PLOS digital health最新文献

{"title":"Foundation model-powered deep learning of endometrial histology for predicting the cumulative live birth of an in vitro fertilization cycle.","authors":"Nianbo Xu, Andy Chun Hang Chen, Hanzhang Ruan, Donglin Yang, Renjie Liao, Xiaojuan Qi, Sze Wan Fong, Dandan Cao, Lu Yu, Yuanhua Huang, William Shu Biu Yeung, Ernest Hung Yu Ng, Yin Lau Lee","doi":"10.1371/journal.pdig.0001744","DOIUrl":"10.1371/journal.pdig.0001744","url":null,"abstract":"<p><p>Objective and reproducible assessment of endometrial receptivity is essential for optimizing in vitro fertilization (IVF) success, yet traditional histological dating suffers from observer variability. This study investigated whether deep learning of hematoxylin and eosin histology images could support cumulative live birth prediction in IVF. An end-to-end ResNet-18 was compared with a UNI2-h-based pipeline, using UNI2-h as a frozen feature extractor. Ten-fold cross-validation ensembles were developed from natural-cycle endometrial biopsies and evaluated in an internal held-out cohort with live birth outcomes. Additional phase-based evaluation was performed, which tested the performance in distinguishing LH + 7 versus non-LH + 7 phases. A luminal epithelium (LE)-focused design was also assessed to examine whether concentrating on maternal-embryo interface could improve fertility-oriented learning. The whole-slide ResNet-18 model performed well in internal outcome-based testing but near random in external phase-based testing. In contrast, the UNI2-h model with mean pooling and a multilayer perceptron classifier showed less divergent performance, with ensembled AUROCs of 0.74 ± 0.03 in internal outcome-based evaluation, and 0.90 ± 0.05 in external phase-based testing. Despite a smaller training set, LE-focused models retained comparable performance. In internal outcome-based testing, LE-focused ResNet-18 and UNI2-h models achieved AUROCs of 0.71 ± 0.03 and 0.74 ± 0.09 respectively. External phase-based testing yielded AUROCs of 0.80 ± 0.16 for ResNet-18, and 0.84 ± 0.06 for UNI2-h. Grad-CAM review of LE-focused ResNet-18 models showed attention commonly on LE-alone or mixed with adjacent stroma. In multimodal analyses, integrated models incorporating histology significantly outperformed the clinical metadata-only model. Integrated model's feature weighting showed dominant histology outputs, smaller contributions from estradiol and maternal age, and negligible contributions from progesterone, endometrial thickness, and BMI. These findings support histology-based AI for fertility-oriented endometrial assessment and highlight biologically informed design and repurposed foundation models as promising bases for clinically meaningful prediction.</p>","PeriodicalId":74465,"journal":{"name":"PLOS digital health","volume":"5 9","pages":"e0001744"},"PeriodicalIF":7.7,"publicationDate":"2026-09-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13618903/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149077059","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Independent evaluation of machine learning and deep learning models for breast cancer detection.","authors":"Deepthi, Ashalatha Nayak, Rani Oomman Panicker","doi":"10.1371/journal.pdig.0001747","DOIUrl":"10.1371/journal.pdig.0001747","url":null,"abstract":"<p><p>Breast cancer is one of the leading causes of cancer-related mortality worldwide, and despite advances in clinical diagnosis, challenges such as inconclusive imaging results and inter-observer variability highlight the need for complementary computational approaches to support early and accurate detection. In this study, we provide an overview of several machine learning (ML) and deep learning (DL) algorithms used for breast cancer detection with structured clinical data and imaging data. We applied ML models to structured clinical data which included Wisconsin Breast Cancer Dataset, CSAW-CC dataset and SEER database using Random Forest, Decision Tree, Logistic Regression, Gradient Boosting, and Bayesian classifier models. DL models, specifically Convolutional Neural Networks (CNN), ResNet50, VGG16, and DenseNet, were utilized in mammography, histopathology, and ultrasound datasets. The experimental results show that the DL models outperformed ML models in image classification tasks, obtaining accuracy rates of 97.88% for histopathological images and 92.00% for ultrasound data. However, ensemble-based and probabilistic ML models achieved strong performance on structured clinical data, with predictive accuracy of