Forecasting national and regional influenza-like illness for the USA

Competition between transmission lineages mediated by human mobility shapes seasonal influenza epidemics in the US

Due to its climatic variability, complex mobility networks and geographic expanse, the United States represents a compelling setting to explore the transmission processes that lead to heterogeneous yearly seasonal influenza epidemics. By analyzing genomic and epidemiological data collected in the US from 2014 to 2023, we show that epidemics consisted of multiple co-circulating transmission lineages that could emerge from all regions and often rapidly expanded. Lineage spread was characterized by strong spatiotemporal hierarchies and lineage size correlated with timing of establishment in the US. Mechanistic epidemic simulations, supported by phylogeographic analyses, suggest that competition between lineages on a network of human mobility consistent with commuting flows drove lineage dynamics. Our results suggest that the processes that disseminate viruses nationwide are highly structured, but variability in the short-term processes that determine the locations, timing, and explosiveness of initial epidemic sparks limits predictability of regional and national epidemics.

A prospective real-time transfer learning approach to estimate influenza hospitalizations with limited data

Accurate, real-time forecasts of influenza hospitalizations would facilitate prospective resource allocation and public health preparedness. State-of-the-art machine learning methods are a promising approach to produce such forecasts, but they require extensive historical data to be properly trained. Unfortunately, data on influenza hospitalizations, for the 50 states in the United States, are only available since the beginning of 2020. In addition, the data are far from perfect as they were under-reported for several months before health systems began consistently submitting their data. To address these issues, we propose a transfer learning approach. We extend the currently available two-season dataset for state-level influenza hospitalizations by an additional ten seasons. Our method leverages influenza-like illness (ILI) data to infer historical estimates of influenza hospitalizations. This data augmentation enables the implementation of advanced machine learning techniques, multi-horizon training, and an ensemble of models to improve hospitalization forecasts. We evaluated the performance of our machine learning approaches by prospectively producing forecasts for future weeks and submitting them in real time to the Centers for Disease Control and Prevention FluSight challenges during two seasons: 2022-2023 and 2023-2024. Our methodology demonstrated good accuracy and reliability, achieving a fourth place finish (among 20 participating teams) in the 2022-23 and a second place finish (among 20 participating teams) in the 2023-24 CDC FluSight challenges. Our findings highlight the utility of data augmentation and knowledge transfer in the application of machine learning models to public health surveillance where only limited historical data is available.

Integrating information from historical data into mechanistic models for influenza forecasting

Seasonal influenza causes significant annual morbidity and mortality worldwide. In France, it is estimated that, on average, 2 million individuals consult their GP for influenza-like-illness (ILI) every year. Traditionally, mathematical models used for epidemic forecasting can either include parameters capturing the infection process (mechanistic or compartmental models) or rely on time series analysis approaches that do not make mechanistic assumptions (statistical or phenomenological models). While the latter make extensive use of past epidemic data, mechanistic models are usually independently initialized in each season. As a result, forecasts from such models can contain trajectories that are vastly different from past epidemics. We developed a mechanistic model that takes into account epidemic data from training seasons when producing forecasts. The parameters of the model are estimated via a first particle filter running on the observed data. A second particle filter is then used to produce forecasts compatible with epidemic trajectories from the training set. The model was calibrated and tested on 35 years' worth of surveillance data from the French Sentinelles Network, representing the weekly number of patients consulting for ILI over the period 1985-2019. Our results show that the new method improves upon standard mechanistic approaches. In particular, when retrospectively tested on the available data, our model provides increased accuracy for short-term forecasts (from one to four weeks into the future) and peak timing and intensity. Our new approach for epidemic forecasting allows the integration of key strengths of the statistical approach into the mechanistic modelling framework and represents an attempt to provide accurate forecasts by making full use of the rich surveillance dataset collected in France since 1985.

