Search Results (1,879)

Artificial intelligence (AI) systems for intensive care units (ICUs) must support early risk prediction while producing explanations that clinicians can inspect, question, and relate to physiological reasoning. Deep neural networks (DNNs) can learn complex temporal patterns from electronic health records (EHRs), but their internal representations are often difficult to translate into clinically actionable explanations. Gradient-boosted decision trees (GBDTs) offer more transparent decision rules, yet they may not capture the full temporal and nonlinear structure of high-dimensional ICU data. This paper presents a causal-pathway-guided DNN–GBDT distillation framework for interpretable ICU decision support. The framework first estimates a directed acyclic graph (DAG), denoted by $\mathcal{G}$, from multivariate ICU time-series data and then uses the graph to guide representation learning in a DNN teacher model through causal gating. The learned teacher is distilled into a GBDT student model using soft predictive targets and a causal attribution-guided split-selection procedure, so that the final model approximates the teacher predictions while prioritizing tree splits aligned with plausible physiological pathways. Experiments using Medical Information Mart for Intensive Care IV (MIMIC-IV) data evaluate sepsis onset and in-hospital mortality prediction through discrimination, precision–recall performance, calibration-oriented reporting, causal consistency, and clinical utility indicators. The proposed causal-aware distilled GBDT achieves stronger predictive performance than conventional interpretable baselines and substantially higher causal consistency than black-box temporal models. The results suggest that causal structure can serve as an inductive bias for converting complex temporal prediction into interpretable rule-based clinical reasoning. The paper also discusses limitations related to observational causal discovery, unmeasured confounding, temporal stationarity, and clinical deployment, following recent reporting expectations for AI-based clinical prediction models. Full article

Background/Objectives: Pancreatic ductal adenocarcinoma (PDAC) carries the poorest prognosis among gastrointestinal malignancies, largely owing to diagnosis at an advanced stage. Hyperglycaemia and new-onset diabetes are increasingly recognised as early manifestations of PDAC, yet long-standing type 2 diabetes is also an established risk factor. Whether admission plasma glucose, together with presenting symptoms, are associated with disease stage at diagnosis remain poorly characterised. We examined the association of admission glucose and presenting symptoms with disease stage in newly diagnosed PDAC. Methods: We performed a retrospective, single-centre analysis of 114 consecutive patients with histologically confirmed PDAC and both recorded admission plasma glucose and imaging-confirmed stage. Random admission plasma glucose, diabetes status, and presenting symptoms were extracted from electronic medical records. Symptoms were categorised as jaundice, abdominal pain, weight loss, or incidental (asymptomatic) detection. Disease stage was classified as metastatic (M1) or non-metastatic (M0) on the basis of imaging at diagnosis. Groups were compared using the Mann–Whitney U test and the χ² test with continuity correction; multivariable logistic regression and receiver operating characteristic (ROC) analysis were performed. Results: The cohort comprised 60 women (52.6%) and 54 men (47.4%); 49 patients (43.0%) had metastatic and 65 (57.0%) had non-metastatic disease. Median age at diagnosis was 68 years (range 27–88). Diabetes was present in 40.4% (46/114), predominantly type 2 (87%), and did not differ by stage (42.9% vs. 38.5%; p = 0.78). Jaundice was more frequent in metastatic than non-metastatic disease (59.2% vs. 38.5%; χ² = 4.02, p = 0.045), whereas abdominal pain (53.1% vs. 36.9%; p = 0.13) and weight loss (24.5% vs. 33.8%; p = 0.38) did not differ significantly. Mean admission glucose was 145.5 ± 59.8 mg/dL and was higher in patients with than without diabetes (178.6 ± 68.7 vs. 123.2 ± 40.1 mg/dL; p < 0.0001). Glucose analysed continuously did not differ by stage (148.4 vs. 143.3 mg/dL; p = 0.29), whereas hyperglycaemia ≥126 mg/dL was more frequent in metastatic disease (63.3% [95% CI 49.3–75.3] vs. 41.5% [95% CI 30.4–53.7]; χ² = 4.44, p = 0.035; odds ratio 2.42, 95% CI 1.13–5.19). In multivariable logistic regression adjusting for age, sex, and diabetes, hyperglycaemia remained independently associated with metastatic stage (adjusted odds ratio 2.59, 95% CI 1.13–5.95; p = 0.025). However, discrimination was poor: the area under the ROC curve for admission glucose was 0.56 (95% CI 0.45–0.66). Among jaundiced patients, admission glucose did not differ significantly by stage (162.8 vs. 152.8 mg/dL; p = 0.41). Conclusions: In this retrospective, single-centre cohort, jaundice at presentation and admission hyperglycaemia ≥126 mg/dL were each associated with metastatic stage, and hyperglycaemia remained associated after adjustment. Nevertheless, the discriminative performance of admission glucose was poor, and these variables cannot be recommended for individual patient stratification. These hypothesis-generating findings require confirmation in larger, prospective, multicentre studies. Full article

