Background: The coronavirus disease (COVID-19) has become an urgent and serious global public health crisis. Community engagement is the first line of defense in the fight against infectious diseases, and general practitioners (GPs) play an important role in it. GPs are facing unique challenges from disasters and pandemics in delivering health care. However, there is still no suitable mobile management system that can help GPs collect data, dynamically assess risks, and effectively triage or follow-up with patients with COVID-19. Objective: The aim of this study is to design, develop, and deploy a mobile-based decision support system for COVID-19 (DDC19) to assist GPs in collecting data, assessing risk, triaging, managing, and following up with patients during the COVID-19 outbreak. Methods: Based on the actual scenarios and the process of patients using health care, we analyzed the key issues that need to be solved and designed the main business flowchart of DDC19. We then constructed a COVID-19 dynamic risk stratification model with high recall and clinical interpretability, which was based on a multiclass logistic regression algorithm. Finally, through a 10-fold cross-validation to quantitatively evaluate the risk stratification ability of the model, a total of 2243 clinical data consisting of 36 dimension clinical features from fever clinics were used for training and evaluation of the model. Results: DDC19 is composed of three parts: mobile terminal apps for the patient-end and GP-end, and the database system. All mobile terminal devices were wirelessly connected to the back end data center to implement request sending and data transmission. We used low risk, moderate risk, and high risk as labels, and adopted a 10-fold cross-validation method to evaluate and test the COVID-19 dynamic risk stratification model in different scenarios (different dimensions of personal clinical data accessible at an earlier stage). The data set dimensions were (2243, 15) when only using the data of patients’ demographic information, clinical symptoms, and contact history; (2243, 35) when the results of blood tests were added; and (2243, 36) after obtaining the computed tomography imaging results of the patient. The average value of the three classification results of the macro–area under the curve were all above 0.71 in each scenario. Conclusions: DCC19 is a mobile decision support system designed and developed to assist GPs in providing dynamic risk assessments for patients with suspected COVID-19 during the outbreak, and the model had a good ability to predict risk levels in any scenario it covered.

Pancreas identification and segmentation is an essential task in the diagnosis and prognosis of pancreas disease. Although deep neural networks have been widely applied in abdominal organ segmentation, it is still challenging for small organs (e.g. pancreas) that present low contrast, highly flexible anatomical structure and relatively small region. In recent years, coarse-to-fine methods have improved pancreas segmentation accuracy by using coarse predictions in the fine stage, but only object location is utilized and rich image context is neglected. In this paper, we propose a novel distance-based saliency-aware model, namely DSD-ASPP-Net, to fully use coarse segmentation to highlight the pancreas feature and boost accuracy in the fine segmentation stage. Specifically, a DenseASPP (Dense Atrous Spatial Pyramid Pooling) model is trained to learn the pancreas location and probability map, which is then transformed into saliency map through geodesic distance-based saliency transformation. In the fine stage, saliency-aware modules that combine saliency map and image context are introduced into DenseASPP to develop the DSD-ASPP-Net. The architecture of DenseASPP brings multi-scale feature representation and achieves larger receptive field in a denser way, which overcome the difficulties brought by variable object sizes and locations. Our method was evaluated on both public NIH pancreas dataset and local hospital dataset, and achieved an average Dice-Sørensen Coefficient (DSC) value of 85.49±4.77% on the NIH dataset, outperforming former coarse-to-fine methods.