Commuting-driven competition between transmission chains shapes seasonal influenza virus epidemics in the United States

Despite intensive study, much remains unknown about the dynamics of seasonal influenza virus epidemic establishment and spread in the United States (US) each season. By reconstructing transmission lineages from seasonal influenza virus genomes collected in the US from 2014 to 2023, we show that most epidemics consisted of multiple distinct transmission lineages. Spread of these lineages exhibited strong spatiotemporal hierarchies and lineage size was correlated with timing of lineage establishment in the US. Mechanistic epidemic simulations suggest that mobility-driven competition between lineages determined the extent of individual lineages' geographical spread. Based on phylogeographic analyses and epidemic simulations, lineage-specific movement patterns were dominated by human commuting behavior. These results suggest that given the locations of early-season epidemic sparks, the topology of inter-state human mobility yields repeatable patterns of which influenza viruses will circulate where, but the importance of short-term processes limits predictability of regional and national epidemics.

Cross-sectional Ct distributions from qPCR tests can provide an early warning signal for the spread of COVID-19 in communities

Background

SARS-CoV-2 PCR testing data has been widely used for COVID-19 surveillance. Existing COVID-19 forecasting models mainly rely on case counts obtained from qPCR results, even though the binary PCR results provide a limited picture of the pandemic trajectory. Most forecasting models have failed to accurately predict the COVID-19 waves before they occur. Recently a model utilizing cross-sectional population cycle threshold (Ct-the number of cycles required for the fluorescent signal to cross the background threshold) values obtained from PCR tests (Ct-based model) was developed to overcome the limitations of using only binary PCR results. In this study, we aimed to improve on COVID-19 forecasting models using features derived from the Ct-based model, to detect epidemic waves earlier than case-based trajectories.

Methods

PCR data was collected weekly at Northeastern University (NU) between August 2020 and January 2022. Campus and county epidemic trajectories were generated from case counts. A novel forecasting approach was developed by enhancing a recent deep learning model with Ct-based features and applied in Suffolk County and NU campus. For this, cross-sectional Ct values from PCR data were used to generate Ct-based epidemic trajectories, including effective reproductive rate (Rt) and incidence. The improvement in forecasting performance was compared using absolute errors and residual squared errors with respect to actual observed cases at the 7-day and 14-day forecasting horizons. The model was also tested prospectively over the period January 2022 to April 2022.

Results

Rt curves estimated from the Ct-based model indicated epidemic waves 12 to 14 days earlier than Rt curves from NU campus and Suffolk County cases, with a correlation of 0.57. Enhancing the forecasting models with Ct-based information significantly decreased absolute error (decrease of 49.4 and 221.5 for the 7 and 14-day forecasting horizons) and residual squared error (40.6 and 217.1 for the 7 and 14-day forecasting horizons) compared to the original model without Ct features.

Conclusion

Ct-based epidemic trajectories can herald an earlier signal for impending epidemic waves in the community and forecast transmission peaks. Moreover, COVID-19 forecasting models can be enhanced using these Ct features to improve their forecasting accuracy. In this study, we make the case that public health agencies should publish Ct values along with the binary positive/negative PCR results. Early and accurate forecasting of epidemic waves can inform public health policies and countermeasures which can mitigate spread.

Consistent pattern of epidemic slowing across many geographies led to longer, flatter initial waves of the COVID-19 pandemic

To define appropriate planning scenarios for future pandemics of respiratory pathogens, it is important to understand the initial transmission dynamics of COVID-19 during 2020. Here, we fit an age-stratified compartmental model with a flexible underlying transmission term to daily COVID-19 death data from states in the contiguous U.S. and to national and sub-national data from around the world. The daily death data of the first months of the COVID-19 pandemic was qualitatively categorized into one of four main profile types: "spring single-peak", "summer single-peak", "spring/summer two-peak" and "broad with shoulder". We estimated a reproduction number R as a function of calendar time tc and as a function of time since the first death reported in that population (local pandemic time, tp). Contrary to the diversity of categories and range of magnitudes in death incidence profiles, the R(tp) profiles were much more homogeneous. We found that in both the contiguous U.S. and globally, the initial value of both R(tc) and R(tp) was substantial: at or above two. However, during the early months, pandemic time R(tp) decreased exponentially to a value that hovered around one. This decrease was accompanied by a reduction in the variance of R(tp). For calendar time R(tc), the decrease in magnitude was slower and non-exponential, with a smaller reduction in variance. Intriguingly, similar trends of exponential decrease and reduced variance were not observed in raw death data. Our findings suggest that the combination of specific government responses and spontaneous changes in behaviour ensured that transmissibility dropped, rather than remaining constant, during the initial phases of a pandemic. Future pandemic planning scenarios should include models that assume similar decreases in transmissibility, which lead to longer epidemics with lower peaks when compared with models based on constant transmissibility.

