PLOS digital health最新文献

{"title":"The Medicine Quality Monitoring Globe: A public health intelligence system for collecting and curating reports of substandard and falsified medical products from publicly available Internet news media.","authors":"Clark C Freifeld, Kerlijn Van Assche, Alberto Olliaro, Martial Battain, Inthapavanh Kitignavong, Vayouly Vidhamaly, Phonepasith Boupha, Kem Boutsamay, Ngan Thi Do, Konnie Bellingham, Viengsavanh Pimxayvong, Thitthiphone Olinh, Ana Rosado Olmo, Jingying Xu, Shan Jiang, Andrea Soult, Southida Insal, Philippe J Guérin, Paul N Newton, Céline Caillet","doi":"10.1371/journal.pdig.0001648","DOIUrl":"10.1371/journal.pdig.0001648","url":null,"abstract":"<p><p>The circulation of substandard and falsified (SF) medical products, including medicines, vaccines, and medical devices, continues to impact health of populations worldwide. Meanwhile, surveillance of SF products and data sharing are limited in most of the world, and incidents are rarely published in accessible databases. We describe here the Medicine Quality Monitoring Globe system (MQM Globe) for capturing, curating, characterising, and disseminating online media reports relating to SF products. The system consists of the curation environment and the Web application. Online reports are acquired from search engines, direct feeds, and targeted extraction, in five languages. They then pass a series of automated content characterisation steps to filter and label them by relevance, location, and product. Reports are then reviewed by a trained curator staff, before being made available through a visualisation Web application. Over the study period from July 2018 through December 2024, we identified 11,000 relevant distinct SF incidents, across a broad range of geography, products, and categories. The automated relevance classifier exhibited precision of 37%, recall 90%, and F1-score 52%, for English-language content. The information captured by the MQM Globe is publicly available and can help inform public health authorities as situations emerge. It is the only publicly available repository of open source intelligence specifically on SF medical products. It can be used as an informational tool to trigger intelligence-led activities across law enforcement, customs and Medicine Regulatory Authorities.</p>","PeriodicalId":74465,"journal":{"name":"PLOS digital health","volume":"5 9","pages":"e0001648"},"PeriodicalIF":7.7,"publicationDate":"2026-09-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13577332/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148978944","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 systematic review of prolonged posture monitoring with accelerometer-based wearable sensors.","authors":"Meixing Liao, Kai Huang, Eva Libeert, Bart Vanrumste, Simon Brumagne, Zhe Chen, Jean-Marie Aerts","doi":"10.1371/journal.pdig.0001538","DOIUrl":"10.1371/journal.pdig.0001538","url":null,"abstract":"<p><p>Prolonged sitting and standing are associated with health risks. Unlike generic sedentary behaviour monitoring based on energy expenditure, prolonged posture monitoring requires posture-specific detection and uninterrupted bout segmentation, yet current accelerometer-based approaches remain heterogeneous. This systematic review synthesises current practices and identifies research gaps across the sensing-to-feedback pipeline, including data collection, preparation, analysis, and user feedback. Ten databases, including Web of Science, Scopus, ProQuest Central, PubMed/MEDLINE, EMBASE, CINAHL, SPORTDiscus, IEEE Xplore, ACM Digital Library, and Engineering Village, were searched. Eligible studies monitored uninterrupted sitting or standing bouts of at least ten minutes during wake time using body-worn accelerometer-based sensors, provided sufficient methodological detail, and were published as peer-reviewed journal articles or full conference proceedings in English. From 27,334 records, fourteen studies were included. Risk of bias was assessed using design-specific tools: JBI checklists for randomised and cross-sectional studies, and WEAR-BOT for wearable-sensor validation or algorithm-development studies. Prolonged sitting was the primary target, while prolonged standing was not assessed as a dedicated bout-level outcome. Thigh-worn activPAL devices were most frequently used for posture measurement or as reference criteria. Bout duration thresholds and interruption rules varied substantially. Deep learning models reported the strongest performance for sitting recognition and transition detection. Four studies included feedback components, but none used real-time posture monitoring outcomes to trigger feedback. Substantial heterogeneity precluded quantitative synthesis, and the evidence base was concentrated in Western, office-based populations. Overall, accelerometer-based wearable sensors provide a feasible basis for monitoring prolonged sitting. Key gaps include inconsistent definitions, limited specific detection, heterogeneous data preparation and validation reporting, and a lack of translation from monitoring to actionable feedback. Future research should prioritise transparent bout definitions, reproducible pipelines, inclusive external validation, and interpretable feedback systems. PROSPERO registration: CRD42025637387.