PLOS Digital Health

Background: Antimicrobial resistance (AMR) is a growing global health threat, and India's high antibiotic consumption makes public knowledge, attitudes, and practices (KAP) a critical determinant of stewardship outcomes. Existing Indian KAP evidence is drawn from small, localised cohorts, leaving a gap in nationwide, digitally-sourced data. This study aimed to describe antibiotic-related knowledge, attitudes and self-reported practices among users of a digital healthcare platform in India, examine demographic and regional variation, and identify potential areas for future stewardship interventions. Methods: A cross-sectional survey was conducted among adult users of the Tata 1mg platform (December 2025-February 2026), with participants invited via an in-app banner. Knowledge and attitude were each scored on a 0-4 scale (4 items); practice was scored as a percentage of applicable items answered favourably, with scoring adjusted for branching logic based on antibiotic-access route. Of 10,726 respondents who started the survey, 4,883 completed it (45.5%). City of residence was mapped to one of six regions (India's five statutory Zonal Council zones plus a North-Eastern region) for 4,187 respondents. Associations with demographic characteristics were assessed using multivariable linear regression, and regional differences were assessed using one-way ANOVA. A parallel survey of 256 physicians was conducted to enable patient-provider comparison. Results: Mean scores were: knowledge 2.28 (SD 1.19), attitude 0.91 (SD 1.14), practice 44.9% (SD 33.4); attitude, not knowledge, was the weakest domain. Older age was associated with higher scores across all domains (practice: B=12.52, p<0.001 for age [&ge;]65 vs. 18-34); city tier was independently associated with lower knowledge and attitude (Tier 3 vs. Tier 1: p=0.010 and p<0.001, respectively). Regional differences were non-significant for knowledge and practice but significant for attitude (F=2.56, p=0.026, ^2=0.003), driven by lower scores in the North-Eastern region. In the parallel physician survey, 55.5% of respondents estimated that more than half of their patients self-medicated before consultation. Conclusion: In this platform-based sample, attitudes were the weakest domain, not knowledge, and 42.0% said their last antibiotic did not come on a doctor's recommendation. The findings may inform digital stewardship touchpoints, such as point-of-purchase guidance and prescription verification, in line with India's National Action Plan on Antimicrobial Resistance 2.0. As a cross-sectional survey of platform users, it cannot establish the relative contributions of knowledge, beliefs and access pathways to antibiotic use.

The explosive growth of ambient AI scribes in healthcare has focused almost exclusively on relieving clinician documentation burdens and enhancing clinician efficiency and reducing burnout. While most research has focused on the clinician experience, these same types of tools, such as a patient held AI Personal Health Assistant, can be put in the hands of patients to facilitate communication with healthcare providers, to provide greater autonomy and activation for patients in managing their own healthcare, and to facilitate the task of receiving and retaining complex new information. However, to ensure these tools serve as more than just administrative aids, we partnered with a large technology company to conceptually codesign how such a tool might be studied in real world settings in a way that reflected outcomes of greatest importance to patients and families. We used the Nominal Group Technique and engaged thirty three patients and caregivers and ten multidisciplinary clinicians in structured virtual sessions to identify the outcomes on which subsequent studies should hinge. Our findings reveal a significant misalignment between current industry trends and patient priorities. While the market emphasizes the scribe function for billing, stakeholders overwhelmingly prioritized the AIPHAs role in longitudinal care management and post visit coordination. The most critical outcome identified was the impact on communication and respect, suggesting that patients view AI as a mechanism to improve the human to human dynamic and ensure a shared understanding of the care plan. Our findings further highlighted that the primary hurdle for the use of patient held AI is not technical utility but clinician trust and the need for formal consent processes. Importantly, as this technology is new and integration into the clinical experience is still being explored, patients cannot assume they will have access to the transcript of their own encounter when that record is created and held by the health system. This gap highlights the potential value of a patient held tool, preserving discussions from their visit and follow up steps in clearer detail than they may receive in their clinical note or traditional visit summary. From a system perspective, this transition requires a strategic shift from measuring workflow focused efficiency to measuring patient side advocacy and engagement. By centering the evaluation of AI on these patient defined metrics and fundamentally designing subsequent AI investigations using a partnership model, health systems can move beyond simple automation toward a model of care delivery that empowers patients to navigate complex clinical journeys with greater autonomy and clarity.