up to 98%. All performance metrics are reported with 95% confidence intervals.Model calibration was assessed using the Expected Calibration Error (ECE) and reliability diagrams. Logistic Regression, Gradient Boosting, and Random Forest produced the best-calibrated probability estimates among the structured-data classifiers, while the transfer-learning-based deep convolutional architectures were well calibrated on the histopathology imaging datasets. Calibration was comparatively weaker for Decision Tree and Naive Bayes among the structured-data classifiers, as well as for the from-scratch CNN baseline and some deep learning models evaluated on the smaller BUSI dataset. The results indicate that model performance is significantly influenced by data modality and that artificial intelligence may serve as a valuable computational tool to support clinical breast cancer diagnosis workflow. Further prospective validation using clinical data across diverse patient populations will be essential to confirm the robustness, interpretability, and generalizability of these findings.</p>","PeriodicalId":74465,"journal":{"name":"PLOS digital health","volume":"5 9","pages":"e0001747"},"PeriodicalIF":7.7,"publicationDate":"2026-09-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13618913/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149077125","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Exploring needs and priorities in digital health management for rare disease patients and their caregivers: A mixed-methods study.","authors":"Anita Burgun, Christina Khnaisser, Roxanne Dault, Jean-François Ethier","doi":"10.1371/journal.pdig.0001240","DOIUrl":"10.1371/journal.pdig.0001240","url":null,"abstract":"<p><p>Rare diseases affect millions worldwide and are associated with long diagnostic delays, limited access to treatments, and substantial challenges in daily care and coordination. Digital health technologies, including mobile apps, tele health, and data‑sharing platforms, offer opportunities to improve care and quality of life for people living with rare diseases. As these tools rapidly expand, this study examines the needs, expectations, and conditions for successful adoption of patient‑centered digital solutions among individuals living with rare diseases and their families. Using a mixed‑methods design, we surveyed 149 patients and caregivers, and conducted follow-up focus groups with 15 participants. Our findings highlight the essential role of digital tools in supporting people with rare diseases and their families. Key priorities include centralized health data, support for patient‑generated data, and improved communication and information exchange with clinicians. Participants strongly emphasized the value of telehealth to reduce travel and simplify daily life, as well as patient‑centered tools for diagnosis and emergency situations. Future digital solutions should integrate system‑wide data, incorporate AI, and provide support during stressful situations, ultimately reducing patient burden despite persistent structural challenges. Respondents expressed strong interest in technologies that place patients at the center of care and improve coordination across providers. Overall, our study identifies actionable targets for innovation and highlights technological, regulatory, and resource‑related barriers that must be addressed to advance patient‑centered digital solutions for rare diseases and guide future research and policy development.</p>","PeriodicalId":74465,"journal":{"name":"PLOS digital health","volume":"5 9","pages":"e0001240"},"PeriodicalIF":7.7,"publicationDate":"2026-09-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13618911/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149077041","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"A pre-post pilot feasibility study of the MindBia web-based nutrition intervention for individuals with severe mental illness: Study protocol.","authors":"Ciara O'Sullivan, Tara Coppinger, Indika Dhanapala, Diarmuid Boyle, Alison Merrotsy","doi":"10.1371/journal.pdig.0001756","DOIUrl":"https://doi.org/10.1371/journal.pdig.0001756","url":null,"abstract":"<p><p>People living with severe mental illness experience substantial health inequalities and high rates of premature mortality, largely driven by modifiable lifestyle-related risk factors, including poor diet quality. Digital technologies may provide scalable and flexible opportunities to support healthier behaviours; however, few interventions have specifically targeted nutrition-related