Journal of Pathology Informatics

Background: Store-and-forward teledermatology commonly relies on several patient-submitted photographs of the same concern, but most dermatology artificial intelligence models classify single images independently. Objective: To develop and internally validate a case-level diagnostic-support model that aggregates multiple patient-submitted photographs for common dermatologic conditions. Methods: We conducted a retrospective diagnostic-modeling study using the Skin Condition Image Network, a public dataset of deidentified self-taken dermatology images from US adults. We curated 2,336 cases comprising 5,041 images across 10 common inflammatory, allergic, and infectious conditions. Cases were split at the submission level into training, validation, and held-out test sets. Frozen general-purpose and dermatology-specific encoders were compared with image-level classifiers and a gated-attention multiple instance learning model that generated one case-level output from 1-3 images. Results: The strongest image-level baseline, dermatology-specific embeddings with random forest classification, achieved macro/micro ROC-AUCs of 0.797/0.854. Case-level aggregation improved discrimination, with dermatology-specific embeddings plus multiple instance learning achieving mean macro/micro ROC-AUCs of 0.819/0.863 across repeated stratified experiments. The locked final model achieved macro/micro ROC-AUCs of 0.800/0.849 on the held-out test set. Balanced-threshold sensitivity/specificity examples were 0.702/0.688 for eczema and 0.818/0.826 for urticaria. Limitations: Internal validation used a 10-condition subset from a US volunteer dataset; external validation, calibration, subgroup performance analysis, and prospective workflow studies are required. Conclusion: Modeling the teledermatology submission as a multi-image case better reflects asynchronous dermatology workflow than single-image classification. The model is preliminary clinician-facing support for structured review and triage, not autonomous diagnosis.

Teledermatology expands access to dermatologic expertise in rural settings, yet diagnostic uncertainty persists in low-resource primary care. This retrospective study evaluated MedGemma-4B-IT, a compact multimodal vision-language model, as adjunctive clinical decision support for challenging diagnostic cases. We analyzed 77 zero-concordance cases (360 clinical photographs) from a Dermatology Extension for Community Healthcare Outcomes (ECHO) tele-mentoring program (2016-2021). Zero-concordance cases showed no overlap between primary clinician provisional diagnosis and dermatologist-confirmed diagnosis. The model was prompted using dermatologist-style format to generate ranked differential diagnoses. Performance was assessed using strict case-level top-k exact-match accuracy and relaxed matching criteria based on fuzzy string similarity. MedGemma achieved 0.0% strict top-1 accuracy, 1.3% top-3 accuracy, 3.9% top-5 accuracy, and 3.9% top-10 accuracy. Relaxed concept-level matching achieved 28.6% top-1, 63.6% top-5, and 67.5% top-10 accuracy. Image-level accuracy was 44.2% (159/360, 95% CI 39.0-49.5%). The model surfaced the correct diagnosis within differential lists in 45.5% of cases despite no exact top-1 matches, suggesting utility for differential expansion rather than definitive diagnosis. Performance varied across diagnostic categories, with highest accuracy in Other categories (54.5%) and lowest in neoplastic conditions (0.0%). Common errors included confusion between inflammatory and other diagnostic groupings. These findings characterize MedGemma performance on real-world teledermatology cases and inform safe, clinician-in-the-loop integration into teledermatology workflows where specialist oversight remains essential.

