Insulin resistance (IR), a primary precursor to type 2 diabetes, is characterized by impaired insulin action in tissues1. However, diagnostic methods remain expensive and inaccessible, which hinders early intervention2,3. Here we present the WEAR-ME study, a large, remotely conducted study of IR (n = 1,165 participants; median body mass index (BMI) = 28 kg m−2, median age = 45 years, median haemoglobin A1c (HbA1c) = 5.4%) that uses time-series data from wearable devices and routine blood biomarkers to train deep neural networks against a ground-truth measure of IR (homeostatic model assessment of IR; HOMA-IR). Using a HOMA-IR cut-off of 2.9, our multimodal model achieved robust performance (area under the receiver operating characteristic curve (AUROC) = 0.80, sensitivity = 76%, specificity = 84%) with data from wearable devices, together with demographic and routine blood biomarker data. To enhance the use of time-series data from wearables, we fine-tuned a wearable foundation model (WFM) pretrained on 40 million hours of sensor data. In an independent validation cohort (n = 72), a model integrating WFM-derived representations with demographic data surpassed a demographics-only baseline (AUROC = 0.75 versus 0.66). Moreover, adding WFM-derived representations to a model with demographics, fasting glucose and a lipid panel substantially improved performance, compared with an identical model without data from wearables (AUROC = 0.88 versus 0.76). We integrate IR prediction into a large language model to contextualize the results and facilitate personalized recommendations. This work establishes a scalable, accessible framework for the early detection of metabolic risk, which could enable timely lifestyle interventions to prevent progression to type 2 diabetes. A machine-learning model that integrates data from wearable devices (such as smartwatches) with blood biomarkers and demographic data can predict whether someone has insulin resistance, enabling timely lifestyle interventions to prevent progression to type 2 diabetes.

Evaluating the reasoning ability of language models (LMs) is complicated by their extensive parametric world knowledge, where benchmark performance often reflects factual recall rather than genuine reasoning. Existing datasets and approaches (e.g., temporal filtering, paraphrasing, adversarial substitution) cannot cleanly separate the two. We present SynthWorlds, a framework that disentangles task reasoning complexity from factual knowledge. In SynthWorlds, we construct parallel corpora representing two worlds with identical interconnected structure: a real-mapped world, where models may exploit parametric knowledge, and a synthetic-mapped world, where such knowledge is meaningless. On top of these corpora, we design two mirrored tasks as case studies: multi-hop question answering and page navigation, which maintain equal reasoning difficulty across worlds. Experiments in parametric-only (e.g., closed-book QA) and knowledge-augmented (e.g., retrieval-augmented) LM settings reveal a persistent knowledge advantage gap, defined as the performance boost models gain from memorized parametric world knowledge. Knowledge acquisition and integration mechanisms reduce but do not eliminate this gap, highlighting opportunities for system improvements. Fully automatic and scalable, SynthWorlds provides a controlled environment for evaluating LMs in ways that were previously challenging, enabling precise and testable comparisons of reasoning and memorization.

While wearables, such as Fitbit devices, are specifically designed to measure physical activity and physiological functioning, they have less consumer uptake than smartphones, which contain many similar sensors that can also measure physical activity (although not cardiac activity). This study’s objective was to assess the agreement between physical and cardiac activity measures collected from Fitbit wearable devices and physical activity measures derived from the naturalistic use of smartphones over 28 days in a large, diverse sample (n = 10,085). Agreement between smartphone and Fitbit features was quantified via intraclass correlation coefficients (ICC; two-way random effects, absolute agreement) computed within participants across aligned days. Although the level of agreement between smartphone and Fitbit metrics was statistically significant in all comparisons examined, the effect sizes ranged from ‘poor’ to ‘good’ across metrics and individuals. The strongest agreement was observed for measurements of smartphone activity categories with strongly related categories derived from Fitbit exercise data. Some effects were stronger when data was filtered to focus on participants with higher activity levels in both smartphone and Fitbit usage. Although variations in data agreement and missingness highlight the need for careful interpretation of results, these findings do provide support for the use of smartphone sensors to measure certain aspects of physical activity at the population level when the scalability of the measurement is critical.

