Monday, Aug 4: 8:30 AM - 10:20 AM
4038
Contributed Papers
Music City Center
Room: CC-202A
Main Sponsor
Section on Statistics in Epidemiology
Presentations
Influenza infection remains a major cause of morbidity and mortality in the United States and worldwide. According to the Centers for Disease Control and Prevention (CDC), between 2010 and 2020, influenza was responsible for 9 to 41 million cases of illness annually, resulting in 140,000 to 710,000 hospitalizations and 12,000 to 52,000 deaths. Beyond its direct health impact, influenza imposes a substantial economic burden on the U.S.
This study integrates influenza-like illness (ILI) data from the CDC with air quality metrics from the U.S. Environmental Protection Agency (EPA) and weather data from the National Oceanic and Atmospheric Administration (NOAA). We developed a novel GRU-CNN hybrid deep learning model to forecast influenza trends using these environmental and meteorological inputs. The model incorporates Yeo-Johnson scaling, self-normalizing activations, and adaptive learning rate scheduling to enhance accuracy and stability. We evaluated the hybrid model's performance and compared it to traditional INGARCH and INGARCHX models. The GRU-CNN approach demonstrated superior predictive accuracy while maintaining computational efficiency, underscoring its potential as a powerful tool for real-time influenza surveillance.
Keywords
Deep learning
Influenza-like illness
INGARCH
INGARCHX
LSTM
CO,NO2, O3, PM2.5, PM10, SO2
We propose a modified population-based susceptible-exposed-infectious-recovered (SEIR) compartmental model for a retrospective study of the COVID-19 transmission dynamics. We extend the conventional SEIR methodology to account for the complexities of COVID-19 infection, its multiple symptoms, and transmission pathways. In particular, we consider a time-dependent transmission rate to account for governmental controls (e.g., national lockdown) and individual behavioral factors (e.g., social distancing, mask-wearing, personal hygiene, and self-quarantine). An essential feature of COVID-19 that is different from other infections is the significant contribution of asymptomatic and pre-symptomatic cases to the transmission cycle. A Bayesian method is used to calibrate the proposed SEIR model using publicly available data (daily new tested positive, death, and recovery cases) from several states. The uncertainty of the parameters is naturally expressed as the posterior probability distribution. The calibrated model is used to estimate undetected cases and study the effect of different initial intervention policies, screening rates, and public behavior factors.
Keywords
Bayesian inference
Infectious disease modeling
Markov chain Monte Carlo
compartmental SEIR model
Seasonal influenza forecasting is critical for public health and individual decision making. We investigate whether the inclusion of data about influenza activity in neighboring states can improve point predictions and distribution forecasting of influenza-like illness (ILI) in each US state using statistical regression models. Using CDC FluView ILI data from 2010-2019, we forecast weekly ILI in each US state with quantile, linear, and Poisson autoregressive models fit using different combinations of ILI data from the target state, neighboring states, and the US population-weighted average. Scoring with root mean squared error and weighted interval score indicated that the covariate sets including neighbors and/or the US weighted average ILI showed slightly higher accuracy than models fit only using lagged ILI in the target state, on average. Additionally, the improvement in performance when including neighbors was similar to the improvement when including the US average instead, suggesting the proximity of the neighboring states is not the driver of the slight increase in accuracy. There is also clear within- and between-season variability in the effect of spatial information.
Keywords
Forecasting
Spatial
Influenza-like Illness
Autoregressive
Infectious Disease
The spatiotemporal dynamics of SARS-CoV-2 transmission are influenced by factors such as human mobility, cumulative incidence, and vaccination coverage. Understanding how these factors shape transmission is essential for designing effective public health strategies. Traditional phylogenetic methods for inferring viral spread are often computationally intensive and unreliable when genetic divergence is low. This research introduces a tree-free modelling framework that leverages Hamming distance-a direct measure of genetic similarity between viral sequences-to estimate the impact of epidemiological factors on transmission and susceptibility. We introduce a likelihood-based framework and conduct simulation studies to demonstrate that the model reliably estimates parameters associated with these factors while addressing key challenges of phylogenetic approaches. Applying this method to SARS-CoV-2 genomic data from Washington State during the Delta wave of 2021, we find that higher cumulative incidence and vaccination rates substantially reduce population susceptibility, though with diminishing returns at higher levels. This work demonstrates the effectiveness of Hamming distance.
Keywords
Infectious disease
COVID-19
Statistical Modelling
Hamming distance
Spatial transmission
Infectious diseases in animal farms pose a significant threat, often resulting in mass livestock mortality and substantial economic losses. The spread of diseases is driven by the interplay of multiple factors, including the recurrence of the virus within a farm, the transportation of livestock between farms, and environmental conditions. We regard the occurrence of infection events in each farm as time-to-event data on network nodes. Existing methods often assume that the interactions between nodes are static. However, the transportation between farms changes dynamically. Therefore, we propose a new version of the Hawkes Process that accounts for evolving transportation networks. Our method also allows the incorporation of time-varying environmental covariates. We further develop an Expectation-Maximization algorithm leveraging the branching structure of the model to conduct statistical inference. The algorithm also enables us to distinguish whether the infection event is externally driven or internally driven. The proposed model is validated through extensive simulations and real-world epidemic data from China.
Keywords
Dynamic Network
Hawkes Process
EM algorithm
Disease Spread
Co-Author(s)
Xuening Zhu, Fudan University
Fangda Song, The Chinese University of Hong Kong, Shenzhen
First Author
Wenhao Chen, The Chinese University of Hong Kong, Shenzhen
Presenting Author
Wenhao Chen, The Chinese University of Hong Kong, Shenzhen
The surge in extreme precipitation is forecast to lead to a rise in cryptosporidiosis, a waterborne acute gastrointestinal infection. This study examines the relationship between precipitation and cryptosporidiosis in Tennessee from 2012 to 2021 using time series analysis and Distributed Lag Nonlinear Models (DLNM). The findings reveal significant seasonal trends in cryptosporidiosis cases, with periodic spikes right after rains. DLNM analyses highlight a delayed effect between precipitation and cryptosporidiosis cases, where moderate precipitation (1-2 inches) increases cryptosporidiosis risk over a 2-week period and extreme rainfall (>5 inches) reduces risk in 4-6 weeks. Furthermore, Contour and 3D surface plots illustrate high-risk zones at low-to-moderate precipitation levels with short lags and high precipitation levels correspond to a lower risk at intermediate lags, followed by a renewed increase after 6 weeks. These findings emphasize the importance of incorporating temporal dynamics in assessing weather-related public health concerns and suggest that long-term public health interventions following rainfall events could help mitigate cryptosporidiosis outbreaks.
Keywords
Distributed Lag Models
Time Series
Precipitation
Cryptosporidiosis
Predictive Analysis
R