DIGITAL HEALTH最新文献

Background: Low birth weight (LBW) is a leading cause of death for newborns and increases chronic disease risks later in life. Early identification of LBW risk is crucial.

Aim: The objective of this study was to develop predictive models for LBW using boosting ensemble machine learning, with a focus on features available during early pregnancy, such as pre-pregnancy body mass index, body height, and blood pressure before 20 weeks of pregnancy.

Methods: This is a retrospective cohort study. We used electronic medical records in four hospitals in Taiwan where pregnant women received prenatal care from January 2016 to July 2019, including 6719 pregnant women. Data preprocessing involved normalization, one-hot encoding, and a synthetic minority oversampling technique for class imbalance. Boosting ensemble methods were used to build the LBW predictive models.

Results: The mean diastolic blood pressure (DBP) in early pregnancy (<20 weeks) was 66.5 mmHg, 29.6% had experienced abortion, 8.7% delivered LBW, 12.2% were overweight or obese before pregnancy, and 18.3% had elevated or stage I hypertension before 20 weeks of pregnancy. Lightweight Gradient Boosting Machine was the best-performing LBW model, with an area under curve of 0.96 and an accuracy of 93.4%. Early pregnancy DBP, maternal height, and number of abortions were the most important features.

Conclusions: The LBW prediction model performed well. Nurses could use the model to assess LBW risk and intervene early. Preventive efforts could be directed to blood pressure management starting early pregnancy, nutritional support for short mothers, and self-care for women with a history of abortions.

Objective: To develop an attention-based convolutional neural network (CNN) pipeline for early prediction of pathological complete response (pCR) to neoadjuvant chemotherapy (NAC) in breast cancer, improving feature selection and interpretability in whole slide image (WSI) analysis.

Methods: A retrospective analysis was conducted on 384,076 tiles extracted from 122 Hematoxylin and Eosin-stained WSIs, divided among an investigational cohort (IC, 82 patients enrolled at IRCCS Istituto Tumori "Giovanni Paolo II"), a validation cohort (VC, 20 patients, same Institution), and an external validation cohort (EVC, 20 patients belonging the Yale trastuzumab response cohort public dataset). WSIs were first annotated and then automatically segmented into tiles. Noninformative regions were filtered using Mini-Batch C-Fuzzy K-Means. Remaining tiles were analyzed using a CNN with a Convolutional Block Attention Module, prioritizing both histological features and tiles critical for predicting pCR.

Results: The model achieved robust performance: 81.4% AUC, 81.3% accuracy, 80.0% specificity, and 83.3% sensitivity in IC; 80.9% AUC, 80.0% accuracy, 85.78% specificity, and 66.7% sensitivity in VC; and 76.2% AUC, 70.0% accuracy, 71.4% specificity, and 66.7% sensitivity in EVC. The EVC, consisting of WSIs at 20× magnification compared to the 40× magnification of IC and VC, demonstrated the model's robustness to varying resolutions.

Conclusion: This is an innovative pipeline that not only improves prediction but also enhances the clinical utility of digital pathology.

Background and objective: Generative artificial intelligence (AI) tools such as ChatGPT are increasingly integrated into healthcare, with potential to support clinical decision-making and improve patient outcomes. In palliative care, where access to multidisciplinary expertise is often limited, these tools may provide support for symptom management. This study aimed to systematically compare ChatGPT-4o and ChatGPT-5 for common palliative care symptoms across four key domains: clinical appropriateness, safety, ethical sensitivity, and understandability.

Methods: Clinical scenarios representing 10 key symptoms (pain, anxiety, pressure ulcer, nausea, delirium, dyspnea, constipation, diarrhea, dry mouth, and sleep disturbance) were presented first to ChatGPT-4o and, 1 week later, to ChatGPT-5. Responses were evaluated independently by two physicians using a five-point Likert scale. Inter-rater agreement was analyzed with weighted Cohen's kappa and Spearman's correlation. The statistical analyses in this study were conducted using the Friedman test, Mann-Whitney U test, and Wilcoxon signed-rank test.

Results: Inter-rater agreement was consistently high across all domains (kappa 0.806-0.886, Spearman's rho 0.813-0.888; all p < 0.001). ChatGPT-5 outperformed ChatGPT-4o in clinical appropriateness (p = 0.010), safety (p = 0.002), and understandability (p = 0.011). Ethical sensitivity scores were high for both models, with no significant difference (p = 0.102).

Conclusions: ChatGPT-5 demonstrated measurable improvements over ChatGPT-4o in key domains of palliative care symptom management, while maintaining consistently high ethical sensitivity. These findings provide the first systematic evidence of the potential of generative AI, with the updated ChatGPT-5 model released in August 2025, as a complementary and reliable clinical decision support tool in palliative care.

Objectives: Although radiologists typically rely on imaging characteristics of pulmonary nodules for preoperative evaluation, the inherent subjectivity of this approach often leads to high misdiagnosis rates. This study comparatively analyzed the diagnostic value of non-contrast-enhanced computed tomography (NCECT) and contrast-enhanced computed tomography (CECT) in differentiating benign and malignant pulmonary nodules using multi-regional radiomics and machine learning algorithms.

