Search

Abstract Background Global myopia prevalence poses a substantial public health burden with vision-threatening complications, necessitating effective prevention and control strategies. Precise prediction of spherical equivalent (SE), myopia, and high myopia onset is vital for proactive clinical interventions. Methods We reviewed electronic medical records of pediatric and adolescent patients who underwent cycloplegic refraction measurements at the Eye & Ear, Nose, and Throat Hospital of Fudan University between January 2005 and December 2019. Patients aged 3–18 years who met the inclusion criteria were enrolled in this study. To predict the SE and onset of myopia and high myopia in a specific year, two distinct models, random forest (RF) and the gradient boosted tree algorithm (XGBoost), were trained and validated based on variables such as age at baseline, and SE at various intervals. Outputs included SE, the onset of myopia, and high myopia up to 15 years post-initial examination. Age-stratified analyses and feature importance assessments were conducted to augment the clinical significance of the models. Results The study enrolled 88,250 individuals with 408,255 refraction records. The XGBoost-based SE prediction model consistently demonstrated robust and better performance than RF over 15 years, maintaining an R 2 exceeding 0.729, and a Mean Absolute Error ranging from 0.078 to 1.802 in the test set. Myopia onset prediction exhibited strong area under the curve (AUC) values between 0.845 and 0.953 over 15 years, and high myopia onset prediction showed robust AUC values (0.807–0.997 over 13 years, with the 14th year at 0.765), emphasizing the models' effectiveness across age groups and temporal dimensions on the test set. Additionally, our classification models exhibited excellent calibration, as evidenced by consistently low brier score values, all falling below 0.25. Moreover, our findings underscore the importance of commencing regular examinations at an early age to predict high myopia. Conclusions The XGBoost predictive models exhibited high accuracy in predicting SE, onset of myopia, and high myopia among children and adolescents aged 3–18 years. Our findings emphasize the importance of early and regular examinations at a young age for predicting high myopia, thereby providing valuable insights for clinical practice.

Purpose: To evaluate the performance of four large language models (LLMs)—GPT-4, PaLM 2, Qwen, and Baichuan 2—in generating responses to inquiries from Chinese patients about dry eye disease (DED). Design: Two-phase study, including a cross-sectional test in the first phase and a real-world clinical assessment in the second phase. Subjects: Eight board-certified ophthalmologists and 46 patients with DED. Methods: The chatbots' responses to Chinese patients' inquiries about DED were assessed by the evaluation. In the first phase, six senior ophthalmologists subjectively rated the chatbots’ responses using a 5-point Likert scale across five domains: correctness, completeness, readability, helpfulness, and safety. Objective readability analysis was performed using a Chinese readability analysis platform. In the second phase, 46 representative patients with DED asked the two language models (GPT-4 and Baichuan 2) that performed best in the in the first phase questions and then rated the answers for satisfaction and readability. Two senior ophthalmologists then assessed the responses across the five domains. Main outcome measures: Subjective scores for the five domains and objective readability scores in the first phase. The patient satisfaction, readability scores, and subjective scores for the five-domains in the second phase. Results: In the first phase, GPT-4 exhibited superior performance across the five domains (correctness: 4.47; completeness: 4.39; readability: 4.47; helpfulness: 4.49; safety: 4.47, p

ObjectivesTo employ a drug supply chain information system to optimize drug management practices, reducing costs and improving efficiency in financial and asset management.MethodsA digital artificial intelligence + vendor managed inventory (AI+VMI)-based system for drug supply chain information management in hospitals has been established. The system enables digitalization and intelligentization of purchasing plans, reconciliations, and consumption settlements while generating purchase, sales, inventory reports as well as various query reports. The indicators for evaluating the effectiveness before and after project implementation encompass drug loss reporting, inventory discrepancies, inter-hospital medication retrieval frequency, drug expenditure, and cloud pharmacy service utilization.ResultsThe successful implementation of this system has reduced the hospital inventory rate to approximately 20% and decreased the average annual inventory error rate from 0.425‰ to 0.025‰, significantly boosting drug supply chain efficiency by 42.4%. It has also minimized errors in drug application, allocation, and distribution while increasing adverse reaction reports. Drug management across multiple hospital districts has been standardized, leading to improved access to medicines and enhanced patient satisfaction.ConclusionThe AI+VMI system improves drug supply chain management by ensuring security, reducing costs, enhancing efficiency and safety of drug management, and elevating the professional competence and service level of pharmaceutical personnel.

Abstract Few-shot semantic segmentation (FSS) has been widely used in the field of information medicine and intelligent diagnosis. Due to the high cost of medical data collection and the privacy protection of patients, labeled medical images are difficult to obtain. Compared with other semantic segmentation dataset which can be automatically generated in a large scale, the medical image data tend to be continually generated. Most of the existing FSS techniques require abundant annotated semantic classes for pre-training and cannot deal with its dynamic nature of medical data stream. To deal with this issue, we propose a dynamic few-shot learning framework for medical semantic segmentation, which can fully utilize the features of newly-collected/generated data stream. We introduce a new pseudo-label generation strategy for continuously generating pseudo-labels and avoiding model collapse during self-training. Furthermore, an efficient consistency regularization strategy is proposed to fully utilize the limited data. The proposed framework is iteratively trained on three tasks: abdominal organ segmentation for CT and MRI, and cardiac segmentation for MRI. Experiments results demonstrate significant performance gain on medical data stream mining compared with the baseline method.

