MicroHub Workshop Weekly Report
Current week’s forecast: new SARI cases
The figure shows the number of severe acute respiratory infections (SARI) cases reported through FluID. The black line represents reported cases. Forecasts include estimates for the previous week (hindcast, in purple), the current week (nowcast, in dark blue), and the 3-week ahead forecasts in the blue line with corresponding prediction intervals.
Ensemble forecasts combine forecasts from different models into one forecast. These ensemble forecasts are generated using an ensemble of three previously validated models: models that focus on historical patterns in the data (smoothing and pattern matching approaches) and a model based on a machine learning method (gradient boosting).
Each week is also summarized as one of five trend categories, shown in the colored strip below the figure, based on the direction and magnitude of change in case counts – large increase, increase, stable, decrease, large decrease. For reported weeks, the trend compares each week with the week immediately before it (a week-to-week change). For forecasted weeks, the trend instead compares each week with the most recent reported week; because that reference point stays fixed, the change is measured over a longer window the further ahead the forecast reaches (a one-week change for the first forecasted week, a two-week change for the next, and so on).The thresholds for these categories were determined using data from 2022, or the earliest year available, through the previous year (see appendix for details).
Prior weeks forecast performance
To evaluate the retrospective performance of prior weeks’ hindcast/nowcast/forecasts, we overlay the truth data (i.e., observed SARI cases) from the current week, depicted with the black dots, with the forecast outputs (i.e., predicted SARI cases). By comparing the observed data to what was expected, we can evaluate the performance of our forecast models. The red dots correspond to the data that were available at the time the forecast models were ran each week.
Weighted Interval Score (WIS) table
The Weighted Interval Score (WIS) table displays the interval scores for each forecast model. Forecasts with lower absolute WIS values are considered more accurate than forecasts with higher absolute WIS values. The relative WIS compares each forecast model to the regular baseline model; the regular baseline model simply predicts that influenza activity in the next five weeks will be a straight line with the band of uncertainty expanding each week. Mean absolute error (MAE) values are also reported which is a measure of forecast accuracy, or how well the forecast models are performing. A lower MAE indicates a more accurate forecast.
Weekly reported influenza cases
This interactive chart shows how the influenza case data evolve and stabilize over time as new data are submitted to FluNET. Because FluNET data can be revised or backfilled after initial reporting, this plot is designed to visualize both the availability of data for forecasting and the extent to which these data stabilize over time. Hovering over any element will display the corresponding details.
The colored dots represent successive data submissions. The color indicates how many weeks after the first reportable week each submission was made - that is, weeks since data for that week could first be complete. The yellow dot marks a submission made in the first reportable week (0 weeks), while darker pink/purple dots indicate later submissions as countries update or backfill their data. The maximum number of submissions since first reportable week is capped at five in this chart, as meaningful data revisions beyond that point are uncommon and reduce visual clarity when graphed. To ensure that submissions with identical values remain individually visible rather than collapsing into a single dot, the orange dots are horizontally jittered by a small random offset.
The black line connects the most recently reported data for each epidemiological week and represents the most up-to-date influenza case counts. When the black dot appears close to earlier yellow/orange dots, the data are relatively stable and required little to no revision.
Appendix
Table A1: categorical forecast thresholds
| Age group | Week difference | Horizon | 95th percentile (large increase) |
75th percentile (increase) |
25th percentile (decrease) |
5th percentile (large decrease) |
|---|---|---|---|---|---|---|
| Overall | 1 | -3 | 120.25 | 30.50 | -46.25 | -122.60 |
| Overall | 2 | -2 | 190.90 | 42.50 | -70.00 | -168.50 |
| Overall | 3 | -1 | 257.00 | 47.50 | -105.00 | -207.50 |
| Overall | 4 | 0 | 306.20 | 50.00 | -130.00 | -249.00 |
| Overall | 5 | 1 | 359.85 | 63.25 | -153.25 | -277.70 |
| Overall | 6 | 2 | 424.70 | 74.50 | -180.00 | -282.80 |
| Overall | 7 | 3 | 486.90 | 87.00 | -202.75 | -324.95 |
| Adults | 1 | -3 | 63.65 | 15.25 | -26.50 | -75.85 |
| Adults | 2 | -2 | 87.60 | 19.00 | -34.50 | -121.00 |
| Adults | 3 | -1 | 126.25 | 20.75 | -44.25 | -166.75 |
| Adults | 4 | 0 | 147.40 | 26.00 | -53.00 | -213.40 |
| Adults | 5 | 1 | 169.40 | 31.25 | -65.00 | -228.80 |
| Adults | 6 | 2 | 185.20 | 28.50 | -69.50 | -253.70 |
| Adults | 7 | 3 | 186.95 | 41.25 | -79.00 | -253.85 |
| Pediatrics | 1 | -3 | 83.20 | 25.25 | -25.25 | -71.00 |
| Pediatrics | 2 | -2 | 131.50 | 29.00 | -36.00 | -101.00 |
| Pediatrics | 3 | -1 | 175.25 | 32.00 | -56.00 | -125.50 |
| Pediatrics | 4 | 0 | 219.40 | 44.00 | -65.00 | -147.00 |
| Pediatrics | 5 | 1 | 252.90 | 55.00 | -88.00 | -169.80 |
| Pediatrics | 6 | 2 | 293.50 | 49.50 | -100.00 | -200.90 |
| Pediatrics | 7 | 3 | 322.55 | 72.25 | -116.00 | -227.70 |