outcomes as a primary focus in this population. MindBia is a co-designed, web-based digital nutrition intervention consisting of six structured modules developed for individuals residing in high-support mental health hostels. The intervention was informed by prior studies with service users and mental health staff. This pilot study aims to evaluate the feasibility, acceptability, and preliminary effectiveness of the MindBia intervention over eight weeks. A mixed-method pre-post design will be employed. Participants will complete baseline and post-intervention assessments examining dietary behaviours, digital literacy, and general wellbeing. Engagement with the web-based application will be monitored using embedded usage analytics. Semi-structured interviews conducted post-intervention will explore participants' experiences, perceived impact, and barriers to engagement. Findings from this pilot study will inform refinement of the MindBia web-based application and guide the design of a larger scale trial evaluating its overall effectiveness.</p>","PeriodicalId":74465,"journal":{"name":"PLOS digital health","volume":"5 9","pages":"e0001756"},"PeriodicalIF":7.7,"publicationDate":"2026-09-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13614651/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149066610","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Comparative predictive modeling of pediatric spirometry reference equations in Jordanian children: Complex versus simple models.","authors":"Walid Al-Qerem, Anan Jarab, Judith Eberhardt, Khalda Smairan, Yousef Mimi, Maher Khdour","doi":"10.1371/journal.pdig.0001746","DOIUrl":"https://doi.org/10.1371/journal.pdig.0001746","url":null,"abstract":"<p><p>Spirometric interpretation relies on reference equations, yet equations developed in one population or age range may not transport. Reference equations use distributional models to account for nonlinear growth, their calibration may differ across populations, it remains uncertain whether machine-learning algorithms improve prediction beyond simpler transformed regression models. This two-phase cross-sectional study compared sex-specific predictive models. Phase 1 used the same 1,576-child derivation dataset used to develop the original Jordanian GAMLSS equation (Al-Qerem equation), allowing comparison with predictive modeling strategies. Phase 2 evaluated equations in a validation sample of 1,007 healthy children aged 6-18 years. Candidate models were evaluated on the scale after back-transformation of log-outcome predictions. Models included GBM for FEV1 in both sexes, GLM for FVC in both sexes, GLM for FEV1/FVC in girls, and GBM for FEV1/FVC in boys.FEV1 and FVC were predicted more accurately than FEV1/FVC, whose explained variance from age and height remained low across model classes and established equations. In external validation, the study model had the lowest mean squared error for female FEV1 and FVC, but did not consistently outperform GLI 2012, GLI 2022, or Al-Qerem equations in boys or for FEV1/FVC. Age-stratified analyses showed elevated FEV1 and FVC below-LLN rates in boys younger than 10 years across the study, Al-Qerem, GLI 2012, and GLI 2022 equations, whereas FEV1/FVC below-LLN proportions were generally closer to nominal values. These findings support comparative predictive modeling as a useful development framework, but do not show superiority of complex machine-learning methods or readiness for clinical deployment without further calibration and external validation. Carefully chosen transformations and calibration were more important than model complexity, and the best models differed by outcome and sex.</p>","PeriodicalId":74465,"journal":{"name":"PLOS digital health","volume":"5 9","pages":"e0001746"},"PeriodicalIF":7.7,"publicationDate":"2026-09-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13614654/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149066630","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Optimizing accelerometer implementation in a gerotherapeutic trial: Feasibility, adherence, and operational insights of ABLE.","authors":"Jessica K Lu, Weilan Wang, Yi Ern Chew, Lihuan Guan, Brian K Kennedy, Andrea B Maier","doi":"10.1371/journal.pdig.0001691","DOIUrl":"10.1371/journal.pdig.0001691","url":null,"abstract":"<p><p>The use of wearable technology for digital phenotyping has been limited in gerotherapeutic trials. Thus, it is important to identify insights into the operational feasibility of adopting accelerometers for digital phenotyping in