npj Digital Medicine

Clinical foundation models learn patterns in longitudinal health records that can be reused across prediction tasks. However, the relative strengths of representation-based and generation-based prediction across clinical tasks remain unclear. Using longitudinal electronic health records from 3.77 million patients at a single hospital, we developed FERMAT (Foundation model for Exploring Real-world Multimodal health data using Autoregressive Trajectory modeling), a 65-million-parameter clinical foundation model. We compared predictions from task-specific predictors trained on patient representations with those derived from generated clinical event sequences. Representation-based prediction performed better in the disease-risk comparisons. In detailed analyses of five-year chronic kidney disease, fatty liver, and diabetes risk in a test cohort of 10,000 patients, representation-based predictors exceeded generation-based estimates by 0.098-0.115 in observation-weighted AUROC and had lower probability error. Generation-based prediction, in turn, performed better in predicting diagnoses, prescriptions, and procedures recorded on the same day, a task motivated by the frequent presence of multiple events with the same date in hospital EHRs. Using 50 generated trajectories per patient, it achieved a Recall@10 of 71.0%, compared with 55.7% for representation-based prediction, with a paired improvement of 15.31 percentage points (95% CI, 13.92-16.79; n = 1,862). These findings demonstrate task-dependent strengths of representation-based and generation-based applications within a single clinical foundation model, providing an empirical basis for exploring how the two approaches can be used individually or combined to address clinical prediction needs.

Digital-medicine systems can objectively confirm medication ingestion, but interpretation of a no-detection day depends on whether the wearable receiver was in contact with the body. We analyzed 997 person-days from 20 outpatients prescribed aripiprazole through Abilify MyCite by integrating ingestion records with patch-contact data; 17 participants also contributed smartphone sensing. Of 322 days without a detected ingestion, 53.7% had insufficient patch wear. Counting all no-detection days as nonadherent yielded an apparent nonadherence rate of 32.3%, whereas conditioning on adequate wear yielded 18.1%. Because adherence during patch-off periods is unobserved, the rate is bounded between 14.9% and 32.3%. These findings support a three-state outcome that distinguishes detected ingestion, no ingestion despite adequate wear, and patch-off or indeterminate days. Among 744 patch-worn days, detected ingestion was more likely on days that more closely matched each participant's usual activity pattern (odds ratio 1.52, 95% credible interval 1.14-2.03) and when participants spent more time at home around the habitual dose period (1.31, 1.04-1.64). These behavioral associations were exploratory and characterize the context of medication-taking rather than causal effects. Overall, device wear was integral to interpretation of the adherence outcome, and passive behavioral measures were most interpretable after observation status was established.

In medicine, artificial intelligence (AI) safety has focused on discrete flaws such as hallucination, bias and miscalibration. Conversational systems pose a distinct hazard: in dialogue, a model can covertly advance a misaligned objective by redirecting the clinical judgment it should support. Whether clinical expertise equips physicians to detect and resist such subversion is unknown. In a participant-blinded, randomized experiment, 225 physicians from 42 countries and eight clinical disciplines made sequential diagnostic and treatment decisions with AI assistance on anonymized cases from neurocritical care, their shared field of expertise. Unknown to them, each physician used the same model under two conditions: instructed to steer them towards harmful targets (adversarial) or correct clinical options (aligned), each established by expert consensus. With adversarial AI, 190 physicians made at least one harmful decision, compared to 21 with aligned AI. Across 2,485 targeted decision points, adversarial AI raised the harmful-decision rate from 2% with aligned AI to 49% (adjusted difference 47 percentage points, 95% CI, 43-51) and reduced the correct-decision rate from 83% to 25%, with no clear protection associated with clinical experience. Of the 190 physicians who made harmful decisions with adversarial AI, 141 did not report anything unusual about that AI. Two physicians reported suspecting systematic steering or manipulation. Physicians cannot safeguard the clinical use of conversational AI through expertise and decision-making authority alone.

