Tiantian Zhang

Background: Precision health promotion, which aims to tailor health messages to individual needs, is hampered by the lack of structured metadata in vast digital health resource libraries. This bottleneck prevents scalable, personalized content delivery and exacerbates information overload for the public. Objective: This study aimed to develop, deploy, and validate an automated tagging system using a large language model (LLM) to create the foundational metadata infrastructure required for tailored health communication at scale. Methods: We developed a comprehensive, 3-tier health promotion taxonomy (10 primary, 34 secondary, and 90,562 tertiary tags) using a hybrid Delphi and corpus-mining methodology. We then constructed a hybrid inference pipeline by fine-tuning a Baichuan2-7B LLM with low-rank adaptation for initial tag generation. This was then refined by a domain-specific named entity recognition model and standardized against a vector database. The system’s performance was evaluated against manual annotations from nonexpert staff on a test set of 1000 resources. We used a “no gold standard” framework, comparing the artificial intelligence–human (A-H) interrater reliability (IRR) with a supplemental human-human (H-H) IRR baseline and expert adjudication for cases where artificial intelligence provided additional tags (“AI Additive”). Results: The A-H agreement was moderate (Cohen κ=0.54, 95% CI 0.53-0.56; Jaccard similarity coefficient=0.48, 95% CI 0.46-0.50). Critically, this was higher than the baseline nonexpert H-H agreement (Cohen κ=0.32, 95% CI 0.29-0.35; Jaccard similarity coefficient=0.35, 95% CI 0.27-0.43). A granular analysis of disagreements revealed that in 15.9% (159/1000) of the cases, the “AI Additive” tags were not identified by human annotators. Expert adjudication of these cases confirmed that the “AI Additive” tags were correct and relevant with a precision of 90% (45/50; 95% CI 78.2%-96.7%). Conclusions: A fine-tuned LLM, integrated into a hybrid pipeline, can function as a powerful augmentation tool for health content annotation. The system’s consistency (A-H κ=0.54) was found to be superior to the baseline human workflow (H-H κ=0.32). By moving beyond simple automation to reliably identify relevant health topics missed by manual annotators with high, expert-validated accuracy, this study provides a robust technical and methodological blueprint for implementing artificial intelligence to enhance precision health communication in public health settings.

BACKGROUND Precision health promotion, which aims to tailor health messages to individual needs, is hampered by the lack of structured metadata in vast digital health resource libraries. This bottleneck prevents scalable, personalized content delivery and exacerbates information overload for the public. OBJECTIVE This study aimed to develop, deploy, and validate an automated tagging system using a large language model (LLM) to create the foundational metadata infrastructure required for tailored health communication at scale. METHODS We developed a comprehensive, 3-tier health promotion taxonomy (10 primary, 34 secondary, and 90,562 tertiary tags) using a hybrid Delphi and corpus-mining methodology. We then constructed a hybrid inference pipeline by fine-tuning a Baichuan2-7B LLM with low-rank adaptation for initial tag generation. This was then refined by a domain-specific named entity recognition model and standardized against a vector database. The system’s performance was evaluated against manual annotations from nonexpert staff on a test set of 1000 resources. We used a “no gold standard” framework, comparing the artificial intelligence–human (A-H) interrater reliability (IRR) with a supplemental human-human (H-H) IRR baseline and expert adjudication for cases where artificial intelligence provided additional tags (“AI Additive”). RESULTS The A-H agreement was moderate (Cohen κ=0.54, 95% CI 0.53-0.56; Jaccard similarity coefficient=0.48, 95% CI 0.46-0.50). Critically, this was higher than the baseline nonexpert H-H agreement (Cohen κ=0.32, 95% CI 0.29-0.35; Jaccard similarity coefficient=0.35, 95% CI 0.27-0.43). A granular analysis of disagreements revealed that in 15.9% (159/1000) of the cases, the “AI Additive” tags were not identified by human annotators. Expert adjudication of these cases confirmed that the “AI Additive” tags were correct and relevant with a precision of 90% (45/50; 95% CI 78.2%-96.7%). CONCLUSIONS A fine-tuned LLM, integrated into a hybrid pipeline, can function as a powerful augmentation tool for health content annotation. The system’s consistency (A-H κ=0.54) was found to be superior to the baseline human workflow (H-H κ=0.32). By moving beyond simple automation to reliably identify relevant health topics missed by manual annotators with high, expert-validated accuracy, this study provides a robust technical and methodological blueprint for implementing artificial intelligence to enhance precision health communication in public health settings.