gerotherapeutic trials, participant adherence to device and intervention, and study dropout risk, particularly in middle-aged adults. This exploratory analysis utilized data from 120 generally healthy, middle-aged adults participating in the Alpha-ketoglutarate (AKG) supplementation and BiologicaL agE in middle-aged adults (ABLE) double-blind, randomized, placebo-controlled trial (ClinicalTrials.gov identifier: NCT05706389). Participants were instructed to wear an ActiGraph wGT3X-BT accelerometer for seven days at four study visits, adhering to a six-month 1 g/day oral AKG supplement regimen with a three-month follow up. Device operational concerns, accelerometer wear behaviour, device wear and supplement use adherence, and study dropouts were evaluated. Associations between device wear (time or adherence) and dropout were analysed using Cox proportional hazards models using study visit as the time scale to account for when participants dropped out from the study. Participants (mean age 50.6 ± 6.0 years, 50% female, n = 59 AKG, n = 61 placebo) showed high adherence: median accelerometer wear was 24 hours per day, > 80% achieved ≥ 4 valid days per visit, and supplement use adherence >90%, with minimal device-related operational concerns across all four visits. Overall study dropout was 17%. In Cox models across study visits, one minute higher in wearing time (hazard ratio (HR): 0.998; 95% confidence interval (CI): 0.997-0.999) and being male (HR: 0.208; CI: 0.065-0.663) were significantly associated with lower dropout risk. Accelerometer deployment proved highly feasible in this gerotherapeutic trial with middle-aged adults. Higher device wearing time and being male were significantly associated with lower study dropout, informing strategies to enhance future trial success.</p>","PeriodicalId":74465,"journal":{"name":"PLOS digital health","volume":"5 9","pages":"e0001691"},"PeriodicalIF":7.7,"publicationDate":"2026-09-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13614584/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149066583","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Artificial intelligence-related perceptions and E-health literacy among nursing students in Palestine: A cross-sectional study.","authors":"Samar Thabet Jallad, Maysa Kassabry, Adel Taher Takruri","doi":"10.1371/journal.pdig.0001737","DOIUrl":"https://doi.org/10.1371/journal.pdig.0001737","url":null,"abstract":"<p><p>Artificial intelligence (AI) and e-health literacy are increasingly relevant to nursing education; however, the relationship between nursing students' perceptions and acceptance of AI and their perceived ability to locate, evaluate, and apply online health information remains insufficiently examined in Palestine. This cross-sectional study examined e-health literacy, AI-related technology-acceptance perceptions, Internet-use factors associated with these constructs, and whether overall AI-related perceptions were independently associated with e-health literacy among 113 third- and fourth-year nursing students. Participants completed the 8-item eHealth Literacy Scale (eHEALS) and a 28-item adapted technology-acceptance questionnaire covering seven AI-related perception domains, with negatively worded items reverse scored where required. Descriptive statistics, regrouped one-way analyses of variance, Pearson correlations with Holm adjustment for multiple testing, and multiple linear regression were performed. The mean eHEALS score was 3.53 (SD 0.59), and the overall AI-related perceptions score was 3.35 (SD 0.34). Overall AI-related perceptions were positively correlated with eHEALS (r = 0.387, Holm-adjusted p < .001). Facilitating conditions showed the largest domain-level correlation with eHEALS (r = 0.450, Holm-adjusted p < .001), whereas anxiety was not significantly associated with eHEALS. After adjustment for year of study, perceived Internet skills, and perceived importance of the Internet, overall AI-related perceptions remained independently associated with eHEALS (B = 0.643, 95% CI 0.289-0.998, p < .001; model R2 = 0.177). Among the regrouped Internet-use comparisons, only perceived usefulness of the Internet remained significant after Holm correction. Overall, AI-related perceptions and perceived e-health literacy were positively associated in this sample, although the cross-sectional design precludes conclusions regarding directionality or causality. Further multi-site, longitudinal, and intervention studies are warranted, and domain-level findings should be interpreted cautiously because several adapted subscales demonstrated low internal consistency.