Stroke is a leading cause of mortality and disability in Uganda, with a prevalence of approximately 2.5% and a 25-30% 30-day mortality rate, exacerbated by limited access to predictive tools and advanced diagnostics. Machine learning (ML) offers potential for early stroke risk stratification, but its application in resource-constrained settings like Uganda requires tailored models and rigorous performance evaluation, as well as assessment of misclassification patterns. This study aimed to develop and internally validate an ML-based predictive model for stroke using clinical and socio-demographic characteristics among patients at Mulago National Referral Hospital (MNRH) and residents of Wakiso District, Uganda, identify key predictors and assess misclassification patterns to enhance targeted prevention. A retrospective case-control study utilized secondary data from 1,016 participants (254 stroke cases, 762 controls) across three prior studies conducted between 2006 and 2015. Eight ML algorithms were evaluated using 5-fold cross-validation, with performance metrics including accuracy, sensitivity, positive predictive value, F1-score, and area under receiver operator characteristics curve (AU-ROC). McNemars test compared model classification performance of the top 2 models. Feature importance identified key predictors, and misclassification patterns were analyzed using classification matrix and line graphs. The Random Forest (RF) model achieved the highest performance (95.18% accuracy, 90.10% F1-score, 97% AU-ROC). McNemars test confirmed RFs superior classification (p=0.01). Age, systolic blood pressure, total cholesterol, and waist-hip ratio were the top drivers of RF stroke predictions. RF misclassified 8.3% of stroke cases and 3.3% of non-stroke cases, due to overlaps in risk profiles (e.g., lower cholesterol, systolic blood pressure and age in false negatives). The RF model demonstrates strong potential for stroke risk prediction in Uganda, leveraging routine data to prioritize high-risk individuals. Misclassification analysis highlights areas for model refinement, supporting its potential for integration into clinical workflows for early intervention and stroke prevention in resource-limited settings. Future external validation across diverse populations is recommended.

Background and Objective: The volume of patient portal messages has grown substantially and artificial intelligence (AI) can support clinical staff and improve the timeliness of patient care. Our objective is to test if an AI tool for managing the electronic inbox In-Basket reduces response times to patient portal messages in neurology. Methods: We conducted an implementation study from January to May 2026 in the neurology department at two clinical sites within a regional medical system. A single AI tool was implemented to auto-categorize and auto-route all patient messages received in the electronic inbox. The tool also generates optional draft replies to messages. Primary process measures were the median weekly time (hours) from message creation to the first action and to closure. Impact was analyzed using a controlled interrupted time-series (ITS) design, comparing message-volume-matched control divisions. Results: Among 24,142 messages analyzed, pre-intervention median weekly time to first action was 11.3 hours [IQR: 5.6, 15.1] (Site A) and 16.1 hours [IQR: 13.8, 19.0] (Site B), falling post-intervention to 1.0 hour [IQR: 0, 3.1] (Site A) and 0.6 hours [IQR: 0.2, 2.4] (Site B). Controlled interrupted time-series analysis showed a significant immediate differential level change in median weekly hours to first action at Site A (-92.5%, 95% CI: -97.5% to -78.0%), and a significant downward trend at Site B (-43.5%, 95% CI: -65.2% to -8.2%). Changes in time to closure were directionally favorable but not statistically significant at either site. There was low use of the generated draft replies, only 0.5% at Site A and 3% at Site B. Discussion: Implementation of AI-assisted message categorization and routing was associated with a substantially reduced initial action time for patient In-Basket messages in Epilepsy. The utilization of AI-generated draft-replies to messages was low. These findings support the feasibility and effectiveness of AI-augmented workflows to improve inbox management.

