Forecast census from a hubverse admission forecast

Convolves a hubverse admission quantile forecast with a length of stay survival vector to produce a hubverse census quantile forecast.

Usage

fcast_census(admission_fcast, los, admission_history)

Arguments

admission_fcast

Hubverse model output tibble of admission quantile forecasts (output of hubData::collect_hub()). One row per (model, location, reference_date, horizon, quantile).

los

Numeric survival vector P(LOS > d) of length max_stay + 1. Use spec_los() for literature priors or fit_los() for a data driven fit.

admission_history

Hubverse target time series tibble of observed admissions (output of hubData::connect_target_timeseries() then collected). Must cover at least max_stay time steps immediately before each reference_date in admission_fcast.

Value

A tibble with the same hubverse columns as admission_fcast, with value now representing census instead of admissions.

Details

Census on the first forecast day depends on admissions before the forecast starts, because patients already in hospital are still occupying beds. admission_history supplies those pre forecast admissions so the convolution can see the full max_stay window. Groups with fewer than max_stay historical observations before their reference_date are dropped silently.