COVID-19 deaths: Which explanatory variables matter the most?

More than a year since the appearance of Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2), many questions about the disease COVID-19 have been answered; however, many more remain poorly understood. Although the situation continues to evolve, it is crucial to understand what factors may be driving transmission through different populations, both for potential future waves, as well as the implications for future pandemics. In this report, we compiled a database of more than 28 potentially explanatory variables for each of the 50 U.S. states through early May 2020. Using a combination of traditional statistical and modern machine learning approaches, we identified those variables that were the most statistically significant, and, those that were the most important. These variables were chosen to be fiduciaries of a range of possible drivers for COVID-19 deaths in the USA. We found that population-weighted population density (PWPD), some "stay at home" metrics, monthly temperature and precipitation, race/ethnicity, and chronic low-respiratory death rate, were all statistically significant. Of these, PWPD and mobility metrics dominated. This suggests that the biggest impact on COVID-19 deaths was, at least initially, a function of where you lived, and not what you did. However, clearly, increasing social distancing has the net effect of (at least temporarily) reducing the effective PWPD. Our results strongly support the idea that the loosening of "lock-down" orders should be tailored to the local PWPD. In contrast to these variables, while still statistically significant, race/ethnicity, health, and climate effects could only account for a few percent of the variability in deaths. Where associations were anticipated but were not found, we discuss how limitations in the parameters chosen may mask a contribution that might otherwise be present.

Applying the Spatial Transmission Network to the Forecast of Infectious Diseases Across Multiple Regions

Objective

Timely and accurate forecast of infectious diseases is essential for achieving precise prevention and control. A good forecasting method of infectious diseases should have the advantages of interpretability, feasibility, and forecasting performance. Since previous research had illustrated that the spatial transmission network (STN) showed good interpretability and feasibility, this study further explored its forecasting performance for infectious diseases across multiple regions. Meanwhile, this study also showed whether the STN could overcome the challenges of model rationality and practical needs.

Methods

The construction of the STN framework involved three major steps: the spatial kluster analysis by tree edge removal (SKATER) algorithm, structure learning by dynamic Bayesian network (DBN), and parameter learning by the vector autoregressive moving average (VARMA) model. Then, we evaluated the forecasting performance of STN by comparing its accuracy with that of the mechanism models like susceptible-exposed-infectious-recovered-susceptible (SEIRS) and machine-learning algorithm like long-short-term memory (LSTM). At the same time, we assessed the robustness of forecasting performance of STN in high and low incidence seasons. The influenza-like illness (ILI) data in the Sichuan Province of China from 2010 to 2017 were used as an example for illustration.

Results

The STN model revealed that ILI was likely to spread among multiple cities in Sichuan during the study period. During the whole study period, the forecasting accuracy of the STN (mean absolute percentage error [MAPE] = 31.134) was significantly better than that of the LSTM (MAPE = 41.657) and the SEIRS (MAPE = 62.039). In addition, the forecasting performance of STN was also superior to those of the other two methods in either the high incidence season (MAPE = 24.742) or the low incidence season (MAPE = 26.209), and the superiority was more obvious in the high incidence season.

Conclusion

This study applied the STN to the forecast of infectious diseases across multiple regions. The results illustrated that the STN not only had good accuracy in forecasting performance but also indicated the spreading directions of infectious diseases among multiple regions to a certain extent. Therefore, the STN is a promising candidate to improve the surveillance work.

Fast and accurate influenza forecasting in the United States with Inferno

Infectious disease forecasting is an emerging field and has the potential to improve public health through anticipatory resource allocation, situational awareness, and mitigation planning. By way of exploring and operationalizing disease forecasting, the U.S. Centers for Disease Control and Prevention (CDC) has hosted FluSight since the 2013/14 flu season, an annual flu forecasting challenge. Since FluSight’s onset, forecasters have developed and improved forecasting models in an effort to provide more timely, reliable, and accurate information about the likely progression of the outbreak. While improving the predictive performance of these forecasting models is often the primary objective, it is also important for a forecasting model to run quickly, facilitating further model development and improvement while providing flexibility when deployed in a real-time setting. In this vein I introduce Inferno, a fast and accurate flu forecasting model inspired by Dante, the top performing model in the 2018/19 FluSight challenge. When pseudoprospectively compared to all models that participated in FluSight 2018/19, Inferno would have placed 2nd in the national and regional challenge as well as the state challenge, behind only Dante. Inferno, however, runs in minutes and is trivially parallelizable, while Dante takes hours to run, representing a significant operational improvement with minimal impact to performance. Forecasting challenges like FluSight should continue to monitor and evaluate how they can be modified and expanded to incentivize the development of forecasting models that benefit public health.