BACKGROUND Timely cause-of-death (COD) information guides health policy, yet many deaths in low- and middle-income countries are not medically certified. In Africa, fewer than 10% of deaths are. Verbal autopsy (VA) yields a structured symptom questionnaire and a free-text narrative. Mortality surveillance increasingly records the narrative alongside the questionnaire, but the algorithms used to produce VA-based statistics were designed for structured responses only, so narratives inform physician COD assignment but do not contribute to algorithmic COD assignment. We evaluated whether supervised models could use both modalities to improve COD assignment. METHODS We analyzed 4,999 VA records for deaths at ages 12 years or older from a national sample of registered deaths in South Africa (RSA), using physician-assigned CODs as the reference standard. We trained unimodal and multimodal pipelines on the labelled records: pretrained language models (PLMs) fine-tuned on narratives alone, machine learning (ML) models on structured responses alone, and the two combined at the data, feature, and decision levels. We also prompted a locally deployed generative large language model (LLM) on the same unimodal and multimodal inputs, without parameter updates. All were compared with InSilicoVA, the operational comparator representing current automated practice. We evaluated individual-level assignment, population-level agreement, calibration, and generalization across provinces in RSA, reporting these metrics rather than ranking pipelines on any one of them. RESULTS Every supervised pipeline improved on InSilicoVA by 8.5-15.9 percentage points in top-1 accuracy and 3.5-7.1 points in cause-specific mortality fraction (CSMF) accuracy. The narrative and structured pipelines performed similarly in aggregate but differed by cause: 63.9% of deaths received identical predictions, 11.6% were correct only from the narrative pipeline, and 13.6% only from the questionnaire pipeline, with the narrative pipeline more accurate for noncommunicable causes and the questionnaire pipeline for communicable causes. Against the unimodal approaches, the decision-fusion pipelines added 4.2-5.4 percentage points in top-1 accuracy, concentrated in the most common causes, whereas feature fusion with cross-modal attention improved the less common causes, adding 6.6 points (95% CI, 1.7-11.5) in balanced accuracy and 5.5 points (95% CI, 0.6-10.3) in macro F1. Feature fusion reached the highest CSMF accuracy (95.7%). Paired differences from the better unimodal representative ranged from -0.2 to +3.4 points, with all intervals including zero. The LLM exceeded InSilicoVA on individual-level metrics and reached the highest balanced accuracy of any pipeline (60.6%), while the supervised pipelines retained higher CSMF accuracy. Supervised pipelines had lower expected calibration error than InSilicoVA. Supervised gains held across all nine held-out provinces, including every top-1 comparison and nearly every macro F1 and CSMF comparison. CONCLUSIONS PLMs using narratives alone substantially improved on current best practice and matched ML pipelines using structured responses, while multiple multimodal strategies produced further gains in individual COD assignment. The zero-shot LLM showed particular strength on rare causes, whereas supervised pipelines better recovered the population cause distribution. These findings support incorporating narratives as a core input to automated VA systems and evaluating representative fusion strategies for the intended output. External validation across settings and age groups will support broader implementation.

Background: Conventional performance metrics do not indicate whether artificial intelligence (AI)-platforms and radiologists are right or wrong on the same examinations, nor whether their non-agreement is more often beneficial or harmful. Objective: To develop, illustrate, and apply directional metrics quantifying non-agreement between radiologists and AI-platforms against a common reference standard, and to adapt a fragility-based framework for quantifying the statistical robustness of these metrics. Methods: We defined Positive Non-Agreement (PNA, AI-platform correct, radiologist incorrect), Negative Non-Agreement (NNA, radiologist correct, AI-platform incorrect), and Net Non-Agreement (NENA, their difference). Metrics were illustrated in three hypothetical scenarios of 1,000 examinations each and applied to three previously published AI-radiologist studies on pulmonary embolism and mammography. Bootstrap confidence intervals were calculated, and a reclassification fragility index (FI) quantified the minimum case reclassification needed to remove statistical support for NENA's direction. Results: In illustration, NENA was positive in two scenarios and negative in one. In the three published studies, NENA ranged from -4.18% to +10.74% and decreased consistently with increasing radiologist experience (junior residents, +6.27%; consultants, +1.45%), a gradient driven primarily by declining PNA with modest increase in NNA. The reclassification FI was defined for all three illustrative scenarios (range, 0.10%-1.70%) and for six of the seven comparisons from these studies (range, 0.34%-4.31%), but not for the weakest comparison, whose confidence interval for NENA included zero. Conclusions: PNA, NNA, and NENA provide a directional, case-level complement to conventional performance metrics for AI-radiologist non-agreement. The accompanying reclassification FI offers a statistically grounded, generalizable method for quantifying their robustness.