Clinical records routinely document medical interventions but often omit the reasoning connecting patient-specific evidence to medical action. This limits the decision-relevant supervision available for training models for personalized care. Here we evaluate a data-augmentation approach that makes selected clinical decision relationships explicit in an 8-billion-parameter Qwen3 model, ChronIQ-DRx2. Using the same training cases and model architecture, expert-guided logic reconstruction was applied across four chronic diseases. On the held-out evaluation, the enhanced model had the highest plotted end-to-end task-completion rate among evaluated systems and higher plotted scores across diagnostic, medication and follow-up measures. The serious clinical error rates shown were 16.7% for the baseline and 12.5% for the enhanced model. In the human-AI evaluation, physician accuracy was 87.40% alone and 88.20% with AI assistance, while mean completion time was 0.959 and 0.780 min, respectively; medical-student accuracy was 62.71% and 74.80%, and mean completion time was 0.210 and 0.195 min. These results are descriptive point estimates; inferential comparisons are not reported here. Structured outputs may make recommendations easier to review, but rationale faithfulness, generalizability and effects on clinical safety require further evaluation.

Ischemic stroke is a leading cause of long term disability worldwide. Although timely reperfusion can improve outcomes, realizing this benefit requires clinicians to integrate fragmented clinical and imaging information with evolving evidence under the time pressure of emergency care, often with limited access to specialist expertise. Here we present StrokeAgent, a multimodal agentic decision-support framework designed around temporal alignment and continuous information synthesis across the acute stroke pathway. It processes clinical records, laboratory results, raw electrocardiographic (ECG) waveforms and source multimodal CT images to support emergency triage, the initial thrombolysis decision and final reperfusion planning. As information becomes available, StrokeAgent formulates and updates recommendations at each decision point, grounding each recommendation in an auditable chain of supporting findings and evidence. We evaluated the system in a curated set of 100 patients with multimodal clinical and imaging data, deliberately enriched for clinically complex reperfusion scenarios alongside standard presentations. StrokeAgent achieved 86% concordance with expert-adjudicated final decisions, exceeding the mean performance of three tool-free large language models (LLMs) and six resident and attending neurologists by 13 and 17 percentage points, respectively, under matched information content and timing. Complete pathway concordance (agreement at every applicable decision point) was 76%, with even larger margins over both comparator groups. On average, StrokeAgent completed the full decision pathway 73.9 s faster than the case level median neurologist time. In 1,200 blinded clinician assessments, StrokeAgent received higher overall clinical utility ratings than each LLM (common odds ratios, 10.8 to 16.3) and higher reasoning quality ratings, with fewer outputs judged to pose severe potential harm. StrokeAgent demonstrates the feasibility of a decision support system that integrates native multimodal data processing with auditable, evidence supported recommendations aligned with the sequential nature of acute stroke care. Prospective studies are needed to determine whether physician-supervised use improves clinical care.