</p>","PeriodicalId":74465,"journal":{"name":"PLOS digital health","volume":"5 9","pages":"e0001737"},"PeriodicalIF":7.7,"publicationDate":"2026-09-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13614630/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149066563","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Strategies for integrating artificial intelligence and cognitive assessment to predict disability progression in relapsing-remitting multiple sclerosis: A model development study.","authors":"Chao Zhu, Xin Zhang, Anneke van der Walt, Bruce Taylor, Mastura Monif, Tomas Kalincik, Jeanette Lechner-Scott, Katherine Buzzard, Trevor Kilpatrick, Michael Barnett, Zhen Zhou, Vilija Jokubaitis, Melissa Gresle, David Darby, Deval Mehta, Zongyuan Ge, Helmut Butzkueven, Daniel Merlo","doi":"10.1371/journal.pdig.0001722","DOIUrl":"10.1371/journal.pdig.0001722","url":null,"abstract":"<p><p>Serial cognitive assessment in multiple sclerosis can help identify patients at higher risk of disability progression, but high dimensionality of reaction-time data makes analysis challenging. We developed a method to convert longitudinal reaction-time data into images and applied deep survival modelling to predict disability progression in relapsing-remitting MS (RRMS). In this cohort study, clinical data were obtained from the MSBase registry and cognitive data from the MSReactor computerised cognitive battery between February 2016 and September 2022, with a median follow-up of 3.2 years. The serial reaction-time data from three tasks (psychomotor function [R], attention [G], and working memory [B]) were converted into multicolour images using RGB encoding. We extracted the key features from the images using a convolutional neural network and combined them with clinical variables in a Transformer-based survival model (MS-TranSurv) to predict time to confirmed disability progression. Discrimination, calibration, and accuracy were assessed using the C-index, integrated Brier score (iBS), and time-dependent area under the receiver operating characteristic (tAUROC). Performance was compared with other machine learning models, including Dynamic DeepHit, Recurrent Deep Survival Machines, a Cox proportional hazards model in traditional statistics using clinical variables only, and a version of MS-TransSurv using mean test values. A total of 746 RRMS patients were included. MS-TranSurv showed slightly higher discrimination compared with benchmark models, with a C-index of 0.61 (95% CI 0.54-0.68) and tAUROC of 0.74 (95% CI 0.63-0.85), as well as comparable calibration, with an iBS of 0.24 (95% CI 0.16-0.32). Using individual test-level data provided slightly better performance than using summary measures. Longitudinal cognitive reaction-time data can be used for survival-based prediction of disability progression in RRMS. This framework supports remote cognitive monitoring and may improve risk stratification and clinical trial enrichment, with potential applicability to other high-dimensional digital biomarkers.</p>","PeriodicalId":74465,"journal":{"name":"PLOS digital health","volume":"5 9","pages":"e0001722"},"PeriodicalIF":7.7,"publicationDate":"2026-09-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13614581/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149066558","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Augmenting patient safety surveillance in radiation oncology with large language model-based root cause analysis: A proof-of-concept study.","authors":"Yuntao Wang, Mariluz De Ornelas, Matthew T Studenski, Elizabeth Bossart, Siamak P Nejad-Davarani, Yunze Yang","doi":"10.1371/journal.pdig.0001740","DOIUrl":"10.1371/journal.pdig.0001740","url":null,"abstract":"<p><p>We investigated the potential utility of large language models (LLMs) in supporting patient safety efforts. Specifically, we evaluated the reasoning capabilities of LLMs in performing root cause analysis (RCA) of radiation oncology incidents using narrative reports from the Radiation Oncology Incident Learning System (RO-ILS). We prompted four state-of-the-art LLMs, Gemini 2.5 Pro, GPT-4o, o3, and Grok 3, with the \"Background and Incident Overview\" sections from 19 publicly available RO-ILS cases. Each model was instructed to perform RCA and generate root causes, lessons learned, and suggested actions using a standardized prompt based on AAPM RCA guidelines. Model outputs were evaluated using a combination of objective semantic similarity metrics (cosine