</p>","PeriodicalId":74465,"journal":{"name":"PLOS digital health","volume":"5 9","pages":"e0001538"},"PeriodicalIF":7.7,"publicationDate":"2026-09-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13577576/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148978999","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":"Reliable detection and continuous monitoring of memory dysfunction in mild cognitive impairment and healthy aging through adaptive computational phenotyping.","authors":"Holly S Hake, Maarten van der Velde, Thomas J Grabowski, Hedderik van Rijn, Andrea Stocco","doi":"10.1371/journal.pdig.0001686","DOIUrl":"10.1371/journal.pdig.0001686","url":null,"abstract":"<p><p>With the rising prevalence of age-related memory impairments, efficiently detecting and monitoring decline is increasingly urgent. Unfortunately, traditional assessment methods fall short of these needs, as they typically require in-person administration and cannot be repeated frequently. Here, we demonstrate that remote identification and monitoring of abnormal memory function is possible by combining an online assessment platform with computational phenotyping, allowing repeatable, unsupervised remote observations from patients. Fifty-one well-characterized older individuals, including 24 patients with amnestic mild cognitive impairment and 27 age- and education-matched healthy controls, completed a series of longitudinal, unsupervised, remote weekly 8-minute online memory assessments for up to one year. Weekly test data were fit to a formal model of memory consolidation and forgetting, yielding an individualized index of memory function, the Seattle-Groningen Memory Assessment (SGMA) score. The SGMA score was found to be reliable, with a mean correlation of r = 0.70 across assessments. The score was also found to be stable across different study materials, and only barely affected by practice effects, which averaged to a 0.2% increase per assessment. Finally, the SGMA score was found to be diagnostic, being capable of detecting mild cognitive impairment with up to 87% accuracy. These findings show that model-based, adaptive assessments can support high-frequency, remote detection and scalable longitudinal monitoring of early memory decline, providing a new way to assess memory decline trajectories in healthy aging and dementia.</p>","PeriodicalId":74465,"journal":{"name":"PLOS digital health","volume":"5 9","pages":"e0001686"},"PeriodicalIF":7.7,"publicationDate":"2026-09-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13577568/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148978966","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":"From Who experiences stress to When It Occurs: Revisiting Lazarus' transactional model of stress using an ecological momentary assessment framework.","authors":"Dennis John, Hannah van Alebeek, Rüdiger Pryss, Carsten Vogel, Johannes Schobel, Teresa O'Rourke, Anne-Kathrin Helten, Jennifer Apolinário-Hagen, Thomas Probst","doi":"10.1371/journal.pdig.0001716","DOIUrl":"10.1371/journal.pdig.0001716","url":null,"abstract":"<p><p>According to the Transactional Model of Stress, stress responses arise from one's perceptions of situational relevance (primary appraisal) and coping resources (secondary appraisal) as being insufficient. Although the model assumes moment-to-moment within-person variability, most research evidence stems from cross-sectional study designs. Therefore, this study applies Ecological Momentary Assessment (EMA) to test the dynamic interplay between primary and secondary appraisals and fluctuations in perceived stress. We used EMA data from the TrackYourStress platform to examine both interindividual (\"who\") and intraindividual (\"when\") associations between primary and secondary appraisals and perceived stress. Bayesian multilevel models were applied to 2,908 observations from 164 participants. These models showed that higher appraisal of situational relevance was associated with increased stress, whereas higher appraisal of coping resources was associated with lower stress levels. Importantly, the association between relevance appraisals and stress was stronger under low coping resource appraisals. Findings were present at the inter- and intraindividual level. These findings support the model's dynamic assumptions and highlight that appraisals not only account for interindividual differences in stress levels but also for moment-to-moment variation in stress responses in daily life.