Doctor–patient relationships (DPRs) in China have been straining. With the emergence of the COVID-19 pandemic, the relationships and interactions between patients and doctors are changing. This study investigated how patients’ attitudes toward physicians changed during the pandemic and what factors were associated with these changes, leading to insights for improving management in the healthcare sector. This paper collected 58,600 comments regarding Chinese doctors from three regions from the online health platform Good Doctors Online (haodf.com, accessed on 13 October 2022). These comments were analyzed using text mining techniques, such as sentiment and word frequency analyses. The results showed improvements in DPRs after the pandemic, and the degree of improvement was related to the extent to which a location was affected. The findings also suggest that administrative services in the healthcare sector need further improvement. Based on these results, we summarize relevant recommendations at the end of this paper.

Natural language-processing tasks have been improved greatly by large language models (LLMs). However, numerous parameters make their execution computationally expensive and difficult on resource-constrained devices. For this problem, as well as maintaining accuracy, some techniques such as distillation and quantization have been proposed. Unfortunately, current methods fail to integrate model pruning with downstream tasks and overlook sentence-level semantic modeling, resulting in reduced efficiency of distillation. To alleviate these limitations, we propose a novel distilled lightweight model for BERT named MicroBERT. This method can transfer the knowledge contained in the “teacher” BERT model to a “student” BERT model. The sentence-level feature alignment loss (FAL) distillation mechanism, guided by Mixture-of-Experts (MoE), captures comprehensive contextual semantic knowledge from the “teacher” model to enhance the “student” model’s performance while reducing its parameters. To make the outputs of “teacher” and “student” models comparable, we introduce the idea of a generative adversarial network (GAN) to train a discriminator. Our experimental results based on four datasets show that all steps of our distillation mechanism are effective, and the MicroBERT (101.14%) model outperforms TinyBERT (99%) by 2.24% in terms of average distillation reductions in various tasks on the GLUE dataset.

Background: Accurately measuring the health care needs of patients with different diseases remains a public health challenge for health care management worldwide. There is a need for new computational methods to be able to assess the health care resources required by patients with different diseases to avoid wasting resources. Objective: This study aimed to assessing dissatisfaction with allocation of health care resources from the perspective of patients with different diseases that can help optimize resource allocation and better achieve several of the Sustainable Development Goals (SDGs), such as SDG 3 (“Good Health and Well-being”). Our goal was to show the effectiveness and practicality of large language models (LLMs) in assessing the distribution of health care resources. Methods: We used aspect-based sentiment analysis (ABSA), which can divide textual data into several aspects for sentiment analysis. In this study, we used Chat Generative Pretrained Transformer (ChatGPT) to perform ABSA of patient reviews based on 3 aspects (patient experience, physician skills and efficiency, and infrastructure and administration)00 in which we embedded chain-of-thought (CoT) prompting and compared the performance of Chinese and English LLMs on a Chinese dataset. Additionally, we used the International Classification of Diseases 11th Revision (ICD-11) application programming interface (API) to classify the sentiment analysis results into different disease categories. Results: We evaluated the performance of the models by comparing predicted sentiments (either positive or negative) with the labels judged by human evaluators in terms of the aforementioned 3 aspects. The results showed that ChatGPT 3.5 is superior in a combination of stability, expense, and runtime considerations compared to ChatGPT-4o and Qwen-7b. The weighted total precision of our method based on the ABSA of patient reviews was 0.907, while the average accuracy of all 3 sampling methods was 0.893. Both values suggested that the model was able to achieve our objective. Using our approach, we identified that dissatisfaction is highest for sex-related diseases and lowest for circulatory diseases and that the need for better infrastructure and administration is much higher for blood-related diseases than for other diseases in China. Conclusions: The results prove that our method with LLMs can use patient reviews and the ICD-11 classification to assess the health care needs of patients with different diseases, which can assist with resource allocation rationally.

