BMC Medical Informatics and Decision Making最新文献

{"title":"Quantifying optimal inner limiting membrane peeling in macular hole surgery: a machine learning framework for predictive modeling and schematic visualization.","authors":"Xiang Zhang, Hongjie Ma, Song Lin, Ledong Zhao, Lu Chen, Zetong Nie, Zhaoxiong Wang, Chang Liu, Xiaorong Li, Wenbo Li, Bojie Hu","doi":"10.1186/s12911-025-03140-2","DOIUrl":"10.1186/s12911-025-03140-2","url":null,"abstract":"<p><strong>Purpose: </strong>Internal limiting membrane (ILM) peeling in macular hole (MH) surgery is critical but challenging, and current practices lack standardized tools for quantifying and visualizing optimal peeling dimensions.This study aimed to develop a machine learning framework to recommend surgeon-specific ILM peeling radius during macular hole surgery, integrating predictive modeling with schematic visualization to guide operative planning.</p><p><strong>Methods: </strong>This retrospective study analyzed data from 95 patients with idiopathic MH who underwent vitrectomy with ILM peeling. Preoperative and postoperative optical coherence tomography (OCT) images were used to measure key MH parameters, including minimum diameter (MIN), base width (BASE), temporal length (T), nasal length (N), and height (H). The dataset was preprocessed by addressing missing values and applying Z-score normalization. 10 regression models were trained and evaluated using an 80 - 20 train-test split. Model performance was assessed using root mean squared error (RMSE), mean squared error (MSE), mean absolute error (MAE), and the coefficient of determination (R²). A graphical user interface (GUI) was developed to generate ILM peeling schematic diagrams from OCT data.</p><p><strong>Results: </strong>The Ridge Regression model demonstrated the best performance, with an RMSE of 0.0320, MSE of 0.0010, MAE of 0.0209, and R² of 0.9427. The generated ILM peeling schematic diagrams provided clear visual representations, aiding surgical planning and education.</p><p><strong>Conclusion: </strong>The Ridge Regression model effectively predicts the optimal ILM peeling radius. The integration of schematic diagram generation enhances surgical planning and provides valuable educational resources, highlighting the potential of machine learning and visualization tools in improving MH surgery outcomes.</p>","PeriodicalId":9340,"journal":{"name":"BMC Medical Informatics and Decision Making","volume":"25 1","pages":"309"},"PeriodicalIF":3.8,"publicationDate":"2025-08-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12355892/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144854648","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Hypercubes to identify geomarkers of rapid cystic fibrosis lung disease progression.","authors":"Yizi Cheng, Cole Brokamp, Erika Rasnick Manning, Elizabeth L Kramer, Patrick H Ryan, Rhonda D Szczesniak, Emrah Gecili","doi":"10.1186/s12911-025-03097-2","DOIUrl":"10.1186/s12911-025-03097-2","url":null,"abstract":"<p><strong>Background: </strong>Prior research has shown that place-based environmental exposures and community characteristics, known as geomarkers, are associated with accelerated lung function decline and increased mortality in individuals with cystic fibrosis (CF). Although geomarkers have been linked to pulmonary outcomes in other respiratory diseases, it is unknown which have the greatest predictive power for rapid lung function decline in CF.</p><p><strong>Methods: </strong>We adapted an existing statistical procedure, which arranges candidate variables in a k-dimensional hypercube, where the hypercube forms a set of variables for a multi-stage selection process involving complex longitudinal data. We embedded the hypercube within a dynamic prediction model of rapid lung function decline, in order to accommodate complexity in lung function trajectories. This practical approach simultaneously selects a handful of genuinely predictive markers among candidates and accounts for complex correlations in longitudinal marker data. Our method is applied to actual geomarker and lung-function outcomes data from the existing Cystic Fibrosis Patient Registry and Cincinnati Cystic Fibrosis Center datasets.</p><p><strong>Results: </strong>We applied a 4 × 4 × 4 3-D hypercube to the national and local datasets and selected a subset of geomarkers using p-values from testing coefficients of the association between each geomarker and lung function decline in the dynamic prediction model. Based on the national data analyses, some road density-related geomarkers were selected, including some air pollution-related and greenspace-related variables. Simulations showed the proposed method's variable selection efficacy and robust performance in identifying true predictors, particularly under weak correlation (ρ≤0.6), although performance dipped with stronger correlations (ρ=0.9).