Infectious disease forecasting, if accurate, timely, and reliable, can assist decision makers with resource allocation planning in an attempt to curb the negative impacts of an outbreak. Forecasting challenges, like the U.S. Centers for Disease Control and Prevention’s flu forecasting challenge, FluSight, provide a space for teams to develop and operationalize real-time forecasting models that benefit public health, with weekly forecasts made at the state-level, Health and Human Services region-level, and the United States. The ultimate goal of these models is to produce accurate forecasts within the constraints of the forecasting challenge. Having a forecasting model that runs quickly is also important for future scalability, model development, and operational flexibility. In this paper, I present a fast and accurate flu forecasting model, Inferno. Through retrospective comparisons with FluSight-participating models, Inferno was shown to be a leading forecasting model in the field. Inferno, however, runs in minutes not hours, as other leading forecasting models do. This reduction in runtime constitutes an advancement in flu forecasting, positioning Inferno to scale to more granular geographic units, like counties or health care providers.

The impact of long-term non-pharmaceutical interventions on COVID-19 epidemic dynamics and control: the value and limitations of early models

Mathematical models of epidemics are important tools for predicting epidemic dynamics and evaluating interventions. Yet, because early models are built on limited information, it is unclear how long they will accurately capture epidemic dynamics. Using a stochastic SEIR model of COVID-19 fitted to reported deaths, we estimated transmission parameters at different time points during the first wave of the epidemic (March-June, 2020) in Santa Clara County, California. Although our estimated basic reproduction number ([Formula: see text]) remained stable from early April to late June (with an overall median of 3.76), our estimated effective reproduction number ([Formula: see text]) varied from 0.18 to 1.02 in April before stabilizing at 0.64 on 27 May. Between 22 April and 27 May, our model accurately predicted dynamics through June; however, the model did not predict rising summer cases after shelter-in-place orders were relaxed in June, which, in early July, was reflected in cases but not yet in deaths. While models are critical for informing intervention policy early in an epidemic, their performance will be limited as epidemic dynamics evolve. This paper is one of the first to evaluate the accuracy of an early epidemiological compartment model over time to understand the value and limitations of models during unfolding epidemics.

Accurate influenza forecasts using type-specific incidence data for small geographic units

Influenza incidence forecasting is used to facilitate better health system planning and could potentially be used to allow at-risk individuals to modify their behavior during a severe seasonal influenza epidemic or a novel respiratory pandemic. For example, the US Centers for Disease Control and Prevention (CDC) runs an annual competition to forecast influenza-like illness (ILI) at the regional and national levels in the US, based on a standard discretized incidence scale. Here, we use a suite of forecasting models to analyze type-specific incidence at the smaller spatial scale of clusters of nearby counties. We used data from point-of-care (POC) diagnostic machines over three seasons, in 10 clusters, capturing: 57 counties; 1,061,891 total specimens; and 173,909 specimens positive for Influenza A. Total specimens were closely correlated with comparable CDC ILI data. Mechanistic models were substantially more accurate when forecasting influenza A positive POC data than total specimen POC data, especially at longer lead times. Also, models that fit subpopulations of the cluster (individual counties) separately were better able to forecast clusters than were models that directly fit to aggregated cluster data. Public health authorities may wish to consider developing forecasting pipelines for type-specific POC data in addition to ILI data. Simple mechanistic models will likely improve forecast accuracy when applied at small spatial scales to pathogen-specific data before being scaled to larger geographical units and broader syndromic data. Highly local forecasts may enable new public health messaging to encourage at-risk individuals to temporarily reduce their social mixing during seasonal peaks and guide public health intervention policy during potentially severe novel influenza pandemics.