Objectives: To demonstrate the feasibility of developing machine learning models for the prediction of hearing impairment in humans exposed to complex non-Gaussian industrial noise. Design: Audiometric and noise exposure data were collected on a population of screened workers (N = 1,113) from 17 factories located in Zhejiang province, China. All the subjects were exposed to complex noise. Each subject was given an otologic examination to determine their pure-tone hearing threshold levels and had their personal full-shift noise recorded. For each subject, the hearing loss was evaluated according to the hearing impairment definition of the National Institute for Occupational Safety and Health. Age, exposure duration, equivalent A-weighted SPL (L_Aeq), and median kurtosis were used as the input for four machine learning algorithms, that is, support vector machine, neural network multilayer perceptron, random forest, and adaptive boosting. Both classification and regression models were developed to predict noise-induced hearing loss applying these four machine learning algorithms. Two indexes, area under the curve and prediction accuracy, were used to assess the performances of the classification models for predicting hearing impairment of workers. Root mean square error was used to quantify the prediction performance of the regression models. Results: A prediction accuracy between 78.6 and 80.1% indicated that the four classification models could be useful tools to assess noise-induced hearing impairment of workers exposed to various complex occupational noises. A comprehensive evaluation using both the area under the curve and prediction accuracy showed that the support vector machine model achieved the best score and thus should be selected as the tool with the highest potential for predicting hearing impairment from the occupational noise exposures in this study. The root mean square error performance indicated that the four regression models could be used to predict noise-induced hearing loss quantitatively and the multilayer perceptron regression model had the best performance. Conclusions: This pilot study demonstrated that machine learning algorithms are potential tools for the evaluation and prediction of noise-induced hearing impairment in workers exposed to diverse complex industrial noises.

Non-used clinical information has negative implications on healthcare quality. Clinicians pay priority attention to clinical information relevant to their specialties during routine clinical practices but may be insensitive or less concerned about information showing disease risks beyond their specialties, resulting in delayed and missed diagnoses or improper management. In this study, we introduced an electronic health record (EHR)-oriented knowledge graph system to efficiently utilize non-used information buried in EHRs. EHR data were transformed into a semantic patient-centralized information model under the ontology structure of a knowledge graph. The knowledge graph then creates an EHR data trajectory and performs reasoning through semantic rules to identify important clinical findings within EHR data. A graphical reasoning pathway illustrates the reasoning footage and explains the clinical significance for clinicians to better understand the neglected information. An application study was performed to evaluate unconsidered chronic kidney disease (CKD) reminding for non-nephrology clinicians to identify important neglected information. The study covered 71,679 patients in non-nephrology departments. The system identified 2,774 patients meeting CKD diagnosis criteria and 10,377 patients requiring high attention. A follow-up study of 5,439 patients showed that 82.1% of patients who met the diagnosis criteria and 61.4% of patients requiring high attention were confirmed to be CKD positive during follow-up research. The application demonstrated that the proposed approach is feasible and effective in clinical information utilization. Additionally, it's valuable as an explainable artificial intelligence to provide interpretable recommendations for specialist physicians to understand the importance of non-used data and make comprehensive decisions.

Evidence-based medicine (EBM) represents a paradigm of providing patient care grounded in the most current and rigorously evaluated research. Recent advances in large language models (LLMs) offer a potential solution to transform EBM by automating labor-intensive tasks and thereby improving the efficiency of clinical decision-making. This study explores integrating LLMs into the key stages in EBM, evaluating their ability across evidence retrieval (PICO extraction, biomedical question answering), synthesis (summarizing randomized controlled trials), and dissemination (medical text simplification). We conducted a comparative analysis of seven LLMs, including both proprietary and open-source models, as well as those fine-tuned on medical corpora. Specifically, we benchmarked the performance of various LLMs on each EBM task under zero-shot settings as baselines, and employed prompting techniques, including in-context learning, chain-of-thought reasoning, and knowledge-guided prompting to enhance their capabilities. Our extensive experiments revealed the strengths of LLMs, such as remarkable understanding capabilities even in zero-shot settings, strong summarization skills, and effective knowledge transfer via prompting. Promoting strategies such as knowledge-guided prompting proved highly effective (e.g., improving the performance of GPT-4 by 13.10% over zero-shot in PICO extraction). However, the experiments also showed limitations, with LLM performance falling well below state-of-the-art baselines like PubMedBERT in handling named entity recognition tasks. Moreover, human evaluation revealed persisting challenges with factual inconsistencies and domain inaccuracies, underscoring the need for rigorous quality control before clinical application. This study provides insights into enhancing EBM using LLMs while highlighting critical areas for further research. The code is publicly available on Github.