Forecasting national and regional influenza-like illness for the USA.

Health planners use forecasts of key metrics associated with influenza-like illness (ILI); near-term weekly incidence, week of season onset, week of peak, and intensity of peak. Here, we describe our participation in a weekly prospective ILI forecasting challenge for the United States for the 2016-17 season and subsequent evaluation of our performance. We implemented a metapopulation model framework with 32 model variants. Variants differed from each other in their assumptions about: the force-of-infection (FOI); use of uninformative priors; the use of discounted historical data for not-yet-observed time points; and the treatment of regions as either independent or coupled. Individual model variants were chosen subjectively as the basis for our weekly forecasts; however, a subset of coupled models were only available part way through the season. Most frequently, during the 2016-17 season, we chose; FOI variants with both school vacations and humidity terms; uninformative priors; the inclusion of discounted historical data for not-yet-observed time points; and coupled regions (when available). Our near-term weekly forecasts substantially over-estimated incidence early in the season when coupled models were not available. However, our forecast accuracy improved in absolute terms and relative to other teams once coupled solutions were available. In retrospective analysis, we found that the 2016-17 season was not typical: on average, coupled models performed better when fit without historically augmented data. Also, we tested a simple ensemble model for the 2016-17 season and found that it underperformed our subjective choice for all forecast targets. In this study, we were able to improve accuracy during a prospective forecasting exercise by coupling dynamics between regions. Although reduction of forecast subjectivity should be a long-term goal, some degree of human intervention is likely to improve forecast accuracy in the medium-term in parallel with the systematic consideration of more sophisticated ensemble approaches.

Early Characterization of the Severity and Transmissibility of Pandemic Influenza Using Clinical Episode Data from Multiple Populations.

The potential rapid availability of large-scale clinical episode data during the next influenza pandemic suggests an opportunity for increasing the speed with which novel respiratory pathogens can be characterized. Key intervention decisions will be determined by both the transmissibility of the novel strain (measured by the basic reproductive number R0) and its individual-level severity. The 2009 pandemic illustrated that estimating individual-level severity, as described by the proportion pC of infections that result in clinical cases, can remain uncertain for a prolonged period of time. Here, we use 50 distinct US military populations during 2009 as a retrospective cohort to test the hypothesis that real-time encounter data combined with disease dynamic models can be used to bridge this uncertainty gap. Effectively, we estimated the total number of infections in multiple early-affected communities using the model and divided that number by the known number of clinical cases. Joint estimates of severity and transmissibility clustered within a relatively small region of parameter space, with 40 of the 50 populations bounded by: pC, 0.0133-0.150 and R0, 1.09-2.16. These fits were obtained despite widely varying incidence profiles: some with spring waves, some with fall waves and some with both. To illustrate the benefit of specific pairing of rapidly available data and infectious disease models, we simulated a future moderate pandemic strain with pC approximately ×10 that of 2009; the results demonstrating that even before the peak had passed in the first affected population, R0 and pC could be well estimated. This study provides a clear reference in this two-dimensional space against which future novel respiratory pathogens can be rapidly assessed and compared with previous pandemics.

Multiple estimates of transmissibility for the 2009 influenza pandemic based on influenza-like-illness data from small US military populations.

Rapidly characterizing the amplitude and variability in transmissibility of novel human influenza strains as they emerge is a key public health priority. However, comparison of early estimates of the basic reproduction number during the 2009 pandemic were challenging because of inconsistent data sources and methods. Here, we define and analyze influenza-like-illness (ILI) case data from 2009-2010 for the 50 largest spatially distinct US military installations (military population defined by zip code, MPZ). We used publicly available data from non-military sources to show that patterns of ILI incidence in many of these MPZs closely followed the pattern of their enclosing civilian population. After characterizing the broad patterns of incidence (e.g. single-peak, double-peak), we defined a parsimonious SIR-like model with two possible values for intrinsic transmissibility across three epochs. We fitted the parameters of this model to data from all 50 MPZs, finding them to be reasonably well clustered with a median (mean) value of 1.39 (1.57) and standard deviation of 0.41. An increasing temporal trend in transmissibility ([Formula: see text], p-value: 0.013) during the period of our study was robust to the removal of high transmissibility outliers and to the removal of the smaller 20 MPZs. Our results demonstrate the utility of rapidly available - and consistent - data from multiple populations.