Cardiovascular-Kidney-Metabolic (CKM) syndrome represents a growing public health crisis, yet its subclinical heterogeneity remains underexplored. Early detection of physiological deviation is critical for preventing irreversible organ damage. Here, we characterize the prevalence and interplay of CKM impairment in a US cohort (N=819) by integrating continuous wearable data with clinical biomarkers. We assessed cardiovascular, kidney via clinical biomarkers, namely Chol/HDL, eGFR, as well as metabolic health risk through Homeostatic Model Assessment of Insulin Resistance (HOMA-IR). We show that while metabolic and cardiovascular disruptions are significantly associated (r=0.26, p<0.001), early-stage kidney impairment manifests independently. Utilizing a normalized deviance score, we identified significant health impairments in 29.0

Language models excel at diagnostic assessments on curated medical case-studies and vignettes, performing on par with, or better than, clinical professionals. However, existing studies focus on complex scenarios with rich context making it difficult to draw conclusions about how these systems perform for patients reporting symptoms in everyday life. We deployed SymptomAI, a set of conversational AI agents for end-to-end patient interviewing and differential diagnosis (DDx), via the Fitbit app in a study that randomized participants (N=13,917) to interact with five AI agents. This corpus captures diverse communication and a realistic distribution of illnesses from a real world population. A subset of 1,228 participants reported a clinician-provided diagnosis, and 517 of these were further evaluated by a panel of clinicians during over 250 hours of annotation. SymptomAI DDx were significantly more accurate (OR = 2.56, p < 0.001) than those from independent clinicians given the same dialogue in a blinded randomized comparison. Moreover, agentic strategies which conduct a dedicated symptom interview that elicit additional symptom information before providing a diagnosis, perform substantially better than baseline, user-guided conversations (p < 0.001). An auxiliary analysis on 1,509 conversations from a general US population panel validated that these results generalize beyond wearable device users. We used SymptomAI diagnoses as labels for all 13,917 participants to analyze over 500,000 days of wearable metrics across nearly 400 unique conditions. We identified strong associations between acute infections and physiological shifts (e.g., OR > 7 for influenza). While limited by self-reported ground truth, these results demonstrate the benefits of a dedicated and complete symptom interview compared to a user-guided symptom discussion, which is the default of most consumer LLMs.

The LLM-empowered personal health agents with user health (sensor) metrics have offered a promising pathway to alleviate global disparities in healthcare access. However, large-scale clinical deployment remains constrained by an open-ended evaluation bottleneck: physician annotation is reliable but costly and unscalable, while LLM-as-a-judge evaluators are scalable but subjective, inconsistent, and sometimes clinically misaligned. We introduce RubricsTree, a scalable evaluation framework with an expert-aligned hierarchical taxonomy of over 100 atomic, clinically-verifiable Boolean rubrics, evolving from the insights of 4,000 real user queries through an iterative human-in-the-loop curation protocol with an expertise panel led by an experienced physician. A context-aware adaptive router activates only the relevant auto-weighted rubric subset per query, providing the throughput needed for scalable evaluation with expert-aligned quality. Through a systematic meta-evaluation, we show that RubricsTree (i) substantially exceeds a strong large-scale evaluation baseline in expert alignment on challenging open-ended queries; (ii) reliably penalizes contextually degraded responses; and (iii) when used as structured instructions, text feedback, or training rewards for performance optimization, yields up to 66