</p><p><strong>Conclusions: </strong>The proposed method is a useful approach for selecting a small set of truly relevant demographic, clinical, and place-based predictors of rapid lung function decline while accounting for the complex correlations inherent in longitudinal lung-function data. We found that selection results differed according to spatial resolution of the geomarkers. Our findings have potential to improve care decisions for people with CF.</p>","PeriodicalId":9340,"journal":{"name":"BMC Medical Informatics and Decision Making","volume":"25 1","pages":"304"},"PeriodicalIF":3.8,"publicationDate":"2025-08-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12344988/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144844510","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Challenging assumptions: a tripartite assessment of medical quality, resource utilization, and equity concerns in pediatric telemedicine.","authors":"Motti Haimi, Ruslan Sergienko, Tzipi Hornik-Lurie, Dov Albukrek","doi":"10.1186/s12911-025-03124-2","DOIUrl":"10.1186/s12911-025-03124-2","url":null,"abstract":"&lt;p&gt;&lt;strong&gt;Abactstr: &lt;/strong&gt;BACKGROUND: Telemedicine has expanded healthcare accessibility, particularly during the COVID-19 pandemic. However, evidence regarding its economic efficiency and clinical quality remains inconclusive, with some studies suggesting increased costs, service utilization, and inappropriate antibiotic prescribing compared to traditional care modalities.&lt;/p&gt;&lt;p&gt;&lt;strong&gt;Objective: &lt;/strong&gt;This study evaluated pediatric telemedicine services across three dimensions: clinical outcomes (antibiotic prescribing patterns), resource utilization implications (healthcare utilization and emergency department visits), and equity considerations (sociodemographic distribution of services).&lt;/p&gt;&lt;p&gt;&lt;strong&gt;Methods: &lt;/strong&gt;We conducted a retrospective cohort study analysis of 1,500 children under 19 years within Israel's Clalit Health Services (Shron-Shomron District) from January 2021 to January 2022. Participants were randomly sampled from three groups based on their telemedicine utilization patterns: in-person primary care physician (PCP) visits only (n = 500); PCP plus phone/video telemedicine (n = 500); and PCP plus phone/video plus Tyto device telemedicine (n = 500). These children were classified as discrete groups, not only as solitary visits. We assessed emergency department (ED) admission rates, antibiotic prescription frequencies for common pediatric conditions, and sociodemographic characteristics across the groups.&lt;/p&gt;&lt;p&gt;&lt;strong&gt;Results: &lt;/strong&gt;Analysis of 21,968 visits revealed striking socioeconomic disparities in telemedicine utilization. While 58.1% of all participants had high socioeconomic status (SES), this increased to 68.6% among Tyto device users. Conversely, low-SES patients comprised 26.4% of in-person-only visits but only 3% of Tyto users. ED admission rates for all groups combined were 4.0%, 1.6% for the \"PCP only\" group, 5.4% for the \"PCP + phone/video\" group (12% after phone/video visits), and 3.6% for the \"PCP + phone/video + Tyto\" group (4.1% after Tyto only). Antibiotic prescribing rates were similarly divergent: mean rates of antibiotic prescriptions after PCP visits were 10.3%, after telephone/video visits were 16.8%, and after Tyto visits were 21.4% (p &lt; 0.001). Multivariable analysis confirmed those findings and demonstrated higher ED utilization and antibiotic prescribing among low-SES patients and those in remote locations, independent of visit type.&lt;/p&gt;&lt;p&gt;&lt;strong&gt;Conclusions: &lt;/strong&gt;Our findings challenge prevailing assumptions about telemedicine benefits, revealing unexpected associations between telemedicine services and higher ED utilization and antibiotic prescribing compared to traditional care, particularly among vulnerable populations. These associations may reflect complex interactions between patient characteristics, provider behavior, and care modalities that warrant further investigation. Healthcare systems should reconsider the implementation of telemedicine services to ensure they de","PeriodicalId":9340,"journal":{"name":"BMC Medical Informatics and Decision Making","volume":"25 1","pages":"303"},"PeriodicalIF":3.8,"publicationDate":"2025-08-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12344953/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144834109","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Applying text-mining to clinical notes: the identification of patient characteristics from electronic health records (EHRs).","authors":"Simone Ten Hoope, Koen Welvaars, Kylian van Geijtenbeek, Mellanie Klok-Everaars, Sander van Schaik, Fatma Karapinar-Çarkit","doi":"10.1186/s12911-025-03137-x","DOIUrl":"10.1186/s12911-025-03137-x","url":null,"abstract":"<p><strong>Background: </strong>Clinical notes contain information on critical patient characteristics, which, if overlooked, could escalate the risk of adverse events as well as miscommunication between the healthcare professional and the patient. This study investigates the feasibility of employing text-mining to extract patient characteristics from Electronic Health Records (EHRs) and compares the effectiveness of text-mining against human intelligence for identifying four patient characteristics: language barrier, living alone, cognitive frailty and non-adherence.</p><p><strong>Methods: </strong>A manual \"golden\" standard was created from 1,120 patient files (878 patients) that had unplanned hospital readmissions. Each patient was categorized in one (or multiple) of the four characteristics with supporting clinical notes extracted from their EHRs. For simple terminology, a rule-based (RB) SQL query was used, and for complex terms, Named Entity Recognition (NER) models were used. Model performance was compared to the manual standard. The primary outcomes were recall, specificity, precision, negative predictive value (NPV) and F1-score.</p><p><strong>Results: </strong>Performance of each patient characteristic was evaluated using a separate train/test dataset. An additional validation dataset was used for the NER models. Within the train/test set, the language barrier RB query achieved a recall of 0.99 (specificity of 0.96). The living alone NER model achieved a recall of 0.86 (specificity of 0.94) on the train/test set and a recall of 0.81 (specificity of 1.00) on the validation set. In that same order, the cognitive frailty model yielded a recall of 0.59 (specificity 0.76) on the train/test set and a recall of 0.73 (specificity 0.96) on the validation set. The NER model for non-adherence achieved a recall of 0.75 (specificity of 0.99) on the train/test set, and a recall of 0.90 (specificity of 0.99) on the validation set. The models showed the tendency to overestimate the presence of patient characteristics such as identifying a family member's language barrier as the patient's.</p><p><strong>Conclusion: </strong>This study successfully demonstrated the feasibility of applying text-mining to identify patient characteristics from EHRs. Also, it seems for more complex terminology, NER models outperform the rule-based option. Future work involves refining these models for broader application in clinical settings.</p><p><strong>Clinical trial number: </strong>Not applicable.</p>","PeriodicalId":9340,"journal":{"name":"BMC Medical Informatics and Decision Making","volume":"25 1","pages":"302"},"PeriodicalIF":3.8,"publicationDate":"2025-08-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12344823/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144834108","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Evaluating gender bias in large language models in long-term care.","authors":"Sam Rickman","doi":"10.1186/s12911-025-03118-0","DOIUrl":"10.1186/s12911-025-03118-0","url":null,"abstract":"<p><strong>Background: </strong>Large language models (LLMs) are being used to reduce the administrative burden in long-term care by automatically generating and summarising case notes. However, LLMs can reproduce bias in their training data. This study evaluates gender bias in summaries of long-term care records generated with two state-of-the-art, open-source LLMs released in 2024: Meta's Llama 3 and Google Gemma.</p><p><strong>Methods: </strong>Gender-swapped versions were created of long-term care records for 617 older people from a London local authority. Summaries of male and female versions were generated with Llama 3 and Gemma, as well as benchmark models from Meta and Google released in 2019: T5 and BART. Counterfactual bias was quantified through sentiment analysis alongside an evaluation of word frequency and thematic patterns.</p><p><strong>Results: </strong>The benchmark models exhibited some variation in output on the basis of gender. Llama 3 showed no gender-based differences across any metrics. Gemma displayed the most significant gender-based differences. Male summaries focus more on physical and mental health issues. Language used for men was more direct, with women's needs downplayed more often than men's.