Recent studies have used Bayesian methods to predict timing of influenza epidemics many weeks in advance, but there is no documented evaluation of how such forecasts might support the day-to-day operations of public health staff.Methods:
During the 2015 influenza season in Melbourne, Australia, weekly forecasts were presented at Health Department surveillance unit meetings, where they were evaluated and updated in light of expert opinion to improve their accuracy and usefulness.Results:
Predictive capacity of the model was substantially limited by delays in reporting and processing arising from an unprecedented number of notifications, disproportionate to seasonal intensity. Adjustment of the predictive algorithm to account for these delays and increased reporting propensity improved both current situational awareness and forecasting accuracy.Conclusions:
Collaborative engagement with public health practitioners in model development improved understanding of the context and limitations of emerging surveillance data. Incorporation of these insights in a quantitative model resulted in more robust estimates of disease activity for public health use.Implications for public health:
In addition to predicting future disease trends, forecasting methods can quantify the impact of delays in data availability and variable reporting practice on the accuracy of current epidemic assessment. Such evidence supports investment in systems capacity.