Accurate influenza forecasting is essential for public health preparedness, yet many models require future covariates, provide limited interpretability, and degrade under post-pandemic regime shifts. We propose a Stage-Aware Multimodal Neural Network (SAMNN) that integrates multimodal temporal features using configurations adapted to heterogeneous transmission regimes and supports season-ahead scenario forecasting using adversarially synthesized covariates. Using 13 years of national A/H1N1 surveillance data from mainland China (2011∼2024), SAMNN was evaluated across three epidemiologically distinct seasons spanning ∼20-fold differences in peak intensity. SAMNN achieved R 2 values of 0.82-0.95, 0.82-0.97, and 0.55-0.97 for 1-week-, 2-week-, and 4-week-ahead forecasting, respectively, and generally maintained competitive performance relative to five baseline models across forecast horizons. SHAP attribution showed that epidemiological signals dominated predictions, with context-dependent contributions from climatic and social-context features. To support prospective scenario forecasting, we used an adversarial synthetic covariate-generation pipeline to produce season-ahead forecasts for 2024/25 using information available at the prespecified forecast origin.
Background:Antimicrobial resistance (AMR) is typically interpreted as a process of gradual accumulation of resistance traits. However, resistance systems may instead undergo structural transitions under changing selective pressures. We investigated whether long-term AMR dynamics in Corynebacterium striatum reflect accumulation or reorganization of a constrained phenotypic state space. Methods:Longitudinal resistance data (2013-2025) were analyzed using segmented regression, multinomial modeling, entropy-based diversity metrics, and Markov transition analysis to characterize temporal dynamics, architecture distributions, and system-level behavior. Results:A significant structural breakpoint was identified around 2020 (Davies' test p < 0.001). The estimated breakpoint location showed limited uncertainty based on model-based confidence interval estimation. Prior to 2020, aminoglycoside-containing resistance architectures were prevalent, whereas after 2020 they collapsed and were replaced by aminoglycoside-negative, fluoroquinolone-associated configurations. Phenotypic diversity declined significantly over time, indicating contraction of the resistance state space. Markov analysis demonstrated convergence toward a stable stationary distribution dominated by two related architectures, with increased dynamical stability after the breakpoint. Conclusions:AMR dynamics in C. striatum reflect a structural transition and restructuring within the simplified state space rather than progressive accumulation of resistance traits. The system converges toward a low-diversity phenotypic attractor, highlighting the importance of dynamical systems approaches in understanding resistance evolution.
Motivated by the impact of environmental variability on within-host HIV dynamics, we develop a stochastic HIV model incorporating latent cell activation. Infection and activation rates are modeled as logarithmic mean-reverting Ornstein-Uhlenbeck processes, ensuring positive parameters and biologically realistic fluctuations. We prove the existence and uniqueness of a global positive solution and derive stochastic thresholds characterizing persistence and extinction. If the persistence threshold exceeds one, the model admits a stationary distribution, indicating sustained viral presence; conversely, sufficiently strong environmental noise can drive latent cells, actively infected cells, and free virus to extinction almost surely, even when the deterministic model predicts persistence. Analysis of the associated Fokker-Planck equation provides explicit expressions for the covariance structure near quasi-equilibrium, quantifying stochastic fluctuations. Numerical simulations support the theoretical findings and illustrate how environmental noise reshapes HIV dynamics. Our results highlight the crucial role of stochastic variability in modulating viral persistence and latent reservoir dynamics.
Large clusters in HIV-1 molecular networks contain a substantial proportion of people living with HIV and dominate local epidemics; however, their large size and complex structure pose a challenge for effective public health interventions. We developed an analytical framework to partition the large cluster into small groups for precise intervention. In the HIV-1 CRF07_BC molecular transmission network in Guangzhou, China (2008-2020), a giant component (681 members) was partitioned into 34 communities with dense internal and sparse external links. All 378 inter-community links involved high-centrality members from Community 1 (P < 0.001) and phylogenetic analysis identified Community 1 as the most likely ancestral source of the giant component (marginal probability = 0.989). Exponential random graph models (ERGMs) revealed significant homophily effect among members with specific characteristics in the giant component and its large communities, highlighting potential outbreaks within specific subgroups. Partitioning giant components into communities may be a promising approach to help develop community-level interventions and could improve the effectiveness of interventions targeted at large clusters.
