Weibing Wang

This study aimed to develop an efficient and cost-saving diagnostic approach using natural language processing and explainable machine learning models. 11,863 Influenza-like illness cases from Huzhou City, China for four common respiratory viruses were collected: SARS-CoV-2, influenza, respiratory syncytial virus, and adenovirus. Natural language processing techniques were employed to extract and normalize symptom features from unstructured clinical text. Five machine learning algorithms were evaluated using AUC, accuracy, sensitivity, and specificity to select the best-performing model. Subgroup analyses by age, sex, and fever status assessed model robustness, and SHAP values were calculated for interpretability. Compared with existing diagnostic tools, our model demonstrated higher accuracy and better predictive performance, with AUCs of 0.856 (95% CI: 0.830-0.881) for SARS-CoV-2, 0.737 (95% CI: 0.713-0.760) for Influenza, 0.801 (95% CI: 0.744-0.857) for RSV, and 0.782 (95% CI: 0.748-0.816) for adenovirus, showing particularly high capability for SARS-CoV-2 and RSV. Subgroup analyses showed particularly excellent discriminative accuracy in pediatric or afebrile patients. This study demonstrates the feasibility of integrating natural language processing and machine learning techniques for identification of respiratory viruses based solely on symptoms, and offers a low-cost and efficient alternative to PCR testing, which can reduce reliance on resource-intensive testing and enhance early detection in clinical practice. This approach can support early screening and resource allocation in both clinical and public health settings.

Based on our clinical trial benefits of soluble dietary fibre in chronic kidney disease (CKD), we investigated the effects of 20% high-dose inulin (INU) in adenine-induced CKD mice. As expected, INU alleviated renal injury, reduced serum creatinine, and downregulated renal expression of inflammatory, fibrotic and STING-related molecules. It also improved serum lipid profiles, upregulated renal lipid metabolism genes, and remodelled serum lipidomic composition. Notably, inulin failed to elevate the main short-chain fatty acids (SCFAs) levels or beneficially reshape gut microbiota, and even enriched potential opportunistic pathogens. Therefore, the renoprotective and lipid-regulating effects of high-dose inulin are not primarily mediated by the gut microbiota-SCFAs axis. High-dose inulin ameliorates renal and lipid damage but shows limitations in optimizing gut microecology, providing experimental evidence for rational prebiotic application in CKD.

Background: Tuberculosis (TB) progression involves immune remodeling, but the regulatory mechanisms underlying functional alterations in CD4⁺ central memory T (TCM) cells remain unclear. We aimed to characterize CD4⁺ TCM remodeling across TB disease states.Methods: We integrated whole-blood RNA sequencing from 432 individuals with single-cell RNA sequencing from 37 individuals spanning healthy controls, tuberculosis infection, and active TB. Weighted gene co-expression network analysis, transcription factor regulon inference, and regulatory-velocity modeling were used to characterize CD4⁺ TCM regulatory programs and state-transition dynamics.Findings: A T-cell memory-associated transcriptional module was negatively correlated with disease progression and enriched in CD4⁺ TCM cells. Despite largely preserved functional-state proportions, CD4⁺ TCM cells exhibited altered regulatory programs and state-transition dynamics in active TB. KLF2 expression and regulon activity were reduced in active TB, accompanied by weakened coupling between KLF2 expression and its inferred regulatory output and attenuation of sphingosine-1-phosphate-associated trafficking programs. KLF2 regulon activity was negatively correlated with regulatory-velocity time, and in silico KLF2 perturbation predicted increased fate probabilities toward a quiescent memory state.Interpretation: Active TB is associated with KLF2-related regulatory remodeling and altered state-transition dynamics within a relatively preserved CD4⁺ TCM compartment. These findings identify KLF2-associated regulation as a candidate contributor to memory T-cell dysfunction during TB progression, warranting longitudinal and experimental validation.Funding: National Science and Technology Major Project of China; Shanghai Municipal Science and Technology Major Project; National Natural Science Foundation of China; and Healthcare Talents Youth Program of Shanghai Pudong New Area.

Background: Acute respiratory infections (ARIs) in children under 5 years represent a high-risk area for inappropriate antibiotic use. However, relevant research lacks systematic characterisation of the complete pathway from symptom onset to clinical antibiotic prescription and assessment of pathway-level interventions on prescribing appropriateness.Methods: We conducted a multicentre cross-sectional survey of children under 5 years with ARIs.. Caregiver-reported data were collected via structured questionnaires. A Bayesian Markov chain model was constructed to simulate care-seeking pathway, calculate state transition probabilities, identify high-risk pathways, and evaluate the impact of four macro-level interventions on inappropriate antibiotic prescribing rates in mild and severe groups.Results: A total of 2,572 children were enrolled, with 75.8% classified as non-severe and 24.2% as severe. The overall antibiotic prescribing rate was 50.0%, with inappropriate use accounting for 47.1% of all prescriptions. The Markov chain model demonstrated excellent fit (severe group r = 0.996; non-severe group r = 0.980). High-risk pathways, defined by presentation at a high‑level facility without diagnostic testing, were identified in both groups. Counterfactual simulations revealed that all four interventions reduced inappropriate prescribing in non-severe cases. In severe cases, however, lowering first-visit care level and reducing infection frequency paradoxically increased inappropriate use. Mandating in-hospital diagnostic testing was the most powerful and consistent intervention across both groups. Conclusions: With half of antibiotic prescriptions inappropriate in children under 5, policy should prioritise high-risk pathways—high-level care without testing. Counterfactuals show mild cases benefit from primary care referral; severe cases require mandatory diagnostics. Stratified, pathway-targeted interventions are essential to curb overuse.