Methods: This retrospective collection included 194 patients who underwent NCECT and CECT scans. Radiomics features were extracted by identifying the intra-nodular and peri-nodular 5 mm area as the region of interest. Six different machine learning classifiers were used to select the most effective classifier to create a predictive model. The efficacy of the models was measured by the area under the curve, further analysis of the combined model was conducted through calibration curves and decision Curve Analysis curves. Additionally, 78 patients were collected as an external validation cohort.

Results: The logistic regression classifier showed the best stability. In the single-region analysis, the model developed based on features extracted from the intra-nodular regions of interest in contrast-enhanced CT scans yielded a significantly higher AUC value compared to the other three single-region models. The combined regions of non-contrast CT achieved an AUC of 0.901, similar to the contrast-enhanced CT combined regions. Furthermore, the NCECT model achieved an AUC of 0.863 in external validation, further confirming its robustness.

Conclusions: The multiple regional features model of intra-nodular and peri-nodular outperformed single-region models in differentiating malignant from benign nodules. Furthermore, the combined model of NCECT demonstrated comparable efficacy to CECT.

Objective: This study aimed to assess whether a second course of digital cognitive behavioral therapy for insomnia (dCBT-I) could benefit patients with chronic insomnia who had suboptimal responses to the initial fully self-guided dCBT-I.

Methods: Building on a previous randomized controlled trial (RCT), participants with an Insomnia Severity Index (ISI) score ≥8 at 6 months after completing the initial digital CBT-I therapy were invited to enroll in the second phase of the study. Among the 20 eligible participants, 9 received a second round of digital dCBT-I therapy, 6 participated in follow-up only. Primary outcomes included changes in ISI scores and ISI remission rates (defined as ISI <8). Secondary outcomes included sleep attitudes and beliefs, as well as mental health questionnaire scores.

Results: Among patients who did not achieve full remission after the first course of digital CBT-I, 44.4% and 62.5% of participants experienced a reduction in their ISI scores to <8 following the second course of treatment and at the 3-month follow-up, respectively. Following the second treatment, ISI scores showed a trend of continued decrease, but no statistically significant difference was observed compared to the baseline before the second treatment (p > 0.05). Sleep-related attitudes and beliefs, as measured by the DBAS-16 scale, also significantly improved after the second round of treatment (p < 0.05). At the 3-month follow-up, the second treatment group showed a greater improvement in ISI scores (3.00(1.25, 5.00)) compared to the follow-up only group (-1.50(-3.25, 0.50)) (p < 0.05).

Conclusion: The present study suggests that a second course of digital CBT-I may benefit chronic insomnia patients who initially fail to achieve ISI remission after the first round of self-guided digital CBT-I. However, larger randomized controlled trials are needed to definitively assess its effectiveness.

Objectives: Inadequate meal frequency (IMF) among children aged 6-23 months remains a pressing public health issue in Somalia, contributing to widespread malnutrition and hindering progress toward Sustainable Development Goals 2 (Zero Hunger) and 3 (Good Health and Well-being). This study investigates the most influential factors associated with IMF to inform targeted public health interventions.

Methods: Data from 4066 children were extracted from the 2020 Somalia Demographic and Health Survey, employing Five machine learning algorithms, Logistic Regression, Decision Tree, Random Forest, Support Vector Machine, and Gradient Boosting, and assessed for predictive performance using accuracy and area under the receiver operating characteristic curve (AUC-ROC) metrics. Feature importance was analyzed to identify key predictors of IMF.

Results: The prevalence of IMF was alarmingly high at 78.51%. The Gradient Boosting model outperformed other models with an accuracy of 89.55% and an AUC-ROC of 92.77%. Birth order emerged as the most dominant predictor across all models, accounting for 74.07% of the Gini importance in the Gradient Boosting model. Other significant predictors included child age, breastfeeding status, maternal education, household wealth, and region of residence.

Conclusion: The high prevalence of IMF highlights an urgent need for targeted interventions. Strategies focusing on families with higher birth order children, maternal education, and poverty reduction may be crucial for improving child nutrition in Somalia. These findings demonstrate the potential of machine learning approaches in informing public health strategies and predictive screening in resource-limited settings.

Purpose: Synthetic data has emerged as a promising solution to overcome the shortage of clinical datasets needed for training healthcare artificial intelligence (AI) models. This study examined how synthetic data can support AI development in Africa's healthcare by analyzing its technical performance, fidelity limitations, and governance implications within low-resource health systems.

Methods: A Critical Literature Review was conducted on scholarly and technical literature focused on the use of synthetic data for AI in healthcare across African settings. Databases searched included Scopus, Web of Science, PubMed, and Google Scholar. Thematic analysis identified trends in synthetic data generation, fidelity, domain adaptation, and adoption challenges in African healthcare AI.