We develop a mathematical modeling framework to address the challenge in launching an effective staged vaccination campaign during a typical viral infection season to avoid an overwhelmed healthcare system for the entire season. Using the COVID-19 pandemic following its acute phase as a motivating example, our model takes into account the willingness of the public to receive vaccines, as well as the uncertainty of vaccination delivery and administration, to achieve the objective of optimizing the timing and distribution of the vaccination campaign, subject to the constraint that hospitalized cases do not exceed healthcare capacity. The integration of a dynamic transmission model with a scenario tree–based stochastic optimization framework enables the evaluation of future scenarios characterized by uncertainty in vaccination rates, contact mixing, and public adherence to safety measures. Accounting for these future scenarios facilitates the identification of strategies to dynamically adjust the timing and scale of vaccine and contact mixing during distinct phases of the viral season. Our study demonstrates that a well–timed, well–phased vaccination campaign, along with other public health interventions, can prevent overcrowding in hospitalization in a typical viral infection season.
The effectiveness of vaccination campaigns depends on vaccine efficacy, vaccination coverage, and vaccine allocation. A fundamental yet often overlooked question is whether higher vaccine efficacy or higher coverage should be prioritised. Here, within a leaky vaccine framework, we provide a theoretical analysis of their relative impact on reducing the epidemic final size in both homogeneous and heterogeneous populations, under proportional and optimal vaccine allocations. We prove that, when vaccine efficacy (in reducing susceptibility) and coverage are traded off so that their product P remains fixed but below the herd immunity threshold, the epidemic final size decreases as efficacy increases and increases as coverage increases, implying that efficacy should be prioritised. Mechanistically, under a fixed efficacy-coverage product, increasing efficacy amplifies population heterogeneity in susceptibility, which in turn reduces the final size. The difference between prioritising efficacy and prioritising coverage can be substantial, particularly when R 0 is large. For example, when the basic reproduction number R 0 = 5.0 and the product P = 0.6 , the final size under prioritising coverage can be 250% as large as that under prioritising efficacy, and up to 400% in heterogeneous populations. These results highlight the critical importance of vaccine efficacy and could provide useful guidance for vaccination policy and epidemic control.
There are several approaches to controlling the spread of infectious diseases. Among the most widely used measures are the protection of susceptible individuals and the isolation of infected ones. This raises a critical question: which strategy, protecting susceptible individuals, isolating infected ones, or combining both, most effectively mitigates disease transmission? The work proposed here, address this question by analyzing protection and isolation strategies within the framework of the classical SIR (Susceptible-Infected-Recovered) epidemic model. Three scenarios are considered: (i) combined protection and isolation, (ii) protection alone, and (iii) isolation alone. Using Optimal Control Theory, we show that, in the short-term context of epidemic control, combining protection and isolation generally provides the most effective strategy under the considered conditions. However, when the combination is unavailable or unnecessary, isolation tends to outperform protection, particularly under challenging conditions such as high transmission rates or low recovery rates. Protection alone can still be effective, but primarily in settings where recovery rates are sufficiently high. These results contrast with those obtained from autonomous models, in which, in the long-term dynamics, protecting susceptible individuals proved to be more effective than isolating the infected.
Background Household contact patterns significantly influence the household transmission dynamics of respiratory pathogens. Previous studies have shown ignoring the heterogeneity of household contact patterns can lead to an underestimation of the relative susceptibility and infectivity of children to influenza. However, there is a lack of studies on household contact patterns across all age groups and family structures in China. Methods We conducted two cross-sectional studies via face-to-face, paper-based surveys in Anhua County, Hunan Province (June–July and October 2021) and Kunming City, Yunnan Province (February–April 2023). Factors associated with household contact patterns were explored using linear mixed-effects models. Role-specific contact matrices were established, stratified by household structures and crowding levels. Additionally, using influenza as a representative example, we developed a discrete-time, individual-based stochastic model to investigate household transmission dynamics initiated by different index case roles and to evaluate various role-targeted vaccination strategies. Results The mean number of household contacts reported by participants was 2.58 (95% CI, 2.50–2.66) in Anhua County and 2.26 (95% CI, 2.19–2.33) in Kunming City. Household contact patterns were mainly associated with family roles, household structures, and the number of bedrooms. Contact duration was significantly longer between father–mother pairs, between mother–child pairs in two-generation households and mother–grandchild pairs in three-generation households, and among siblings. Mothers as index cases generated the highest number of secondary cases, whereas grandfathers caused the fewest. When grandparents or parents were the index cases, transmission most frequently occurred to children or grandchildren, whereas when children or grandchildren were the index cases, infection most commonly spread among siblings. Crowded living environments increased transmission risk, leading to a higher average number of secondary cases across all index case roles. Prioritizing vaccination of children or grandchildren yielded the greatest reduction in secondary cases, with further improvement when mothers were also vaccinated. Although vaccinating grandparents had minimal impact on reducing household transmission, it remained essential for their personal protection. Notably, vaccinating both grandparents and grandchildren together was the most efficient four-dose strategy for reducing infection risk for grandparents. Conclusions Given the heterogeneity of household contact patterns, to reduce household secondary cases, priority should be given to vaccinating children or grandchildren. However, if the aim is to reduce the infection risk among grandparents, directly vaccinating them is more effective. Unlike traditional age-based vaccination strategies, our findings provide a new perspective to optimize vaccine allocation.