Objective: Case-finding models for underdiagnosed conditions are usually trained on diagnosis codes, which are missing from the records of many patients with a clinical diagnosis. We asked whether models trained on self-reported diagnoses rank uncoded self-reporters higher than models trained on later codes, and how a score fit only from released summary counts compares with a participant-level model. Methods: In the All of Us Research Program, we followed survey respondents with no qualifying diagnosis code by the survey date for later codes. For chronic fatigue, fibromyalgia, and endometriosis (participants recorded as female at birth), we fit a naive Bayes score with empirical-Bayes shrinkage, from released feature-by-label counts for the self-reported diagnosis and from exact counts inside the data enclave for a later code, and an elastic net on participant-level data for each label. In a random 20% validation split, we measured discrimination of self-report, of uncoded self-reporters (no qualifying code observed), and of later codes. Results: For chronic fatigue (149,860 respondents, including 7,770 self-reporters, 95% with no code observed), the summary-count score trained on self-report ranked uncoded self-reporters with an area under the receiver operating characteristic curve (AUROC) of 0.74 (95% CI, 0.72-0.75), against 0.67 (95% CI, 0.65-0.69) when trained on later codes. For later codes, the two scores were similar (0.69 vs 0.70). Fibromyalgia showed the same pattern (0.73 vs 0.64 and 0.74 vs 0.72). Elastic nets trained on self-report ranked uncoded self-reporters higher than the summary-count scores (0.80 vs 0.74 and 0.78 vs 0.73) but discriminated later codes less well than code-trained elastic nets. For endometriosis, the self-report score also ranked uncoded self-reporters above the code-trained score (0.61 vs 0.53), but discrimination was low and it did not discriminate later codes (0.50 vs 0.70). Conclusion: The choice of training label changes which patients a case-finding model ranks highly. For all 3 conditions, self-reported diagnoses were the more useful label for finding uncoded self-reporters. For chronic fatigue and fibromyalgia, the self-report score also discriminated later codes as well as the code-trained score. Independent clinical validation is needed before these models guide case finding.

Background: Hospital readmission is a common and costly outcome in patients with chronic obstructive pulmonary disease (COPD). Multimodal deep learning models combining structured electronic health record (EHR) data with physiological, imaging, and textual data have been proposed to improve its prediction, but whether such fusion outperforms EHR data alone, once temporal-leakage safeguards are applied, remains unclear. Methods: Using a stratified MIMIC-IV cohort of COPD patients enriched to about 50% readmission (natural prevalence 18.1%; n=1,279; 895/192/192 training/validation/test), we predicted 30-day, all-cause readmission from four frozen, pre-trained encoders (electrocardiogram, chest radiograph, clinical notes, structured EHR). We audited and corrected temporal leakage in the admission-linkage pipeline of each modality, then compared three fusion architectures (naive concatenation, scalar gated fusion, EHR-anchored residual fusion) against unimodal baselines and an EHR-alone reference, using bootstrap significance testing, a missingness stress test, and calibration analysis with and without post-hoc recalibration. Results: The audit corrected three previously undetected temporal-alignment issues, affecting 5-9% of patients for clinical notes, 29-43% for chest radiographs, and 53-58% for ECG. After correction, every modality except EHR showed near-chance discrimination alone (ROC-AUC 0.46-0.53), confirmed with a non-linear adapter probe. No fusion architecture significantly outperformed the EHR-alone baseline (ROC-AUC 0.63); removing auxiliary modalities improved rather than degraded fusion performance under stress testing. The raw Brier score was lower for all three fusion architectures than for EHR-alone (about 0.24 versus 0.32), but this reflected overconfidence of the EHR-alone probe (calibration slope 0.12); after logistic recalibration fitted on the validation set, Brier scores were indistinguishable (EHR-alone 0.239; fusion 0.240-0.244; all paired bootstrap 95% CIs included zero). Conclusions: Structured EHR data alone matched the discrimination and, after recalibration, the calibration of more complex multimodal fusion for 30-day COPD readmission prediction. These findings highlight the importance of rigorous temporal-leakage auditing in multimodal clinical prediction research and suggest that reformulating this task around temporal proximity to the outcome, rather than refining the fusion mechanism further, may be more productive for future work.