Background/Objectives: Metabolic dysregulation has emerged as a potential modifier of disease activity and disability in multiple sclerosis (MS). However, the relationship between glycemic status and clinical outcomes in patients with MS remains incompletely understood. This study aimed to investigate the associations of glycemic regulation with disease severity, relapse activity, inflammatory markers, and metabolic characteristics in patients with MS. Methods: In this retrospective single-center observational study, 455 patients with MS aged 18–65 years were included. Patients were categorized into normoglycemia (n = 241), prediabetes (n = 98), and diabetes (n = 116) according to American Diabetes Association criteria and documented diabetes diagnosis. Demographic characteristics, Expanded Disability Status Scale (EDSS) scores, annualized relapse rate (ARR), disease-modifying therapy (DMT), hematological indices, inflammatory markers, and metabolic parameters were retrieved from electronic medical records. Associations between glycemic status and clinical outcomes were evaluated using group comparisons, Spearman correlation analyses, and multivariable linear regression models. Results: Significant differences were observed among glycemic groups for age, disease duration, EDSS score, ARR, MS phenotype, platelet count, C-reactive protein (CRP), erythrocyte sedimentation rate (ESR), fasting plasma glucose (FPG), glycated hemoglobin (HbA1c), lipid parameters, estimated glomerular filtration rate, alanine aminotransferase, gamma-glutamyl transferase, triglyceride–glucose (TyG) index, and atherogenic index of plasma (all p < 0.05). HbA1c, FPG, TyG index, lipid parameters, CRP, and ESR were positively correlated with EDSS scores (all p < 0.05). In contrast, these markers showed inverse correlations with ARR. However, after adjustment for demographic and clinical variables, including age, sex, disease duration, MS phenotype, and disease-modifying therapy (DMT), neither prediabetes nor diabetes remained independently associated with disability, as measured by the EDSS score, or with ARR. Progressive MS phenotype, older age, and longer disease duration were independently associated with higher disability, whereas age, disease duration, and progressive phenotype were independently associated with lower ARR. Conclusions: Impaired glycemic status is associated with greater disability, unfavorable metabolic profiles, and increased systemic inflammation in patients with MS. Nevertheless, glycemic status was not independently associated with disability or relapse activity after adjustment for major clinical confounders. These findings suggest that metabolic dysregulation accompanies more severe clinical characteristics but may not independently drive disease progression in MS. Full article

Background/Objectives: Children with atopic dermatitis (AD) are at increased risk of food allergy (FA). This study aimed to develop and evaluate machine learning (ML) models for predicting FA in children aged <2 years with AD and to assess feature contribution. Methods: This retrospective study included children aged 1 month to 2 years who were diagnosed with AD and underwent evaluation for FA. Demographic characteristics, clinical features, and laboratory findings were extracted from electronic medical records. Four ML algorithms—CatBoost, XGBoost, LightGBM, and logistic regression—were evaluated for predicting FA. Results: A total of 435 children with AD were included, of whom 185 (42.5%) were female. The median age at presentation was 7.4 months (IQR, 4.8–10.8), and FA was present in 101 patients (23.2%). Among all evaluated models, CatBoost using clinical features alone demonstrated the best overall performance, achieving an AUC of 0.91 (95% CI, 0.84–0.97), with a sensitivity of 0.85 and a specificity of 0.91. Using combined clinical and laboratory features, XGBoost achieved an AUC of 0.88 (95% CI, 0.79–0.95), with a sensitivity of 0.70 and a specificity of 0.96. Using clinical features alone, LightGBM and logistic regression achieved AUCs of 0.89 (95% CI, 0.82–0.96) and 0.86 (95% CI, 0.75–0.95), respectively, with sensitivities of 0.85 and 0.65 and specificities of 0.82 and 0.88. Conclusions: CatBoost demonstrated the best overall performance in predicting FA in children aged <2 years with AD using clinical features alone. ML models may help identify children with AD who require further diagnostic evaluation for FA. Full article