Precision medicine aims to advance our ability from a "one-size-fits-all" approach to personalized and predictive healthcare across diverse populations. It promotes integration of multi-omics and phenotypic data to understand disease mechanisms and discover novel biomarkers and risk factors, which could be used to predict and prevent critical diseases in individual patients across diverse populations. The potential implications of precision medicine approach can accelerate our ability to classify patients at higher risk of developing critical diseases, improve diagnostic capabilities, develop deeper understanding of individual risk, investigate racial differences and demographic characteristics, and find relationships between genetic variants, expressions, and diseases. This study focuses on implementing an innovative and data driven framework of translational bioinformatics and Machine Learning (ML) techniques to analyze multi-omics, including RNA-seq and Whole-Genome Sequencing (WGS) data, generated using blood samples of randomly consented patients. First, we utilized bioinformatics pipelines to identify differentially expressed genes and their pathogenic and likely pathogenic variants for the downstream data analysis, annotation, and visualization. Then, applied a nexus of ML models for multi-omics biomarker discovery, disease prediction, density-based clustering, single-patient profiling, and pathogenicity classification. WGS data analysis supported the exploration of genetic variation and diversity among patients to identify known and novel biomarkers, whereas RNA-seq data analysis improved our understanding of functional and biological pathways that underlying disease states. We classified and clustered pathogenic variants and expressions across various genes and discovered numerous diseases leading risk factors. Our results include gene-disease associations and captured common pathways across the broader population, demonstrating a level of sensitivity and accuracy that has broad clinical implications. We validated our results through clinical records, and state of the science literature. This study delves into the strengths of multi-omics data integration and capabilities of ML application in genetically diverse and complex patient cohorts. Our approach has the potential to elucidate complex gene-disease interactions for genetically diverse populations, which can support earlier diagnoses for patients in many disease realms.

Quantitative eye movement analysis is important for neuro- logical diagnostics, yet existing video-oculography (VOG) systems typ- ically require calibration, device-specific settings, or accurate gaze la- bels. We present VOGeo-Gaze, a real-time, calibration-free, geometry- aware neural network that estimates gaze by reconstructing anatomi- cally meaningful eyeball parameters from image features. The method combines segmentation-driven projection geometry, a refraction-aware pupil correction module, and temporal anatomical stabilization, so gaze is derived from interpretable eye geometry rather than direct angular regression. Trained only on the public TEyeD dataset with weak gaze supervision, VOGeo-Gaze was evaluated on 116 clinical recordings from 17 patients and 19 healthy subjects using EyeSeeCam, a clinical gold- standard VOG system. It achieved median absolute angular errors of 0.33{whitebullet} horizontally and 0.35{whitebullet} vertically, with nearly 92% of recordings below 1{whitebullet} error while operating at >300 FPS. These results demonstrate sub-degree clinical gaze estimation without subject-specific calibration, camera intrinsics, or accurate gaze labels, providing a scalable and inter- pretable alternative to conventional VOG pipelines. Code is available at https://github.com/DSGZ-MotionLab/VOGeo-Gaze.

Background: Consensus continuous glucose monitoring (CGM) metrics, including time in range (TIR), time above range (TAR), time below range (TBR), mean glucose, glucose management indicator, and glycemic variability, are essential for modern glucose assessment. However, these whole-day summaries do not explicitly partition nocturnal basal from daytime ambulatory glycemic burden. Objective: To develop and evaluate a complementary domain-based CGM framework that quantifies basal and daytime ambulatory glycemic exposure across oral glucose tolerance test (OGTT)-derived dysglycemia phenotypes. Methods: In this observational, clinic-based study, 253 individuals underwent OGTT with insulin measurement and CGM. Participants were classified using a prespecified OGTT-derived phenotyping algorithm, implemented through a deterministic rules-based web calculator, and collapsed into five groups: NoDM, Increased insulin resistance, Midzone Glycemia, Prediabetes, and Diabetes. CGM files were uniformly reprocessed by selecting the latest contiguous episode and retaining the most recent 15 calendar days with data. The 24-hour profile was partitioned into nocturnal basal (00:00 to <06:00) and daytime ambulatory (06:00 to <24:00) domains. Derived indices included Area of Basal Glycemia (ABG), Area of Prandial/Daytime Ambulatory Glycemia (APG), incremental ABG (iABG), incremental APG (iAPG), and exploratory deficit indices dABG and dAPG. Results: The final dataset contributed 3,647 analyzable CGM days. APG remained higher than ABG across all groups. Mean ABG/APG increased from 80.45/86.38 mg/dL in NoDM to 111.96/124.70 mg/dL in Diabetes. Mean iABG/iAPG increased from 5.65/6.60 to 34.12/38.91 mg/dL, whereas dABG/dAPG declined as dysglycemia worsened. Conclusions: The ABG/APG framework provides interpretable, domain-resolved CGM burden metrics that separate basal from daytime ambulatory exposure and distinguish total burden from above-threshold excess. These indices are proposed as adjunctive metrics to support dysglycemia phenotyping, early risk recognition, and treatment monitoring, but are not intended to replace established consensus CGM metrics or diagnostic criteria. External, prospective validation is required.