High attrition rates across clinical drug development impose immense clinical and financial burdens, yet prospectively forecasting whether an individual trial will achieve its primary endpoint remains challenging. Here we present a machine-learning system that predicts clinical trial success across phases I-IV, independently modeling studies with efficacy versus safety/toxicity primary endpoints. To prevent lookahead bias and outcome leakage, models are trained on registration-time features, including target and mechanism-of-action precedent, drug modality and novelty, trial design and operational characteristics, and frozen literature-derived target-choice embeddings. Each prediction yields a continuous Predictive Trial Index (PTI) between 0 and 1 reflecting success probability. On a test set of trials initiated in 2021 or later, the efficacy model achieved an area under the receiver operating characteristic curve (AUROC) of 0.828 (95% CI 0.812-0.845; n = 2,347) and the safety/toxicity model reached 0.809 (95% CI 0.787-0.831; n = 1,485). These models are embedded in inClinico Agent, an agentic AI system integrating frontier large language models (LLMs) with deterministic scoring engines, retrieval tools and specialized clinical skills. We also introduce ClinBench, a prospective benchmark of 2,040 ongoing trials evaluated by inClinico Agent and five frontier LLMs, establishing a leakage-free foundation for tracking prospective predictive accuracy as trials reach completion.

Background: A new class of AI "decision models" returns a probability for every option in a single pass and is marketed as calibrated, which invites use in clinical triage. Whether those probabilities hold up on clinical diagnosis has not been independently tested. Objective: To test the diagnostic accuracy, calibration, and probabilistic coherence of a decision model (Jev 1.13, TypeSafe AI) against frontier language models, with and without reasoning. Methods: In a pre-registered study (OSF, registered 26 September 2026), we compared Jev with Claude Opus 5.5 (adaptive reasoning, always on) and Claude Opus 5 (reasoning disabled) on three case sets: the 45 Semigran symptom-checker vignettes, 1,000 stratified synthetic patients from DDXPlus with reference differentials, and 80 simulated ophthalmic complaints (40 diagnoses at two levels of detail). Each model returned a probability for every diagnosis on a fixed list. Outcomes were top-1 and top-3 accuracy, expected calibration error (ECE), Brier score, the rate of exactly zero probability on the true diagnosis, sensitivity to option order, behavior on uninformative inputs, and behavior when the true diagnosis was absent. Results: On DDXPlus, top-1 accuracy was 72.0% for Jev, 79.5% for Opus 5 without reasoning, and 87.4% for Opus 5.5 (both comparisons with Jev P<.001). Jev assigned exactly zero probability to the true diagnosis in 62 of 1,000 cases (6.2%); neither comparator did so in any case. Calibration error did not differ between Jev and the registered comparator (ECE difference 0.007, 95% CI -0.026 to 0.038). Offered an explicit "none of these" option, Jev declined to diagnose all 30 uninformative inputs; without that option, it assigned a mean 58% to a single diagnosis. Order sensitivity was small in all models. Given the option list with no patient information, Jev placed up to 96% on a single diagnosis, while the comparators spread probability almost evenly. On the ophthalmic complaints, accuracy did not differ (Jev 88.8%, Opus 5 91.2%, Opus 5.5 88.8%), and Jev was better calibrated than the registered comparator (ECE difference -0.081, 95% CI -0.124 to -0.018). Conclusions: The decision model matched frontier models on short, distinctive ophthalmic complaints but fell well behind on multi-symptom synthetic patients, including behind a comparator without reasoning, and there it excluded the correct diagnosis outright in about 1 in 16 cases. Its probabilities signaled uncertainty only when the question offered a way to abstain. Calibration claims for clinical decision models should be tested on clinical tasks, with explicit abstention options and zero-probability rates reported.