</p>","PeriodicalId":74465,"journal":{"name":"PLOS digital health","volume":"5 9","pages":"e0001716"},"PeriodicalIF":7.7,"publicationDate":"2026-09-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13577512/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148978995","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":"Reconstructing healthcare networks from patient transfer data: A systematic review.","authors":"Elise Hodbert, Pascal Crépey, Pauline Clin, Victor Renault, Gabriel Birgand, Laura Temime","doi":"10.1371/journal.pdig.0001718","DOIUrl":"10.1371/journal.pdig.0001718","url":null,"abstract":"<p><p>Healthcare facilities are interconnected via patient transfers, facilitating pathogen spread. We systematically reviewed studies reconstructing patient-sharing networks. Following PRISMA 2020 guidelines and a preregistered protocol (CRD42024559127), we searched PubMed, Scopus and IEEE Xplore up to October 1, 2025. We included articles presenting networks of at least three healthcare facilities or units connected by observed patient transfers, excluding those restricted to specific patient groups. Seventy-nine articles were included from 5,721 screened, describing 50 distinct networks published across multiple disciplines. Overall confidence in study findings, assessed through a modified QuADS tool, was good (mean total score of 21/33), although precision in the description of underlying patient transfer data was often limited. Most networks were reconstructed in Europe and North America. The majority were restricted to hospitals, with nursing homes and other healthcare facilities rarely integrated, and 10 (20%) used temporal network approaches. While most articles reported descriptive statistics and network metrics, they focused mainly on density and degree (and its variants), with other metrics only rarely calculated. Only 27 (35%) articles developed mathematical models of pathogen transmission, including 12 that simulated public health interventions. Overall, this review highlights the value of patient-sharing networks to characterize healthcare connectivity and understand pathogen spread. Clearly articulating the methodological assumptions underlying network reconstruction and tailoring them to specific research questions is key to fully leveraging the reconstructed networks for public health purposes. In particular, including a broader range of facility types and more temporal modelling can help develop more realistic models of pathogen spread over these networks. This review was funded by the French National Research Agency (ANR 22-PAMR-0003).</p>","PeriodicalId":74465,"journal":{"name":"PLOS digital health","volume":"5 9","pages":"e0001718"},"PeriodicalIF":7.7,"publicationDate":"2026-09-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13577407/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148978978","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":"Evaluating an AI-assisted triage workflow for retinal diseases.","authors":"Doohyun Park, Richul Oh, Jihyeon Baek, Kunho Bae","doi":"10.1371/journal.pdig.0001700","DOIUrl":"10.1371/journal.pdig.0001700","url":null,"abstract":"<p><p>The clinical value of artificial intelligence (AI) in fundus photography depends on workflow efficiency as well as diagnostic performance. We evaluated an AI-assisted negative-screening workflow for diabetic retinopathy (DR), retinal vein occlusion (RVO), and age-related macular degeneration (AMD) using 6,904 color fundus photographs independently graded by three retina specialists. In this workflow, AI-negative images were classified as negative without ophthalmologist review, whereas AI-positive images were referred for human interpretation. For each reader-disease pair, the reference standard was agreement between the other two readers; discordant cases were excluded (DR, 1.2%-2.3%; RVO, 0.3%-0.6%; AMD, 6.1%-11.6%). The workflow reduced direct ophthalmologist review to 7.3%-8.0% of images for DR, 11.7%-11.9% for RVO, and 13.5%-16.8% for AMD. For DR, sensitivity decreased significantly (0.9168 to 0.8838; difference, -0.0330; 95% confidence interval [CI], -0.0495 to -0.0165; p < 0.001), whereas specificity did not change significantly (0.9961 to 0.9983; p = 0.183). For RVO, sensitivity was unchanged (0.9807) and specificity did not change significantly (0.9985 to 0.9988; p = 0.320). For AMD, sensitivity decreased significantly (0.8570 to 0.8445; difference, -0.0124; 95% CI, -0.0215 to -0.0033; p = 0.008), whereas specificity did not change significantly (0.9685 to 0.9843; p = 0.324). In an exploratory image-level analysis for at least one target disease, review decreased to 27.7%-30.1%; sensitivity changed from 0.9122 to 0.9039 (difference, -0.0083; 95% CI, -0.0131 to -0.0035; p < 0.001) and specificity from 0.9659 to 0.9804 (p = 0.334). AI-assisted negative-screening can reduce ophthalmologist workload across multiple retinal diseases, but with disease-specific safety trade-offs: because AI-negative images are not reviewed, reduced sensitivity for DR and AMD means a small proportion of true-positive cases would go undetected, potentially delaying diagnosis and treatment for sight-threatening disease. These results support disease-specific implementation strategies that balance workload reduction against missed positive cases.