Purpose: To compare the performance of three large language models (LLMs), ChatGPT-o1, Grok 3, and Gemini 2.0 Pro, on the American College of Prosthodontists' National Prosthodontic Resident Examination (NPRE). Materials and Methods: Official NPRE examinations from years 2015-2024 were retrieved from the American College of Prosthodontists website and used to prompt each LLM. Paid access tiers and advanced reasoning were enabled. Responses from LLMs were recorded and compared against the official answer keys for accuracy, variability across years, and consistency of responses. Duplicate questions were retained for the primary analysis and removed in a sensitivity analysis. Accuracy (percent correct) was compared using generalized estimating equations binary logistic regression (primary); Cochran's Q with Bonferroni-adjusted McNemar tests (sensitivity). Year-to-year variability (SD of annual accuracy) was compared using weighted Levene tests. Consistency across duplicate items was quantified as the percentage of perfect matches and compared using Cochran's Q with Bonferroni-adjusted McNemar tests. Results: Across 1,487 questions, overall accuracy differed among models (global P < 0.001). ChatGPT-o1 achieved 78.4% correct, Grok 3 76.4%, and Gemini 2.0 Pro 73.2%; ChatGPT-o1 and Grok 3's accuracy significantly exceeded Gemini 2.0 Pro (both P [&le;] 0.016), with no significant difference between ChatGPT-o1 and Grok 3 (P = 0.099). After removing duplicates (n = 1,200), results were similar: ChatGPT-o1, Grok 3, and Gemini 2.0 Pro were 77.1%, 76.6%, and 71.5% correct, respectively; ChatGPT-o1 and Grok 3's accuracy was significantly higher than Gemini 2.0 Pro (both P < 0.001). Variability across years did not differ significantly (P [&ge;] 0.097). Grok 3 showed significantly higher perfect-match consistency (79.2%) than ChatGPT-o1 (69.4%) and Gemini 2.0 Pro (64.5%) (P [&le;] 0.010). Conclusions: Contemporary LLMs showed 70-80% accuracy on postgraduate prosthodontic multiple-choice content. While promising for educational support, residual error and inconsistency warrant human oversight and transparent methods reporting before curricular adoption.

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.

Importance: Conditions that are hard to diagnose, several more common in women, may be missing from electronic health records (EHRs), hiding patients from research cohorts and case finding. Objective: To measure how often the EHR contains a diagnosis code for chronic conditions that participants report, and to compare this with national survey benchmarks. Design, Setting, and Participants: This cross-sectional study used the All of Us Research Program Registered Tier Dataset v9. Participants were adults with EHR visits in both 12-month periods from July 2021 to June 2023 who completed a health history survey. We analyzed the data in October 2026. Exposures: Self-reported clinician diagnosis of 10 conditions with ongoing visits or treatment, as well as sex at birth, race and ethnicity, age, and years of EHR records. Main Outcomes and Measures: Share of participants reporting current care who had a qualifying diagnosis code, alone and relative to hypertension and type 2 diabetes, and the ratio of age-standardized coded shares to national survey prevalence. Results: Of 161,480 respondents, 114,610 (73,870 female [64.5%], median [IQR] age 63 [50-72] years) had at least 5 years of EHR records before the survey. Among those reporting current care, 84.6% (95% CI, 84.2%-85.0%) with hypertension and 87.0% (95% CI, 86.5%-87.6%) with type 2 diabetes had a code, compared with 41.3% (95% CI, 38.9%-43.8%) for endometriosis, 39.4% (95% CI, 34.3%-44.8%) for Lyme disease, and 24.4% (95% CI, 23.2%-25.6%) for chronic fatigue. Counting ICD-9-CM and other mapped codes, 55.3%, 51.5%, and 28.3%, respectively, had any qualifying record. Women had lower coded shares than men for hypertension, type 2 diabetes, obesity, depression, and anxiety (crude ratios 0.93-0.96, within-health-system ratios 0.98-1.02) and higher shares for migraine and fibromyalgia (within health systems, 1.11 [95% CI, 1.07-1.15] and 1.41 [95% CI, 1.29-1.57]). Age-standardized coded shares among active patients were 1.1 to 2.4 times national prevalence of the same self-reported diagnoses. Conclusions and Relevance: In this cross-sectional study, fewer than half of participants reporting ongoing care for chronic fatigue, Lyme disease, or endometriosis had a qualifying ICD-10-CM code. Linking each participant's report to their own record measured this gap directly.