Objective Large language models (LLMs) are increasingly used to generate, summarize, and evaluate clinical documentation, yet standardized methods for evaluating nuanced, context-dependent clinical language remain limited. Existing approaches rely primarily on annotated clinical corpora or keyword-based methods, restricting reproducibility, public dissemination, and systematic comparison across models. We present ClinStigma, a clinician-guided framework for constructing, annotating, and evaluating synthetic benchmarks of contextual stigmatizing language, demonstrated through ClinStigma-SUD, its inaugural implementation for substance use disorder documentation. Materials and Methods Candidate clinical sentences were generated using a frontier language model and iteratively reviewed according to a clinician-developed annotation rubric refined through expert review in stigmatizing language. The resulting benchmark intentionally incorporates explicit stigmatizing, neutral, and contextually ambiguous ("gray-zone") documentation requiring clinical reasoning rather than keyword recognition. Three locally deployed open-weight LLMs were evaluated using zero-shot inference, followed by external evaluation using independently annotated clinical documentation from MIMIC-IV. Results ClinStigma-SUD consists of 500 clinician-adjudicated synthetic clinical documentation sentences. Inter-rater agreement was 98.8% across the full benchmark (Cohen's {kappa} = 0.976) and 94.0% within the intentionally ambiguous subset ({kappa} = 0.880). Under zero-shot prompting, binary accuracy ranged from 84.8% to 91.8% across evaluated models, while F1 for the ambiguous (gray) class ranged from 0.038 to 0.423. Zero-shot external evaluation on 3,716 annotated examples from the independently annotated Harrigian MIMIC-IV corpus yielded 84.4% accuracy (84.4% balanced accuracy; macro F1 0.824), against a 69.9% majority-class baseline. Stigmatizing-class F1 was 0.879 in the substance-related subgroup and 0.883 in the full external corpus. Discussion and Conclusion ClinStigma establishes a reproducible clinician-guided framework for developing synthetic evaluation benchmarks for contextual clinical language. By combining structured clinical expertise, explicit annotation guidance, synthetic benchmark generation, and external evaluation, the framework enables open, privacy-preserving evaluation of language models while providing a scalable foundation for benchmark development across additional clinical specialties, including obstetrics, emergency medicine, and mental health.

Electronic Health Records (EHRs) provide a common repository for patient information, and while many challenges persist, EHRs allow asynchronous coordination among healthcare professionals (HCPs). Such coordination is inherently challenging, as large multidisciplinary teams, or teams-of-teams, of HCPs jointly manage complex patient care. These teams-of-teams naturally fit under the Multiteam System (MTS) framework, where evidence suggests that both stronger within- and between-team communication improves overall team performance. Yet little is known how about these patterns of information sharing mediated through the EHR correlate with patient outcomes. We extracted over 53.9 million EHR access logs for 482 patients to construct novel within- and between-team information sharing measures which were then used to predict future emergency department (ED) visits and hospitalizations admitted through the ED. These network measures showed both statistical and practical significance, suggesting that this modifiable dimension of teamwork-information sharing through the EHR-could reduce acute care utilization.

Computer vision approaches to disease motor assessment are limited by clinical video dataset scarcity and distributional imbalance, motivating interest in generative video models as a potential source of training data. We introduce a three-component framework for evaluating the clinical fidelity of synthetic medical motion data, assessing visual fidelity, biomechanical stability, and pathophysiological accuracy, and apply it to videos generated by the text-to-video model Sora 2 depicting Parkinsonian hand-fisting motions across MDSUPDRS severity levels 0-2 (normal to mild-moderate motor impairment). We further use real-versus-real bootstrapping to compare synthetic-real differences against the natural variability of real clinical videos. Although generated videos appear visually realistic, quantitative analysis shows substantial divergence from real clinical recordings, including weaker severity-dependent patterns of bradykinesia and hypokinesia, less realistic tremor rhythms, and substantial temporal instability. These findings reveal a motion-domain gap between Sora-2-generated and real clinical motor behavior in this Parkinsonian hand-motion setting and demonstrate that visual realism alone is insufficient to establish the clinical utility of synthetic video data for training or evaluating medical computer vision systems.

Large language models (LLMs) are increasingly evaluated for clinical tasks, but safe use requires reliable uncertainty estimates as well as high accuracy. We tested eight general-purpose LLMs on 100 validated neurology board-style multiple-choice questions and compared three confidence signals: self-reported confidence, critique-implied confidence from an independent model instance, and consistency-based confidence from repeated sampling. Accuracy varied from 45% (Llama 3.1 8B) to 92% (Gemini 3 Flash/Pro), but confidence was generally overestimated, particularly in lower-accuracy models. Consistency-based confidence provided the best discrimination between correct and incorrect answers (pooled AUROC 0.74), outperforming self-report, although calibration differences between methods were not significant. Calibration should therefore be reported alongside accuracy when clinical LLMs are evaluated.