BackgroundConversational agents (CAs) have been developed in outpatient departments to improve physician-patient communication efficiency. As end users, patients’ continuance intention is essential for the sustainable development of CAs. ObjectiveThe aim of this study was to facilitate the successful usage of CAs by identifying key factors influencing patients’ continuance intention and proposing corresponding managerial implications. MethodsThis study proposed an extended expectation-confirmation model and empirically tested the model via a cross-sectional field survey. The questionnaire included demographic characteristics, multiple-item scales, and an optional open-ended question on patients’ specific expectations for CAs. Partial least squares structural equation modeling was applied to assess the model and hypotheses. The qualitative data were analyzed via thematic analysis. ResultsA total of 172 completed questionaries were received, with a 100% (172/172) response rate. The proposed model explained 75.5% of the variance in continuance intention. Both satisfaction (β=.68; P

Background: Recently, internet hospitals have been emerging in China, saving patients time and money during the COVID-19 pandemic. In addition, pharmacy services that link doctors and patients are becoming essential in improving patient satisfaction. However, the existing internet hospital pharmacy service mode relies primarily on manual operations, making it cumbersome, inefficient, and high-risk.Objective: To establish an internet hospital pharmacy service mode based on artificial intelligence (AI) and provide new insights into pharmacy services in internet hospitals during the COVID-19 pandemic.Methods: An AI-based internet hospital pharmacy service mode was established. Initially, prescription rules were formulated and embedded into the internet hospital system to review the prescriptions using AI. Then, the “medicine pick-up code,” which is a Quick Response (QR) code that represents a specific offline self-pick-up order, was created. Patients or volunteers could pick up medications at an offline hospital or drugstore by scanning the QR code through the window and wait for the dispensing machine or pharmacist to dispense the drugs. Moreover, the medication consultation function was also operational.Results: The established internet pharmacy service mode had four major functional segments: online drug catalog search, prescription preview by AI, drug dispensing and distribution, and AI-based medication consultation response. The qualified rate of AI preview was 83.65%. Among the 16.35% inappropriate prescriptions, 49% were accepted and modified by physicians proactively and 51.00% were passed after pharmacists intervened. The “offline self-pick-up” mode was preferred by 86% of the patients for collecting their medication in the internet hospital, which made the QR code to be fully applied. A total of 426 medication consultants were served, and 48.83% of them consulted outside working hours. The most frequently asked questions during consultations were about the internet hospital dispensing process, followed by disease diagnosis, and patient education. Therefore, an AI-based medication consultation was proposed to respond immediately when pharmacists were unavailable.Conclusion: The established AI-based internet hospital pharmacy service mode could provide references for pharmacy departments during the COVID-19 pandemic. The significance of this study lies in ensuring safe/rational use of medicines and raising pharmacists’ working efficiency.

Objective: Tinnitus is highly prevalent in the population, but there are currently few effective therapeutic interventions. Mobile applications (apps) might be helpful in tinnitus diagnosis and treatment by offering sound or music tools as well as questionnaires. We assessed the efficacy of a free, publicly available smartphone app (Fudan Tinnitus Relieving System, FTRS) for self-management of tinnitus and related symptoms. Methods: Among a total of 3564 participants recruited primarily online, 2744 patients had complete information at baseline and were an average of 37 years old and were 59.84 % male. Web-administered self-report measures THI, HADS, AIS, and other multi-dimensional scales were conducted at baseline and at 1 month and 2 months following treatment. Data from 54 participants who completed continuous follow-up were used for the final efficacy analysis and longitudinal analysis. Results: Following the intent-to-treat principle, t-tests revealed that the distribution of patients and the tinnitus features of patients of different genders were heterogeneous. One-way ANOVA showed that after using the FTRS app, THI scores showed a decreasing trend (p

Early detection and treatment of visual impairment diseases are critical and integral to combating avoidable blindness. To enable this, artificial intelligence–based disease identification approaches are vital for visual impairment diseases, especially for people living in areas with a few ophthalmologists. In this study, we demonstrated the identification of a large variety of visual impairment diseases using a coarse-to-fine approach. We designed a hierarchical deep learning network, which is composed of a family of multi-task & multi-label learning classifiers representing different levels of eye diseases derived from a predefined hierarchical eye disease taxonomy. A multi-level disease–guided loss function was proposed to learn the fine-grained variability of eye disease features. The proposed framework was trained for both ocular surface and retinal images, independently. The training dataset comprised 7,100 clinical images from 1,600 patients with 100 diseases. To show the feasibility of the proposed framework, we demonstrated eye disease identification on the first two levels of the eye disease taxonomy, namely 7 ocular diseases with 4 ocular surface diseases and 3 retinal fundus diseases in level 1 and 17 subclasses with 9 ocular surface diseases and 8 retinal fundus diseases in level 2. The proposed framework is flexible and extensible, which can be inherently trained on more levels with sufficient training data for each subtype diseases (e.g., the 17 classes of level 2 include 100 subtype diseases defined as level 3 diseases). The performance of the proposed framework was evaluated against 40 board-certified ophthalmologists on clinical cases with various visual impairment diseases and showed that the proposed framework had high sensitivity and specificity with the area under the receiver operating characteristic curve ranging from 0.743 to 0.989 in identifying all identified major causes of blindness. Further assessment of 4,670 cases in a tertiary eye center also demonstrated that the proposed framework achieved a high identification accuracy rate for different visual impairment diseases compared with that of human graders in a clinical setting. The proposed hierarchical deep learning framework would improve clinical practice in ophthalmology and broaden the scope of service available, especially for people living in areas with a few ophthalmologists.