Background:Africa bears the highest global burden of HIV, with marked regional inequalities in prevalence, incidence and clinical outcomes. Mapping the spatial and temporal evolution of the epidemic is essential to guide targeted interventions and anticipate future trends. Methods:We conducted a retrospective analysis of UNAIDS annual estimates for adults aged 15-49 years across 49 African countries (2014-2023). We described spatiotemporal patterns in HIV prevalence, incidence, adults living with HIV (ALHIV) and AIDS-related deaths, and quantified temporal trends using annual percentage change and linear regression. For the ten highest-burden countries, we forecast prevalence to 2033 using an ensemble of machine learning models. Hierarchical and k-means clustering, supported by principal component analysis, were applied to identify epidemic archetypes based on average prevalence levels and temporal trajectories. Results:Southern Africa remained the epicentre of the epidemic, with mean adult prevalence of 19.97% versus <1.3% in Northern and Western Africa. From 2014 to 2023, prevalence and incidence declined in all regions, with the steepest reductions in Southern (prevalence -19.5%; incidence -68.4%) and Eastern Africa (-22.2% and -65.6%, respectively). Despite falling rates, the absolute number of ALHIV increased in several regions, while AIDS-related deaths decreased by more than 44% in Central and Western Africa. Forecasts for the highest-burden countries indicate a continued, gradual decline in prevalence. Cluster analysis identified a hyperendemic group of six Southern African countries (mean prevalence 15.6%) and a second cluster of 41 countries with moderate-to-low prevalence (2.1%) and mainly stable or declining trajectories. Conclusions:The African HIV epidemic is increasingly heterogeneous and evolving rather than uniformly controlled. Combining machine learning forecasts and clustering with routine surveillance can support differentiated, data-driven strategies that intensify prevention and treatment in hyperendemic settings while sustaining gains elsewhere.
In light of the ongoing 2022-2025 HPAI outbreak in the U.S., which affects millions of commercial and backyard flocks, disrupts egg and meat production, and causes significant economic losses, it is important to develop biological models and optimization algorithms that conform to the U.S.-specific patterns of HPAI transmission and reflect control and prevention measures adopted in the USA. In this study, we introduce a partially stochastic network compartmental model to ascertain the progression and potential containment of HPAI virus in commercial flocks and wildlife. Parameters of the model get estimated using available data on wild bird migration, HPAI poultry outbreaks, and poultry farm inventory in different states of the U.S. The new model simulates HPAI virus transmission driven by wildlife dynamic, seasonality, and farm-to-farm relations. Unlike many prior global models, this framework is closely tailored to the U.S. HPAI statistics, farm structure, and current mitigation practices. The proposed network model, along with HPAI surveillance data, are used to analyze optimal control strategies aimed at lowering HPAI spread from wild birds to poultry. Our numerical experiments illustrate that the above control strategy is very powerful. In the absence of prevalent vaccination, these relatively inexpensive separation measures, such as covered runs and secure housing, help to prevent environmental contamination and the risk of HPAI transmission to domestic birds, thus protecting the flock and reducing depopulation.
Accurate modelling of infectious disease transmission often requires capturing how individuals adjust their behaviours in response to evolving epidemic conditions. While recently developed behavioural change epidemic models attempt to acknowledge such dynamics, the role of memory in shaping perceived risk has been treated in an ad hoc fashion. This study develops a data-driven framework of memory enhanced behavioural change individual-level models (MEBC-ILMs) that incorporate four distinct memory specifications: memoryless, sliding window, power-law decay, and exponential decay. These models allow behavioural responses to reflect varying assumptions about how past epidemiological information informs risk perception, with memory features inferred from epidemic data. Simulation experiments show that MEBC-ILMs can reliably recover key transmission and behavioural parameters under different memory settings and exhibit robust predictive performance even when the assumed memory structure differs from the true process. In contrast, the basic BC-ILM can perform poorly when memory affects present behaviour but is not accounted for in the model. Applying our framework to data from the 2001 U.K. foot and mouth disease epidemic illustrates how it can represent behavioural effects and explore plausible memory structures when fitted to real-world data.
