Development and implementation of a nowcasting method for the syndromic surveillance of severe acute respiratory infections (SARI) in Germany

BackgroundIn Germany, diagnosis-based hospital surveillance of severe acute respiratory infections (SARI) incidence is operated weekly. However, diagnosis data from the most recent weeks is provisional and may be subject to systematic changes that we wish to account for to achieve a timely assessment of SARI trends. MethodsWe developed a novel nowcasting method based on (i) the already available data for the current week and (ii) historic changes in the data in weeks following initial reporting. The available weekly data were rescaled using multiplicative factors based on the historic patterns, to obtain the expected final incidence. We fit a parametric Beta distribution for observed upward changes in reported incidence, and a mixture distribution in case of the existence of downward changes, e.g., due to later corrections in diagnoses. We evaluated the model performance for the respiratory winter season 2024/2025. ResultsThe average weekly SARI incidence upon initial estimation in the first week after hospitalisation was 79.4% (median) of the incidence five weeks after hospitalisation, which was considered final. It increased to 95.1% in the second week after hospitalisation. The nowcast predictions matched the final incidences well and were able to indicate trends in real-time. Prediction intervals were well-calibrated and nowcasting can be performed in subgroups given sufficient case numbers. ConclusionsThe SARI surveillance in Germany achieved a high completeness already in the initial weeks after hospitalisation. The newly-developed nowcasting method yielded robust results with reliable uncertainty quantification. In combination, this facilitates a close to real-time assessment of SARI hospitalisation incidence in Germany.