Results: Drawing on interdisciplinary evidence, the analysis demonstrates how addressing technical challenges, improving synthetic data fidelity, leveraging domain adaptation techniques, and confronting practical adoption barriers are critical to enhancing the reliability and applicability of synthetic data for AI-driven healthcare in Africa. Four themes emerged from the analysis. First, hybrid synthetic-real datasets consistently outperform synthetic-only models. Second, fidelity gaps introduced bias risk and misclassification. Third, domain adaptation remains underused in low-resource contexts. Fourth, infrastructure gaps, weak regulation, and clinician skepticism hindered the adoption of synthetic data.

Conclusion: Synthetic data can enhance AI-enabled healthcare in Africa if it is embedded within regulatory frameworks, validated through hybrid modeling, and supported by investment in infrastructure and capacity-building. This study highlights the intersection of synthetic data, healthcare AI, data fidelity, domain adaptation, and governance considerations in African health systems, underscoring the need for robust health technology assessment processes.

Objective: To examine the association between Health Information System (HIS) performance and maternal health service (MHS) utilization in the Oromia and Gambella regions, Ethiopia.

Methods: A comparative cross-sectional study was conducted (15-25 October 2023) among 840 mothers in catchment areas of health facilities categorized as model (high HIS performance) or candidate (lower HIS performance). HIS performance was evaluated based on infrastructure (30%), data quality (30%), and data use (40%). MHS utilization was measured using a modified composite coverage index (CCI) integrating 10 essential interventions. Multivariable logistic regression (Stata/MP 17.0) identified predictors, reporting adjusted odds ratios (AORs) and 95% confidence intervals (CIs).

Results: MHS utilization was 60.3%, with higher crude odds in model facility areas (COR = 2.03, 95% CI [1.4-3.0]). After adjustment, this association attenuated (AOR = 1.4, 95% CI [0.92-2.15]). Key barriers included poverty (poorest quintile AOR = 0.45, 95% CI [0.30-0.68]) and limited transport access (AOR = 0.21, 95% CI [0.15-0.29]), which were associated with significantly lower MHS utilization. Sensitivity analyses confirmed robustness, and transport access modified the effect of facility type.

Conclusion: HIS performance alone did not independently predict MHS utilization after accounting for structural inequities. Transportation and economic barriers disproportionately hinder access, even in high-performing systems. Integrating HIS strengthening with poverty-sensitive interventions (e.g., transport support, financial protection) is critical to achieving equitable maternal health outcomes.

Objective: This study aimed to evaluate the quality and reliability of information presented in short videos related to anterior cruciate ligament (ACL) injuries on two major Chinese social media platforms, TikTok and Bilibili.

Methods: A systematic search using the keyword "ACL injuries" was conducted to identify the top 100 Chinese videos on TikTok and Bilibili, respectively. The Global Quality Score (GQS) and the modified DISCERN evaluation scale were employed to assess video content reliability and quality. Videos characteristics-including engagement metrics, uploader identity, video length, and content type-were also gathered. Statistical analyses were conducted to examine differences and correlations between platforms, uploader categories, and video quality.

Results: Out of 200 videos reviewed, 175 met inclusion criteria. The most common content theme was treatment, found in 59 videos (33.71%). TikTok videos attracted higher user engagement than Bilibili. However, the overall video quality on both platforms was moderate. TikTok videos scored higher on GQS and modified DISCERN than on Bilibili. Engagement on TikTok showed no positive correlation with content quality, while that on Bilibili demonstrated a moderate positive correlation. Videos uploaded by healthcare professionals were more popular but often tended to be shorter in duration. Notably, videos uploaded by individual users often achieved quality scores comparable to, or even exceeding, those of medical professionals and science communicators.

Conclusion: TikTok demonstrated higher engagement than Bilibili, but both platforms showed limited quality and reliability in ACL injury-related video content. No strong correlation was observed between video content quality and engagement. These findings highlight the need for improved oversight of ACL injury-related information disseminated through short video platforms in China.

Fraudulent participation is a growing challenge in digital health research, particularly in online studies where duplicate identities, automated responses, and coordinated sign-ups can distort recruitment, compromise validity, and divert resources. Safeguards intended to prevent fraud might also risk excluding legitimate participants, raising concerns about sample representativeness and study generalizability. Although a wide range of technical and behavioral strategies exists, guidance is lacking on how to organize these methods and report outcomes consistently across studies. To address this gap, we introduce the Configure, Assess, Triage, Corroborate, and Hone (CATCH) framework, a hybrid fraud detection-mitigation model with actionable recommendations for investigators. CATCH begins with pre-study configuration to prepare for fraud mitigation and proceeds through systematic assessment of fraud risk, triage of candidates into risk categories, and corroboration of inconclusive cases, while honing strategies through ongoing monitoring. The framework emphasizes transparent documentation and reporting of actions and outcomes to facilitate comparability across studies and continuous methodological refinement. As fraudulent participation grows and emerging technologies act as both risks and solutions, CATCH can help guide investigators' efforts to maximize data integrity in digital health research. By synthesizing existing fraud-mitigation strategies into a unified, staged framework, CATCH offers practical guidance for structuring decisions, documenting actions, and balancing data integrity with inclusivity.