nowcasting {EpiLPS}

Nowcasting and estimation of occurred but not yet reported events

Description

This routine can be used to estimate cases that have not yet been reported also known as occurred-but-not-yet-reported-events. Daily cases are typically subject to reporting delays, so that the reported number of infected individuals is not always reflecting the true epidemic status. Nowcasting aims to correct this underreporting phenomenon by estimating the number of infections that have occurred but that have not yet been reported. The latter number is then combined with the already reported cases and interpreted as a nowcast or prediction for the true epidemic status regarding the number of daily cases. The routine is anchored around Laplacian-P-splines in an epidemic context (Gressani et al. 2022) and the detailed methodology can be found in Sumalinab et al. (2023).

Usage

nowcasting(data, day.effect = TRUE, ref.day = "Monday", verbose = TRUE)

Arguments

data

A data frame containing the data for each time and delay combination with the following 6 columns. The first column is a numeric variable associated to the calendar date. The second column is a numeric variable indicating the delay of reporting. The third column corresponds to the calendar date of the event (e.g. death) and the fourth column to the calendar date at which the event of interest was reported. The fifth column indicates the number of cases for each time and delay combination. Finally, the sixth column indicates whether the cases are already reported or not yet reported. To see an example of such a data structure type data("cov19mort2021") and then head(cov19mort2021). This will illustrate the data structure used for nowcasting the daily number of cases based on mortality data for Belgium in 2021.

day.effect

If TRUE (defaut), include the day of the week effect.

ref.day

If day.effect = TRUE, then the reference category for the day of the week must be specified. The default is "Monday".

verbose

Show summary output of nowcast in console? Default is TRUE.

Value

A list with the following components:

• data: The data frame used as an input.

• cases.now: A data frame containing the nowcasting results with 95\% prediction intervals.

• delay: A data frame containing the two-dimensional delay distribution.

• lambda_estim: Estimated penalty parameters of the P-splines model.

• phi_estim: Estimated overdispersion parameter from the negative binomial model.

Author(s)

Bryan Sumalinab (writing) and Oswaldo Gressani (editing).

References

Gressani, O., Wallinga, J., Althaus, C. L., Hens, N. and Faes, C. (2022). EpiLPS: A fast and flexible Bayesian tool for estimation of the time-varying reproduction number. Plos Computational Biology, 18(10): e1010618.

Sumalinab, B., Gressani, O., Hens, N. and Faes, C. (2023). Bayesian nowcasting with Laplacian-P-splines. MedRxiv preprint. https://www.medrxiv.org/content/10.1101/2022.08.26.22279249v2

Examples

# data("cov19mort2021")
# ncast <- nowcasting(data = cov19mort2021, day.effect = FALSE)
# plot(ncast) # Show nowcasted cases
# plot(ncast, type = "delay") # Show contour of delay distribution