The surge in medical imaging has spurred the development of vision-language models (VLMs) to alleviate radiologist workloads. However, clinical deployment is hindered by the lack of meaningful evaluation frameworks. Current metrics - ranging from semantic similarity to large language model (LLM) based judges - often fail to distinguish between clinically trivial and critical discrepancies, poorly reflecting real-world clinical judgment. To address this, we introduce DISCERN (Discordance and Significance-aware Entity-level Radiology Report Comparison). DISCERN is a significance-aware framework that weighs report errors based on their potential impact on patient care. Our results demonstrate that DISCERN powered by closed source LLMs aligns more closely with expert radiologist assessments than traditional metrics or current LLM evaluators, providing a more interpretable and clinically relevant benchmark. By modeling radiologist prioritization and entity-level feedback, DISCERN facilitates targeted model refinement and ensures the safer integration of generative AI into clinical workflows.

Objectives: Diabetes affects over 500 million people globally and glycemia is inadequately managed. Metformin is the most frequently prescribed initial treatment for type 2 diabetes globally, yet glycemic response trajectories to metformin in routine real-world care and predictors of treatment response have not been well described. We aimed to identify glycemic response trajectories in adults prescribed metformin monotherapy as initial type 2 diabetes treatment and predictors of poor glycemic response to metformin. Design: Observational cohort study using latent class mixed models to identify hemoglobin A1c (HbA1c) trajectory classes, followed by random forests machine learning to predict trajectory class membership. Setting: US Veterans Affairs Healthcare System Participants: Adults treated with metformin alone for >30 days after diabetes diagnosis with a minimum of two HbA1c measurements from 90 days prior to two years after the first metformin prescription (N=140,413). Exposures: Demographic, laboratory, vital sign, and comorbidity data were included as predictors of metformin response trajectory Main Outcomes and Measures: We included all HbA1c measurements (487,604 total) for two years after metformin initiation to define metformin glycemic response trajectories. Results: We identified three HbA1c trajectories: stably low (89.7% of sample, mean HbA1c decrease from 7.2% to 6.6%), brisk response (7.1% of sample, mean HbA1c decrease from 11.4% to 7.0%), and non-response (3.1% of sample, mean HbA1c increase from 8.9% to 10.8%). Of those in the stably low and brisk response classes at 2 years, 91% maintained HbA1c at approximately 7% on metformin alone for 5 years after drug initiation. Prediction models could accurately predict brisk response (91% accuracy) but not metformin non-response (59% accuracy). Conclusions: Most individuals treated initially with metformin monotherapy have a beneficial and durable glycemic response. Predicting individuals who will not respond to metformin may be challenging but is evident within six months with recommended glycemic surveillance. The findings support current guidelines for HbA1c surveillance when initiating diabetes treatment.

Background: Traditional diagnostic models lack explainability, while multimodal language models prone to hallucination remain unsafe for medical education. An interactive, risk-free artificial intelligence framework is required to serve as a reliable clinical mentor for radiology trainees. Methods: We propose a multi-agent architecture decoupling deterministic image analysis from generative consultation. Specialized computer vision models perform anatomical localization and pathological segmentation. These quantitative outputs are synthesized into a structured payload, which grounds a locally hosted large language model (LLaVA 7B) using strict prompt guardrails and prerequisite protocols. Results: The system effectively eliminates visual hallucinations by intercepting unanchored queries. The artificial intelligence tutor successfully contextualizes spatial anomalies and baseline metrics, generating accurate conversational explanations and formally structured radiology reports while strictly enforcing medical safety disclaimers. Discussion and Conclusion: By anchoring language generation exclusively to verified algorithmic realities, this framework transforms opaque diagnostic models into safe, interactive educational simulators. This establishes a highly reliable paradigm for integrating explainable artificial intelligence into medical training.