Background: The first thousand days are a sensitive window for child health and social health inequalities. When routine care falls short, parents of young children turn to online sources and, recently, to generative artificial intelligence (AI). Evidence on their concerns, child-health literacy, and AI use is scarce. Objective: To describe health informational needs and concerns, child-health literacy, and use of AI for child-health questions, including differences by parental role and parity, among parents of young children. Methods: In this cross-sectional study, two anonymous self-administered online questionnaires were disseminated through a digital health application between May 12 and June 11, 2026, to French-speaking parents of at least one child aged 0 to 47 months in France. Questionnaire 1 covered unmet information needs, concerns across 16 topics, perceived intra-couple concordance, and confidence in routine care. Questionnaire 2 covered a four-item child-health literacy quiz and AI use (frequency, topics, 12-month trend, barriers, trust conditions, perceived effect on consultations, and disclosure). Results: In total, 12,634 parents responded (Questionnaire 1, n=5,410; Questionnaire 2, n=7,224), mostly mothers (94.6% and 92.9%). Most (61.8%) had searched for information without finding a clear, reliable answer, selecting a median of 4 (Interquartile range [IQR] 3-7) over 16 topics. Leading concerns were sleep (51.3%), nutrition (48.1%), and education (i.e., upbringing and behavior) (47.8%). Mothers were nearly twice as likely as fathers to select accident prevention (26.5% vs 16.3%) and parental mental health (38.7% vs 28.6%); first-time parents selected most topics more often than experienced parents, most markedly dental hygiene (43.8% vs 27.6%). The median child-health literacy score was 3 (IQR 3-4) on a 0-to-4 quiz, with 19.9% of parents scoring 2 or below; tooth-brushing age was the weakest item (68.4% correct). Mothers outscored fathers on every item. More than half of parents (54.2%) had used AI for a child-health question, 11.2% weekly; among those reporting a trend, 68.3% had increased AI use for child-health questions over the year and 6.2% decreased it. The main barriers to AI use were a preference for professionals (55.9%) and reliability concerns (48.7%); trust depended most on medical oversight of the tool (76.6%). AI use appeared additive, with 66.8% reporting no change in healthcare visit frequency. One third (35.5%) had told an AI something shared with neither a professional nor their network, rising to 64.1% among regular users. Conclusions: Parental informational needs were widespread and frequently unmet. AI has become a common and fast-rising source for child-health questions, at times a confidential one, used alongside rather than instead of professionals. Child-health literacy was generally high but uneven, with oral health the clearest gap. Strengths include the sample size and early-generative-AI timing; limitations include the cross-sectional design, self-report, digital-recruitment selection bias, and the small proportion of fathers.

Purpose To evaluate the external transportability of source-derived probability calibration for chest radiograph deep learning classifiers under distribution shift and determine its dependence on sample size and disease prevalence. Materials and Methods This retrospective secondary computational analysis of public, de-identified radiographic datasets involved no participant contact or new data collection. Cohort sizes were prespecified by availability without prospective power calculation: 106,686 NIH ChestX-ray14 images (29,323 patients; 1992-2015) for development and internal testing, 20,000 CheXpert images (20,000 patients; October 2002-July 2017), and 14,998 VinBigData images (2018-2020) for external testing. ResNet-18, DenseNet-121, and DeiT-Tiny classifiers for cardiomegaly and pleural effusion were trained across four patient fractions and three seeds (36 fits). Calibrators were fitted exclusively on NIH data. Twelve primary external Brier-score contrasts (temperature-scaled versus raw probabilities) were evaluated using paired cluster-bootstrapping with Holm adjustment. Results Temperature scaling reduced Brier scores in nine of 12 primary contrasts but worsened pleural effusion calibration on VinBigData across all architectures (Brier increases: ResNet-18, 0.001460 [95% CI: 0.001347, 0.001568]; DenseNet-121, 0.001193 [95% CI: 0.001109, 0.001274]; DeiT-Tiny, 0.000498 [95% CI: 0.000417, 0.000577]; all adjusted P=.006). Standardizing class weights to NIH prevalence reversed all CheXpert improvements. Although full versus 10% training improved external AUROC across all comparisons, DeiT-Tiny CheXpert pleural effusion Brier score worsened by 0.007039 despite a 0.032466 AUROC gain. At 80% selective image retention, calibrated ensembles retained only 12.3%-17.2% of VinBigData positive pleural effusion examinations. Conclusion Source-fitted calibration is not uniformly transportable across health systems, and discrimination gains do not guarantee calibrated probabilities. External evaluation must simultaneously assess probability reliability, prevalence shifts, and retained-case composition.