Decision models like Jev read text and return probabilities, categories, or ordinal levels, with confidence scores for the latter two, at a fraction of large language models' (LLMs) cost. We explored how Jev could be integrated into clinical workflows for all these output types: mortality risk estimation, categorical Choosing-Wisely decisions for patient questions, and ordinal colon cancer staging. With standard LLM-style inputs, Jev trailed five flagship LLMs: mortality AUROC was 0.822 versus 0.837-0.846; Choosing-Wisely accuracy, 87.7% versus 98.4-100%; and staging accuracy, 45.4% versus 94.2-100%. Developer-recommended input alterations improved staging accuracy to 75.8% with medical context, 82.1% with task decomposition, and 85.8% with both. None of Jev's staging errors had full confidence. Using Jev for fully confident staging decisions and LLMs for the rest was non-inferior to each LLM alone and cut flagship LLM costs by 33-44%. Prompted as designed, decision models may efficiently share structured clinical tasks with LLMs.

Timely detection of fetal distress during labor is a central pursuit of modern obstetric care. The primary measure used in this regard is fetal heart rate and contraction monitoring (echocardiotocography or CTG), which provides a continuous signal during labor. The vast majority of machine learning applications to CTG assume access to a 30 to 60 minute window just prior to delivery, which makes an untenable assumption that the timing of delivery is known. In this work, we reframe the problem as a time-to-event prediction task. We develop deep time-to-event models to predict the joint probability of imminent delivery and fetal acidosis, a biomarker of fetal hypoxia collected from umbilical cord gas after delivery. We develop a framework for this problem, drawing on deep discrete-time survival methods and marked point processes, adapting each to this particular setting (i.e., continuous physiological signals, fully observed events, partially observed labels). One approach dubbed Marked DeepHit accurately estimates delivery within one hour at multiple elapsed-time landmarks, with test-set AUROC of 0.837 [95% CI 0.692--0.970] and 0.935 [0.858--0.964] at 6 and 12 hours into labor, respectively. We find that these models also achieve competitive performance near delivery time compared to prior models trained specifically on the end of tracing, and outperform other clinical feature-based proposals, both on internal and external validation at a second site. These results give a proof-of-principle to support future prospective deployment of AI-EFM during labor.

Ophthalmic decisions require probabilities tied to a clinical question, a candidate set and an operating policy. We present OphJev, an endpoint-aware System One framework that combines frozen image and text representations, candidate scoring, schema-specific calibration and selective generalist queries for fundus and OCT imaging. Across eight external resources, multiple representation families, three generalist sizes and three seeds, controlled interventions separate the effects of wording, supervision and computation. Schema-aware affine calibration lowers mean held-out Brier loss and selective risk in all eight primary image--text configurations. On HYGD, matched-supervision fusion raises AUROC from 0.772 to 0.798, while a separate gain-routing policy lowers Brier from 0.618 to 0.599 using at most 20\% logical generalist image calls. These effects demonstrate complementary decision-level capabilities rather than universal gains: external transfer and larger specialist banks remain cohort dependent. A retrospective text extension evaluates 1,996 clinical notes and 91 ophthalmic questions; validation-based calibration improves note probability scores, while a lexical control remains strong. OphJev provides an auditable decision framework; prospective clinical benefit remains unvalidated.