</p>","PeriodicalId":74465,"journal":{"name":"PLOS digital health","volume":"5 9","pages":"e0001700"},"PeriodicalIF":7.7,"publicationDate":"2026-09-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13577534/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148979013","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":"The development of a technologic approach to improve access to individualized clinical documentation in caregivers' preferred language.","authors":"Jacqueline Toscano, King Yan Kwok, Meg Simione","doi":"10.1371/journal.pdig.0001721","DOIUrl":"10.1371/journal.pdig.0001721","url":null,"abstract":"<p><p>This pilot study aims to describe the process of developing a template that semi-automatically translates information from English to Spanish to improve caregivers' access to clinical documents in their preferred language. It used human-centered design and the Discover, Design/Build, and Test Framework to improve health literacy outcomes for patients who use a language other than English. The Discover Phase revealed current methods and barriers to clinicians providing patients access to written documentation in their primary language. During the Design/Build Phase, an interprofessional team of a speech-language pathologist and a certified translation specialist developed the template in the electronic health record. In the Test Phase, we evaluated the template's acceptability and feasibility and surveyed speech-language pathologists (SLPs) and patients' caregivers. Clinicians affirmed the importance of the template, but also had concerns regarding feasibility and usability. Caregivers found it helpful to receive their child's health information in their primary language. The results showed that sustainable access to written documentation in patients' preferred language is lacking, and this template is prepared to reduce language barriers in an overwhelmed healthcare system.</p>","PeriodicalId":74465,"journal":{"name":"PLOS digital health","volume":"5 9","pages":"e0001721"},"PeriodicalIF":7.7,"publicationDate":"2026-09-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13568490/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148946278","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":"Explainable machine learning for breast cancer prediction in resource-constrained settings: A multi-algorithmic framework integrating shap-based transparency with clinical decision support.","authors":"Oluwaseun Adebayo Bamodu, Sumaiya Nezam, Chen-Chih Chung","doi":"10.1371/journal.pdig.0001706","DOIUrl":"10.1371/journal.pdig.0001706","url":null,"abstract":"<p><p>Breast cancer remains the most commonly diagnosed malignancy among women globally, with disproportionately higher mortality rates in low- and middle-income countries (LMICs) where diagnostic delays and limited specialist pathology capacity are widespread. While machine learning (ML) approaches achieve strong predictive performance for cancer classification, algorithmic opacity and absence of interpretability frameworks tailored to resource-constrained environments have impeded clinical adoption. This study bridges the translational gap between predictive accuracy and clinical utility by developing an explainable artificial intelligence (XAI) framework specifically designed for breast cancer diagnosis in underserved healthcare settings. Using the Wisconsin Breast Cancer Diagnostic Dataset (569 fine-needle aspirate cytological specimens with 30 nuclear morphometric features), we systematically benchmarked eight supervised classification algorithms: Logistic Regression, Random Forest, XGBoost, LightGBM, Support Vector Machine (SVM), Gradient Boosting, Decision Tree, and K-Nearest Neighbors, using stratified 10-fold cross-validation and an independent hold-out test set (80:20 split). Performance was evaluated across discriminative and probabilistic metrics, including AUC-ROC, F1-score, Matthews Correlation Coefficient (MCC), and Brier score, and interpretability was operationalized through SHapley Additive exPlanations (SHAP) analysis with global feature importance, cross-model consensus ranking, and individual-level dependence characterization. All ensemble and regularized models achieved test-set AUCs above 0.98, with XGBoost and SVM attaining the highest AUC of 0.996, and Logistic Regression the highest accuracy (98.25%) and MCC (0.962). SHAP analysis consistently identified worst perimeter, worst concave points, and worst area as the dominant predictors, with strong concordance across gradient-boosted models (pairwise Spearman rho: XGBoost-LightGBM 0.86, XGBoost-Random Forest 0.82, Random Forest-LightGBM 0.67). Logistic Regression also demonstrated superior probability calibration, a critical requirement for clinical risk stratification. Collectively, these findings deliver a reproducible, transparent framework whose SHAP-derived signatures align with established cytopathological principles, supporting responsible integration of interpretable ML into resource-limited diagnostic workflows and providing a template for equitable AI deployment in global oncology.