Patient-managed Personal Health Records (PHRs) promises to empower patients to better understand their health; but information in the record is complex, potentially hindering insights. In this study, we assess the potential of large language models (LLMs, Gemini 3.0 Flash) to provide helpful answers to user health queries, when provided clinical data from PHRs as context. A total of 2,257 user queries were drawn from 3 different distributions to represent patient questions: shorter web search queries, longer questions derived from templates of chatbot conversations, and questions patients asked to their healthcare team (patient calls). Queries were matched with de-identified PHRs (from a pool of 1,945). Gemini responses were generated (1) without PHR context; (2) with a basic summary of demographics, conditions, and medications; (3) with full, extensive clinical notes. For evaluation, we leveraged an existing rating framework (SHARP), and developed a new framework for specific error modes when interpreting PHRs. Evaluation was performed using autoraters for the full set, and with clinician ratings for a subset (n=95), with both sets of raters knowing the full PHR context. We see significant improvements in the helpfulness of answers to all question types with PHR data (p < 0.001, paired t-test). We also observe potential gains in safety, accuracy, relevance and personalization of answers. Our PHR evaluation framework further identifies gaps in LLM understanding of particular aspects of complex PHRs, such as temporal disorientation, and rare but meaningful confabulations. These results suggest potential for PHR data to help people with a wide range of user needs; and provide a framework for monitoring for gaps in LLM answers based on PHR context. This study motivates further work to assess and realize potential benefits to users from understanding their health records.

Large Language Models (LLMs) have shown strong capability in interpreting multimodal data but remain limited in their ability to natively handle time-series data. Addressing this limitation could enable the translation of longitudinal and wearable sensing data into actionable insights and patient-facing digital health applications. We propose OpenTSLM, a family of Time Series Language Models that integrate time-series as a native modality into pretrained LLMs, enabling natural-language prompting and reasoning over multiple time-series. We implement two OpenTSLM variants based on soft prompting (OpenTSLM-SP) and cross-attention (OpenTSLM-Flamingo). To conduct comprehensive experiments on reasoning over medical text and time-series, we introduce three chain of thought (CoT) datasets: HAR-CoT (human activity recognition), Sleep-CoT (sleep staging), and ECG-QA-CoT (electrocardiogram question answering). Across tasks, OpenTSLM models consistently outperform baselines. OpenTSLMs with time-series encoders trained from scratch achieve 69.88% in sleep staging and 65.44% in HAR, while OpenTSLM combined with time series foundation models (TSFMs) achieve 68.33% and 67.64%, compared to 9.05% and 60.44% for fine-tuned text-only baselines. Additionally, we conduct expert evaluations with cardiologists, which show that OpenTSLMs exhibit strong reasoning capabilities and temporal understanding on raw sensor data for ECG-QA. We further show that OpenTSLM-Flamingo models scale better in memory as the number and length of time-series increase. To facilitate further research, we release all code, datasets, and models as open-source resources.

Disruptions to biobehavioral rhythms are a transdiagnostic feature of mental health conditions, yet the clinical utility of smartphone-derived rhythm metrics remains poorly characterized across different mental health outcomes. Here we quantified biobehavioral rhythms from three smartphone interaction patterns—phone unlocks, general application usage, and social media application usage—across 27 days in 8,341 adults from the Google Digital Well-Being Study. Using spectral analysis and non-parametric rhythm metrics, we examined associations with depression (PHQ-8), anxiety (GAD-7), and sleep-related impairment (PROMIS SRI). Smartphone usage exhibited robust rhythmic structure, with rhythm metrics showing stronger correlations with depression (mean top-five |r| = 0.238) and sleep impairment (|r| = 0.214) than anxiety (|r| = 0.182). XGBoost models combining rhythm features with demographic variables achieved the best classification performance for depression (F1 = 0.53) and regression performance for sleep impairment, while rhythm features provided limited incremental value for anxiety prediction. Relative Amplitude and least-active-5-hour mean (L5) of phone unlocks were the most consistently important features across all outcomes, with Intradaily Variability of social media usage also contributing to depression and anxiety prediction. These findings support the theoretical prediction, grounded in the Social Zeitgeber hypothesis and circadian biology, that biobehavioral rhythm disruption is more closely tied to mood and sleep regulation than to anxiety, and suggest that passively sensed smartphone data may warrant further investigation as a component of mental health assessment, pending longitudinal validation.

