Table 1.
New Reported Cases Over Days.
Fig 1.
Fig A and B indicate outbreak caused by different variants.
Fig 2.
Steps of each method for simulation study 1.
Fig 3.
Panel A and B are adjusted cases for BA.1/2 and XBB outbreaks between different smoothing methods.
Panel C and D are estimated Rt for BA.1/2 and XBB outbreaks between different smoothing methods.
Fig 4.
Comparing different methods in terms of timeliness, the EARS/EpiEstim methods that end with ’NS’ are based on the original data, while the EARS/EpiEstim methods that end with ’S’ are based on data smoothed using the MAH method.
The solid navy-blue horizontal line represents the median timeliness for each method, whereas the dashed navy-blue line indicates the average timeliness.
Fig 5.
Panels A and B examine how the sample size of daily tested individuals (parameter b) and the total number of daily COVID-19 observations (parameter a) affect the performance of the proposed methods with MAH smoothing (PM_MAH), as well as the logistic regression method.
In the logistic regression methods, those ending with ’NS’ are based on the original data, while those ending with ’S’ use data smoothed with MAH. Panels C and D evaluate the effect of total daily COVID-19 observations (parameter a) on the performance of the EARS methods. Similarly, EARS methods ending with ’NS’ use the original data, while those ending with ’S’ apply MAH smoothing. It should be noted that this analysis was not intended to compare timeliness across methods, but rather to evaluate how performance responds to changes in observation scale and sample size.
Fig 6.
Each panel corresponds to a specific model and compares the timeliness of all methods while maintaining the false positive rate at the same level.
For EARS methods, those ending with ’NS’ are based on the original data, while those ending with ’S’ use data smoothed with MAH. Model 1 and 2 rely solely on observed total cases (excluding the logistic regression model for comparison), with the difference being that the DGP in model 1 excluded while the DGP in model 2 included day-of-the-week and public holiday effects. Model 3 is designed to assess performance in stratified analyses (and hence include the logistic regression model), with day-of-the-week and public holiday effects are simulated as well. Further details are provided in the S1 Supplementary Materials.