Background/Objectives: Chronic dust-induced lung diseases, including pneumoconiosis and asbestosis, require long-term monitoring and individualized management. Conventional rehabilitation and follow-up often depend on in-person visits and may not provide continuous assessment outside clinical settings. This scoping review mapped evidence on digital technologies used to monitor patients, assess respiratory symptoms and functional status, and support rehabilitation follow-up in chronic dust-induced lung diseases. Methods: The review followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for Scoping Reviews (PRISMA-ScR) and Joanna Briggs Institute methodology. PubMed, Scopus, and Web of Science were searched from inception to 29 June 2026 without language restrictions. Five eligible publications were included. Results: The evidence was grouped into three categories: wearable monitoring of activity and respiratory symptoms, analysis of data from a digital rehabilitation platform, and electronic medical record (EMR) analysis. Wearable sensors showed high performance in recognizing basic activity states and cough events under controlled conditions. Platform- and EMR-based approaches showed potential for using clinical and functional data to support patient stratification and monitoring. However, most studies were limited to early technical or algorithmic validation and did not assess long-term home use, patient adherence, integration into clinical workflows, or clinical and rehabilitation outcomes. Conclusions: Digital technologies may support objective monitoring and rehabilitation follow-up in chronic dust-induced lung diseases, but the evidence base remains small and clinically immature. Prospective studies in real-world settings should evaluate usability, external validity, workflow integration, and clinically meaningful outcomes. Full article

Keyword-based remote integrity auditing schemes effectively address the integrity of electronic medical records (EMRs) stored in the cloud. In practice, users expect to be able to perform flexible dynamic data updates while also protecting data privacy against a third-party auditor during the auditing process. However, existing schemes fail to simultaneously satisfy both requirements: they either incur prohibitive overhead for block-level updates or disclose to the auditor which EMRs match the target keyword and the number of such EMRs. To address this, we propose a new keyword-based auditing scheme for medical cloud. Specifically, we design a novel authentication identifier set. Unlike the keyword tags in Shen et al.’s scheme, this set aggregates the block hashes and thereby enables the auditor to verify integrity without obtaining sensitive information. Furthermore, we introduce a dynamic hash list. By updating this list during block insertion and deletion, the scheme eliminates the need to recompute the authenticators of subsequent blocks, significantly enhancing the efficiency of dynamic data updates. Security analysis confirms that the proposed scheme is secure. Performance analysis shows our scheme reduces block insertion and deletion overhead by over 60% compared to Gao et al.’s scheme, demonstrating high efficiency and practicality. Full article

Background and Objectives: Prostate cancer (PC) has traditionally been regarded as a disease of older men, most commonly diagnosed after the age of 70 years. Recent epidemiological data indicate an increasing incidence of this malignancy among men younger than 55 years, with important clinical and therapeutic implications. The present study aimed to analyze the epidemiological, clinical, and pathological characteristics of PC in young patients diagnosed at a tertiary referral urology center in western Romania. Materials and Methods: Medical records and the electronic database of the “Pius Brînzeu” County Emergency Clinical Hospital in Timișoara (PBCECHT) were reviewed to identify men aged ≤55 years diagnosed with PC over a 10-year period (2015–2024). Collected data included: the age at diagnosis, serum prostate-specific antigen (PSA) levels, specimen type, histologic type, Gleason score/International Society of Urological Pathology (ISUP) grade group, tumor stage, and additional prognostic factors. Results: Of the 1742 patients diagnosed with PC during the study period, 70 (4.02%) were aged ≤55 years. The average age at diagnosis for the group of young men was 52.47 years. Most patients were from urban areas (64.29%). The PSA was available in 51/70 cases (range: 1.35–5182; median: 15; IQR: 51). The mean Gleason score was 7.15, and the mean ISUP grade group was 2.5. Histologically, acinar adenocarcinoma predominated, while a minority of cases showed mixed acinar and ductal/adenocarcinomas with ductal features. Most tumors (70%) were clinically significant. Conclusions: Although less common, PC in young men often exhibits aggressive features. It is crucial to consider genetic risk factors and a multidisciplinary approach for optimal management. Full article