</p><p><strong>Conclusion: </strong>Care services are allocated on the basis of need. If women's health issues are underemphasised, this may lead to gender-based disparities in service receipt. LLMs may offer substantial benefits in easing administrative burden. However, the findings highlight the variation in state-of-the-art LLMs, and the need for evaluation of bias. The methods in this paper provide a practical framework for quantitative evaluation of gender bias in LLMs. The code is available on GitHub.</p>","PeriodicalId":9340,"journal":{"name":"BMC Medical Informatics and Decision Making","volume":"25 1","pages":"274"},"PeriodicalIF":3.8,"publicationDate":"2025-08-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12337462/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144815771","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Risk factors for tuberculosis treatment outcomes: a statistical learning-based exploration using the SINAN database with incomplete observations.","authors":"Nguyen Ky Phat, Yoonah Lee, Dinh Hoa Vu, Nguyen Phuoc Long, Seongoh Park","doi":"10.1186/s12911-025-03139-9","DOIUrl":"10.1186/s12911-025-03139-9","url":null,"abstract":"<p><strong>Background: </strong>Understanding early predictors of treatment outcomes allows better outcome prediction and resource allocation for efficient tuberculosis (TB) management.</p><p><strong>Objectives: </strong>This study aimed to predict treatment outcomes of TB patients from a real-world population-wide health record dataset with a significant rate of incomplete observations. In addition, potential risk factors associated with death during TB treatment were investigated.</p><p><strong>Methods: </strong>We exploited the upweighting approach and multiple imputation analysis (MIA) to address the extreme imbalance in responses and missing data. Three algorithms were employed for TB treatment outcome prediction, including logistic regression (LOGIT), random forest, and stochastic gradient boosting. The three models exhibited similar performance in predicting the treatment outcomes. Moreover, an interpretation of LOGIT was conducted, adjusted odds ratios (aORs) were computed, and the interpretation results were compared between MIA and complete case analysis (CCA).</p><p><strong>Results: </strong>MIA was an appropriate method for coping with missing data. In addition, compared to CCA, the interpretation results of the MIA-derived LOGIT showed more statistically significant covariates associated with TB treatment outcomes. In MIA, factors such as TB clinical form involving both pulmonary TB and extrapulmonary TB [aOR = 3.077, 95% confidence interval (CI) = 2.994-3.163], retreatment after abandonment (aOR = 2.272, 95% CI = 2.209-2.338), and the absence of isoniazid (aOR = 2.072, 95% CI = 1.892-2.269) or rifampicin (aOR = 1.968, 95% CI = 1.746-2.218) in the treatment regimen were associated with increased odds of death.</p><p><strong>Conclusion: </strong>In conclusion, our results shed light on the potential risk factors for death during TB treatment and suggest the use of simple yet interpretable LOGIT for the prediction of TB treatment outcomes.</p>","PeriodicalId":9340,"journal":{"name":"BMC Medical Informatics and Decision Making","volume":"25 1","pages":"301"},"PeriodicalIF":3.8,"publicationDate":"2025-08-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12341307/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144820617","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Semantics-driven improvements in electronic health records data quality: a systematic review.","authors":"Yirong Wu, Mudan Ren, Na Chen, Liu Yang","doi":"10.1186/s12911-025-03146-w","DOIUrl":"10.1186/s12911-025-03146-w","url":null,"abstract":"<p><strong>Background: </strong>Data quality (DQ) of electronic health record (EHR) is crucial for the advancement of health informatization, yet it remains a significant challenge. Scholars are showing a growing interest in leveraging semantic technologies to enhance EHR data quality. However, previous studies have focused predominantly on specific semantic technologies, scenarios, or objectives-such as interoperability-often overlooking the potential of a various semantic technologies across different scenarios.</p><p><strong>Objective: </strong>This systematic review aimed to explore the potential of employing a range of semantic technologies to improve EHR data quality in a broader spectrum of application scenarios.</p><p><strong>Methods: </strong>Our systematic review followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines. Three databases were searched, including PubMed, IEEE Xplore, and Web of Science Core Collection. The search terms used included \"Semantic*\", \"Quality\", \"Electronic Health Record*\", \"EHR*\", \"Electronic Medical Record*\", and \"EMR*\". These terms were combined via various Boolean operators to formulate multiple search queries.