Multiscale influenza forecasting

Influenza forecasting in the United States (US) is complex and challenging due to spatial and temporal variability, nested geographic scales of interest, and heterogeneous surveillance participation. Here we present Dante, a multiscale influenza forecasting model that learns rather than prescribes spatial, temporal, and surveillance data structure and generates coherent forecasts across state, regional, and national scales. We retrospectively compare Dante's short-term and seasonal forecasts for previous flu seasons to the Dynamic Bayesian Model (DBM), a leading competitor. Dante outperformed DBM for nearly all spatial units, flu seasons, geographic scales, and forecasting targets. Dante's sharper and more accurate forecasts also suggest greater public health utility. Dante placed 1st in the Centers for Disease Control and Prevention's prospective 2018/19 FluSight challenge in both the national and regional competition and the state competition. The methodology underpinning Dante can be used in other seasonal disease forecasting contexts having nested geographic scales of interest.

Optimizing respiratory virus surveillance networks using uncertainty propagation

Infectious disease prevention, control and forecasting rely on sentinel observations; however, many locations lack the capacity for routine surveillance. Here we show that, by using data from multiple sites collectively, accurate estimation and forecasting of respiratory diseases for locations without surveillance is feasible. We develop a framework to optimize surveillance sites that suppresses uncertainty propagation in a networked disease transmission model. Using influenza outbreaks from 35 US states, the optimized system generates better near-term predictions than alternate systems designed using population and human mobility. We also find that monitoring regional population centers serves as a reasonable proxy for the optimized network and could direct surveillance for diseases with limited records. The  proxy method is validated using model simulations for 3,108 US counties and historical data for two other respiratory pathogens - human metapneumovirus and seasonal coronavirus - from 35 US states and can be used to guide systemic allocation of surveillance efforts.

Aggregating forecasts of multiple respiratory pathogens supports more accurate forecasting of influenza-like illness

Influenza-like illness (ILI) is a commonly measured syndromic signal representative of a range of acute respiratory infections. Reliable forecasts of ILI can support better preparation for patient surges in healthcare systems. Although ILI is an amalgamation of multiple pathogens with variable seasonal phasing and attack rates, most existing process-based forecasting systems treat ILI as a single infectious agent. Here, using ILI records and virologic surveillance data, we show that ILI signal can be disaggregated into distinct viral components. We generate separate predictions for six contributing pathogens (influenza A/H1, A/H3, B, respiratory syncytial virus, and human parainfluenza virus types 1-2 and 3), and develop a method to forecast ILI by aggregating these predictions. The relative contribution of each pathogen to the total ILI signal is estimated using a Markov Chain Monte Carlo (MCMC) method upon forecast aggregation. We find highly variable overall contributions from influenza type A viruses across seasons, but relatively stable contributions for the other pathogens. Using historical data from 1997 to 2014 at US national and regional levels, the proposed forecasting system generates improved predictions of both seasonal and near-term targets relative to a baseline method that simulates ILI as a single pathogen. The hierarchical forecasting system can generate predictions for each viral component, as well as infer and predict their contributions to ILI, which may additionally help physicians determine the etiological causes of ILI in clinical settings.

Viral Long-Term Evolutionary Strategies Favor Stability over Proliferation

Viruses are known to have some of the highest and most diverse mutation rates found in any biological replicator, with single-stranded (ss) RNA viruses evolving the fastest, and double-stranded (ds) DNA viruses having rates approaching those of bacteria. As mutation rates are tightly and negatively correlated with genome size, selection is a clear driver of viral evolution. However, the role of intragenomic interactions as drivers of viral evolution is still unclear. To understand how these two processes affect the long-term evolution of viruses infecting humans, we comprehensively analyzed ssRNA, ssDNA, dsRNA, and dsDNA viruses, to find which virus types and which functions show evidence for episodic diversifying selection and correlated evolution. We show that selection mostly affects single stranded viruses, that correlated evolution is more prevalent in DNA viruses, and that both processes, taken independently, mostly affect viral replication. However, the genes that are jointly affected by both processes are involved in key aspects of their life cycle, favoring viral stability over proliferation. We further show that both evolutionary processes are intimately linked at the amino acid level, which suggests that it is the joint action of selection and correlated evolution, and not just selection, that shapes the evolutionary trajectories of viruses—and possibly of their epidemiological potential.