ABSTRACT OBJECTIVE: We performed a systematic review and meta-analysis (SRMA) of post-surgical outcomes, comparing chlorhexidine gluconate (CHG) versus povidone iodine (PI) for vaginal antisepsis of major gynecologic procedures. DATA SOURCES: Ovid Medline, Embase, Scopus, Embase, Cochrane, and Clinicaltrials.gov were searched between 1986 and December 2023, for studies comparing CHG with PI for vaginal antisepsis of major gynecologic operations. STUDY ELIGIBILITY CRITERIA: We included Randomized Controlled Trials (RCTs) and non-RCTs comparing CHG to PI for vaginal antisepsis of major gynecologic operations. The primary outcome was surgical site infections (SSIs) and the secondary outcome was urinary tract infections (UTIs) and vaginal irritation. METHODS: Summary estimates were calculated by fixed effects models when I2 [&le;] 25% and by random effects models when I2 > 25%. Statistical analysis was performed using RevMan 5.4.1. The protocol for this systematic review was registered on PROSPERO (ID CRD42022378101). RESULTS: Nine studies met the inclusion criteria, four of which were randomized controlled trials (RCTs). 9538 patients were included, 4300 (45%) of whom were allocated to CHG and 5238 (55%) to PI. No statistically significant difference in SSI incidence was found for vaginal antisepsis with CHG versus PI in pooled analyses (n= 9538 patients; RR 1.20; 95% CI 0.92-1.57; I2 =0%). In contrast, a significantly higher risk of UTIs was observed for vaginal antisepsis with CHG than with PI (n=6061 patients; RR 1.48 95% CI 1.03-2.14; I2 = 0%). CONCLUSION: In our SRMA, there were no significant differences in SSI risk when either CHG or PI was utilized for antiseptic vaginal preparation. Interestingly, vaginal antisepsis with PI was associated with a lower incidence of post-operative UTIs following major gynecologic surgery. Our findings support current guidelines that form of vaginal antisepsis can be used for SSI prevention. They also suggest that PI may result in fewer postoperative UTIs but further randomized studies are needed to support these findings. Key words: surgical site infection, surgical wound infection, urinary tract infection, urogynecologic surgery, Chlorhexidine, Povidone Iodine, surgical antiseptic,

The Epic Sepsis Model version 2 (ESMv2) is a prediction model embedded into the electronic medical record used to warn clinicians which hospitalized patients are at risk for sepsis. We conducted a retrospective cohort study of 31,951 hospitalizations of 25,760 patients to compare analyses conducted at the commonly used patient-level (where a maximum prediction prior to the onset of sepsis is used to measure performance) vs novel prediction-level (where each prediction is used to measure performance). Sepsis, defined by the Sepsis 3 criteria occurred during 1,049 hospitalizations (3.3%). Patient-level analyses suggested excellent discrimination AUC 0.86; [IQR 0.85, 0.87], whereas prediction-level analyses demonstrated lower performance AUC 0.62; [IQR 0.57, 0.65]. Low estimates of the positive predictive value (14.5% at the patient level vs 4% at the prediction level) imply a high number of false alerts. Common evaluation approaches may overstate the performance of dynamic prediction models and mislead clinical decision-making.