similarity via Sentence Transformer), semi-subjective assessments (precision, recall, F1-score, expert-adjudicated PPV (Positive Predictive Value), hallucination rate and performance criteria including relevance, comprehensiveness, quality of justification and quality of solution), and subjective ratings (reasoning quality and overall performance) by five board-certified medical physicists. LLMs demonstrated satisfactory performance across evaluation metrics. All models exhibited some degree of hallucination, ranging from 11% to 61%. All the evaluated LLMs demonstrated comparable baseline capabilities in objective causal extraction, and Gemini 2.5 Pro exhibited the highest overall performance score among 4 models. Statistically significant differences were observed among models in expert-adjudicated PPV, hallucination rate, and subjective ratings (p < 0.05). LLMs delivered promising results as assistive tools for RCA in radiation oncology, with the ability to generate relevant and accurate analyses aligned with expert expectations. LLMs may support incident analysis and contribute to quality improvement efforts to advance patient safety in clinical radiation oncology practice.</p>","PeriodicalId":74465,"journal":{"name":"PLOS digital health","volume":"5 9","pages":"e0001740"},"PeriodicalIF":7.7,"publicationDate":"2026-09-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13614575/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149066612","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"SYNERGY-VAE: An explainable generative deep learning framework for discovering depression subgroups from multimodal population health data.","authors":"Divya Sharma, Alice Rueda, Qiaowei Lin, Shakila Meshkat, Argyrios Perivolaris, Venkat Bhat","doi":"10.1371/journal.pdig.0001719","DOIUrl":"10.1371/journal.pdig.0001719","url":null,"abstract":"<p><p>Depression represents a complex, multifactorial disorder influenced by the interplay of biological, behavioral, and social determinants. While extensive health surveys, including the National Health and Nutrition Examination Survey (NHANES), gather comprehensive multidomain data, existing studies tend to analyze these domains independently, overlooking the potential to identify integrated, data-driven patterns of depression risk. We have developed SYNERGY-VAE, an interpretable and multimodal deep learning framework that is designed to identify latent health subgroups and facilitate the transparent prediction of depression risk, utilizing NHANES-derived multimodal data from a large analytic cohort that spans five distinct domains from 2005 to 2018. SYNERGY-VAE utilizes a variational autoencoder to learn a shared latent representation derived from five modalities of NHANES, specifically demographic, dietary, examination, laboratory, and questionnaire data. Clustering within this latent space revealed subpopulations exhibiting distinct health signatures. To enhance interpretability, we triangulated insights utilizing encoder weights, standardized mean differences, and permutation feature importance (PFI). Machine learning classifiers were trained within each cluster employing the top 30 PFI-ranked features, with performance evaluated through the receiver operating characteristic area under the curve (AUC) in a 70/30 train-test split. Three latent clusters emerged, each demonstrating markedly different observed prevalence rates of depression within the analytic sample, ranging from 6.8% to 10.9%. Cluster-specific models consistently surpassed pooled models in performance, with the highest predictive accuracy identified in Cluster 2 utilizing XGBoost (AUC = 0.839, 95% CI:0.804-0.874). The importance of features varied across clusters, indicating unique depression risk profiles specific to each subgroup. SYNERGY-VAE demonstrates the power of generative, explainable deep learning for the identification of latent health phenotypes, thereby furthering the objectives of precision mental health. By integrating high-dimensional, multimodal data and facilitating transparent subgroup discovery, our model may inform future stratified and context-aware screening research and contribute to precision psychiatry-oriented research in large health survey datasets.</p>","PeriodicalId":74465,"journal":{"name":"PLOS digital health","volume":"5 9","pages":"e0001719"},"PeriodicalIF":7.7,"publicationDate":"2026-09-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13606994/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149042634","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}