Deriving personalized insights from popular wearable trackers requires complex numerical reasoning that challenges standard LLMs, necessitating tool-based approaches like code generation. Large language model (LLM) agents present a promising yet largely untapped solution for this analysis at scale. We introduce the Personal Health Insights Agent (PHIA), a system leveraging multistep reasoning with code generation and information retrieval to analyze and interpret behavioral health data. To test its capabilities, we create and share two benchmark datasets with over 4000 health insights questions. A 650-hour human expert evaluation shows that PHIA significantly outperforms a strong code generation baseline, achieving 84

The pursuit of artificial general intelligence necessitates robust methods for evaluating the cognitive capabilities of models beyond narrow task performance. Here, we introduce a psychometric framework to assess the cognitive profiles of generative AI, comparing them to human norms and tracking their evolution across generations. Initial evaluation of leading multimodal models using tasks adapted from the Wechsler Adult Intelligence Scale revealed a profoundly uneven cognitive architecture: near-ceiling performance in verbal comprehension and working memory (>98^th percentile) contrasted with near-floor performance in perceptual reasoning (<1^st percentile). To track developmental trajectories beyond human-normed limits, we developed the Artificial Intelligence Quotient (AIQ) Benchmark and applied it to six generations and two model families, revealing significant but asymmetric performance gains. Notably, we uncovered a sharp dissociation between modalities; abstract quantitative reasoning matured far more rapidly when presented linguistically compared to a visually analogous format, indicating an architectural bias towards language-based symbolic manipulation. While abstract visual reasoning improved, visual-perceptual organization remained largely stagnant. Collectively, these findings demonstrate that the cognitive abilities of generative models are evolving unevenly, suggesting that scaling and optimization approaches to AGI development alone may be insufficient to overcome fundamental architectural limitations in achieving balanced, human-like general intelligence.

Wearable sensors enable the continuous acquisition of high-resolution physiological waveforms, such as photoplethysmography and accelerometry, under free-living conditions. However, inferring health-related phenotypes from these signals presents significant challenges due to high sampling frequencies, multimodal dependencies, and extreme sequence lengths (e.g., weeks of recordings), compounded by a scarcity of ground-truth labels. To address these challenges, existing self-supervised learning (SSL) methodologies typically follow two paradigms: (1) learning rich morphological representations from short waveform segments while collapsing longitudinal dynamics through simple aggregation, or (2) modeling behavioral patterns from coarse, hand-crafted features (e.g. heart rate, step counts) spanning longer horizons but foregoing subtle, predictive signatures in raw waveforms. To bridge this gap, we propose WavesFM, a foundation model utilizing a two-stage SSL framework for longitudinal physiological data. Specifically, we decompose the learning problem into two stages: first, a segment-level encoder is pretrained to extract local embeddings from short waveforms; subsequently, a temporal encoder is trained to model the sequence of these embeddings across a multi-day horizon. This hierarchical approach overcomes the computational complexity of high-resolution, long-sequence data, allowing the overall model to capture both local signal semantics and the complex circadian and inter-day variations governing physiological dynamics. Pretrained on over 6.8M hours (N=324k individuals) of recordings for the first stage and 5.3M hours (N=10k) for the second stage, WavesFM demonstrates superior performance across 58 diverse tasks spanning demographics, lifestyle, health conditions, and medications.

Language models are remarkably capable at medical question answering, in some cases surpassing the accuracy of general physicians. However, answering questions about wearable health data remains challenging and understudied, as these ubiquitous sensors produce continuous, high-dimensional, and longitudinal data, which is non-trivial to align with text-centric distributions in LLM pretraining. The diversity of sensor modalities and user intents cannot be effectively handled by a fixed reasoning workflow or a single pretrained foundation model. To address these challenges, we propose WEQA, a query-adaptive agent framework that unifies LLM reasoning with specialized wearable analytical and modeling tools. An LLM controller is employed to synthesize execution plans and dynamically route each query to the appropriate combination of sensor analysis and pretrained models, and perform grounded response auditing with external knowledge. We also curate a benchmark spanning four open wearable datasets comprising analytic and predictive tasks in three different health domains. Experiments show that our framework is 24