Background/Objectives: Meconium ileus (MI) related to cystic fibrosis (CF) presents unique nutrition challenges for newborns. The aim of this study was to present a descriptive case series reviewing the management practice and clinical outcomes of infants with CF and MI at a large, quaternary care center. Methods: A retrospective analysis examined patients born with CF and MI over fifteen years at a large pediatric hospital in the Midwest. Patient demographics, prenatal imaging data, newborn screening results, stool studies, sweat chloride and details of the clinical course were obtained through the electronic medical record (EMR). Outcomes including length of stay (LOS), timing of surgery, and duration of total parenteral nutrition (TPN) needs were compared between infants with and without a surgical ostomy. Associations of MI severity with CF genotype, stool elastase, and prenatal imaging findings were evaluated. Results: A total of 30 neonates (21 female, 9 male) were included. Surgery was required in 26 infants, with 18 requiring ostomy placement. There were no significant correlations between CF genotype, stool elastase, or prenatal imaging findings with need for ostomy placement. Compared to those without ostomy placed, infants with ostomies had significantly longer median [IQR] LOS (71 [61, 105] vs. 22 [16, 32] days; difference 49, 95% CI 34–77) and duration of TPN (52 [41, 74] vs. 8 [6, 12] days; difference 43, 95% CI 25–62) (p < 0.001 for both). Conclusions: This work highlights the complex medical needs of these infants, requiring prolonged hospitalization and TPN for adequate nutrition. We share our institution’s approach to the nutritional management infants with CF and MI, with a focus on collaboration needed with the multidisciplinary team. Full article

The digitalization of healthcare systems has transformed the production, storage, and sharing of clinical information. While electronic health records (EHRs) enhance care efficiency, accessibility, and continuity, they simultaneously introduce complex ethical, legal, and cybersecurity challenges that directly affect patient interests. This narrative review examines the emerging rights associated with digital health records, particularly the right to privacy and the right to be forgotten, alongside threats to confidentiality, cybersecurity, and patient safety. A comprehensive literature search was conducted across PubMed, Web of Science, MEDLINE, and the Cochrane Library. First, the right to be forgotten appears particularly relevant for oncological patients facing financial discrimination and for individuals asserting gender identity rights, yet it cannot be unconditionally extended to genetic data, given its relevance to relatives and future generations. Second, confidentiality risks are amplified by re-identification vulnerabilities, unauthorized access by personnel, and the broad connectivity of digital systems. Third, the secondary use of data from EHRs, including artificial intelligence (AI) integration, commercial exploitation, and large language model training, raises substantial privacy concerns. Fourth, ransomware, phishing, and data breaches can erode patient trust. In this article, we analyze these issues within the evolving European regulatory framework, highlighting the tension between individual privacy rights and broader public interests. We argue that robust data protection must be balanced with scientific progress and that this requires opt-in consent frameworks, staff training, and transparent AI governance. Full article

Digital twins (DTs) are dynamic, virtual representations of individual patients that could support predictive diagnostics and personalized therapeutic optimization. Their development depends on patient-level data from real-world data (RWD) sources and electronic medical record (EMR) data, but major barriers in data-stewardship, interoperability, provenance, and governance persist. This review examines critical bottlenecks within the current healthcare ecosystem, with particular attention to fragmented EMRs, limited longitudinal data continuity, and missing, incomplete, or inaccurate information. We address data availability, stewardship, and provenance rather than model construction itself. We also explore ethical imperatives of mitigating representational bias, where over-representation of European ancestry amplifies existing health inequalities. We evaluate blockchain technology as a decentralized trust anchor to ensure provenance, automate consent, and incentivize longitudinal data stewardship. In parallel, we acknowledge that clinically useful DTs also depend on substantial advances in model specification, calibration, and validation, especially for biologically complex diseases. By synthesizing computational, regulatory, and ethical challenges, we provide a roadmap for developing robust, equitable, and interoperable ecosystems for precision medicine. Full article