Background: Despite the success of combination antiretroviral therapy (cART), vascular comorbidities, including cerebrovascular disease, are more prominent in people living with HIV (PLWH) compared to people without HIV (PWOH). However, quantitative assessments of cerebrovascular morphometry and their associations with cognitive outcomes in the context of HIV are still limited. In this study, we explore this missing link. Methods: Magnetic Resonance Angiography (MRA) data, blood markers, and neurocognitive assessments were collected from 73 PWOH subjects (male: 57, female: 16; age: 53 {+/-} 16) and 99 PLWH subjects (male: 66, female: 30, age: 53 {+/-} 11). Vessel morphometric features were quantified using intraCranial Artery Feature Extraction (iCafe) to investigate associations between vessel morphometry, markers of monocytes, endothelial cell activation, and cognitive performance. Results: HIV status predicted a lower total number of branches ({beta} = -0.224, p = 0.001, d = -0.517) and shorter total distal length ({beta} = -0.173, p = 0.021, d = -0.370) with a moderate effect size. Total branch number was found to be negatively associated with plasma levels of monocyte markers (sCD14: r = -0.167, p = 0.033; sCD163: r = -0.157, p = 0.045) and positively correlated with white matter cerebral blood flow (r = 0.550; p [&le;] 0.05). HIV status was the strongest predictor of overall cognitive performance in ANCOVA model ({beta} = -0.219, p = 0.006, d = -0.453). Conclusions: Our results suggest that cognitive impairment in PLWH is associated with vessel morphology metrics. Monocyte immune activation may contribute to changes in vessel morphology.

Patients with primary immunodeficiency (PID) often face prolonged diagnostic delays and may increasingly turn to large language models (LLMs) to interpret their symptoms during this period. We evaluated whether an LLM could recognize PID from symptom descriptions derived from interviews with 21 PID patients. In a prior study, we showed that GPT-4o identified PID in 96% of cases when prompted with physician-written patient histories (Rider et al., JACI, 2024). Here, when prompted with symptom descriptions in patients' own words, GPT-5 identified PID in only 7 cases (33%), although it more broadly suggested immune system issues in 18 cases (81%). The gap between these findings indicates that LLMs are sensitive to the language and framing of symptom descriptions, performing substantially worse when patients describe their own symptoms in everyday language than when clinicians summarize patient histories in structured medical terms. This study underscores the need to carefully evaluate how LLMs are used in patient-facing applications.

Background Atrial fibrillation (AF) is a leading cause of stroke, cardiovascular morbidity, and mortality. Atrial myopathy, characterized by progressive metabolic, electrical, and structural changes, creates the arrhythmogenic substrate that drives AF. Defining the key drivers of atrial myopathic processes is essential for targeted therapies that can mitigate AF progression. Here we explore how reduced ERBB4 expression contributes to the development of left atrial myopathy. Methods We analyzed the Cleveland Clinic Biobank to compare left atrial ERBB4 levels in patients grouped by AF diagnosis. To investigate the impact of reduced ERBB4 levels on atrial tissue substrate, we created mouse models of cardiac-specific Erbb4 deficiency using Mlc2a (myosin light chain 2a)-Cre. Comprehensive physiological assessments were performed. Transcriptomic analyses of the left atrium were performed in an Erbb4 haploinsufficient mouse model and compared with human atrial datasets. Molecular validation of key dysregulated pathways was performed. Results We found that left atrial ERBB4 levels are reduced in patients with AF. Adult cardiomyocyte-specific Erbb4 heterozygous (Erbb4fl/+;Mlc2a-Cre) mice exhibited prolonged P-wave duration in the absence of ventricular dysfunction. Left atrial transcriptomic analysis in Erbb4 haploinsufficient mice showed upregulation of pathways related to fibrosis, apoptosis, and coagulation, and downregulation of pathways related to fatty acid metabolism and mitochondrial function, mirroring changes observed in pressure overload mouse models. A cross-species transcriptomic comparison revealed significant overlap between ERBB4-correlated gene expression and functional pathways in adult human atria and mice with Erbb4 haploinsufficiency. Validating the transcriptomic data, protein and functional assays demonstrated increased fibrosis, apoptosis, and oxidative stress in the mutant left atrial tissue. Conclusion Left atrial ERBB4 levels are reduced in AF patients. A mouse model of Erbb4 deficiency and human atrial transcriptomic analyses highlight a role for ERBB4 in supporting normal atrial metabolism while protecting against inflammation, apoptosis, and fibrosis.