Sentiment analysis using Large Language Models (LLMs) has gained significant attention in recent research due to its outstanding performance and ability to understand complex texts. However, popular LLMs, such as ChatGPT, are typically closed-source and come with substantial API costs, posing challenges for resource-limited scenarios and raising concerns about privacy. To address this, our study evaluates the feasibility of using small LLMs (sLLMs) as alternatives to GPT for aspect-based sentiment analysis in Chinese healthcare reviews.We compared several Chinese sLLMs of varying sizes with GPT-3.5, using GPT-4o’s results as the benchmark, and assessed their classification accuracy by computing F1 scores for each individual aspect as well as an overall F1 score. Additionally, we examined sLLMs’ instruction-following capabilities, VRAM requirements, generation times, and the impact of temperature settings on the performance of top-performing sLLMs. The results demonstrate that several sLLMs can effectively follow instructions and even surpass GPT-3.5 in accuracy. For instance, InternLM2.5 achieved an F1 score of 0.85 with zero-shot prompting, while the smaller Qwen2.5-3B model performed well despite its minimal size. Prompt strategies significantly influenced smaller and older models like Qwen2.5-1.5B and ChatGLM3.5 but had limited impact on newer models. Temperature settings showed minimal effect, while older models generated responses faster, and newer models offered higher accuracy. This study underscores the potential of sLLMs as resource-efficient, privacy-preserving alternatives to closed-source LLMs in specialized domains. Our work demonstrates versatility, with potential applications across domains such as finance and education, and tasks like sentiment analysis, credit risk assessment, and learning behavior analysis, offering valuable insights for real-world use cases.

Background: Kidney stones, a prevalent urinary disease, pose significant health risks. Factors like insufficient water intake or a high-protein diet increase an individual’s susceptibility to the disease. Social media platforms can be a valuable avenue for users to share their experiences in managing these risk factors. Analyzing such patient-reported information can provide crucial insights into risk factors, potentially leading to improved quality of life for other patients. Objective: This study aims to develop a model KSrisk-GPT, based on a large language model (LLM) to identify potential kidney stone risk factors from web-based user experiences. Methods: This study collected data on the topic of kidney stones on Zhihu in the past 5 years and obtained 11,819 user comments. Experts organized the most common risk factors for kidney stones into six categories. Then, we use the least-to-most prompting in the chain-of-thought prompting to enable GPT-4.0 to think like an expert and ask GPT to identify risk factors from the comments. Metrics, including accuracy, precision, recall, and F1-score, were used to evaluate the performance of such a model. Results: Our proposed method outperforms other models in identifying comments containing risk factors with 95.9% accuracy and F1-score, with a precision of 95.6% and a recall of 96.2%. Out of the 863 comments identified with risk factors, our analysis showed the most mentioned risk factors for kidney stones in Zhihu user discussions, mainly including dietary habits (high protein, high calcium intake), insufficient water intake, genetic factors, and lifestyle. In addition, new potential risk factors were discovered with GPT, such as excessive use of supplements like vitamin C and calcium, laxatives, and hyperparathyroidism. Conclusions: Comments from social media users offer a new data source for disease prevention and understanding patient journeys. Our method not only sheds light on using LLMs to efficiently summarize risk factors from social media data but also on LLMs’ potential to identify new potential factors from the patient’s perspective.

Flat feet and high arches are common foot deformities that significantly affect human foot health. Early recognition has a substantial positive impact on interventions for abnormal arches. Previous research has focused on classifying foot arch types using machine learning methods based on plantar pressure images. However, due to angular deviation and the asymmetry between the left and right feet caused by individual differences, traditional convolutional neural networks (CNNs) struggle to achieve high performance, particularly when addressing variations in foot orientation in plantar pressure images. Additionally, objectively assessing these deep learning models is challenging due to the lack of publicly available image datasets for arch classification. To evaluate existing methods, this study collected 2,327 plantar pressure images from 1,126 participants, which were classified by two graders into the categories of high arches, flat feet, and normal feet. The proposed data augmentation method aligns more closely with assessing the generalization and robustness of these models. A series of improved models referred to as EquArchNets, has been proposed to address the issue of rotation invariance in CNNs for foot arch classification tasks. These models incorporate equivariant modules into traditional CNNs, and their parameters are optimized accordingly. Experimental results indicate that in the task of arch classification based on orientation-varied plantar pressure images, our enhanced framework achieves better performance in terms of accuracy, F1 score, and precision, with maximum increases of 4.02%, 4.12%, and 5.06%, respectively. Additionally, the minimum number of parameters in the optimized model is only 40.41% of the original.