Introduction Current preventive screening reminders use generic due-date notifications that lack personalization. Targeted reminders allow personalized messages and resource allocation, yet the accuracy of behavioral forecasting with cholesterol screening behaviors remains unconfirmed. Therefore, this study aims to examine the accuracy of machine learning in predicting adults’ future cholesterol screening behavior. Methods This study included 8,288 participants from the Health and Retirement Study. Never receiving cholesterol screening over 12 years of follow-up (2007-2018) was identified as the target outcome. Predictors were constructed based on the Social-Ecological Model. Four machine learning (ML) algorithms were compared: logistic regression, random forests, gradient boosting machine (GBM), and eXtreme Gradient Boosting (XGBoost). Model performance was evaluated by Recall, Precision, and false-negative rate (FNR). Results Among the four ML algorithms, XGBoost performed best, with higher Recall (average: 0.8850) and the lowest FNR (0.1311), but suboptimal precision (average: 0.6228). The top five predictors were past cholesterol screening behavior, diagnosis of hypertension, osteoarthritis or rheumatoid arthritis, number of private insurance plans, and gender. Conclusion ML might be used to predict individuals’ cholesterol screening behavior. Future research is needed to examine whether leveraging ML in targeted reminders might enhance cholesterol screening behaviors.
Objective This study aimed to assess the quality and reliability of videos about type 1 diabetes mellitus (T1DM) on TikTok, Bilibili, and YouTube. Methods The keyword “type 1 diabetes mellitus” was used to search for related videos on TikTok, Bilibili, and YouTube. The quality and reliability of 263 videos were assessed using the modified DISCERN (mDISCERN), the Global Quality Scale (GQS), and the Journal of the American Medical Association (JAMA) benchmark criteria. Results Quality scores of videos on YouTube were higher, and videos uploaded by specialists were rated higher than those posted by non-specialists. However, user engagement metrics (likes, comments, saves, and shares) did not correlate with quality scores, indicating that engagement alone does not reflect content quality. Conclusion The quality and reliability of videos related to T1DM on these platforms were insufficient. Video platforms should recommend high-quality videos to guide viewers toward reliable sources.
Objective Nested named entity recognition in Chinese electronic medical records (EMRs) is challenging because entity spans overlap and have ambiguous boundaries. We evaluated whether part-of-speech (POS) cues and automatically extracted entity boundary knowledge could improve a RoBERTa–GlobalPointer framework for this task. Methods This retrospective study used public, de-identified benchmark datasets. We evaluated a RoBERTa-wwm-ext-large–GlobalPointer framework with two enhancement branches: token-aligned POS embeddings derived from jieba. posseg and an entity boundary knowledge base (EBKB) constructed from entity head, tail and length patterns. The primary experiment used Chinese Medical Entity Extraction version 2 (CMeEE-V2), and supplementary test-set analyses used cEHRNER and CCKS2019. Results On the CMeEE-V2 test set, the best configuration achieved 72.45% precision, 73.02% recall and 72.60% F1, outperforming the comparators. Ablation results indicated that boundary knowledge was the principal source of improvement, whereas POS features had a smaller effect. The cEHRNER test-set analysis provided additional cross-dataset evidence. Across five CCKS2019 test-set runs, F1 was 0.8152 ± 0.0096 for the knowledge-enhanced model and 0.7054 ± 0.0435 for the baseline (paired p=0.0043). Conclusion Automatically extracted boundary knowledge was the main driver of improvement in the framework, while POS features acted as auxiliary cues. Findings were consistent across primary and supplementary test-set evaluations.
Objectives To develop and implement an Enhanced Recovery After Surgery (ERAS) nursing pathway for lung cancer patients undergoing day surgery, and to evaluate its impact on perioperative nursing compliance, postoperative outcomes, and nurse satisfaction. Methods A quasi-experimental study was conducted in a tertiary-level hospital in Zhejiang Province from September 2021 to March 2022. A total of 671 lung cancer patients and 57 nurses were included. Patients were assigned to either a control group receiving routine perioperative care or a study group managed under an ERAS nursing pathway. The ERAS intervention included structured training, standardized perioperative nursing procedures, and a visual pathway-based guidance system. Key perioperative compliance indicators and clinical outcomes were collected and compared between groups. Results Implementation of the ERAS nursing pathway significantly improved multiple compliance indicators, including preoperative education rate, early postoperative feeding rate, early ambulation rate, elastic stocking usage rate, ankle pump exercise compliance, and breathing training compliance. In addition, postoperative complication rates and delayed discharge rates were reduced in the ERAS group compared with the control group. Nurses reported a high level of satisfaction with the ERAS system. Conclusion The ERAS nursing pathway effectively improves perioperative nursing compliance and postoperative recovery outcomes in lung cancer day-surgery patients, while reducing complications and delayed discharge. It also enhances nursing workflow standardization and staff satisfaction, supporting its clinical value and broader implementation in surgical nursing practice.
Purpose High-quality data are critical for developing healthcare machine learning (ML) models yet remain a major challenge in low- and middle-income countries (LMICs). This study evaluated the applicability of the Duke Institute for Health Innovation’s (DIHI) Machine Learning Data Quality Assurance (ML-DQA) framework in an LMIC setting using two ML projects at a tertiary care hospital in Pakistan. Methods We applied the ML-DQA framework to two ML projects, Sepsis Watch (SW) and In-hospital Mortality Prediction (IMP). The framework comprises four phases: pre-DQA, data element pre-processing, ML-DQA checks, and ML-DQA adjudication, and was used to evaluate data quality and usability for both models. Results In the pre-DQA phase, required data elements were identified through consultation with clinicians and project teams. During data element pre-processing, transformations and harmonization were performed, including consolidation of differently labelled glucose variables. ML-DQA checks assessed plausibility, conformance, and completeness, identifying elements that met quality standards. In the adjudication phase, reports were reviewed by two independent clinicians to determine suitability for model development. Conclusions The Duke ML-DQA framework effectively identified data limitations and informed model development, demonstrating its utility for assessing data quality and supporting ML implementation in resource-constrained settings.