</p>","PeriodicalId":74465,"journal":{"name":"PLOS digital health","volume":"5 9","pages":"e0001706"},"PeriodicalIF":7.7,"publicationDate":"2026-09-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13568489/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148946303","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":"Impact of mobile health interventions for mental health literacy: A systematic review and meta-analysis of randomized controlled trials.","authors":"Jamin Patel, Sheriff Ibrahim, Tarun Reddy Katapally","doi":"10.1371/journal.pdig.0001688","DOIUrl":"10.1371/journal.pdig.0001688","url":null,"abstract":"<p><p>Mental health literacy (MHL) - the ability to recognize, manage, and prevent mental health problems - is a key determinant of mental health. Mobile health (mHealth) interventions, delivered via portable digital devices such as smartphones, offer scalable, interactive, and user-tailored approaches to improve MHL. This study aimed to evaluate the impact of mHealth interventions on MHL outcomes, compared to control conditions (non-mHealth interventions, waitlist, or no intervention). This systematic review and meta-analysis searched five databases for randomized controlled trials published between January 1, 2010, and April 18, 2025. Eligible studies enrolled participants aged 12 and older and compared mHealth interventions to control groups. Two reviewers independently screened articles, extracted data, and assessed risk of bias using the Cochrane RoB 2 tool. The primary outcome was MHL, assessed using standardized mean differences (SMD) for overall MHL and its components: knowledge, help-seeking, and stigma. A random-effects meta-analysis was conducted. Out of 5,043 records identified, 14 RCTs met all inclusion criteria, encompassing 4,314 participants. mHealth interventions showed higher overall MHL (SMD = 0.13, 95% CI: 0.03-0.23), knowledge (SMD = 0.23, 95% CI: 0.09-0.37), and help-seeking (SMD = 0.26, 95% CI: 0.04-0.47; p = 0.019), compared to controls. High heterogeneity was detected (I² = 93.6%; τ² = 0.03; p < 0.001). Two studies were rated at high risk of bias; the remainder had at least some concerns. mHealth interventions modestly improved MHL outcomes, particularly knowledge and help-seeking. Although the effect sizes were small and heterogeneity was high, these interventions may offer meaningful public health value when delivered at scale.</p>","PeriodicalId":74465,"journal":{"name":"PLOS digital health","volume":"5 9","pages":"e0001688"},"PeriodicalIF":7.7,"publicationDate":"2026-09-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13567821/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148946245","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":"Temporal and cross-site validation of an AI system for self-harm detection.","authors":"Vlada Rozova, Liuliu Chen, Katrina Witt, Mike Conway, Jo Robinson, Karin Verspoor","doi":"10.1371/journal.pdig.0001667","DOIUrl":"10.1371/journal.pdig.0001667","url":null,"abstract":"<p><p>Adequate self-harm surveillance is a key part of suicide prevention. Our previous research demonstrated that an artificial intelligence (AI)-based system could effectively detect self-harm in emergency department triage notes. However, the system was developed using data from a single hospital, raising concerns about its generalisability. Here, we aim to validate the system prospectively and externally to better understand its portability across hospitals. We leveraged emergency department data from two Australian hospitals, with free-text triage notes manually annotated for self-harm. The AI system was developed using data from a major metropolitan hospital in Melbourne, spanning 2012-2017. The system combined extensive text normalisation with a self-harm classification model utilising 1931 selected features. We prospectively validated the model on 329,655 triage notes from the same hospital over the following four years. For external validation, we used 316,877 triage notes from 2012 to 2021 from a regional hospital 150 km outside Melbourne. On the test set, the model achieved an area under the precision-recall curve (PR AUC) of 0.84 with a 95% confidence interval of [0.82, 0.86]. This performance remained stable at the development site with a PR AUC of 0.84, 95% CI [0.83, 0.85]. When applied in the regional context, the model's ability to distinguish self-harm cases declined, resulting in an overall PR AUC of 0.78, 95% CI [0.77, 0.79]. While the text normalisation component was equally effective across the datasets, our analysis revealed that in regional settings, self-harm presentations are more likely to involve medication ingestion. At the metropolitan hospital, the AI system for self-harm detection maintained its epidemiological utility. At the regional hospital, the text normalisation process was effective, but performance was unstable primarily due to linguistic domain shift and differences in self-harm presentations.</p>","PeriodicalId":74465,"journal":{"name":"PLOS digital health","volume":"5 9","pages":"e0001667"},"PeriodicalIF":7.7,"publicationDate":"2026-09-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13567808/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148946272","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}