Wearable devices generate continuous physiological and behavioral data, but converting these signals into clinically reviewable biomarker hypotheses remains labor-intensive. We introduce CoDaS, an AI co-data-scientist that integrates multi-agent hypothesis generation, deterministic statistical analysis, adversarial validation and literature-grounded interpretation under human oversight. Across three wearable cohorts comprising 9,279 participant-observations, CoDaS prioritized candidate associations for mental-health and metabolic endpoints after internal checks for replication, stability, robustness and leakage. The system identified related circadian-instability signals associated with depression, including sleep-duration variability in DWB (ρ = 0.252, p < 0.001) and sleep-onset variability in GLOBEM (ρ = 0.126, p < 0.001), and derived a wearable cardiovascular-fitness index associated with insulin resistance (steps/resting heart rate; ρ = -0.374, p < 0.001). Adding these features to demographic models produced modest gains (ΔR^2 = 0.040 for depression, 0.021 for insulin resistance). In a 12-clinician review totaling approximately 25 active hours, clinician validity judgments aligned with CoDaS confidence tiers (ρ = 0.67, p = 0.005), whereas added clinical value and confidence to act were rated lower. CoDaS supports traceable, hypothesis-generating prioritization of wearable candidate biomarkers.

Agent systems often decompose a task across multiple roles, but these roles are typically specified by prompts rather than enforced by access controls. Without enforcement, a team pass rate can mask whether agents actually coordinated or whether one role effectively did another role's work. We present TeamBench, a benchmark with 851 task templates and 931 seeded instances for evaluating agent coordination under operating system-enforced role separation. TeamBench separates specification access, workspace editing, and final certification across Planner, Executor, and Verifier roles, so that no role can read the full requirements, modify the workspace, and certify the final answer. Prompt-only and sandbox-enforced teams reach statistically indistinguishable pass rates, but prompt-only runs produce 3.6 times more cases where the verifier attempts to edit the executor's code. Verifiers approve 49

While ubiquitous wearable sensors capture a wealth of behavioral and physiological information, effectively transforming these signals into personalized health insights is challenging. Specifically, converting low-level sensor data into representations capable of characterizing higher-level states is difficult due to high phenotypic diversity and variation in individual baseline health, physiology, and lifestyle factors. Moreover, collecting wearable data paired with health outcome annotations is laborious and expensive, and retrospective annotation remains practically unfeasible, contributing to a scarcity of data with high-quality labels. To overcome these limitations, we propose a foundation model for wearable health that is pretrained on more than one trillion minutes of unlabeled sensor signals drawn from a large cohort of five million participants. We demonstrate that the joint scaling of model capacity and pretraining data volume leads to systematic improvements in performance, as evaluated on a diverse set of 35 health prediction tasks, spanning cardiovascular, metabolic, sleep, and mental health, as well as lifestyle choices and demographic factors. We find that this population scale representation unlocks label-efficient few-shot learning and generative capabilities for robust daily metric estimation. To further leverage this learned representation, we deploy a classroom of LLM agents to autonomously search the space of downstream predictive heads built on the model embeddings, showing broad performance improvements that increase with LLM model capacity. Finally, we show how integrating these downstream predictors into a Personal Health Agent can support model responses that are more relevant, contextually aware, and safe, and we validate this via 1,860 ratings from a cohort of clinicians.

The rise of large language models (LLMs) has shifted time series analysis from narrow analytics to general-purpose reasoning. Yet, existing benchmarks cover only a small set of health time series modalities and tasks, failing to reflect the diverse domains and extensive temporal dependencies inherent in real-world physiological modeling. To bridge these gaps, we introduce HEARTS (Health Reasoning over Time Series), a unified benchmark for evaluating hierarchical reasoning capabilities of LLMs over general health time series. HEARTS integrates 16 real-world datasets across 12 health domains and 20 signal modalities, and defines a comprehensive taxonomy of 110 tasks grouped into four core capabilities: Perception, Inference, Generation, and Deduction. Evaluating 14 state-of-the-art LLMs on more than 20K test samples reveals intriguing findings. First, LLMs substantially underperform specialized models, and their performance is only weakly related to general reasoning scores. Moreover, LLMs often rely on simple heuristics and struggle with multi-step temporal reasoning. Finally, performance declines with increasing temporal complexity, with similar failure modes within model families, indicating that scaling alone is insufficient. By making these gaps measurable, HEARTS provides a standardized testbed and living benchmark for developing next-generation LLM agents capable of reasoning over diverse health signals.