In November 2025, Marburg virus caused the first outbreak of Marburg virus disease (MVD) in Ethiopia. By December 15, a total of 14 laboratory-confirmed cases were reported, including 9 deaths, corresponding to a case fatality ratio of 64.3% among confirmed cases. In the absence of licenced vaccines or antivirals, non-pharmaceutical interventions (NPIs) were implemented to control transmission. We developed a stochastic epidemic model incorporating a stochastic exponential growth process with reporting adjustment to assess the effectiveness of NPIs. The reporting ratio was modelled using a beta distribution, and transmission parameters were estimated via Markov chain Monte Carlo with particle filtering for likelihood calculation. Using pre-NPI data (November 12-23), we estimated an exponential growth rate of 0.012 (95% credible interval [CrI]: 0.008, 0.014) per day, corresponding to an initial reproduction number of 1.13 (95% CrI: 1.09, 1.16). We projected 64 cases (95% CrI: 19, 189) and 41 deaths (95% CrI: 12, 122) during post-NPI period (November 24 to December 15) without interventions. Compared with 4 observed cases, NPIs were associated with an estimated effectiveness of 93.7% (95% CrI: 78.9, 97.9). These findings indicated moderate transmissibility of MVD, and a potential transmission risk reduction following the implementation of control measures, underscoring the importance of timely intervention and sustained surveillance regarding local MVD activities.
Convergent cross mapping (CCM) method has been widely applied to investigate environmental drivers of infectious disease dynamics, particularly for seasonal influenza. However, its robustness to temporal gaps and missing observations—common features in surveillance data—remains largely unexplored. Using seasonal influenza surveillance data from Hong Kong, we systematically assessed the sensitivity of inferred environment–disease relationships to different data-exclusion scenarios, including the removal of low-activity periods and targeted time points. We compared CCM with quasi-binomial generalized linear models (GLM) and the Peter–Clark Momentary Conditional Independence (PCMCI+) framework.Across all scenarios, CCM-based inference exhibited pronounced sensitivity to data gaps, with both causal strength and inferred relationships varying substantially across gap configurations. In contrast, GLM and PCMCI+ estimates remained stable, consistently indicating a negative association between ozone and influenza transmission. These findings highlight a critical limitation of CCM when applied to incomplete time series and underscore the need for caution in interpreting causality from gap-affected epidemiological data.
Monkeypox (Mpox) has re-emerged as a serious global public health concern due to its potential human-to-human transmission and persistence in the environment. This study develops a new deterministic model incorporating double-dose vaccination approach to analyze the transmission dynamics and effective interventions for the Mpox outbreak. The model divides the human population in six groups and includes an environmental reservoir, capturing both direct transmission from infectious individuals and indirect transmission through environmental contamination. Global stability results of the equilibria are examined utilizing a standard Lyapunov function approach. The parameters are estimated using cumulative Mpox cases reported during 2022 outbreak in the United States. Further, normalized sensitivity analysis is performed to indicate the key parameters influencing disease transmission and eradication. Pontryagin’s maximum principle is applied to formulate an optimal control model using time-dependent variables for vaccination, treatment, and environmental disinfection. Simulation illustrates that double-dose vaccination substantially reduces new infections, particularly when coupled with timely treatment and environmental clearance measures. The findings of the present study highlight the importance of double-dose vaccination combined with complementary controlling measures in managing Mpox incidence and offer practical insights for public health planning.
Evaluating vector control interventions through randomized trials is often challenging because clinical endpoints, such as infection incidence, are rare and heterogeneous even areas at high risk, resulting in large required sample sizes. Antibodies to mosquito salivary proteins (MSPs) have emerged as promising markers of exposure to mosquito bites, yet their utility as trial endpoints is not fully understood. To address this gap, we developed a mechanistic modeling framework to describe MSP antibody dynamics in response to mosquito biting and to compare the statistical power of serological, clinical, and entomological trial endpoints. We introduce a new stochastic model for anti-MSP antibody dynamics—the Antibody Non-Homogeneous Poisson Process (ANPP) model—which incorporates seasonal variation in exposure and inter-individual heterogeneity. With a temperature-driven SEIR–SEI transmission model to create a unified simulation platform. Using this framework, we systematically compare sample size requirements for each endpoint under a range of simulated vector control strategies. Our results reveal a pronounced efficiency hierarchy: serological endpoints can reduce sample size needs by several orders of magnitude compared with clinical endpoints, especially in low-incidence settings. We also show that endpoints tied directly to mosquito population dynamics, such as antibody levels or trap counts, display seasonal patterns that mirror fluctuations in mosquito abundance, whereas infection-based endpoints remain comparatively flat. These findings provide a quantitative foundation for incorporating serological markers into trial design and highlight their potential to accelerate the evaluation and deployment of vector control tools.