Maternal anemia during pregnancy may adversely affect fetal growth, including neonatal head circumference (HC). This retrospective cohort study aimed to identify predictors of small HC among infants born to women with third-trimester anemia. Secondary data were obtained from the HosXP electronic medical database and delivery records of pregnant women who delivered between October 2020 and September 2023. A total of 934 women met the eligibility criteria and were included in the analysis. Maternal characteristics were summarized using descriptive statistics. Associations between maternal factors and neonatal HC were examined using the Chi-square test, Fisher’s exact test or Mann–Whitney U test, and independent predictors were identified using binary logistic regression analysis. The mean neonatal HC was 32.76 ± 1.50 cm, and 43.90% of infants were classified as having a small HC. Gravidity, pre-pregnancy body mass index (BMI), and gestational weight gain (GWG) were significantly associated with neonatal HC. Multivariable analysis revealed that women with two pregnancies (aOR = 0.51, 95% CI: 0.352–0.734) and those with three or more pregnancies (aOR = 0.54, 95% CI: 0.373–0.773) had lower odds of delivering infants with small HC than primigravida women. Pre-pregnancy overweight or obesity was associated with lower odds of small HC (aOR = 0.59, 95% CI: 0.426–0.805), whereas inadequate GWG increased the likelihood of this outcome (aOR = 1.38, 95% CI: 1.010–1.894). Maternal gravidity, pre-pregnancy BMI, and GWG may influence neonatal HC. These findings may support nurses in identifying women at risk, monitoring GWG, and providing targeted nutritional counseling during antenatal care. Full article

Background/Objectives: Bariatric surgery is the most effective treatment for sustained weight loss; however, early postoperative complications remain a significant concern. Routine biochemical markers obtained in the immediate postoperative period may enable early risk stratification, yet their prognostic value remains incompletely defined. This study aimed to evaluate the association between early postoperative biochemical alterations and composite adverse outcomes after bariatric surgery. Methods: In this retrospective observational study, 260 adult patients undergoing elective bariatric surgery were included. Preoperative and postoperative day 1 (POD1) laboratory parameters were retrieved from electronic medical records. The primary endpoint was a composite adverse outcome comprising postoperative intensive care unit admission, infection, anastomotic leak, or 30-day readmission. Univariable and multivariable logistic regression analyses were performed to evaluate independent associations, and receiver operating characteristic (ROC) curve analysis was used to assess discriminative performance. Results: Composite adverse outcomes occurred in 51 patients (19.6%). In multivariable analysis, longer operative duration (OR 1.17 per 10 min, 95% CI 1.04–1.31; p = 0.009), POD1 AST (OR 1.15 per 10 U/L, 95% CI 1.07–1.24; p < 0.001), and POD1 lactate-to-albumin ratio (LAR ×100) (OR 1.20, 95% CI 1.04–1.38; p = 0.014) were independently associated with adverse outcomes. POD1 AST demonstrated the highest discriminative ability (AUC 0.733), followed by glucose (AUC 0.688) and ALT (AUC 0.680), whereas LAR showed limited stand-alone performance (AUC 0.594). Conclusions: Longer operative duration and higher POD1 AST and LAR levels were independently associated with the composite adverse perioperative outcome. AST showed the highest, although moderate, individual discrimination, whereas LAR demonstrated limited stand-alone performance. Because immediate postoperative ICU transfer preceded POD1 biomarker sampling and dominated the composite outcome, these findings should be interpreted as associations rather than prospective prediction. Full article