Background: Self-supervised learning may improve segmentation when labelled data are limited, but its benefit may depend on model and initialisation strategy. This study evaluated in-domain masked-reconstruction pretraining across limited annotation budgets. Methods: The International Skin Imaging Collaboration (ISIC) 2016 development set was partitioned into 765 images for self-supervised pretraining, 94 for fine-tuning and 41 for validation. Five nested annotation budgets (12-94 images) and five seeds were evaluated on the 379-image test set. A custom U-Net compared random and in-domain self-supervised initialisation; a ResNet-34 U-Net compared random, ImageNet, in-domain self-supervised and sequential ImageNet-to-self-supervised initialisation. Models trained with 94 labelled images were evaluated without adaptation on ISIC 2017. A separate experiment examined restoration-based pretraining. Results: In the custom U-Net, self-supervised pretraining achieved higher mean Dice than random initialisation at all five budgets, including 0.664 versus 0.615 with 12 labels and 0.811 versus 0.768 with 94 labels. Seed-level 95% confidence intervals for paired differences excluded zero at 47 and 94 labels but not at the other budgets. With 47 labels, the self-supervised model achieved mean Dice of 0.789, compared with 0.768 for random initialisation with 94 labels. In ResNet-34, ImageNet initialisation achieved higher mean Dice than in-domain self-supervised pretraining at all five budgets, and adding self-supervised pretraining after ImageNet reduced mean Dice at four of five budgets. All six 94-label configurations achieved lower mean Dice on ISIC 2017 than ISIC 2016. In the restoration experiment, mean Dice was 0.686 with random initialisation and 0.638-0.642 following pretraining. Conclusions: In-domain masked-reconstruction pretraining improved mean Dice relative to random initialisation in the custom U-Net, but not consistently in the ResNet-34 U-Net, where ImageNet initialisation performed better at every budget. Adding self-supervised pretraining after ImageNet provided no consistent gain. These findings are specific to the dermoscopic lesion-segmentation experiments and do not establish melanoma-specific diagnostic or clinical performance.

Background. Research data enclaves such as the All of Us Research Program do not allow participant-level data, or models fitted to them, to leave, which makes it hard to deliver case-finding models, which flag patients with possibly uncoded conditions for clinician review, to patients and clinicians. Objectives. We describe and technically evaluate a system that fits models outside an enclave from released summary counts and delivers them through standard health information technology interfaces. Methods. We exported dissemination-compliant summary counts from the All of Us Researcher Workbench and, outside the enclave, fitted versioned models for Long COVID, chronic fatigue, fibromyalgia, endometriosis, Lyme disease, lupus, migraine, depression, and anxiety. A model contract (a model file and a written specification), a Python reference implementation, and synthetic Fast Healthcare Interoperability Resources (FHIR) test patients define the expected outputs. One JavaScript scorer runs in a patient demonstration, a SMART on FHIR clinician application, and a CDS Hooks service. An atlas displays aggregate results. We evaluated the system on synthetic records with automated tests, cross-implementation output comparison, launches through the public SMART sandbox, and a counterexample review of the scoring and display rules. Results. The deployed Long COVID model had an area under the receiver operating characteristic curve of 0.67 in a random validation split. The eight self-report models reached 0.61 to 0.82 in identifying self-reporters without a code. The JavaScript scorer reproduced all 162 reference outputs exactly, and all 437 automated tests passed. A counterexample review found 5 classes of failure, including scoring ineligible patients and treating unavailable data as absent, all fixed and added as tests. Conclusions. Models fitted from released summary counts can be delivered through FHIR, SMART on FHIR, and CDS Hooks with reproducible scoring on synthetic records. Real-record interoperability, usability, and prospective clinical performance remain to be evaluated.