Quantifying transmission distance helps to understand infectious disease spread patterns, but few studies have assessed this for (re-)emerging respiratory infectious diseases. The COVID-19 pandemic's extensive surveillance and high-quality data in China offer a strong model for studying the spread pattern of respiratory infections. We collected COVID-19 outbreak data from June 2020 to December 2022 in mainland China to estimate transmission distances, examine the association between urban characteristics and transmission distance, and explore its implications in outbreak response. The mean transmission distance of all outbreaks was 18.94 km. This distance varied by SARS-CoV-2 variants, with wild-type outbreaks showing shortest transmission distance (3.08 km) and Omicron outbreaks showing longest transmission distance (20.17 km). Transmission distance was positively associated with city size and total length of roads. Potential transmission areas identified as areas within transmission distance of the first 100 cases contained 63.84–99.61% of cases. Providing testing services to residents in these areas helped to control the outbreak. Our results suggest that transmission distances can help predict where emerging respiratory infectious diseases may spread next spatially and guide the implementation of targeted interventions.

Background: Public health initiatives increasingly target multiple overlapping high-risk groups to maximize impact. However, a common challenge in modelling these initiatives is to capture these overlapping risk factors, leading to potential misallocation of resources and biased effectiveness estimates. Using China’s tuberculosis (TB) control program as an example, this study explores different possible frameworks to account for population heterogeneity and risk overlap. Methods: We examined four risk allocation frameworks: ( i ) Direct Summation (DS), a simple additive benchmark; ( ii ) Probabilistic Union Deduplication (PUD), using inclusion-exclusion principles; ( iii ) Risk population combination (RPC), modeling interaction effects; and (iv) Agent-Based Framework (ABF), a granular microsimulation. To show how these frameworks could be used in epidemiological modelling, we embedded each within a deterministic transmission model of TB epidemiology in China, to simulate the impact of China’s National Tuberculosis Strategic Plan (NTSP). We explored each framework when implemented in both static and dynamic versions. We compared them using methodological principles and indicators of intervention cost (screening volume) and benefits (cases/deaths averted). Results: Under the static version, the detection yield of active cases followed a consistent hierarchy: DS > PUD > RPC ≈ ABF. The DS method systematically overestimated yields by double-counting overlapping populations, while PUD corrected for overlap but ignored interaction. The RPC and ABF methods provided the most granular estimates by incorporating Risk population combinations. Additionally, comparing static versus dynamic versions revealed that for the same multi-risk screening framework, mortality reductions remained stable and incidence reductions varied significantly. Conclusion: This study presents potential screening frameworks for overlapping risk populations. The RPC method offers optimal balance of real-world plausibility and computational efficiency. We propose the dynamic RPC method as the preferred tool for routine analysis where multimorbidity and intersectional risks exist, providing a robust evidence base for optimizing resource allocation in heterogeneous populations.

Background: In settings with high antibiotic pressure, pneumococcal conjugate vaccines (PCVs) offer an underutilized strategy for antimicrobial resistance mitigation. In China, PCVs remain excluded from the National Immunization Program (NIP), providing a unique epidemiological context to evaluate the impact of suboptimal vaccine coverage on the carriage and transmission of resistant Streptococcus pneumoniae.Methods: In the multicenter HAPEC study, we collected nasopharyngeal swabs from 1,695 children under five years with clinically diagnosed pneumonia and 1,467 healthy community children across Hainan Province between April 2023 and November 2024, and characterized pneumococci by multiplex PCR, culture, serotyping and antimicrobial susceptibility testing.Findings: Pneumococcus was detected in 41.8% of pneumonia cases and 33.1% of community children, despite widespread pre-sampling antibiotic exposure. PCV13-type serotypes accounted for 53.5% and 42.3% of carriage, respectively, and were more frequently resistant than non-PCV13 serotypes to one or more (67.5% vs 54.7%; p=0.01) and two or more (21.6% vs 12.4%; p=0.02) clinically used antibiotics. Only 10–20% of children had received any PCV. Receipt of two or more doses, but not a single dose, was associated with lower PCV13-type carriage (aOR=0.44, p<0.01).Interpretation: Broader PCV13 coverage with full-schedule completion and government-supported access could reduce antibiotic-resistant pneumococcal carriage and transmission in both hospital and community settings.