Artificial intelligence (AI) is rapidly transforming cardiovascular medicine, driven by the increasing availability of large-scale clinical data and advances in machine learning. Early computational applications in cardiology were primarily limited to rule-based electrocardiogram interpretation systems. Over time, these approaches have evolved into sophisticated deep learning models capable of analysing complex cardiovascular signals and imaging data. In parallel with the broader development of digital health technologies, including wearable devices, electronic health records, and remote monitoring systems, AI applications have expanded across multiple domains of cardiovascular care. These now include electrocardiographic (ECG) and electrophysiological analysis, cardiovascular imaging, surgical planning, and multimodal risk prediction. More recently, multimodal AI models have emerged that integrate heterogeneous data sources such as imaging, physiological signals, clinical records, and genomic information, enabling more comprehensive characterisation of cardiovascular disease. Beyond diagnostic applications, AI is increasingly influencing system-level aspects of cardiovascular medicine, including clinical decision support, workflow optimisation, medical education, and clinical trial design. This narrative review traces the historical and clinical evolution of artificial intelligence in cardiovascular medicine from early automated ECG interpretation systems to contemporary multimodal and system-level applications. It highlights key technological developments, current clinical applications, translational challenges, and the emerging role of AI within digital cardiovascular health ecosystems, with particular emphasis on early disease detection, risk stratification, prognostic modelling, and personalised cardiovascular care. Full article

Background: Warfarin remains widely used in specific clinical situations. Its management is complicated by multiple factors that affect anticoagulant response, particularly during the early period after hospital discharge. This study examined the prevalence, patterns, and factors associated with nontherapeutic international normalized ratio (INR) among patients discharged on warfarin from a tertiary-care hospital. Methods: Electronic health records of adult patients discharged home with warfarin who had at least one INR measurement within 90 days (N = 1222) were retrospectively analyzed. Nontherapeutic INR was defined as INR outside the therapeutic range: 2.5–3.5 for mitral valve replacement and 2.0–3.0 otherwise. All available INR measurements were included. Major warfarin potential drug–drug interactions (pDDIs) were defined as DDIs with major severity according to the Micromedex^® database. Factors associated with nontherapeutic INR were examined using repeated-measures generalized estimating equations (GEEs), with generalized linear mixed models (GLMMs) as confirmatory analyses. Results: Of 3704 INR measurements within 90 days after discharge, 49.4% were subtherapeutic, while 30.5% were therapeutic and 20.1% were supratherapeutic. The proportion of therapeutic INR values did not show a substantial improvement over time. In GEEs, discharge from surgical service (adjusted odds ratio (aOR) 1.24, 95%CI: 1.05–1.48, p = 0.014) and presence of major warfarin pDDIs at discharge (aOR 1.36, 95%CI: 1.11–1.67, p = 0.003) were associated with nontherapeutic INR. GLMM analyses produced consistent results with the GEE model. Conclusions: Suboptimal INR control was prevalent within 90 days post-discharge. Discharge from surgical services and presence of major warfarin pDDIs at discharge were associated with nontherapeutic INRs. Major warfarin pDDIs may serve as markers of medication complexity at discharge and may help identify patients requiring closer anticoagulation monitoring. Full article

Background: Increasing access to Electronic Health Records (EHRs) has enabled the development of Machine Learning (ML) models to predict early cardiovascular disease (CVD) risk. Nevertheless, real EHR data is usually limited in availability and distribution because of confidentiality issues, legal limitations, and insufficient access. Synthetic data generation has become a promising approach to overcome such difficulties. Objectives: This study proposes a Large Language Model (LLM)-guided retrieval-aware framework for generating realistic synthetic EHR data on a large scale to train ML models to predict cardiovascular risks accurately. Methods: The framework uses a Tabular Denoising Diffusion Probabilistic Model to learn the underlying distribution of the original dataset and generate an initial synthetic dataset. To improve the clinical plausibility of the generated data, a knowledge-guided refinement module with LLaMA 2 13B combined with Retrieval-Augmented Generation (RAG) is introduced. The LLM analyzes statistical trends in real and synthetic data while retrieving relevant medical information to detect and correct clinically implausible correlations. Results: Experimental findings show that the optimized synthetic dataset preserves important statistical features of the original data and achieves high performance in predicting CVD. Conclusions: Thus, the framework offers a scalable, privacy-conservative method for generating realistic synthetic healthcare data suitable for medical research. Full article