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.

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.

Purpose Autistic adults have diverse communication characteristics that are not well captured by standard demographic measures. Communication affects self-determination, relationships, access to healthcare, and services. Characterizing communication differences may improve research and intervention targeting for meaningful subgroups of autistic adults. Methods The Academic Autism Spectrum Partnership in Research and Education (AASPIRE) Outcomes Project team used a community based participatory research (CBPR) approach to develop a self- or caregiver-report module to characterize communication among autistic adults. The 10-item module asks about receptive and expressive communication, literacy, speech reliability, usage of digital communication, use of alternatives to speech, and communication fluency. It can be completed directly, with or without support, or via a caregiver. We used baseline survey data from 844 autistic adults to evaluate the module and derived a five-level variable, Communication Profile, from selected module items to stratify participants. We examined completion feasibility, response consistency, and construct validity. We also assessed test-retest reliability in a 15% random subsample who completed a 2-week retest following the 6-month survey, with oversampling of caregiver-report respondents. Results The Communication Characteristics module demonstrated high completion feasibility, minimal systematic error, and strong construct validity. The Communication Profile variable stratified participants into five meaningful communication profiles and demonstrated strong construct validity and substantial test-retest reliability ({kappa} = .69). Conclusion The Communication Characteristics module and Communication Profile variable provide a valid approach for characterizing communication diversity among autistic adults for research, clinical, and service settings. The full AASPIRE Measurement Toolkit is available at https://aaspire.org/measurement. Lay Summary Autistic people communicate in many different ways, and their communication may change in different situations. We developed and tested a set of survey questions to describe an individuals communication. We also used the results to describe 5 groups of autistic people based on how they communicate, without treating these groups as different types of autism.

Large language models (LLMs) are increasingly consulted for health advice, but they can produce fluent statements that are false or unsupported, known as hallucinations, and this risk is poorly characterised in low-resource languages such as Bangla. We measured hallucinations of locally deployable open-weight LLMs in Bangla diabetes question answering and tested whether retrieval-augmented generation (RAG) grounded in Bangladeshs National Guideline on Diabetes Mellitus reduces them. Forty patient questions spanning six clinical sub-domains were each answered twice, under zero-shot prompting, by Qwen3.5-9B, TigerLLM-9B, Qwen3.6-35B-A3B and Gemma-4-26B-A4B, and by BanglaDiaRAG, which adds hybrid dense (BGE-M3) and sparse (BM25) retrieval from a physician-reviewed Bangla guideline knowledge base to Qwen3.5-9B. Two physicians rated the 400 de-identified responses by consensus for medical accuracy, completeness, safety, readability and reliability on a 0-1 scale, and responses with accuracy below 0.70 were classified as hallucinated. Hallucination rates were 2.5% for Gemma-4-26B-A4B, 3.8% for TigerLLM-9B, 22.5% for Qwen3.6-35B-A3B and 77.5% for Qwen3.5-9B. Guideline grounding lowered the rate of Qwen3.5-9B to 11.3% (difference -66.3 percentage points; 95% CI -76.3 to -55.0) and raised every criterion by 0.24 to 0.38 points (Holm-adjusted p < 0.001). BanglaDiaRAG outscored Qwen3.6-35B-A3B on all five criteria and matched or exceeded TigerLLM-9B, but it remained below Gemma-4-26B-A4B in safety, readability and reliability, and its residual errors were specific, confident and potentially harmful. Guideline grounding can turn a weak general-purpose model into a much safer Bangla diabetes assistant, but the choice of base model mattered as much as retrieval and no configuration was free of hallucination, so clinical oversight remains essential before patient-facing use.

