pkgdown/extra.css

Skip to contents

Detect batch reporting (backlog dumps) on the report-date axis

detect_report_batches.Rd

[Experimental]

Usage

detect_report_batches(
  x,
  ...,
  signals = c("volume", "delay"),
  threshold = 3,
  backlog_delay = NULL,
  backlog_quantile = 0.9,
  baseline_window = NULL,
  gap_window = 3,
  min_reports = 1,
  by_strata = FALSE,
  strata = NULL
)

Arguments

x

A tbl_now object.

...

Unused.

signals

Character vector of the signals to require (AND-ed). Any of "volume", "delay", "span", "gap". Defaults to c("volume", "delay").

threshold

Robust-z cutoff a signal must reach to fire (default 3).

backlog_delay

Delay at/above which a report counts as backlog (for the backlog column). NULL (default) uses the backlog_quantile of the observed delays.

backlog_quantile

Delay quantile used to set backlog_delay when it is NULL (default 0.9).

baseline_window

Rolling-baseline width in report periods. NULL (default) uses roughly two months.

gap_window

Number of preceding periods assessed by the "gap" signal (default 3).

min_reports

Minimum reports for a period to be eligible (default 1).

by_strata

Logical (default FALSE). Detect batches separately per stratum.

strata

Character vector of columns to group on when by_strata = TRUE. NULL (default) uses the object's strata.

Value

A tibble with one row per candidate report date (n_reports > 0), carrying report_date (and strata when grouped), the features (n_reports, n_event_dates, mean_delay, median_delay, max_delay, backlog), the robust scores (score_volume, score_delay, score_span, score_gap) and the batch flag.

Details

Laboratories sometimes withhold results and then release a whole backlog at once. Operationally such a batch is a report date that is anomalous on the report axis. This scans each report-date period and flags batches using up to four robust signals, of which you choose the combination to require:

  • "volume" — unusually many cases reported that period (robust z of the report count vs a rolling median/MAD baseline).

  • "delay" — the reports that period have unusually long delays (a cleared backlog). This is the signal that distinguishes a batch from an epidemic peak: an epidemic peak also inflates the volume, but its cases are still reported with the normal, short delay distribution, so its delay score stays low.

  • "span" — the period covers an unusually large number of distinct event dates.

  • "gap" — the period is preceded by a run of unusually low reporting (the "silence then dump" cadence).

A report date is flagged when all the activated signals fire (each score at or above threshold). Requiring "delay" (or "span") alongside "volume" is what keeps epidemic peaks from being flagged.

See also

plot_report_batches()

Examples

data(denguedat)
dengue <- tbl_now(denguedat,
  event_date = "onset_week", report_date = "report_week", verbose = FALSE
)
batches <- detect_report_batches(dengue)
batches[batches$batch, ]
#> # A tibble: 2 × 12
#>   report_date n_reports n_event_dates mean_delay median_delay max_delay backlog
#>   <date>          <dbl>         <dbl>      <dbl>        <dbl>     <dbl>   <dbl>
#> 1 1993-09-27         65             8       2.38            2         8      22
#> 2 1996-02-12         46             6       2.98            3         6      26
#> # ℹ 5 more variables: score_volume <dbl>, score_delay <dbl>, score_span <dbl>,
#> #   score_gap <dbl>, batch <lgl>