</p><p><strong>Results: </strong>Thirty-seven papers that met the inclusion criteria between 2008 and 2024 were analyzed. Six semantic techniques were identified as instrumental in improving EHR DQ: EHR standardization, controlled vocabulary, ontology, semantic web, knowledge graph, and natural language processing (NLP). These technologies were further mapped to 16 core data quality indicators and the FAIR principles (Findable, Accessible, Interoperable, and Reusable), highlighting their contributions across both technical and governance dimensions.</p><p><strong>Conclusions: </strong>The six identified semantic technologies can be categorized into three levels: foundational, general, and advanced. These technologies show significant potential in enhancing EHR DQ, particularly in the areas of conformance, portability, usability, and applicability, and they are suitable for a variety of contexs beyond interoperability, aligning with FAIR-aligned best practices in data management and reuse.</p>","PeriodicalId":9340,"journal":{"name":"BMC Medical Informatics and Decision Making","volume":"25 1","pages":"298"},"PeriodicalIF":3.8,"publicationDate":"2025-08-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12337493/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144820618","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Leveraging BERT for embedding ICD codes from large scale cardiovascular EMR data to understand patient diagnostic patterns.","authors":"Minkyoung Kim, Yunha Kim, Hee Jun Kang, Hyeram Seo, Heejung Choi, JiYe Han, Gaeun Kee, Soyoung Ko, HyoJe Jung, Byeolhee Kim, Boeun Choi, Tae Joon Jun, Young-Hak Kim","doi":"10.1186/s12911-025-03145-x","DOIUrl":"10.1186/s12911-025-03145-x","url":null,"abstract":"<p><p>The integration of electronic medical records (EMRs) with artificial intelligence (AI) is enhancing medical research, particularly in real-world evidence (RWE) studies. Extracting insights from coded medical data, such as ICD-10 codes, is essential for patient characterization. Traditional techniques, such as one-hot encoding (OHE), face limitations, particularly in managing high-dimensional data. In this study, a Bidirectional Encoder Representations from Transformers (BERT) approach is introduced to encode ICD-10 diagnostic codes, significantly improving model performance and reducing dimensionality. Data from 495,269 patients who visited the Cardiology Department at Asan Medical Center between 2000 and 2020 were used. The performance of models trained with OHE and ClinicalBERT embeddings was compared. For predicting major adverse cardiovascular events within one year following percutaneous coronary intervention (PCI) or coronary artery bypass grafting (CABG), the ClinicalBERT (code-embedded) model outperformed OHE. It achieved an AUC of 0.746 compared to 0.719, while also significantly reducing the dimensionality from 2,492 to 128. This method, which integrates diagnostic and medication data, provides valuable insights into patient care, enhancing the precision of predictions and supporting healthcare professionals in making more informed decisions.</p>","PeriodicalId":9340,"journal":{"name":"BMC Medical Informatics and Decision Making","volume":"25 1","pages":"300"},"PeriodicalIF":3.8,"publicationDate":"2025-08-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12337513/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144820616","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Early detection of vascular catheter-associated infections employing supervised machine learning - a case study in Lleida region.","authors":"Radu Spaimoc, Jordi Mateo, Francesc Solsona, Alfredo Jover-Sáenz, Fernando Barcenilla, María Ramírez-Hidalgo, Marcos Serrano, Miquel Mesas, Dídac Florensa","doi":"10.1186/s12911-025-03113-5","DOIUrl":"10.1186/s12911-025-03113-5","url":null,"abstract":"<p><p>Healthcare-associated infections (HAIs), particularly Vascular Catheter-Associated Infections (VCAIs), are a significant concern, accounting for over 7% of all infections and are often linked to medical devices. Early detection of VCAIs before invasive infection is crucial for improving hospital care and reducing antibiotic use. This study retrospectively developed and evaluated machine learning models to classify VCAIs from patient medical records, excluding fever and antibiotic prescription indicators. The dataset, collected from the group of public hospitals of the Lleida health region in Catalonia (Spain) between 2011 and 2019, consisted of 24,239 episodes with 150 features related to vascular catheter use. After validation, processing and feature engineering, the dataset showed an imbalance, with 94.46% (10,090) non-catheter episodes and 