Intro: Characteristics of the pulse wave transmitted through the carotid arteries are predictive of cognitive decline and cerebrovascular health in humans. This study aimed to identify risk factor trajectories in childhood, adolescence and early adulthood that are associated with forward compression wave intensity (FCWI) in the common carotid artery in adults aged 28 years. Methods: Systolic blood pressure (SBP), body mass index (BMI) and fasting blood glucose (FBG) measured at multiple time-points when participants were aged between 8-20 years were included in a trajectory analysis. At age 28 years, FCWI was measured in 402 (M=206, F=196) participants who underwent a Duplex ultrasound assessment of the common carotid artery. Statistical analysis assessed differences in FCWI between each trajectory group for males and females separately. Results: In males, four trajectory groups were identified for BMI, three for SBP, and two for FBG. In females, three trajectory groups were identified for BMI, SBP, and FG. In males, having higher BMI (P=0.006), SBP (P=0.021) and FBG (P=0.002) from ages 8-20 years was associated with greater FCWI at age 28 years. In females, no associations were found between FCWI at age 28-years and trajectory groups for BMI (P=0.185), SBP (P=0.289) or FBG (P=0.070). Conclusion: Having high BMI, SBP and FBG throughout childhood, adolescence and early adulthood was associated with higher FCWI in the carotid artery at age 28 years in males, but not females. This may have a direct impact on the etiology of cognitive decline and cerebrovascular disease in later life.

Lead is a toxic metal ubiquitous in our environment. While dramatic reductions in lead sources have paralleled equivalent decreases in lead-poisoning rates, chronic lead exposure remains a critical public health concern. Childhood lead exposure (at its lowest levels) is liked to changes in cognitive development but less is known about lead's effects on children's brain structure, especially as a result of in utero exposure. We measured prenatal and early-postnatal lead exposure in shed deciduous teeth of 448 9- and 10-year-old children (from 20 United States cities) and linked those lead levels to childhood brain structure, cognition/behavior, and neighborhood- and family-level socioeconomic characteristics. Here we show negative associations between tooth-lead levels and the thickness of the brain's cortex, particularly in regions linked to language processing. With increasing tooth-lead levels, children of lower-income (versus higher-income) families showed steeper declines in receptive vocabulary. Caregiver-reported behavioral problems exhibited similar associations. With in utero exposure linked to adverse neurodevelopmental outcomes (well before lead exposure and its risks are evaluated by healthcare professionals), prenatal screening of maternal lead levels/exposure, coupled with recommended strategies to reduce its placental transmission, may help reduce lead's effects on future generations.

Background: Conventional psychiatric screening instruments summarize symptoms within individual scales and prioritize cases with high single-instrument additive score severity. This design treats items as independent within instruments and ignores cross-instrument covariance structure, making it insensitive to respondents whose responses are distributed across multiple domains in unusual combinations that remain below threshold on every individual scale. Methods: We analyzed two cohorts spanning older and younger adults. Item prompts from depression, stress, anxiety, and sleep instruments were embedded into a shared semantic space using a pretrained sentence encoder. Principal component analysis of the item-prompt embeddings alone---with no use of respondent data at this stage---was used to construct a low-dimensional subspace retaining 80\% of variance in the item embedding matrix. Normalized participant responses were then projected into this subspace, with Jaccard-based stability analysis used as a check on dimensional robustness. Multivariate deviation from the cohort norm was quantified with Mahalanobis distance using Ledoit-Wolf covariance regularization. Candidate outliers were defined by the empirical 95th percentile of the cohort-specific distance distribution. To isolate response configurations not already captured by conventional single-instrument extreme-value logic, we excluded all outlier respondents who had endorsed any individual item at the maximum value of its Likert scale on any instrument. For the remaining outliers, anomalous components were backtracked to their original item loadings for interpretation. Results: In the older-adult Health and Retirement Study (HRS) cohort, principal component analysis of 27 item-prompt embeddings showed that a 10-dimensional subspace provided a stable representation of cross-instrument semantic structure. In the younger-adult Xinxiang cohort the corresponding stable solution was 16-dimensional. In each cohort, seven respondents remained as multivariate outliers despite falling below every single-instrument extreme-value threshold. These cases were not characterized by uniformly severe symptom scores but by unusual cross-domain response configurations that became visible only in the shared semantic covariance subspace. The response structure of the retained configurations differed across cohorts: older-adult cases more often involved weak endorsement of mood-labeled items alongside nonzero body- and sleep-related responses, whereas younger-adult cases more often involved incomplete response configurations spanning mood, sleep, stress, and self-harm-related items. Conclusions: A semantically aligned, auditable covariance subspace provides a practical tool for flagging unusual multivariate response configurations that single-instrument additive screening may not flag. The method is interpretable at the level of original item contributions. It should be understood as a hypothesis-generating screen for unusual response configurations requiring further clinical assessment, not as a diagnostic instrument. Outcome validity remains to be established by prospective study.

