Nowcast right-truncated surveillance data

Recent weeks of surveillance data are incomplete because of reporting delays (right truncation). get_ncast estimates their final counts with baselinenowcast, replacing the last max_delay weeks of every series and leaving earlier weeks untouched. Downward revisions are redistributed across earlier delays via preprocess_negative_values.

Usage

get_ncast(x, max_delay = 2, draws = 1000, prop_delay = 0.5, scale_factor = 3)

Arguments

x: An incast_data (get_data / check_data) with revision history; use get_data(revisions = TRUE).
max_delay: Integer. Number of recent weeks treated as right-truncated. Default 2.
draws: Integer. Number of posterior samples. Default 1000.
prop_delay: Numeric in (0, 1). Proportion of reference times used for delay estimation. Default 0.5.
scale_factor: Numeric. Multiplier on max_delay for the estimation window. Default 3.

Value

An incast_ncast object with the shared backbone (key, target, window, interval, history) plus:

data: Corrected series. observation holds the nowcast median for corrected weeks; ncast_lower / ncast_upper (95\ propagate nowcast uncertainty.
meta: The nowcast settings (max_delay, draws, prop_delay, scale_factor) and ncast_summary, one tidy table of per-series weekly nowcasts (median, CrI bounds and the reported-so-far observed count) plotted by autoplot.

Details

Weekly data only; other cadences are rejected (the rest of the pipeline is cadence-agnostic).

Author

Cyril Geismar

Examples

if (FALSE) { # \dontrun{
x <- get_data(pathogen = "covid", geo_value = "ca", revisions = TRUE)
ncast <- get_ncast(x)
autoplot(ncast)
} # }