Catastrophic health expenditure (CHE) is a key indicator for studying the economic burden of diseases. Previous study mainly focused on exploring the relationship between severe diseases such as cancer and multimorbidity with CHE. However, there is a lack of research exploring the differential effects of single and multiple chronic diseases on CHE. This study aims to analyse the longitudinal association between different chronic disease conditions and CHE among elderly households in China, and investigate whether this association is influenced by other demographic and socioeconomic variables. A total of 6,483 older adults obtained from four waves (2013, 2015, 2018, 2020) of the China Health and Retirement Longitudinal Study (CHARLS) were analysed (elderly household without CHE at baseline were included). CHE was defined as out-of-pocket health expenditures exceeding 40% of household non-food expenditures. Mixed-effects logistic regression models were employed to explore the longitudinal relationship between different chronic diseases conditions and CHE, and stratification analyses were conducted to explore the potential moderation effect. The analysis indicates that both single chronic diseases(P < 0.001) and multimorbidity(P < 0.001) significantly increase the risk of CHE in elderly households. A stratified analysis based on population and socio-economic variables showed similar results. Besides, when the threshold for CHE was set at 40% of household non-food expenditures, the risk increase for CHE was similar for both elderly individuals with single and multiple chronic diseases. However, when the threshold was lowered to 10%, 25% of total household expenditures or 25% of household non-food expenditures, elderly individuals with multiple chronic diseases had a higher risk of CHE compared to those with a single chronic disease or without chronic disease. The influencing factors of CHE included age, education level, marital status, hukou, and place of residence. In particular, basic medical insurance did not significantly reduce the risk of CHE. Both single and multiple chronic diseases significantly increase the risk of catastrophic health expenditure among elderly households in China, and the existing basic medical insurance coverage has not effectively alleviated this burden. The findings highlight the urgent need for improving social health insurance programs, as well as conducting differentiated financial protection strategies for elderly individuals with different chronic disease statuses to reduce the financial vulnerability of elderly households with chronic diseases.

Background: Due to the expanding coverage of medical insurance and the growth of medical expenses, the ability to assess the performance of designated medical institutions (DMIs) in supporting the delivery of high-quality patient care and the standardized use of funds represents a priority in China. Despite such interest, there has yet to be an operable standard and labor-saving method for assessing DMIs in China.Objective: The main objectives include two aspects: (1) establishing an evaluation index system for DMIs based on contracts; (2) designing and developing an online evaluation platform.Methods: A group of 20 experts with theoretical and practical expertise in medical insurance regulation and performance evaluation were invited to select available indicators. A combination weighting method based on analytic hierarchy process and entropy method was used to determine the weight coefficient. Shanghai was taken as the sample area, and 760 DMIs were included in the empirical research. The test-retest reliability method and criterion-related validity method was used to test the reliability and the validity of the evaluation result.Results: An assessment index system that included 6 domains and 56 indicators was established in this study. Furthermore, we developed an online platform to assist in the implementation of the assessment. The results showed that the average score of assessment was 94.39, the median was 96.92. The test-retest reliability value was 0.96 (P ≤ 0.01), which indicated high stability of the assessment. In addition, there was a significant negative relationship between assessment score and the penalty amount of DMIs (R = −0.133, P < 0.001). After adjusting for the basic characteristics of medical institutions, the number of visits and revenue, the negative relationship was still significant (B = −0.080, P < 0.05). These results are consistent with expectations, indicating that the assessment had good criterion-related validity.Conclusions: This study established an operable assessment measure and developed an online platform to assess the performance of DMIs. The results showed good feasibility and reliability in empirical research. Our research findings provided an operable Chinese solution for DMI assessment that saves manpower and time, which would have good enlightening significance in other regions of China and in low-income and middle-income countries internationally.

Background: Dyslipidemia is a major risk factor for cardiovascular diseases, but its prevalence and determinants among sedentary occupational population are poorly understood. This study aimed to investigate the prevalence and associated factors for dyslipidemia among a sedentary occupational population in Shanghai, China. Methods: We collected data from 35,950 sedentary occupational workers on their demographics, social, medical, and family history, lifestyle, anthropometry, and biochemistry. We used the 2016 Chinese guideline for the management of dyslipidemia in adults to define dyslipidemia and its subtypes. We performed multivariate logistic regression to examine the factors associated with dyslipidemia. Results: The prevalence of dyslipidemia was 29.10%, with 15.86% for high triglycerides (TG), 6.43% for high total cholesterol (TC), 5.37% for high low-density lipoprotein cholesterol (LDL-C), and 14.68% for low high-density lipoprotein cholesterol (HDL-C). Men had a significantly higher prevalence of dyslipidemia than women (39.64% vs. 12.43%, P < 0.01). Factors associated with dyslipidemia included older age, being married, longer sedentary time while resting, frequent intake of animal viscera, current smoking, hypertension, diabetes, and obesity. Current drinking was associated with a 1.24 times higher prevalence of high TG (P < 0.01). Current smokers were less likely to have low HDL-C than non-smokers. Conclusions: Our present study, in a population of 35,950 sedentary occupational workers from Shanghai, demonstrated a prevalence of dyslipidemia, but lower than in other previous studies without the limitation of occupational characteristics. Prevention and control measures for dyslipidemia should take into account the characteristics and related factors for this population group.