In this paper, we introduce a novel dengue model on a weighted network to investigate the roles of human mobility, seasonal temperature shifts, and spatial heterogeneity. First, the positivity, global existence, and ultimate boundedness of the model solutions have also been discussed. Then, by utilizing the next generation operator theory, we derived the basic reproduction number R0 for the model, which determines the threshold dynamics of the system: When R0<1, the model possesses a unique disease-free periodic solution that is globally asymptotically stable; whereas when R0>1, dengue fever persists in both humans and mosquitoes, with at least one endemic periodic solution existing. Moreover, we formulate a corresponding optimal control model for dengue disease and derive the optimal prevention strategies with minimal implementation costs. Finally, several numerical examples are conducted to validate the theoretical results and the visualization outcomes demonstrate that spatial heterogeneity leads to heterogeneous disease transmission, and the total number of infected humans and infected mosquitoes increasing as the diffusion rate of infected humans rises and the peak of the temperature difference in the season is greater, and the epidemic is more likely to break out earlier, with more infected humans and mosquitoes. Additionally, we also present the control effects under different combinations of control measures and the spatiotemporal evolution of the optimal control solution.
Background World Health Organization recommends that the individuals with HIV infection take HIV detection as early as possible to avoid the potential transmission risk in the community, because HIV infection often exhibit no any symptoms and being neglected by the individuals at the acute phase. In China, a series of policies relating to HIV detection have been announced by the Chinese Center for Disease Control and Prevention for effectively controlling the HIV/AIDS infection scale. Methods The next generation matrix method is governed to derive the basic reproduction number of SIDMA model with HIV detection in this study. The least squares method is used to perform the optimal fitting against the surveillance data of HIV/AIDS. The sensitivity index and partial rank correlation coefficients are used to investigate the impacts of key parameters on the HIV/AIDS incidence. The ARIMA model method is applied to predict the gender and age distributions of the Fujian HIV/AIDS epidemics. The Gini index method is for the spatial heterogeneity of all diagnosed cases. Results The numerical simulation results show that the HIV/AIDS incidence depends on two significant parameters: detection rate and effective contact rate, and that basic reproduction number relies on effective contact rate. This study reveals that the long awareness delay extends the infection scale and prevalence risk of HIV/AIDS, and that the spatial heterogeneity of diagnosed cases in nine cities of Fujian Province is relatively uniform. The 2023-2030 tendency predictions with scenarios indicate that detection rate of SIDMA model is most critical for the prevalence of the Fujian HIV/AIDS epidemics. The gender and age predictions by ARIMA model indicate that females and 60 years old and over face the potential high HIV infection risks. Conclusions The HIV detection rate is the most important contributor of SIDMA model for suppressing the HIV/AIDS incidence. The declining trend of basic reproduction number of SIDMA model in five phases indicates that HIV detection in Fujian Province have achieved remarkable achievements. The 2023-2030 tendency predictions of the Fujian HIV/AIDS epidemics provide the estimations with low-level incidence, meeting the control goal from the Chinese government. The four insights are recommended to the policy-makers and the local governments for fighting against the HIV/AIDS prevalence.
Wildlife aggregate for many reasons (e.g. reproduction, feeding) and at times these aggregations can be extreme, with host densities increasing several orders of magnitude. While the impact of seasonality on infectious disease dynamics is well studied, few-if any-studies have explicitly examined how extreme aggregation affects key epidemiological outcomes. Here we consider an epidemic in a closed SIR (Susceptible-Infectious-Recovered) metapopulation with a hub-satellite structure, where seasonal movement into the hub follows a modified Gaussian function. We numerically explore how aggregation duration and timing shape two outcomes: final size and peak prevalence. We find a narrow set of circumstances and pathogens for which even extreme aggregation materially alters these outcomes. When aggregation coincides with, or begins just prior to, infection introduction, aggregation can strongly affect pathogens with R 0 ≈ 1 or R 0 < 1 , enabling epidemics that would otherwise fade. Effects are strongest under density-dependent transmission, where contact rate scales with local density; frequency-dependent transmission renders aggregation negligible. High transmissibility ( R 0 ≫ 2 ) minimises aggregation's impact because most susceptibles are infected regardless of density changes.
Aim This study aimed to develop an efficient and cost-saving diagnostic approach using natural language processing and explainable machine learning models. Subject and Methods 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. Results 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. Conclusions 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.