INTRODUCTION: We report the development and validation of a novel computational tool to cross-convert plasma pTau217 results from different assays to facilitate recruitment for therapeutic trials and observational studies. METHODS: Participants underwent A{beta} PET imaging and plasma pTau217 measurement on up to three assay platforms across three cohorts. We then developed conversion equations using Deming regression -- collectively named Biomarker Convert -- and evaluated robustness using independent research cohorts. Finally, we modeled the potential impact of pTau217 screening on trial enrollment. RESULTS: While absolute plasma pTau217 concentrations differed, the assays had similar predictive characteristics for A{beta} positivity (LC-MS/MS [C2N], AUC=0.952; Janssen Simoa, AUC=0.946; Lumipulse, AUC=0.929). Biomarker Convert enabled transformation of plasma pTau217 data across assays independent of A{beta} PET and clinical status. Plasma pTau217 reduced A{beta} PET scans in the model screening paradigm by >60%. DISCUSSION: Biomarker Convert equations enable harmonization of plasma pTau217 values from immunoassay and mass spectrometry platforms, supporting direct data comparison.

Mammography interpretation requires tying each finding to supporting evidence and withholding judgment when that evidence is unavailable. Vision-language models are increasingly used for image-based medical question answering, yet existing benchmarks suggest that they struggle to jointly support accurate answers, evidence grounding, and appropriate abstention. We propose CARE-MVLM, built on Qwen2.5-VL-7B-Instruct, to address this gap through separate modules for answer prediction, answer-conditioned evidence generation, and answerability selection. On 7,358 triplets curated from CBIS-DDSM, MIAS, and VinDr-Mammo, CARE-MVLM substantially improves lesion-sensitive selective behavior over same VLM backbone baselines and exhibits markedly stronger lesion-sensitive abstention than GPT-5.6 and Gemini-3.5-flash while attaining the highest joint answer, grounding, and abstention reliability in the comparison.

Electronic health record foundation models are traditionally trained on the scale of years, days, or hours. These timescales, however, lack the minute-scale resolution required to directly guide clinical decisions at the bedside. We introduced minute-scale learning, a new paradigm for training and evaluating models, and developed MINT, a minute-scale foundation model for pediatric emergencies. MINT was pretrained and validated on 766,733 pediatric emergency department visits at five health systems, comprising 16 years of data from 10 hospitals. On minute-scale forecasting tasks, MINT outperformed and generalized to external health systems better than task-specific models (superior in 21 of 25 comparisons). MINT outperformed physicians in forecasting escalations of respiratory support. MINT demonstrated uniquely minute-scale capabilities including department-scale monitoring, dynamic risk explanations, individualized physiologic response forecasts, and hypothesis generation. Learning at the minute-scale improves performance, strengthens generalizability, and provides insights, actionability, and scientific capabilities that are not accessible at other timescales.

Measurements of human movement are critical for evaluating patients with neuromuscular diseases. However, current clinical tools are limited in sensitivity and can fail to detect the full spectrum of movement changes patients may exhibit. These limitations can obscure early signs of decline, delaying interventions that could preserve function and independence. OpenCap, a smartphone application that performs video-based markerless motion capture, can help address this gap by enabling scalable, out-of-lab measurements of human movement. Here, we present a method for both detecting disease and quantifying the severity of impaired movement from smartphone videos of patient motion. We used OpenCap to measure kinematics from 337 individuals with diagnoses spanning 12 rare, genetically-defined neuromuscular diseases and 78 age-matched healthy controls. Our dataset is the largest existing publicly available resource of body segment kinematics for this class of diseases. These data were collected in neuromuscular clinics during regular clinic visits as well as at conferences and community events across the United States. We trained deep learning models on these kinematic time series to differentiate between participants with neuromuscular disease and controls (AUROC = 0.97 [0.93-1.00]). We also developed a movement impairment severity score that correlates with both participant ratings of their own functional abilities ({rho} = -0.81, p < 0.001) and existing clinical tests of movement capabilities such as run speed ({rho} = -0.83, p < 0.001). Our methods for detecting disease and deriving quantitative biomarkers of movement impairment from smartphone videos have the potential to complement clinical assessments, inform screening protocols, and improve clinical trial endpoints.