Hybrid mechanistic-statistical models offer interpretability and adaptability for short-term seasonal epidemic forecasting, but it remains unclear whether their accuracy depends more on increased biological complexity or on the assimilation of richer data. Using eight retrospective influenza seasons in North Carolina, we evaluate whether training on historical data and assimilating auxiliary emergency department (ED) visit data improves four-week-ahead hospital admission forecasts more than adding biological complexity (multi-subtype structure and cross-season immunity). Hierarchical Bayesian training on historical data improves accuracy by 22.4 % (95 % CI: 16.4-28.1 %), and inclusion of ED visit data yields a further 5.3 % (95 % CI: 3.0-7.6 %) improvement, whereas added biological complexity produces diminishing or null gains. We further observe a substitution effect in which ED visit data partially compensates for omitted biological structure. We deployed a simplified model variant in the 2025-2026 CDC FluSight Challenge and ranked among the top ensemble performers, supporting the robustness of Bayesian hierarchical training in real time. Together, these findings indicate that short-term forecast accuracy is driven more by historical learning and assimilating auxiliary signals than by biological fidelity, with implications for how forecasting systems should balance mechanistic complexity.

Background: Non-communicable diseases (NCDs) represent a critical public health challenge in Kenya, responsible for over 50% of inpatient admissions and 40% of deaths. While digital health tools and artificial intelligence offer promising ways to improve prevention, diagnosis, and management, little is known about how these tools are perceived and used in practice. There is limited research exploring the views and lived experiences of young people in Kenya, who are a strategic priority for NCD prevention because behavioral risk factors are established in this window, and for Community Health Providers (CHPs) who provide health services within the community. This study aims to address this gap by examining the perspectives of the burden of non-communicable diseases and the potential role of digital health technologies, including artificial intelligence, for preventing and managing these conditions in these specific populations. Methods: A qualitative research design using focus group discussions (FGDs) was employed in Nairobi (urban) and Busia (rural) counties between March and July 2024. Eight FGDs were conducted with 60 participants purposively sampled from three stakeholder groups: community health promoters (CHPs), healthcare workers (HCWs), and youth aged 18-35 years. A semi-structured guide, co-developed with a Community Advisory Board, explored beliefs about NCDs, health-seeking behaviors, lifestyle practices, and attitudes toward digital health and AI. Audio recordings were transcribed verbatim, translated where necessary, and analyzed thematically using grounded theory principles on NVivo software (v12). Results: Six consolidated themes emerged: (1) understanding of NCDs and perceived risk; (2) barriers to NCD prevention and care; (3) the role of CHPs; (4) adoption of AI tools for NCD management; (5) trust, ethics and access concerns; and (6) community-driven recommendations for AI integration. Significant barriers including stigma, economic constraints, and barriers to care were documented alongside enthusiasm for AI tools among youth and CHPs in both urban and rural areas. Conclusion: This study shows that AI tools are being used for NCD prevention and management through spontaneous community adoption. However, it emphasizes the need for culturally relevant, equitable, and community-driven solutions. Effective scaling requires the identification and bridging of digital literacy gaps, the establishment of affordable infrastructure, the protection of data privacy, and the integration of artificial intelligence tools into existing community health frameworks. This process should involve the collaboration of trusted intermediaries, such as CHPs and community leaders, to ensure successful outcomes. Future initiatives should prioritize participatory design, policy frameworks for ethical governance, and targeted capacity building to enhance acceptance and sustainability of digital health innovations in low- and middle-income country settings.

