Retrospective forecasting of the 2010–2014 Melbourne influenza seasons using multiple surveillance systems

Abstract

SUMMARYAccurate forecasting of seasonal influenza epidemics is of great concern to healthcare providers in temperate climates, since these epidemics vary substantially in their size, timing and duration from year to year, making it a challenge to deliver timely and proportionate responses. Previous studies have shown that Bayesian estimation techniques can accurately predict when an influenza epidemic will peak many weeks in advance, and we have previously tailored these methods for metropolitan Melbourne (Australia) and Google Flu Trends data. Here we extend these methods to clinical observation and laboratory-confirmation data for Melbourne, on the grounds that these data sources provide more accurate characterizations of influenza activity. We show that from each of these data sources we can accurately predict the timing of the epidemic peak 4-6 weeks in advance. We also show that making simultaneous use of multiple surveillance systems to improve forecast skill remains a fundamental challenge. Disparate systems provide complementary characterizations of disease activity, which may or may not be comparable, and it is unclear how a 'ground truth' for evaluating forecasts against these multiple characterizations might be defined. These findings are a significant step towards making optimal use of routine surveillance data for outbreak forecasting.

Epidemic forecasts as a tool for public health: interpretation and (re)calibration

“…In comparison, the highest peak observed in previous seasonal influenza epidemics was only 560 notified cases. By this point it had become clear that our forecast settings – based on retrospective forecasts of the 2010–14 influenza seasons using the same data source 6 – were not consistent with the 2015 data, and we immediately explored adjustments to these settings (see next section).…”

“…Values for p id and k were informed by retrospective forecasts using notifications data from previous seasons, 6 while the background notification rate p bg was estimated from out‐of‐season notification levels in March and April 2015. We previously used this same method to estimate background notification rates for the 2010–14 seasons, which ranged from 15 to 46 cases per week, 6 but the estimated value was higher in 2015 (≈60 cases per week).…”

“…public health messaging, control measures, investigation efforts, surge capacity planning, staff rosters). In recent years, a number of studies have evaluated epidemic forecasting methods with a particular focus on seasonal influenza epidemics, 1–5 and we have previously demonstrated that seasonal influenza forecasts for the city of Melbourne (Australia) can be obtained using any one of several surveillance systems 6 . These methods could provide valuable support for public health preparedness and response, but a number of challenges remain to be addressed, including: “assessment of model calibration; integration of subjective expert input; operational research in pilot, real‐world applications; and improved mutual understanding among modelers and public health officials” 1 …”

Evaluation of Nowcasting for Detecting and Predicting Local Influenza Epidemics, Sweden, 2009–2014

“…Hence, we consider these categories to be reliable. A longer prospective evaluation period would have increased the possibility of drawing valid conclusions concerning the outcome of the evaluation, and it would be preferable to have corresponding local data from other cities or regions ( 22 ). Evaluating our nowcasting method for epidemics from several other regions would enable conclusions to be drawn about the generalizability of the method.…”

Epidemiological and statistical basis for detection and prediction of influenza epidemics

“…In upcoming versions of nowcasting methods, both regular updates and more stable sources than telenursing data (with regard to the time lag to influenza-diagnosis data) may become accessible and could be used to improve the peak timing predictions and thus also the peak intensity predictions. Finally, a longer prospective evaluation period would have increased the possibility of drawing valid conclusions regarding the outcome of the evaluation, and it would be preferable to have corresponding local data from other cities or regions (Moss et al 2017). Evaluations of the nowcasting method on epidemics from several other regions would allow conclusions to be drawn about the generalizability of the method.…”