INTRODUCTION: The digital clock drawing test (dCDT) is a cognitive screening tool capturing high-resolution pen-movement data. Existing approaches for classifying cognitive impairment from dCDTs rely on selected features or image-based methods. We propose representing dCDTs as structural networks to classify cognitive impairment and characterize spatial organization in clock drawings. METHODS: Weighted networks were constructed from 3,362 Framingham Heart Study participants, with and without cognitive impairment. Network representations were generated separately for command and copy conditions, with nodes representing pre-defined clock components (digits, hands, clock face) and edges capturing pairwise spatial relationships between components. Random forest models and graph convolutional networks were trained to predict cognitive impairment from the networks. Models were benchmarked against demographics-only and established summary feature baselines. RESULTS: Network-based models achieved AUC values comparable to the established summary feature baseline across command-only and copy-only conditions. Differences in spatial organization between the cognitively intact and impaired groups were greatest in the upper-right clock region, involving digits 1 and 2, the clock hands, and the surrounding clock face. Differences between the command and copy networks were larger among cognitively impaired participants. DISCUSSION: The proposed network representations provide an objective, scalable, and flexible framework for representing clock drawings for use in classification models. dCDT networks matched the predictive performance of established summary feature-based approaches and identified clock regions where spatial organization differed by cognitive status. After external validation, this approach has the potential to be incorporated within a clinical workflow to support automated and interpretable cognitive screening.

Objectives To evaluate discrepancies between radiologists and referring physicians in the interpretation of diagnostic probabilities conveyed by common Japanese expressions used in radiology reports. Materials and Methods Fifteen expressions commonly used to convey diagnostic probabilities in radiology reports were selected by five attending radiologists. An anonymous online survey was distributed to 26 radiologists and 33 referring physicians at a single institution. Participants assigned a diagnostic probability (from 0% to 100% in 10% increments) to each expression. Differences in median probabilities between radiologists and referring physicians were analyzed using the Mann-Whitney U test. Variability in interpretation was assessed using interquartile range (IQR), with IQR [&ge;] 30% defined as substantial variability. Results All radiologists (26/26) and 97% of referring physicians (32/33) completed the survey. Seven of the 15 expressions showed statistically significant differences in interpretation between radiologists and referring physicians: "c/w" (80% vs. 60%; p = 0.031), the Japanese equivalents of "present" (100% vs. 100%; p = 0.024), "suspect of" (80% vs. 70%; p = 0.007), "possibly" (50% vs. 70%; p < 0.001), "unlikely" (10% vs. 10%; p = 0.022), "cannot be ruled out" (25% vs. 40%; p < 0.001), and "not inconsistent with" (80% vs. 90%; p = 0.008). Substantial variability in interpretation (IQR [&ge;] 30%) was observed for four expressions: "cannot be ruled out" (IQR, 10%-40%), "s/o" (IQR, 60%-90%), "c/w" (IQR, 50%-90%), and "not inconsistent with" (IQR, 70%-100%). Conclusion This study revealed that nearly half of the Japanese expressions commonly used to convey diagnostic probabilities were interpreted differently by radiologists and referring physicians. These discrepancies highlight the need for a standardized lexicon to improve clarity and consistency in radiology reporting.

Objective: to estimate reconsultation risk and trajectories after a new episode of musculoskeletal pain in children and young people (CYP), and to determine routinely recorded prognostic factors associated with reconsultation. Design: Population based retrospective cohort study. Setting: Clinical Practice Research Datalink (CPRD) Aurum, 2005-2021. Participants: CYP aged 8-18 years consulting primary care for a new musculoskeletal pain episode (no musculoskeletal consultation in prior 12-months). Main Outcome Measures: Musculoskeletal pain reconsultation within 12-months, and 3-year musculoskeletal reconsultation trajectories derived using latent class growth analysis. Prognostic factors selected from a literature search and two-year symptom histories (sample prevalence >1%) were examined using regression models (conditional logistic; multinomial). Results: Among 313,830 CYP with >12 months' follow-up, 32% reconsulted for musculoskeletal pain within 12 months. Among 240,384 CYP with >3 years' follow-up, three trajectories were observed: single episode (62%), early reconsultation (18%), and persistent reconsultation (20%). Older age; female gender; hip, knee, multisite and back pain on index; previous headache; previous abdominal pain; and analgesic prescribing were associated with higher reconsultation risk. Asian or Black ethnicity, and residing in more deprived areas, were associated with lower recorded reconsultation. Conclusion: Most CYP had no further recorded consultation, but 38% followed early or persistent reconsultation trajectories. Routinely recorded prognostic factors, particularly index pain side and prior pain/symptom history, may differentiate groups who could benefit from proactive follow-up. Lower recorded reconsultation among Asian and Black individuals, and living in more deprived areas, should not imply lower need and warrants equity-sensitive monitoring of access, health-seeking, and coding.