5.53% (591) catheter infection cases. Machine learning classifiers demonstrated significant challenges in classifying imbalanced datasets, particularly in the context of VCAIs. While most models achieved high accuracy and specificity (approximately 97%), they frequently exhibited limited sensitivity, reaching only around 60% in the best-performing cases. Among the evaluated classifiers, the Gradient Boosting (GB) model outperformed others, attaining the highest balanced accuracy (82.5%) and sensitivity (67%), underscoring its potential utility for early VCAI detection. Additionally, the analysis examined the impact of oversampling techniques on model performance. Although these methods enhanced metrics for some classifiers, they did not consistently outperform models trained on the original dataset. Therefore, if the improvement is not significant, it is preferable to use the original dataset. This study highlights that strategic feature engineering with the GB classifier is sufficient to obtain robust VCAI detection before the appearance of a probable sepsis.</p>","PeriodicalId":9340,"journal":{"name":"BMC Medical Informatics and Decision Making","volume":"25 1","pages":"299"},"PeriodicalIF":3.8,"publicationDate":"2025-08-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12337550/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144820615","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Developing an AI-powered wound assessment tool: a methodological approach to data collection and model optimization.","authors":"Alessio Stefanelli, Sofia Zahia, Guillaume Chanel, Rania Niri, Swann Pichon, Sebastian Probst","doi":"10.1186/s12911-025-03144-y","DOIUrl":"10.1186/s12911-025-03144-y","url":null,"abstract":"<p><strong>Background: </strong>Chronic wounds (CWs) represent a significant and growing challenge in healthcare due to their prolonged healing times, complex management, and associated costs. Inadequate wound assessment by healthcare professionals (HCPs), often due to limited training and high clinical workload, contributes to suboptimal treatment and increased risk of complications. This study aimed to develop an artificial intelligence (AI)-powered wound assessment tool, integrated into a mobile application, to support HCPs in diagnosis, monitoring, and clinical decision-making.</p><p><strong>Methods: </strong>A multicenter observational study was conducted across three healthcare institutions in Western Switzerland. Researchers compiled a hybrid dataset of approximately 4,000 wound images through both retrospective extraction from clinical records and prospective collection using a standardized mobile application. The prospective data included high-resolution images, short videos, and 3D scans, along with structured clinical metadata. Retrospective data were anonymized and manually annotated by wound care experts. All images were labeled for wound segmentation and tissue classification to train and validate deep learning models.</p><p><strong>Results: </strong>The resulting dataset represented a broad spectrum of wound types (acute and chronic), anatomical locations, skin tones, and healing stages. The AI-based wound segmentation model, developed using the Deeplabv3 + architecture with a ResNet50 backbone, achieved a DICE score of 92% and an Intersection-over-Union (IOU) score of 85%. Tissue classification yielded a preliminary mean DICE score of 78%, although accuracy varied across tissue types, especially fibrin and necrosis. The models were optimized for mobile implementation through quantization, achieving real-time inference with an average processing time of 0.3 seconds and only a 0.3% performance reduction. The dual approach to data collection, prospective and retrospective-ensured both image standardization and real-world variability, enhancing the model's generalizability.</p><p><strong>Conclusions: </strong>This study laid the foundation for an AI-driven digital tool to assist clinical wound assessment and education. The integration of robust datasets and AI models demonstrated the potential to improve diagnostic precision, support personalized care, and reduce wound-related healthcare costs. Although challenges remained, particularly in tissue classification, this work highlighted the promise of AI in transforming wound care and advancing clinical training.</p><p><strong>Trial registration: </strong>Not applicable.</p>","PeriodicalId":9340,"journal":{"name":"BMC Medical Informatics and Decision Making","volume":"25 1","pages":"297"},"PeriodicalIF":3.8,"publicationDate":"2025-08-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12335118/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144811832","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}