lack basic primary care. Workforce shortages of more than 80,000 physicians and 100,000 nurses are projected by the late 2030s. Wait times for new primary and specialty care appointments average about a month and continue to rise. The country spends nearly 18% of GDP on health care while wasting roughly a quarter of it. Provider-controlled solutions, including the most heavily funded AI tools, aim to make clinicians marginally more efficient. This approach, although exciting, cannot close gaps of this magnitude. Personal Health Care is a conceptual system architecture focused on what patients can do for themselves between, or instead of, clinical encounters. It is condition-agnostic across acute and chronic care, AI-powered, and designed for use by patients across the entire care pathway without requiring a clinician in the loop. Personal Health Care does not replace physicians, nurses, hospitals, or clinics. It complements them by eliminating fragmentation in access, while escalating higher-risk events to traditional care with full context. The Personal Health Care system rests on five capabilities. Sensing captures the right data at the right moments. Deciding and Acting reasons over that data with stated confidence and takes action within validated scope. Monitoring, Governance, and Safety validates performance, surfaces drift, and learns at both population and individual levels. Escalation routes higher-risk scenarios to clinical care. External Integration connects the system to existing health system records, prescribing, and clinicians. These five capabilities map to the Institute of Medicine's Six Aims for healthcare quality and the ACGME's Six Core Competencies for physician training. Examples already in market (direct-to-consumer medications, behavioral health apps, AI-enabled home diagnostics, continuous physiologic monitoring, and emerging diagnostic-reasoning systems) are pieces and not yet a system. Personal Health Care is the architecture that turns these pieces into a coherent, patient-controlled whole. This Digital Biomarkers series explores the realities and possibilities of each capability.
Introduction: Vision-based digital biomarkers have emerged as promising tools for objectively (and possibly remotely) assessing Parkinson’s disease (PD) motor signs, addressing inherent limitations of traditional clinical scales like the Unified Parkinson’s Disease Rating Scale (UPDRS). However, real-world deployment is hindered by variability in video quality, particularly in uncontrolled home environments. This study aims to quantify the impact of video quality parameters on the accuracy of human pose estimation (HPE) and downstream clinical assessments, including finger-tapping event detection and UPDRS scoring. Methods: We analyzed videos of PD patients performing finger-tapping tasks across two settings: high-resolution recordings collected in controlled clinical environments (n = 227: “clinical dataset”) and patient-recorded videos from home settings (n = 88: “home-recorded dataset”). To evaluate the effect of video quality on assessment accuracy, we introduced systematic degradations and assessed key parameters, including resolution, frame rate, lighting conditions, and hand visibility. Performance was measured using mean per joint position error, percentage of correct keypoints, and estimated landmark failed frames. Results: Low frame rates and inadequate hand coverage within the frame significantly reduced the accuracy of HPE-based assessments. In contrast, video resolution had a less pronounced effect than expected. Conclusion: Frame rate and proper visibility of body parts are more critical than resolution for reliable at-home motor signs evaluation in PD. We establish practical thresholds for video quality in remote PD assessments and provide actionable guidelines for optimizing remote monitoring systems.
Introduction: Wearable devices such as smartphones and smartwatches collect large volumes of biometric data on patients, yet their utility for admitting inpatient providers remains largely unexplored. Methods: This exploratory pilot screened patients admitted to Stanford University Hospital for use of smartphones or smartwatches. For patients reporting wearable use, preadmission biometric data were collected and analyzed in relation to inpatient outcomes, including admitting diagnosis and discharge disposition. Results: Among 137 screened patients, 27 reported smartphone or smartwatch use, and 16 had adequate preadmission wearable data for analysis. Step count was the most consistently recorded metric. Among those analyzed, a steep decline in step count prior to hospitalization was associated with discharge to a skilled nursing facility or need for home health services. Patients with cardiac diagnoses also exhibited more pronounced declines in step count compared to those with noncardiac conditions. Conclusion: Despite low rates of wearable usage, our findings suggest that trends in prehospital biometric data may offer early insights into patient functional status and discharge needs. As wearable adoption increases, such data could enhance inpatient decision-making and discharge planning.