While multimodal large language models offer a promising solution to the "black box" nature of health AI by generating interpretable reasoning traces, verifying the validity of these traces remains a critical challenge. Existing evaluation methods are either unscalable, relying on manual clinician review, or superficial, utilizing proxy metrics (e.g. QA) that fail to capture the semantic correctness of clinical logic. In this work, we introduce a reproducible framework for evaluating reasoning in ECG signals. We propose decomposing reasoning into two distinct, components: (i) Perception, the accurate identification of patterns within the raw signal, and (ii) Deduction, the logical application of domain knowledge to those patterns. To evaluate Perception, we employ an agentic framework that generates code to empirically verify the temporal structures described in the reasoning trace. To evaluate Deduction, we measure the alignment of the model's logic against a structured database of established clinical criteria in a retrieval-based approach. This dual-verification method enables the scalable assessment of "true" reasoning capabilities.

Large language models (LLMs) have emerged as powerful tools for analyzing and interpreting complex datasets. Recent studies demonstrate their potential to generate useful, personalized responses when provided with patient-specific health information that encompasses lifestyle, biomarkers, and context. As LLM-driven health applications are increasingly adopted, rigorous and efficient one-sided evaluation methodologies are crucial to ensure response quality across multiple dimensions, including accuracy, personalization, relevance and safety. However, current evaluation practices, particularly for open-ended text responses, heavily rely on human experts. This approach introduces human factors (perspectives, potential biases, inconsistencies) and is often cost-prohibitive, labor-intensive, and hinders scalability, especially in complex domains like healthcare where response assessment necessitates domain expertise and considers multifaceted patient data, which is often nuanced and diverse. In this work, we introduce Adaptive Precise Boolean rubrics: an evaluation framework that aims to streamline human and automated evaluation of open-ended questions by identifying critical gaps in model responses using a minimal set of targeted rubric questions. Our approach is based on recent work in more general evaluation settings that contrasts a smaller set of complex evaluation targets with a larger set of more precise, granular targets answerable with simple Boolean responses. We validate this approach in metabolic health, a domain encompassing diabetes, cardiovascular disease, and obesity. Our results demonstrate that Adaptive Precise Boolean rubrics yield substantially higher inter-rater agreement among both expert and non-expert human evaluators, as well as in automated assessments, compared to traditional Likert scales, while requiring approximately half the evaluation time of Likert-based methods. This enhanced efficiency and scalability, particularly through automated evaluation and non-expert contributions, paves the way for more extensive and cost-effective evaluation of LLMs in health.

Resting heart rate (RHR) is a key biomarker of cardiovascular health and mortality1–3, but passively tracking it longitudinally generally requires a wearable device, limiting its availability. Here we present passive heart-rate monitoring (PHRM), a deep-learning system that uses facial video-based photoplethysmography for passive measurements of heart rate (HR) and RHR during everyday smartphone interactions. Our system was developed using 192,353 videos from 485 participants and validated on 162,546 videos from 211 participants in laboratory and free-living conditions, representing, to our knowledge, the largest validation study of its kind. PHRM outperformed state-of-the-art methods on our benchmarks. Compared with reference electrocardiograms, PHRM achieved a mean absolute percentage error (MAPE) lower than 10% for HR measurements across three skin-tone groups of light, medium and dark pigmentation, meeting industry accuracy standards; MAPE for each skin-tone group was non-inferior versus the others. Daily RHR measured by PHRM had a mean absolute error of less than five beats per minute, compared with a wearable HR tracker, and was associated with known risk factors for cardiovascular disease. These results highlight the potential of smartphones for enabling passive and equitable monitoring of heart health. To facilitate further research, we publicly release a large, annotated smartphone video dataset along with a pre-trained HR model. A machine-learning model that uses smartphone cameras to measure heart rate in the background during normal daily phone use and subsequently estimate resting heart rate could make it easier for people to monitor heart health.