Background: Thalamic low-intensity transcranial focused ultrasound (tFUS) has shown promise for increasing behavioral responsiveness in disorders of consciousness (DOC), but no study has examined whether it can causally modulate the well-validated behavioral, electrophysiological, and metabolic biomarkers of DOC impairment. Methods: Sixteen adult patients (44% Female; Age, M=37.81, SD=15.97) with a chronic DOC (Time Since Injury, M=3.39, SD=1.94 years) secondary to severe brain injury (TBI 44%, non-TBI 56%) underwent a 10-day inpatient, longitudinal, single-arm, open-label protocol. tFUS was delivered in a single session targeting the left central thalamus. Well-known behavioral (CRS-R), electrophysiological (EEG {delta}/{beta} ratio), metabolic (18F-FDG PET), and polysomnographic outcomes were assessed at baseline and after sonication. Results: The maximum CRS-R total score increased significantly following tFUS compared to baseline (M=13.27 vs. M=10.33; t(14)=7.407, p<0.001, d=1.913), as did the global EEG {delta}/{beta} ratio (N=14; W=17, p=0.025, r=0.68), with the degree of frontal slowing positively predicting behavioral gains ({tau}b=0.51, p=0.016). Glucose metabolism decreased bilaterally in thalamus and frontal, temporal, and parietal cortices at both post-tFUS timepoints compared to baseline. Finally, N2 sleep increased by 33% following tFUS (N=11; t(10)=2.386, p=0.038, d=0.72), though this did not survive correction. No severe adverse events were observed. Conclusion: Thalamic tFUS can causally modulate well-validated behavioral, electrophysiological, and metabolic biomarkers of DOC. The convergent inhibitory signature across these measures suggests a thalamocortical reset mechanism, complementing existing excitatory neuromodulation approaches and providing the mechanistic foundation for a large, randomized sham-controlled trial.

Objective: Hispanic/Latinx populations in the U.S. experience higher rates of chronic disease linked to physical inactivity, yet digital health interventions remain largely inaccessible to more than 16 million Hispanic/Latinx adults with limited English proficiency. While large language models (LLMs) offer scalable personalization, their use in non-English behavioral coaching is unexplored. This study introduces MHC-Coach-ES, a Spanish-language LLM fine-tuned on the Transtheoretical Model (TTM) of behavior change. Materials and Methods: We fine-tuned Llama 3-70B-Instruct using a two-stage pipeline. First, the model was adapted to Spanish health and motivational language using a 2.21-million-token corpus. Second, it was instruction-tuned on 3,268 translated human written messages to align the model with the Transtheoretical Model (TTM) of Behavioral Change. We compared MHC-Coach-ES with Llama 3-70B-Instruct and translated human-expert messages using a forced-choice preference survey (N = 77) and blinded expert review (N = 2). Results: Spanish-speaking participants significantly preferred MHC-Coach-ES messages over translated human-expert messages (81% preference, P<0.001). Linguistic analysis showed that MHC-Coach-ES produced more temporally anchored messages than the base model (65% vs. 20%), while maintaining readability. In blinded evaluation, clinical experts rated MHC-Coach-ES higher for alignment with Transtheoretical Model stages than human-expert messages (4.83 vs. 4.38 out of 5). The base model also outperformed translated expert messages across preference and expert ratings. Conclusions: Generative AI can operationalize behavioral science frameworks in Spanish, offering a scalable approach to reducing health disparities. The strong performance of both MHC-Coach-ES and the base model highlights the promise of generative and personalized approaches over translation-based localization for theory-driven behavioral interventions.